xref: /openbmc/linux/net/mptcp/protocol.c (revision 871808fd)
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 struct mptcp_skb_cb {
40 	u64 map_seq;
41 	u64 end_seq;
42 	u32 offset;
43 	u8  has_rxtstamp:1;
44 };
45 
46 #define MPTCP_SKB_CB(__skb)	((struct mptcp_skb_cb *)&((__skb)->cb[0]))
47 
48 enum {
49 	MPTCP_CMSG_TS = BIT(0),
50 	MPTCP_CMSG_INQ = BIT(1),
51 };
52 
53 static struct percpu_counter mptcp_sockets_allocated ____cacheline_aligned_in_smp;
54 
55 static void __mptcp_destroy_sock(struct sock *sk);
56 static void __mptcp_check_send_data_fin(struct sock *sk);
57 
58 DEFINE_PER_CPU(struct mptcp_delegated_action, mptcp_delegated_actions);
59 static struct net_device mptcp_napi_dev;
60 
61 /* If msk has an initial subflow socket, and the MP_CAPABLE handshake has not
62  * completed yet or has failed, return the subflow socket.
63  * Otherwise return NULL.
64  */
65 struct socket *__mptcp_nmpc_socket(const struct mptcp_sock *msk)
66 {
67 	if (!msk->subflow || READ_ONCE(msk->can_ack))
68 		return NULL;
69 
70 	return msk->subflow;
71 }
72 
73 /* Returns end sequence number of the receiver's advertised window */
74 static u64 mptcp_wnd_end(const struct mptcp_sock *msk)
75 {
76 	return READ_ONCE(msk->wnd_end);
77 }
78 
79 static bool mptcp_is_tcpsk(struct sock *sk)
80 {
81 	struct socket *sock = sk->sk_socket;
82 
83 	if (unlikely(sk->sk_prot == &tcp_prot)) {
84 		/* we are being invoked after mptcp_accept() has
85 		 * accepted a non-mp-capable flow: sk is a tcp_sk,
86 		 * not an mptcp one.
87 		 *
88 		 * Hand the socket over to tcp so all further socket ops
89 		 * bypass mptcp.
90 		 */
91 		sock->ops = &inet_stream_ops;
92 		return true;
93 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
94 	} else if (unlikely(sk->sk_prot == &tcpv6_prot)) {
95 		sock->ops = &inet6_stream_ops;
96 		return true;
97 #endif
98 	}
99 
100 	return false;
101 }
102 
103 static int __mptcp_socket_create(struct mptcp_sock *msk)
104 {
105 	struct mptcp_subflow_context *subflow;
106 	struct sock *sk = (struct sock *)msk;
107 	struct socket *ssock;
108 	int err;
109 
110 	err = mptcp_subflow_create_socket(sk, &ssock);
111 	if (err)
112 		return err;
113 
114 	msk->first = ssock->sk;
115 	msk->subflow = ssock;
116 	subflow = mptcp_subflow_ctx(ssock->sk);
117 	list_add(&subflow->node, &msk->conn_list);
118 	sock_hold(ssock->sk);
119 	subflow->request_mptcp = 1;
120 
121 	/* This is the first subflow, always with id 0 */
122 	subflow->local_id_valid = 1;
123 	mptcp_sock_graft(msk->first, sk->sk_socket);
124 
125 	return 0;
126 }
127 
128 static void mptcp_drop(struct sock *sk, struct sk_buff *skb)
129 {
130 	sk_drops_add(sk, skb);
131 	__kfree_skb(skb);
132 }
133 
134 static void mptcp_rmem_charge(struct sock *sk, int size)
135 {
136 	mptcp_sk(sk)->rmem_fwd_alloc -= size;
137 }
138 
139 static bool mptcp_try_coalesce(struct sock *sk, struct sk_buff *to,
140 			       struct sk_buff *from)
141 {
142 	bool fragstolen;
143 	int delta;
144 
145 	if (MPTCP_SKB_CB(from)->offset ||
146 	    !skb_try_coalesce(to, from, &fragstolen, &delta))
147 		return false;
148 
149 	pr_debug("colesced seq %llx into %llx new len %d new end seq %llx",
150 		 MPTCP_SKB_CB(from)->map_seq, MPTCP_SKB_CB(to)->map_seq,
151 		 to->len, MPTCP_SKB_CB(from)->end_seq);
152 	MPTCP_SKB_CB(to)->end_seq = MPTCP_SKB_CB(from)->end_seq;
153 	kfree_skb_partial(from, fragstolen);
154 	atomic_add(delta, &sk->sk_rmem_alloc);
155 	mptcp_rmem_charge(sk, delta);
156 	return true;
157 }
158 
159 static bool mptcp_ooo_try_coalesce(struct mptcp_sock *msk, struct sk_buff *to,
160 				   struct sk_buff *from)
161 {
162 	if (MPTCP_SKB_CB(from)->map_seq != MPTCP_SKB_CB(to)->end_seq)
163 		return false;
164 
165 	return mptcp_try_coalesce((struct sock *)msk, to, from);
166 }
167 
168 static void __mptcp_rmem_reclaim(struct sock *sk, int amount)
169 {
170 	amount >>= SK_MEM_QUANTUM_SHIFT;
171 	mptcp_sk(sk)->rmem_fwd_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
172 	__sk_mem_reduce_allocated(sk, amount);
173 }
174 
175 static void mptcp_rmem_uncharge(struct sock *sk, int size)
176 {
177 	struct mptcp_sock *msk = mptcp_sk(sk);
178 	int reclaimable;
179 
180 	msk->rmem_fwd_alloc += size;
181 	reclaimable = msk->rmem_fwd_alloc - sk_unused_reserved_mem(sk);
182 
183 	/* see sk_mem_uncharge() for the rationale behind the following schema */
184 	if (unlikely(reclaimable >= SK_RECLAIM_THRESHOLD))
185 		__mptcp_rmem_reclaim(sk, SK_RECLAIM_CHUNK);
186 }
187 
188 static void mptcp_rfree(struct sk_buff *skb)
189 {
190 	unsigned int len = skb->truesize;
191 	struct sock *sk = skb->sk;
192 
193 	atomic_sub(len, &sk->sk_rmem_alloc);
194 	mptcp_rmem_uncharge(sk, len);
195 }
196 
197 static void mptcp_set_owner_r(struct sk_buff *skb, struct sock *sk)
198 {
199 	skb_orphan(skb);
200 	skb->sk = sk;
201 	skb->destructor = mptcp_rfree;
202 	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
203 	mptcp_rmem_charge(sk, skb->truesize);
204 }
205 
206 /* "inspired" by tcp_data_queue_ofo(), main differences:
207  * - use mptcp seqs
208  * - don't cope with sacks
209  */
210 static void mptcp_data_queue_ofo(struct mptcp_sock *msk, struct sk_buff *skb)
211 {
212 	struct sock *sk = (struct sock *)msk;
213 	struct rb_node **p, *parent;
214 	u64 seq, end_seq, max_seq;
215 	struct sk_buff *skb1;
216 
217 	seq = MPTCP_SKB_CB(skb)->map_seq;
218 	end_seq = MPTCP_SKB_CB(skb)->end_seq;
219 	max_seq = atomic64_read(&msk->rcv_wnd_sent);
220 
221 	pr_debug("msk=%p seq=%llx limit=%llx empty=%d", msk, seq, max_seq,
222 		 RB_EMPTY_ROOT(&msk->out_of_order_queue));
223 	if (after64(end_seq, max_seq)) {
224 		/* out of window */
225 		mptcp_drop(sk, skb);
226 		pr_debug("oow by %lld, rcv_wnd_sent %llu\n",
227 			 (unsigned long long)end_seq - (unsigned long)max_seq,
228 			 (unsigned long long)atomic64_read(&msk->rcv_wnd_sent));
229 		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_NODSSWINDOW);
230 		return;
231 	}
232 
233 	p = &msk->out_of_order_queue.rb_node;
234 	MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOQUEUE);
235 	if (RB_EMPTY_ROOT(&msk->out_of_order_queue)) {
236 		rb_link_node(&skb->rbnode, NULL, p);
237 		rb_insert_color(&skb->rbnode, &msk->out_of_order_queue);
238 		msk->ooo_last_skb = skb;
239 		goto end;
240 	}
241 
242 	/* with 2 subflows, adding at end of ooo queue is quite likely
243 	 * Use of ooo_last_skb avoids the O(Log(N)) rbtree lookup.
244 	 */
245 	if (mptcp_ooo_try_coalesce(msk, msk->ooo_last_skb, skb)) {
246 		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOMERGE);
247 		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOQUEUETAIL);
248 		return;
249 	}
250 
251 	/* Can avoid an rbtree lookup if we are adding skb after ooo_last_skb */
252 	if (!before64(seq, MPTCP_SKB_CB(msk->ooo_last_skb)->end_seq)) {
253 		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOQUEUETAIL);
254 		parent = &msk->ooo_last_skb->rbnode;
255 		p = &parent->rb_right;
256 		goto insert;
257 	}
258 
259 	/* Find place to insert this segment. Handle overlaps on the way. */
260 	parent = NULL;
261 	while (*p) {
262 		parent = *p;
263 		skb1 = rb_to_skb(parent);
264 		if (before64(seq, MPTCP_SKB_CB(skb1)->map_seq)) {
265 			p = &parent->rb_left;
266 			continue;
267 		}
268 		if (before64(seq, MPTCP_SKB_CB(skb1)->end_seq)) {
269 			if (!after64(end_seq, MPTCP_SKB_CB(skb1)->end_seq)) {
270 				/* All the bits are present. Drop. */
271 				mptcp_drop(sk, skb);
272 				MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
273 				return;
274 			}
275 			if (after64(seq, MPTCP_SKB_CB(skb1)->map_seq)) {
276 				/* partial overlap:
277 				 *     |     skb      |
278 				 *  |     skb1    |
279 				 * continue traversing
280 				 */
281 			} else {
282 				/* skb's seq == skb1's seq and skb covers skb1.
283 				 * Replace skb1 with skb.
284 				 */
285 				rb_replace_node(&skb1->rbnode, &skb->rbnode,
286 						&msk->out_of_order_queue);
287 				mptcp_drop(sk, skb1);
288 				MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
289 				goto merge_right;
290 			}
291 		} else if (mptcp_ooo_try_coalesce(msk, skb1, skb)) {
292 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOMERGE);
293 			return;
294 		}
295 		p = &parent->rb_right;
296 	}
297 
298 insert:
299 	/* Insert segment into RB tree. */
300 	rb_link_node(&skb->rbnode, parent, p);
301 	rb_insert_color(&skb->rbnode, &msk->out_of_order_queue);
302 
303 merge_right:
304 	/* Remove other segments covered by skb. */
305 	while ((skb1 = skb_rb_next(skb)) != NULL) {
306 		if (before64(end_seq, MPTCP_SKB_CB(skb1)->end_seq))
307 			break;
308 		rb_erase(&skb1->rbnode, &msk->out_of_order_queue);
309 		mptcp_drop(sk, skb1);
310 		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
311 	}
312 	/* If there is no skb after us, we are the last_skb ! */
313 	if (!skb1)
314 		msk->ooo_last_skb = skb;
315 
316 end:
317 	skb_condense(skb);
318 	mptcp_set_owner_r(skb, sk);
319 }
320 
321 static bool mptcp_rmem_schedule(struct sock *sk, struct sock *ssk, int size)
322 {
323 	struct mptcp_sock *msk = mptcp_sk(sk);
324 	int amt, amount;
325 
326 	if (size < msk->rmem_fwd_alloc)
327 		return true;
328 
329 	amt = sk_mem_pages(size);
330 	amount = amt << SK_MEM_QUANTUM_SHIFT;
331 	msk->rmem_fwd_alloc += amount;
332 	if (!__sk_mem_raise_allocated(sk, size, amt, SK_MEM_RECV)) {
333 		if (ssk->sk_forward_alloc < amount) {
334 			msk->rmem_fwd_alloc -= amount;
335 			return false;
336 		}
337 
338 		ssk->sk_forward_alloc -= amount;
339 	}
340 	return true;
341 }
342 
343 static bool __mptcp_move_skb(struct mptcp_sock *msk, struct sock *ssk,
344 			     struct sk_buff *skb, unsigned int offset,
345 			     size_t copy_len)
346 {
347 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
348 	struct sock *sk = (struct sock *)msk;
349 	struct sk_buff *tail;
350 	bool has_rxtstamp;
351 
352 	__skb_unlink(skb, &ssk->sk_receive_queue);
353 
354 	skb_ext_reset(skb);
355 	skb_orphan(skb);
356 
357 	/* try to fetch required memory from subflow */
358 	if (!mptcp_rmem_schedule(sk, ssk, skb->truesize))
359 		goto drop;
360 
361 	has_rxtstamp = TCP_SKB_CB(skb)->has_rxtstamp;
362 
363 	/* the skb map_seq accounts for the skb offset:
364 	 * mptcp_subflow_get_mapped_dsn() is based on the current tp->copied_seq
365 	 * value
366 	 */
367 	MPTCP_SKB_CB(skb)->map_seq = mptcp_subflow_get_mapped_dsn(subflow);
368 	MPTCP_SKB_CB(skb)->end_seq = MPTCP_SKB_CB(skb)->map_seq + copy_len;
369 	MPTCP_SKB_CB(skb)->offset = offset;
370 	MPTCP_SKB_CB(skb)->has_rxtstamp = has_rxtstamp;
371 
372 	if (MPTCP_SKB_CB(skb)->map_seq == msk->ack_seq) {
373 		/* in sequence */
374 		WRITE_ONCE(msk->ack_seq, msk->ack_seq + copy_len);
375 		tail = skb_peek_tail(&sk->sk_receive_queue);
376 		if (tail && mptcp_try_coalesce(sk, tail, skb))
377 			return true;
378 
379 		mptcp_set_owner_r(skb, sk);
380 		__skb_queue_tail(&sk->sk_receive_queue, skb);
381 		return true;
382 	} else if (after64(MPTCP_SKB_CB(skb)->map_seq, msk->ack_seq)) {
383 		mptcp_data_queue_ofo(msk, skb);
384 		return false;
385 	}
386 
387 	/* old data, keep it simple and drop the whole pkt, sender
388 	 * will retransmit as needed, if needed.
389 	 */
390 	MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
391 drop:
392 	mptcp_drop(sk, skb);
393 	return false;
394 }
395 
396 static void mptcp_stop_timer(struct sock *sk)
397 {
398 	struct inet_connection_sock *icsk = inet_csk(sk);
399 
400 	sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
401 	mptcp_sk(sk)->timer_ival = 0;
402 }
403 
404 static void mptcp_close_wake_up(struct sock *sk)
405 {
406 	if (sock_flag(sk, SOCK_DEAD))
407 		return;
408 
409 	sk->sk_state_change(sk);
410 	if (sk->sk_shutdown == SHUTDOWN_MASK ||
411 	    sk->sk_state == TCP_CLOSE)
412 		sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_HUP);
413 	else
414 		sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
415 }
416 
417 static bool mptcp_pending_data_fin_ack(struct sock *sk)
418 {
419 	struct mptcp_sock *msk = mptcp_sk(sk);
420 
421 	return !__mptcp_check_fallback(msk) &&
422 	       ((1 << sk->sk_state) &
423 		(TCPF_FIN_WAIT1 | TCPF_CLOSING | TCPF_LAST_ACK)) &&
424 	       msk->write_seq == READ_ONCE(msk->snd_una);
425 }
426 
427 static void mptcp_check_data_fin_ack(struct sock *sk)
428 {
429 	struct mptcp_sock *msk = mptcp_sk(sk);
430 
431 	/* Look for an acknowledged DATA_FIN */
432 	if (mptcp_pending_data_fin_ack(sk)) {
433 		WRITE_ONCE(msk->snd_data_fin_enable, 0);
434 
435 		switch (sk->sk_state) {
436 		case TCP_FIN_WAIT1:
437 			inet_sk_state_store(sk, TCP_FIN_WAIT2);
438 			break;
439 		case TCP_CLOSING:
440 		case TCP_LAST_ACK:
441 			inet_sk_state_store(sk, TCP_CLOSE);
442 			break;
443 		}
444 
445 		mptcp_close_wake_up(sk);
446 	}
447 }
448 
449 static bool mptcp_pending_data_fin(struct sock *sk, u64 *seq)
450 {
451 	struct mptcp_sock *msk = mptcp_sk(sk);
452 
453 	if (READ_ONCE(msk->rcv_data_fin) &&
454 	    ((1 << sk->sk_state) &
455 	     (TCPF_ESTABLISHED | TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2))) {
456 		u64 rcv_data_fin_seq = READ_ONCE(msk->rcv_data_fin_seq);
457 
458 		if (msk->ack_seq == rcv_data_fin_seq) {
459 			if (seq)
460 				*seq = rcv_data_fin_seq;
461 
462 			return true;
463 		}
464 	}
465 
466 	return false;
467 }
468 
469 static void mptcp_set_datafin_timeout(const struct sock *sk)
470 {
471 	struct inet_connection_sock *icsk = inet_csk(sk);
472 	u32 retransmits;
473 
474 	retransmits = min_t(u32, icsk->icsk_retransmits,
475 			    ilog2(TCP_RTO_MAX / TCP_RTO_MIN));
476 
477 	mptcp_sk(sk)->timer_ival = TCP_RTO_MIN << retransmits;
478 }
479 
480 static void __mptcp_set_timeout(struct sock *sk, long tout)
481 {
482 	mptcp_sk(sk)->timer_ival = tout > 0 ? tout : TCP_RTO_MIN;
483 }
484 
485 static long mptcp_timeout_from_subflow(const struct mptcp_subflow_context *subflow)
486 {
487 	const struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
488 
489 	return inet_csk(ssk)->icsk_pending && !subflow->stale_count ?
490 	       inet_csk(ssk)->icsk_timeout - jiffies : 0;
491 }
492 
493 static void mptcp_set_timeout(struct sock *sk)
494 {
495 	struct mptcp_subflow_context *subflow;
496 	long tout = 0;
497 
498 	mptcp_for_each_subflow(mptcp_sk(sk), subflow)
499 		tout = max(tout, mptcp_timeout_from_subflow(subflow));
500 	__mptcp_set_timeout(sk, tout);
501 }
502 
503 static inline bool tcp_can_send_ack(const struct sock *ssk)
504 {
505 	return !((1 << inet_sk_state_load(ssk)) &
506 	       (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_TIME_WAIT | TCPF_CLOSE | TCPF_LISTEN));
507 }
508 
509 void __mptcp_subflow_send_ack(struct sock *ssk)
510 {
511 	if (tcp_can_send_ack(ssk))
512 		tcp_send_ack(ssk);
513 }
514 
515 void mptcp_subflow_send_ack(struct sock *ssk)
516 {
517 	bool slow;
518 
519 	slow = lock_sock_fast(ssk);
520 	__mptcp_subflow_send_ack(ssk);
521 	unlock_sock_fast(ssk, slow);
522 }
523 
524 static void mptcp_send_ack(struct mptcp_sock *msk)
525 {
526 	struct mptcp_subflow_context *subflow;
527 
528 	mptcp_for_each_subflow(msk, subflow)
529 		mptcp_subflow_send_ack(mptcp_subflow_tcp_sock(subflow));
530 }
531 
532 static void mptcp_subflow_cleanup_rbuf(struct sock *ssk)
533 {
534 	bool slow;
535 
536 	slow = lock_sock_fast(ssk);
537 	if (tcp_can_send_ack(ssk))
538 		tcp_cleanup_rbuf(ssk, 1);
539 	unlock_sock_fast(ssk, slow);
540 }
541 
542 static bool mptcp_subflow_could_cleanup(const struct sock *ssk, bool rx_empty)
543 {
544 	const struct inet_connection_sock *icsk = inet_csk(ssk);
545 	u8 ack_pending = READ_ONCE(icsk->icsk_ack.pending);
546 	const struct tcp_sock *tp = tcp_sk(ssk);
547 
548 	return (ack_pending & ICSK_ACK_SCHED) &&
549 		((READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->rcv_wup) >
550 		  READ_ONCE(icsk->icsk_ack.rcv_mss)) ||
551 		 (rx_empty && ack_pending &
552 			      (ICSK_ACK_PUSHED2 | ICSK_ACK_PUSHED)));
553 }
554 
555 static void mptcp_cleanup_rbuf(struct mptcp_sock *msk)
556 {
557 	int old_space = READ_ONCE(msk->old_wspace);
558 	struct mptcp_subflow_context *subflow;
559 	struct sock *sk = (struct sock *)msk;
560 	int space =  __mptcp_space(sk);
561 	bool cleanup, rx_empty;
562 
563 	cleanup = (space > 0) && (space >= (old_space << 1));
564 	rx_empty = !__mptcp_rmem(sk);
565 
566 	mptcp_for_each_subflow(msk, subflow) {
567 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
568 
569 		if (cleanup || mptcp_subflow_could_cleanup(ssk, rx_empty))
570 			mptcp_subflow_cleanup_rbuf(ssk);
571 	}
572 }
573 
574 static bool mptcp_check_data_fin(struct sock *sk)
575 {
576 	struct mptcp_sock *msk = mptcp_sk(sk);
577 	u64 rcv_data_fin_seq;
578 	bool ret = false;
579 
580 	if (__mptcp_check_fallback(msk))
581 		return ret;
582 
583 	/* Need to ack a DATA_FIN received from a peer while this side
584 	 * of the connection is in ESTABLISHED, FIN_WAIT1, or FIN_WAIT2.
585 	 * msk->rcv_data_fin was set when parsing the incoming options
586 	 * at the subflow level and the msk lock was not held, so this
587 	 * is the first opportunity to act on the DATA_FIN and change
588 	 * the msk state.
589 	 *
590 	 * If we are caught up to the sequence number of the incoming
591 	 * DATA_FIN, send the DATA_ACK now and do state transition.  If
592 	 * not caught up, do nothing and let the recv code send DATA_ACK
593 	 * when catching up.
594 	 */
595 
596 	if (mptcp_pending_data_fin(sk, &rcv_data_fin_seq)) {
597 		WRITE_ONCE(msk->ack_seq, msk->ack_seq + 1);
598 		WRITE_ONCE(msk->rcv_data_fin, 0);
599 
600 		sk->sk_shutdown |= RCV_SHUTDOWN;
601 		smp_mb__before_atomic(); /* SHUTDOWN must be visible first */
602 
603 		switch (sk->sk_state) {
604 		case TCP_ESTABLISHED:
605 			inet_sk_state_store(sk, TCP_CLOSE_WAIT);
606 			break;
607 		case TCP_FIN_WAIT1:
608 			inet_sk_state_store(sk, TCP_CLOSING);
609 			break;
610 		case TCP_FIN_WAIT2:
611 			inet_sk_state_store(sk, TCP_CLOSE);
612 			break;
613 		default:
614 			/* Other states not expected */
615 			WARN_ON_ONCE(1);
616 			break;
617 		}
618 
619 		ret = true;
620 		mptcp_send_ack(msk);
621 		mptcp_close_wake_up(sk);
622 	}
623 	return ret;
624 }
625 
626 static bool __mptcp_move_skbs_from_subflow(struct mptcp_sock *msk,
627 					   struct sock *ssk,
628 					   unsigned int *bytes)
629 {
630 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
631 	struct sock *sk = (struct sock *)msk;
632 	unsigned int moved = 0;
633 	bool more_data_avail;
634 	struct tcp_sock *tp;
635 	bool done = false;
636 	int sk_rbuf;
637 
638 	sk_rbuf = READ_ONCE(sk->sk_rcvbuf);
639 
640 	if (!(sk->sk_userlocks & SOCK_RCVBUF_LOCK)) {
641 		int ssk_rbuf = READ_ONCE(ssk->sk_rcvbuf);
642 
643 		if (unlikely(ssk_rbuf > sk_rbuf)) {
644 			WRITE_ONCE(sk->sk_rcvbuf, ssk_rbuf);
645 			sk_rbuf = ssk_rbuf;
646 		}
647 	}
648 
649 	pr_debug("msk=%p ssk=%p", msk, ssk);
650 	tp = tcp_sk(ssk);
651 	do {
652 		u32 map_remaining, offset;
653 		u32 seq = tp->copied_seq;
654 		struct sk_buff *skb;
655 		bool fin;
656 
657 		/* try to move as much data as available */
658 		map_remaining = subflow->map_data_len -
659 				mptcp_subflow_get_map_offset(subflow);
660 
661 		skb = skb_peek(&ssk->sk_receive_queue);
662 		if (!skb) {
663 			/* if no data is found, a racing workqueue/recvmsg
664 			 * already processed the new data, stop here or we
665 			 * can enter an infinite loop
666 			 */
667 			if (!moved)
668 				done = true;
669 			break;
670 		}
671 
672 		if (__mptcp_check_fallback(msk)) {
673 			/* if we are running under the workqueue, TCP could have
674 			 * collapsed skbs between dummy map creation and now
675 			 * be sure to adjust the size
676 			 */
677 			map_remaining = skb->len;
678 			subflow->map_data_len = skb->len;
679 		}
680 
681 		offset = seq - TCP_SKB_CB(skb)->seq;
682 		fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
683 		if (fin) {
684 			done = true;
685 			seq++;
686 		}
687 
688 		if (offset < skb->len) {
689 			size_t len = skb->len - offset;
690 
691 			if (tp->urg_data)
692 				done = true;
693 
694 			if (__mptcp_move_skb(msk, ssk, skb, offset, len))
695 				moved += len;
696 			seq += len;
697 
698 			if (WARN_ON_ONCE(map_remaining < len))
699 				break;
700 		} else {
701 			WARN_ON_ONCE(!fin);
702 			sk_eat_skb(ssk, skb);
703 			done = true;
704 		}
705 
706 		WRITE_ONCE(tp->copied_seq, seq);
707 		more_data_avail = mptcp_subflow_data_available(ssk);
708 
709 		if (atomic_read(&sk->sk_rmem_alloc) > sk_rbuf) {
710 			done = true;
711 			break;
712 		}
713 	} while (more_data_avail);
714 
715 	*bytes += moved;
716 	return done;
717 }
718 
719 static bool __mptcp_ofo_queue(struct mptcp_sock *msk)
720 {
721 	struct sock *sk = (struct sock *)msk;
722 	struct sk_buff *skb, *tail;
723 	bool moved = false;
724 	struct rb_node *p;
725 	u64 end_seq;
726 
727 	p = rb_first(&msk->out_of_order_queue);
728 	pr_debug("msk=%p empty=%d", msk, RB_EMPTY_ROOT(&msk->out_of_order_queue));
729 	while (p) {
730 		skb = rb_to_skb(p);
731 		if (after64(MPTCP_SKB_CB(skb)->map_seq, msk->ack_seq))
732 			break;
733 
734 		p = rb_next(p);
735 		rb_erase(&skb->rbnode, &msk->out_of_order_queue);
736 
737 		if (unlikely(!after64(MPTCP_SKB_CB(skb)->end_seq,
738 				      msk->ack_seq))) {
739 			mptcp_drop(sk, skb);
740 			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
741 			continue;
742 		}
743 
744 		end_seq = MPTCP_SKB_CB(skb)->end_seq;
745 		tail = skb_peek_tail(&sk->sk_receive_queue);
746 		if (!tail || !mptcp_ooo_try_coalesce(msk, tail, skb)) {
747 			int delta = msk->ack_seq - MPTCP_SKB_CB(skb)->map_seq;
748 
749 			/* skip overlapping data, if any */
750 			pr_debug("uncoalesced seq=%llx ack seq=%llx delta=%d",
751 				 MPTCP_SKB_CB(skb)->map_seq, msk->ack_seq,
752 				 delta);
753 			MPTCP_SKB_CB(skb)->offset += delta;
754 			MPTCP_SKB_CB(skb)->map_seq += delta;
755 			__skb_queue_tail(&sk->sk_receive_queue, skb);
756 		}
757 		msk->ack_seq = end_seq;
758 		moved = true;
759 	}
760 	return moved;
761 }
762 
763 /* In most cases we will be able to lock the mptcp socket.  If its already
764  * owned, we need to defer to the work queue to avoid ABBA deadlock.
765  */
766 static bool move_skbs_to_msk(struct mptcp_sock *msk, struct sock *ssk)
767 {
768 	struct sock *sk = (struct sock *)msk;
769 	unsigned int moved = 0;
770 
771 	__mptcp_move_skbs_from_subflow(msk, ssk, &moved);
772 	__mptcp_ofo_queue(msk);
773 	if (unlikely(ssk->sk_err)) {
774 		if (!sock_owned_by_user(sk))
775 			__mptcp_error_report(sk);
776 		else
777 			__set_bit(MPTCP_ERROR_REPORT,  &msk->cb_flags);
778 	}
779 
780 	/* If the moves have caught up with the DATA_FIN sequence number
781 	 * it's time to ack the DATA_FIN and change socket state, but
782 	 * this is not a good place to change state. Let the workqueue
783 	 * do it.
784 	 */
785 	if (mptcp_pending_data_fin(sk, NULL))
786 		mptcp_schedule_work(sk);
787 	return moved > 0;
788 }
789 
790 void mptcp_data_ready(struct sock *sk, struct sock *ssk)
791 {
792 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
793 	struct mptcp_sock *msk = mptcp_sk(sk);
794 	int sk_rbuf, ssk_rbuf;
795 
796 	/* The peer can send data while we are shutting down this
797 	 * subflow at msk destruction time, but we must avoid enqueuing
798 	 * more data to the msk receive queue
799 	 */
800 	if (unlikely(subflow->disposable))
801 		return;
802 
803 	ssk_rbuf = READ_ONCE(ssk->sk_rcvbuf);
804 	sk_rbuf = READ_ONCE(sk->sk_rcvbuf);
805 	if (unlikely(ssk_rbuf > sk_rbuf))
806 		sk_rbuf = ssk_rbuf;
807 
808 	/* over limit? can't append more skbs to msk, Also, no need to wake-up*/
809 	if (__mptcp_rmem(sk) > sk_rbuf) {
810 		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_RCVPRUNED);
811 		return;
812 	}
813 
814 	/* Wake-up the reader only for in-sequence data */
815 	mptcp_data_lock(sk);
816 	if (move_skbs_to_msk(msk, ssk))
817 		sk->sk_data_ready(sk);
818 
819 	mptcp_data_unlock(sk);
820 }
821 
822 static bool __mptcp_finish_join(struct mptcp_sock *msk, struct sock *ssk)
823 {
824 	struct sock *sk = (struct sock *)msk;
825 
826 	if (sk->sk_state != TCP_ESTABLISHED)
827 		return false;
828 
829 	/* attach to msk socket only after we are sure we will deal with it
830 	 * at close time
831 	 */
832 	if (sk->sk_socket && !ssk->sk_socket)
833 		mptcp_sock_graft(ssk, sk->sk_socket);
834 
835 	mptcp_propagate_sndbuf((struct sock *)msk, ssk);
836 	mptcp_sockopt_sync_locked(msk, ssk);
837 	return true;
838 }
839 
840 static void __mptcp_flush_join_list(struct sock *sk)
841 {
842 	struct mptcp_subflow_context *tmp, *subflow;
843 	struct mptcp_sock *msk = mptcp_sk(sk);
844 
845 	list_for_each_entry_safe(subflow, tmp, &msk->join_list, node) {
846 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
847 		bool slow = lock_sock_fast(ssk);
848 
849 		list_move_tail(&subflow->node, &msk->conn_list);
850 		if (!__mptcp_finish_join(msk, ssk))
851 			mptcp_subflow_reset(ssk);
852 		unlock_sock_fast(ssk, slow);
853 	}
854 }
855 
856 static bool mptcp_timer_pending(struct sock *sk)
857 {
858 	return timer_pending(&inet_csk(sk)->icsk_retransmit_timer);
859 }
860 
861 static void mptcp_reset_timer(struct sock *sk)
862 {
863 	struct inet_connection_sock *icsk = inet_csk(sk);
864 	unsigned long tout;
865 
866 	/* prevent rescheduling on close */
867 	if (unlikely(inet_sk_state_load(sk) == TCP_CLOSE))
868 		return;
869 
870 	tout = mptcp_sk(sk)->timer_ival;
871 	sk_reset_timer(sk, &icsk->icsk_retransmit_timer, jiffies + tout);
872 }
873 
874 bool mptcp_schedule_work(struct sock *sk)
875 {
876 	if (inet_sk_state_load(sk) != TCP_CLOSE &&
877 	    schedule_work(&mptcp_sk(sk)->work)) {
878 		/* each subflow already holds a reference to the sk, and the
879 		 * workqueue is invoked by a subflow, so sk can't go away here.
880 		 */
881 		sock_hold(sk);
882 		return true;
883 	}
884 	return false;
885 }
886 
887 void mptcp_subflow_eof(struct sock *sk)
888 {
889 	if (!test_and_set_bit(MPTCP_WORK_EOF, &mptcp_sk(sk)->flags))
890 		mptcp_schedule_work(sk);
891 }
892 
893 static void mptcp_check_for_eof(struct mptcp_sock *msk)
894 {
895 	struct mptcp_subflow_context *subflow;
896 	struct sock *sk = (struct sock *)msk;
897 	int receivers = 0;
898 
899 	mptcp_for_each_subflow(msk, subflow)
900 		receivers += !subflow->rx_eof;
901 	if (receivers)
902 		return;
903 
904 	if (!(sk->sk_shutdown & RCV_SHUTDOWN)) {
905 		/* hopefully temporary hack: propagate shutdown status
906 		 * to msk, when all subflows agree on it
907 		 */
908 		sk->sk_shutdown |= RCV_SHUTDOWN;
909 
910 		smp_mb__before_atomic(); /* SHUTDOWN must be visible first */
911 		sk->sk_data_ready(sk);
912 	}
913 
914 	switch (sk->sk_state) {
915 	case TCP_ESTABLISHED:
916 		inet_sk_state_store(sk, TCP_CLOSE_WAIT);
917 		break;
918 	case TCP_FIN_WAIT1:
919 		inet_sk_state_store(sk, TCP_CLOSING);
920 		break;
921 	case TCP_FIN_WAIT2:
922 		inet_sk_state_store(sk, TCP_CLOSE);
923 		break;
924 	default:
925 		return;
926 	}
927 	mptcp_close_wake_up(sk);
928 }
929 
930 static struct sock *mptcp_subflow_recv_lookup(const struct mptcp_sock *msk)
931 {
932 	struct mptcp_subflow_context *subflow;
933 	struct sock *sk = (struct sock *)msk;
934 
935 	sock_owned_by_me(sk);
936 
937 	mptcp_for_each_subflow(msk, subflow) {
938 		if (READ_ONCE(subflow->data_avail))
939 			return mptcp_subflow_tcp_sock(subflow);
940 	}
941 
942 	return NULL;
943 }
944 
945 static bool mptcp_skb_can_collapse_to(u64 write_seq,
946 				      const struct sk_buff *skb,
947 				      const struct mptcp_ext *mpext)
948 {
949 	if (!tcp_skb_can_collapse_to(skb))
950 		return false;
951 
952 	/* can collapse only if MPTCP level sequence is in order and this
953 	 * mapping has not been xmitted yet
954 	 */
955 	return mpext && mpext->data_seq + mpext->data_len == write_seq &&
956 	       !mpext->frozen;
957 }
958 
959 /* we can append data to the given data frag if:
960  * - there is space available in the backing page_frag
961  * - the data frag tail matches the current page_frag free offset
962  * - the data frag end sequence number matches the current write seq
963  */
964 static bool mptcp_frag_can_collapse_to(const struct mptcp_sock *msk,
965 				       const struct page_frag *pfrag,
966 				       const struct mptcp_data_frag *df)
967 {
968 	return df && pfrag->page == df->page &&
969 		pfrag->size - pfrag->offset > 0 &&
970 		pfrag->offset == (df->offset + df->data_len) &&
971 		df->data_seq + df->data_len == msk->write_seq;
972 }
973 
974 static void __mptcp_mem_reclaim_partial(struct sock *sk)
975 {
976 	int reclaimable = mptcp_sk(sk)->rmem_fwd_alloc - sk_unused_reserved_mem(sk);
977 
978 	lockdep_assert_held_once(&sk->sk_lock.slock);
979 
980 	if (reclaimable > SK_MEM_QUANTUM)
981 		__mptcp_rmem_reclaim(sk, reclaimable - 1);
982 
983 	sk_mem_reclaim_partial(sk);
984 }
985 
986 static void mptcp_mem_reclaim_partial(struct sock *sk)
987 {
988 	mptcp_data_lock(sk);
989 	__mptcp_mem_reclaim_partial(sk);
990 	mptcp_data_unlock(sk);
991 }
992 
993 static void dfrag_uncharge(struct sock *sk, int len)
994 {
995 	sk_mem_uncharge(sk, len);
996 	sk_wmem_queued_add(sk, -len);
997 }
998 
999 static void dfrag_clear(struct sock *sk, struct mptcp_data_frag *dfrag)
1000 {
1001 	int len = dfrag->data_len + dfrag->overhead;
1002 
1003 	list_del(&dfrag->list);
1004 	dfrag_uncharge(sk, len);
1005 	put_page(dfrag->page);
1006 }
1007 
1008 static void __mptcp_clean_una(struct sock *sk)
1009 {
1010 	struct mptcp_sock *msk = mptcp_sk(sk);
1011 	struct mptcp_data_frag *dtmp, *dfrag;
1012 	bool cleaned = false;
1013 	u64 snd_una;
1014 
1015 	/* on fallback we just need to ignore snd_una, as this is really
1016 	 * plain TCP
1017 	 */
1018 	if (__mptcp_check_fallback(msk))
1019 		msk->snd_una = READ_ONCE(msk->snd_nxt);
1020 
1021 	snd_una = msk->snd_una;
1022 	list_for_each_entry_safe(dfrag, dtmp, &msk->rtx_queue, list) {
1023 		if (after64(dfrag->data_seq + dfrag->data_len, snd_una))
1024 			break;
1025 
1026 		if (unlikely(dfrag == msk->first_pending)) {
1027 			/* in recovery mode can see ack after the current snd head */
1028 			if (WARN_ON_ONCE(!msk->recovery))
1029 				break;
1030 
1031 			WRITE_ONCE(msk->first_pending, mptcp_send_next(sk));
1032 		}
1033 
1034 		dfrag_clear(sk, dfrag);
1035 		cleaned = true;
1036 	}
1037 
1038 	dfrag = mptcp_rtx_head(sk);
1039 	if (dfrag && after64(snd_una, dfrag->data_seq)) {
1040 		u64 delta = snd_una - dfrag->data_seq;
1041 
1042 		/* prevent wrap around in recovery mode */
1043 		if (unlikely(delta > dfrag->already_sent)) {
1044 			if (WARN_ON_ONCE(!msk->recovery))
1045 				goto out;
1046 			if (WARN_ON_ONCE(delta > dfrag->data_len))
1047 				goto out;
1048 			dfrag->already_sent += delta - dfrag->already_sent;
1049 		}
1050 
1051 		dfrag->data_seq += delta;
1052 		dfrag->offset += delta;
1053 		dfrag->data_len -= delta;
1054 		dfrag->already_sent -= delta;
1055 
1056 		dfrag_uncharge(sk, delta);
1057 		cleaned = true;
1058 	}
1059 
1060 	/* all retransmitted data acked, recovery completed */
1061 	if (unlikely(msk->recovery) && after64(msk->snd_una, msk->recovery_snd_nxt))
1062 		msk->recovery = false;
1063 
1064 out:
1065 	if (cleaned && tcp_under_memory_pressure(sk))
1066 		__mptcp_mem_reclaim_partial(sk);
1067 
1068 	if (snd_una == READ_ONCE(msk->snd_nxt) &&
1069 	    snd_una == READ_ONCE(msk->write_seq)) {
1070 		if (mptcp_timer_pending(sk) && !mptcp_data_fin_enabled(msk))
1071 			mptcp_stop_timer(sk);
1072 	} else {
1073 		mptcp_reset_timer(sk);
1074 	}
1075 }
1076 
1077 static void __mptcp_clean_una_wakeup(struct sock *sk)
1078 {
1079 	lockdep_assert_held_once(&sk->sk_lock.slock);
1080 
1081 	__mptcp_clean_una(sk);
1082 	mptcp_write_space(sk);
1083 }
1084 
1085 static void mptcp_clean_una_wakeup(struct sock *sk)
1086 {
1087 	mptcp_data_lock(sk);
1088 	__mptcp_clean_una_wakeup(sk);
1089 	mptcp_data_unlock(sk);
1090 }
1091 
1092 static void mptcp_enter_memory_pressure(struct sock *sk)
1093 {
1094 	struct mptcp_subflow_context *subflow;
1095 	struct mptcp_sock *msk = mptcp_sk(sk);
1096 	bool first = true;
1097 
1098 	sk_stream_moderate_sndbuf(sk);
1099 	mptcp_for_each_subflow(msk, subflow) {
1100 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1101 
1102 		if (first)
1103 			tcp_enter_memory_pressure(ssk);
1104 		sk_stream_moderate_sndbuf(ssk);
1105 		first = false;
1106 	}
1107 }
1108 
1109 /* ensure we get enough memory for the frag hdr, beyond some minimal amount of
1110  * data
1111  */
1112 static bool mptcp_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1113 {
1114 	if (likely(skb_page_frag_refill(32U + sizeof(struct mptcp_data_frag),
1115 					pfrag, sk->sk_allocation)))
1116 		return true;
1117 
1118 	mptcp_enter_memory_pressure(sk);
1119 	return false;
1120 }
1121 
1122 static struct mptcp_data_frag *
1123 mptcp_carve_data_frag(const struct mptcp_sock *msk, struct page_frag *pfrag,
1124 		      int orig_offset)
1125 {
1126 	int offset = ALIGN(orig_offset, sizeof(long));
1127 	struct mptcp_data_frag *dfrag;
1128 
1129 	dfrag = (struct mptcp_data_frag *)(page_to_virt(pfrag->page) + offset);
1130 	dfrag->data_len = 0;
1131 	dfrag->data_seq = msk->write_seq;
1132 	dfrag->overhead = offset - orig_offset + sizeof(struct mptcp_data_frag);
1133 	dfrag->offset = offset + sizeof(struct mptcp_data_frag);
1134 	dfrag->already_sent = 0;
1135 	dfrag->page = pfrag->page;
1136 
1137 	return dfrag;
1138 }
1139 
1140 struct mptcp_sendmsg_info {
1141 	int mss_now;
1142 	int size_goal;
1143 	u16 limit;
1144 	u16 sent;
1145 	unsigned int flags;
1146 	bool data_lock_held;
1147 };
1148 
1149 static int mptcp_check_allowed_size(const struct mptcp_sock *msk, struct sock *ssk,
1150 				    u64 data_seq, int avail_size)
1151 {
1152 	u64 window_end = mptcp_wnd_end(msk);
1153 	u64 mptcp_snd_wnd;
1154 
1155 	if (__mptcp_check_fallback(msk))
1156 		return avail_size;
1157 
1158 	mptcp_snd_wnd = window_end - data_seq;
1159 	avail_size = min_t(unsigned int, mptcp_snd_wnd, avail_size);
1160 
1161 	if (unlikely(tcp_sk(ssk)->snd_wnd < mptcp_snd_wnd)) {
1162 		tcp_sk(ssk)->snd_wnd = min_t(u64, U32_MAX, mptcp_snd_wnd);
1163 		MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_SNDWNDSHARED);
1164 	}
1165 
1166 	return avail_size;
1167 }
1168 
1169 static bool __mptcp_add_ext(struct sk_buff *skb, gfp_t gfp)
1170 {
1171 	struct skb_ext *mpext = __skb_ext_alloc(gfp);
1172 
1173 	if (!mpext)
1174 		return false;
1175 	__skb_ext_set(skb, SKB_EXT_MPTCP, mpext);
1176 	return true;
1177 }
1178 
1179 static struct sk_buff *__mptcp_do_alloc_tx_skb(struct sock *sk, gfp_t gfp)
1180 {
1181 	struct sk_buff *skb;
1182 
1183 	skb = alloc_skb_fclone(MAX_TCP_HEADER, gfp);
1184 	if (likely(skb)) {
1185 		if (likely(__mptcp_add_ext(skb, gfp))) {
1186 			skb_reserve(skb, MAX_TCP_HEADER);
1187 			skb->ip_summed = CHECKSUM_PARTIAL;
1188 			INIT_LIST_HEAD(&skb->tcp_tsorted_anchor);
1189 			return skb;
1190 		}
1191 		__kfree_skb(skb);
1192 	} else {
1193 		mptcp_enter_memory_pressure(sk);
1194 	}
1195 	return NULL;
1196 }
1197 
1198 static struct sk_buff *__mptcp_alloc_tx_skb(struct sock *sk, struct sock *ssk, gfp_t gfp)
1199 {
1200 	struct sk_buff *skb;
1201 
1202 	skb = __mptcp_do_alloc_tx_skb(sk, gfp);
1203 	if (!skb)
1204 		return NULL;
1205 
1206 	if (likely(sk_wmem_schedule(ssk, skb->truesize))) {
1207 		tcp_skb_entail(ssk, skb);
1208 		return skb;
1209 	}
1210 	tcp_skb_tsorted_anchor_cleanup(skb);
1211 	kfree_skb(skb);
1212 	return NULL;
1213 }
1214 
1215 static struct sk_buff *mptcp_alloc_tx_skb(struct sock *sk, struct sock *ssk, bool data_lock_held)
1216 {
1217 	gfp_t gfp = data_lock_held ? GFP_ATOMIC : sk->sk_allocation;
1218 
1219 	if (unlikely(tcp_under_memory_pressure(sk))) {
1220 		if (data_lock_held)
1221 			__mptcp_mem_reclaim_partial(sk);
1222 		else
1223 			mptcp_mem_reclaim_partial(sk);
1224 	}
1225 	return __mptcp_alloc_tx_skb(sk, ssk, gfp);
1226 }
1227 
1228 /* note: this always recompute the csum on the whole skb, even
1229  * if we just appended a single frag. More status info needed
1230  */
1231 static void mptcp_update_data_checksum(struct sk_buff *skb, int added)
1232 {
1233 	struct mptcp_ext *mpext = mptcp_get_ext(skb);
1234 	__wsum csum = ~csum_unfold(mpext->csum);
1235 	int offset = skb->len - added;
1236 
1237 	mpext->csum = csum_fold(csum_block_add(csum, skb_checksum(skb, offset, added, 0), offset));
1238 }
1239 
1240 static void mptcp_update_infinite_map(struct mptcp_sock *msk,
1241 				      struct sock *ssk,
1242 				      struct mptcp_ext *mpext)
1243 {
1244 	if (!mpext)
1245 		return;
1246 
1247 	mpext->infinite_map = 1;
1248 	mpext->data_len = 0;
1249 
1250 	MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPTX);
1251 	mptcp_subflow_ctx(ssk)->send_infinite_map = 0;
1252 	pr_fallback(msk);
1253 	mptcp_do_fallback(ssk);
1254 }
1255 
1256 static int mptcp_sendmsg_frag(struct sock *sk, struct sock *ssk,
1257 			      struct mptcp_data_frag *dfrag,
1258 			      struct mptcp_sendmsg_info *info)
1259 {
1260 	u64 data_seq = dfrag->data_seq + info->sent;
1261 	int offset = dfrag->offset + info->sent;
1262 	struct mptcp_sock *msk = mptcp_sk(sk);
1263 	bool zero_window_probe = false;
1264 	struct mptcp_ext *mpext = NULL;
1265 	bool can_coalesce = false;
1266 	bool reuse_skb = true;
1267 	struct sk_buff *skb;
1268 	size_t copy;
1269 	int i;
1270 
1271 	pr_debug("msk=%p ssk=%p sending dfrag at seq=%llu len=%u already sent=%u",
1272 		 msk, ssk, dfrag->data_seq, dfrag->data_len, info->sent);
1273 
1274 	if (WARN_ON_ONCE(info->sent > info->limit ||
1275 			 info->limit > dfrag->data_len))
1276 		return 0;
1277 
1278 	/* compute send limit */
1279 	info->mss_now = tcp_send_mss(ssk, &info->size_goal, info->flags);
1280 	copy = info->size_goal;
1281 
1282 	skb = tcp_write_queue_tail(ssk);
1283 	if (skb && copy > skb->len) {
1284 		/* Limit the write to the size available in the
1285 		 * current skb, if any, so that we create at most a new skb.
1286 		 * Explicitly tells TCP internals to avoid collapsing on later
1287 		 * queue management operation, to avoid breaking the ext <->
1288 		 * SSN association set here
1289 		 */
1290 		mpext = skb_ext_find(skb, SKB_EXT_MPTCP);
1291 		if (!mptcp_skb_can_collapse_to(data_seq, skb, mpext)) {
1292 			TCP_SKB_CB(skb)->eor = 1;
1293 			goto alloc_skb;
1294 		}
1295 
1296 		i = skb_shinfo(skb)->nr_frags;
1297 		can_coalesce = skb_can_coalesce(skb, i, dfrag->page, offset);
1298 		if (!can_coalesce && i >= sysctl_max_skb_frags) {
1299 			tcp_mark_push(tcp_sk(ssk), skb);
1300 			goto alloc_skb;
1301 		}
1302 
1303 		copy -= skb->len;
1304 	} else {
1305 alloc_skb:
1306 		skb = mptcp_alloc_tx_skb(sk, ssk, info->data_lock_held);
1307 		if (!skb)
1308 			return -ENOMEM;
1309 
1310 		i = skb_shinfo(skb)->nr_frags;
1311 		reuse_skb = false;
1312 		mpext = skb_ext_find(skb, SKB_EXT_MPTCP);
1313 	}
1314 
1315 	/* Zero window and all data acked? Probe. */
1316 	copy = mptcp_check_allowed_size(msk, ssk, data_seq, copy);
1317 	if (copy == 0) {
1318 		u64 snd_una = READ_ONCE(msk->snd_una);
1319 
1320 		if (snd_una != msk->snd_nxt) {
1321 			tcp_remove_empty_skb(ssk);
1322 			return 0;
1323 		}
1324 
1325 		zero_window_probe = true;
1326 		data_seq = snd_una - 1;
1327 		copy = 1;
1328 
1329 		/* all mptcp-level data is acked, no skbs should be present into the
1330 		 * ssk write queue
1331 		 */
1332 		WARN_ON_ONCE(reuse_skb);
1333 	}
1334 
1335 	copy = min_t(size_t, copy, info->limit - info->sent);
1336 	if (!sk_wmem_schedule(ssk, copy)) {
1337 		tcp_remove_empty_skb(ssk);
1338 		return -ENOMEM;
1339 	}
1340 
1341 	if (can_coalesce) {
1342 		skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1343 	} else {
1344 		get_page(dfrag->page);
1345 		skb_fill_page_desc(skb, i, dfrag->page, offset, copy);
1346 	}
1347 
1348 	skb->len += copy;
1349 	skb->data_len += copy;
1350 	skb->truesize += copy;
1351 	sk_wmem_queued_add(ssk, copy);
1352 	sk_mem_charge(ssk, copy);
1353 	WRITE_ONCE(tcp_sk(ssk)->write_seq, tcp_sk(ssk)->write_seq + copy);
1354 	TCP_SKB_CB(skb)->end_seq += copy;
1355 	tcp_skb_pcount_set(skb, 0);
1356 
1357 	/* on skb reuse we just need to update the DSS len */
1358 	if (reuse_skb) {
1359 		TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH;
1360 		mpext->data_len += copy;
1361 		WARN_ON_ONCE(zero_window_probe);
1362 		goto out;
1363 	}
1364 
1365 	memset(mpext, 0, sizeof(*mpext));
1366 	mpext->data_seq = data_seq;
1367 	mpext->subflow_seq = mptcp_subflow_ctx(ssk)->rel_write_seq;
1368 	mpext->data_len = copy;
1369 	mpext->use_map = 1;
1370 	mpext->dsn64 = 1;
1371 
1372 	pr_debug("data_seq=%llu subflow_seq=%u data_len=%u dsn64=%d",
1373 		 mpext->data_seq, mpext->subflow_seq, mpext->data_len,
1374 		 mpext->dsn64);
1375 
1376 	if (zero_window_probe) {
1377 		mptcp_subflow_ctx(ssk)->rel_write_seq += copy;
1378 		mpext->frozen = 1;
1379 		if (READ_ONCE(msk->csum_enabled))
1380 			mptcp_update_data_checksum(skb, copy);
1381 		tcp_push_pending_frames(ssk);
1382 		return 0;
1383 	}
1384 out:
1385 	if (READ_ONCE(msk->csum_enabled))
1386 		mptcp_update_data_checksum(skb, copy);
1387 	if (mptcp_subflow_ctx(ssk)->send_infinite_map)
1388 		mptcp_update_infinite_map(msk, ssk, mpext);
1389 	trace_mptcp_sendmsg_frag(mpext);
1390 	mptcp_subflow_ctx(ssk)->rel_write_seq += copy;
1391 	return copy;
1392 }
1393 
1394 #define MPTCP_SEND_BURST_SIZE		((1 << 16) - \
1395 					 sizeof(struct tcphdr) - \
1396 					 MAX_TCP_OPTION_SPACE - \
1397 					 sizeof(struct ipv6hdr) - \
1398 					 sizeof(struct frag_hdr))
1399 
1400 struct subflow_send_info {
1401 	struct sock *ssk;
1402 	u64 linger_time;
1403 };
1404 
1405 void mptcp_subflow_set_active(struct mptcp_subflow_context *subflow)
1406 {
1407 	if (!subflow->stale)
1408 		return;
1409 
1410 	subflow->stale = 0;
1411 	MPTCP_INC_STATS(sock_net(mptcp_subflow_tcp_sock(subflow)), MPTCP_MIB_SUBFLOWRECOVER);
1412 }
1413 
1414 bool mptcp_subflow_active(struct mptcp_subflow_context *subflow)
1415 {
1416 	if (unlikely(subflow->stale)) {
1417 		u32 rcv_tstamp = READ_ONCE(tcp_sk(mptcp_subflow_tcp_sock(subflow))->rcv_tstamp);
1418 
1419 		if (subflow->stale_rcv_tstamp == rcv_tstamp)
1420 			return false;
1421 
1422 		mptcp_subflow_set_active(subflow);
1423 	}
1424 	return __mptcp_subflow_active(subflow);
1425 }
1426 
1427 #define SSK_MODE_ACTIVE	0
1428 #define SSK_MODE_BACKUP	1
1429 #define SSK_MODE_MAX	2
1430 
1431 /* implement the mptcp packet scheduler;
1432  * returns the subflow that will transmit the next DSS
1433  * additionally updates the rtx timeout
1434  */
1435 static struct sock *mptcp_subflow_get_send(struct mptcp_sock *msk)
1436 {
1437 	struct subflow_send_info send_info[SSK_MODE_MAX];
1438 	struct mptcp_subflow_context *subflow;
1439 	struct sock *sk = (struct sock *)msk;
1440 	u32 pace, burst, wmem;
1441 	int i, nr_active = 0;
1442 	struct sock *ssk;
1443 	u64 linger_time;
1444 	long tout = 0;
1445 
1446 	sock_owned_by_me(sk);
1447 
1448 	if (__mptcp_check_fallback(msk)) {
1449 		if (!msk->first)
1450 			return NULL;
1451 		return sk_stream_memory_free(msk->first) ? msk->first : NULL;
1452 	}
1453 
1454 	/* re-use last subflow, if the burst allow that */
1455 	if (msk->last_snd && msk->snd_burst > 0 &&
1456 	    sk_stream_memory_free(msk->last_snd) &&
1457 	    mptcp_subflow_active(mptcp_subflow_ctx(msk->last_snd))) {
1458 		mptcp_set_timeout(sk);
1459 		return msk->last_snd;
1460 	}
1461 
1462 	/* pick the subflow with the lower wmem/wspace ratio */
1463 	for (i = 0; i < SSK_MODE_MAX; ++i) {
1464 		send_info[i].ssk = NULL;
1465 		send_info[i].linger_time = -1;
1466 	}
1467 
1468 	mptcp_for_each_subflow(msk, subflow) {
1469 		trace_mptcp_subflow_get_send(subflow);
1470 		ssk =  mptcp_subflow_tcp_sock(subflow);
1471 		if (!mptcp_subflow_active(subflow))
1472 			continue;
1473 
1474 		tout = max(tout, mptcp_timeout_from_subflow(subflow));
1475 		nr_active += !subflow->backup;
1476 		pace = subflow->avg_pacing_rate;
1477 		if (unlikely(!pace)) {
1478 			/* init pacing rate from socket */
1479 			subflow->avg_pacing_rate = READ_ONCE(ssk->sk_pacing_rate);
1480 			pace = subflow->avg_pacing_rate;
1481 			if (!pace)
1482 				continue;
1483 		}
1484 
1485 		linger_time = div_u64((u64)READ_ONCE(ssk->sk_wmem_queued) << 32, pace);
1486 		if (linger_time < send_info[subflow->backup].linger_time) {
1487 			send_info[subflow->backup].ssk = ssk;
1488 			send_info[subflow->backup].linger_time = linger_time;
1489 		}
1490 	}
1491 	__mptcp_set_timeout(sk, tout);
1492 
1493 	/* pick the best backup if no other subflow is active */
1494 	if (!nr_active)
1495 		send_info[SSK_MODE_ACTIVE].ssk = send_info[SSK_MODE_BACKUP].ssk;
1496 
1497 	/* According to the blest algorithm, to avoid HoL blocking for the
1498 	 * faster flow, we need to:
1499 	 * - estimate the faster flow linger time
1500 	 * - use the above to estimate the amount of byte transferred
1501 	 *   by the faster flow
1502 	 * - check that the amount of queued data is greter than the above,
1503 	 *   otherwise do not use the picked, slower, subflow
1504 	 * We select the subflow with the shorter estimated time to flush
1505 	 * the queued mem, which basically ensure the above. We just need
1506 	 * to check that subflow has a non empty cwin.
1507 	 */
1508 	ssk = send_info[SSK_MODE_ACTIVE].ssk;
1509 	if (!ssk || !sk_stream_memory_free(ssk))
1510 		return NULL;
1511 
1512 	burst = min_t(int, MPTCP_SEND_BURST_SIZE, mptcp_wnd_end(msk) - msk->snd_nxt);
1513 	wmem = READ_ONCE(ssk->sk_wmem_queued);
1514 	if (!burst) {
1515 		msk->last_snd = NULL;
1516 		return ssk;
1517 	}
1518 
1519 	subflow = mptcp_subflow_ctx(ssk);
1520 	subflow->avg_pacing_rate = div_u64((u64)subflow->avg_pacing_rate * wmem +
1521 					   READ_ONCE(ssk->sk_pacing_rate) * burst,
1522 					   burst + wmem);
1523 	msk->last_snd = ssk;
1524 	msk->snd_burst = burst;
1525 	return ssk;
1526 }
1527 
1528 static void mptcp_push_release(struct sock *ssk, struct mptcp_sendmsg_info *info)
1529 {
1530 	tcp_push(ssk, 0, info->mss_now, tcp_sk(ssk)->nonagle, info->size_goal);
1531 	release_sock(ssk);
1532 }
1533 
1534 static void mptcp_update_post_push(struct mptcp_sock *msk,
1535 				   struct mptcp_data_frag *dfrag,
1536 				   u32 sent)
1537 {
1538 	u64 snd_nxt_new = dfrag->data_seq;
1539 
1540 	dfrag->already_sent += sent;
1541 
1542 	msk->snd_burst -= sent;
1543 
1544 	snd_nxt_new += dfrag->already_sent;
1545 
1546 	/* snd_nxt_new can be smaller than snd_nxt in case mptcp
1547 	 * is recovering after a failover. In that event, this re-sends
1548 	 * old segments.
1549 	 *
1550 	 * Thus compute snd_nxt_new candidate based on
1551 	 * the dfrag->data_seq that was sent and the data
1552 	 * that has been handed to the subflow for transmission
1553 	 * and skip update in case it was old dfrag.
1554 	 */
1555 	if (likely(after64(snd_nxt_new, msk->snd_nxt)))
1556 		msk->snd_nxt = snd_nxt_new;
1557 }
1558 
1559 void mptcp_check_and_set_pending(struct sock *sk)
1560 {
1561 	if (mptcp_send_head(sk))
1562 		mptcp_sk(sk)->push_pending |= BIT(MPTCP_PUSH_PENDING);
1563 }
1564 
1565 void __mptcp_push_pending(struct sock *sk, unsigned int flags)
1566 {
1567 	struct sock *prev_ssk = NULL, *ssk = NULL;
1568 	struct mptcp_sock *msk = mptcp_sk(sk);
1569 	struct mptcp_sendmsg_info info = {
1570 				.flags = flags,
1571 	};
1572 	struct mptcp_data_frag *dfrag;
1573 	int len, copied = 0;
1574 
1575 	while ((dfrag = mptcp_send_head(sk))) {
1576 		info.sent = dfrag->already_sent;
1577 		info.limit = dfrag->data_len;
1578 		len = dfrag->data_len - dfrag->already_sent;
1579 		while (len > 0) {
1580 			int ret = 0;
1581 
1582 			prev_ssk = ssk;
1583 			ssk = mptcp_subflow_get_send(msk);
1584 
1585 			/* First check. If the ssk has changed since
1586 			 * the last round, release prev_ssk
1587 			 */
1588 			if (ssk != prev_ssk && prev_ssk)
1589 				mptcp_push_release(prev_ssk, &info);
1590 			if (!ssk)
1591 				goto out;
1592 
1593 			/* Need to lock the new subflow only if different
1594 			 * from the previous one, otherwise we are still
1595 			 * helding the relevant lock
1596 			 */
1597 			if (ssk != prev_ssk)
1598 				lock_sock(ssk);
1599 
1600 			ret = mptcp_sendmsg_frag(sk, ssk, dfrag, &info);
1601 			if (ret <= 0) {
1602 				mptcp_push_release(ssk, &info);
1603 				goto out;
1604 			}
1605 
1606 			info.sent += ret;
1607 			copied += ret;
1608 			len -= ret;
1609 
1610 			mptcp_update_post_push(msk, dfrag, ret);
1611 		}
1612 		WRITE_ONCE(msk->first_pending, mptcp_send_next(sk));
1613 	}
1614 
1615 	/* at this point we held the socket lock for the last subflow we used */
1616 	if (ssk)
1617 		mptcp_push_release(ssk, &info);
1618 
1619 out:
1620 	/* ensure the rtx timer is running */
1621 	if (!mptcp_timer_pending(sk))
1622 		mptcp_reset_timer(sk);
1623 	if (copied)
1624 		__mptcp_check_send_data_fin(sk);
1625 }
1626 
1627 static void __mptcp_subflow_push_pending(struct sock *sk, struct sock *ssk)
1628 {
1629 	struct mptcp_sock *msk = mptcp_sk(sk);
1630 	struct mptcp_sendmsg_info info = {
1631 		.data_lock_held = true,
1632 	};
1633 	struct mptcp_data_frag *dfrag;
1634 	struct sock *xmit_ssk;
1635 	int len, copied = 0;
1636 	bool first = true;
1637 
1638 	info.flags = 0;
1639 	while ((dfrag = mptcp_send_head(sk))) {
1640 		info.sent = dfrag->already_sent;
1641 		info.limit = dfrag->data_len;
1642 		len = dfrag->data_len - dfrag->already_sent;
1643 		while (len > 0) {
1644 			int ret = 0;
1645 
1646 			/* the caller already invoked the packet scheduler,
1647 			 * check for a different subflow usage only after
1648 			 * spooling the first chunk of data
1649 			 */
1650 			xmit_ssk = first ? ssk : mptcp_subflow_get_send(mptcp_sk(sk));
1651 			if (!xmit_ssk)
1652 				goto out;
1653 			if (xmit_ssk != ssk) {
1654 				mptcp_subflow_delegate(mptcp_subflow_ctx(xmit_ssk),
1655 						       MPTCP_DELEGATE_SEND);
1656 				goto out;
1657 			}
1658 
1659 			ret = mptcp_sendmsg_frag(sk, ssk, dfrag, &info);
1660 			if (ret <= 0)
1661 				goto out;
1662 
1663 			info.sent += ret;
1664 			copied += ret;
1665 			len -= ret;
1666 			first = false;
1667 
1668 			mptcp_update_post_push(msk, dfrag, ret);
1669 		}
1670 		WRITE_ONCE(msk->first_pending, mptcp_send_next(sk));
1671 	}
1672 
1673 out:
1674 	/* __mptcp_alloc_tx_skb could have released some wmem and we are
1675 	 * not going to flush it via release_sock()
1676 	 */
1677 	if (copied) {
1678 		tcp_push(ssk, 0, info.mss_now, tcp_sk(ssk)->nonagle,
1679 			 info.size_goal);
1680 		if (!mptcp_timer_pending(sk))
1681 			mptcp_reset_timer(sk);
1682 
1683 		if (msk->snd_data_fin_enable &&
1684 		    msk->snd_nxt + 1 == msk->write_seq)
1685 			mptcp_schedule_work(sk);
1686 	}
1687 }
1688 
1689 static void mptcp_set_nospace(struct sock *sk)
1690 {
1691 	/* enable autotune */
1692 	set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1693 
1694 	/* will be cleared on avail space */
1695 	set_bit(MPTCP_NOSPACE, &mptcp_sk(sk)->flags);
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 	size_t copied = 0;
1703 	int ret = 0;
1704 	long timeo;
1705 
1706 	/* we don't support FASTOPEN yet */
1707 	if (msg->msg_flags & MSG_FASTOPEN)
1708 		return -EOPNOTSUPP;
1709 
1710 	/* silently ignore everything else */
1711 	msg->msg_flags &= MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL;
1712 
1713 	lock_sock(sk);
1714 
1715 	timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1716 
1717 	if ((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) {
1718 		ret = sk_stream_wait_connect(sk, &timeo);
1719 		if (ret)
1720 			goto out;
1721 	}
1722 
1723 	pfrag = sk_page_frag(sk);
1724 
1725 	while (msg_data_left(msg)) {
1726 		int total_ts, frag_truesize = 0;
1727 		struct mptcp_data_frag *dfrag;
1728 		bool dfrag_collapsed;
1729 		size_t psize, offset;
1730 
1731 		if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN)) {
1732 			ret = -EPIPE;
1733 			goto out;
1734 		}
1735 
1736 		/* reuse tail pfrag, if possible, or carve a new one from the
1737 		 * page allocator
1738 		 */
1739 		dfrag = mptcp_pending_tail(sk);
1740 		dfrag_collapsed = mptcp_frag_can_collapse_to(msk, pfrag, dfrag);
1741 		if (!dfrag_collapsed) {
1742 			if (!sk_stream_memory_free(sk))
1743 				goto wait_for_memory;
1744 
1745 			if (!mptcp_page_frag_refill(sk, pfrag))
1746 				goto wait_for_memory;
1747 
1748 			dfrag = mptcp_carve_data_frag(msk, pfrag, pfrag->offset);
1749 			frag_truesize = dfrag->overhead;
1750 		}
1751 
1752 		/* we do not bound vs wspace, to allow a single packet.
1753 		 * memory accounting will prevent execessive memory usage
1754 		 * anyway
1755 		 */
1756 		offset = dfrag->offset + dfrag->data_len;
1757 		psize = pfrag->size - offset;
1758 		psize = min_t(size_t, psize, msg_data_left(msg));
1759 		total_ts = psize + frag_truesize;
1760 
1761 		if (!sk_wmem_schedule(sk, total_ts))
1762 			goto wait_for_memory;
1763 
1764 		if (copy_page_from_iter(dfrag->page, offset, psize,
1765 					&msg->msg_iter) != psize) {
1766 			ret = -EFAULT;
1767 			goto out;
1768 		}
1769 
1770 		/* data successfully copied into the write queue */
1771 		sk->sk_forward_alloc -= total_ts;
1772 		copied += psize;
1773 		dfrag->data_len += psize;
1774 		frag_truesize += psize;
1775 		pfrag->offset += frag_truesize;
1776 		WRITE_ONCE(msk->write_seq, msk->write_seq + psize);
1777 
1778 		/* charge data on mptcp pending queue to the msk socket
1779 		 * Note: we charge such data both to sk and ssk
1780 		 */
1781 		sk_wmem_queued_add(sk, frag_truesize);
1782 		if (!dfrag_collapsed) {
1783 			get_page(dfrag->page);
1784 			list_add_tail(&dfrag->list, &msk->rtx_queue);
1785 			if (!msk->first_pending)
1786 				WRITE_ONCE(msk->first_pending, dfrag);
1787 		}
1788 		pr_debug("msk=%p dfrag at seq=%llu len=%u sent=%u new=%d", msk,
1789 			 dfrag->data_seq, dfrag->data_len, dfrag->already_sent,
1790 			 !dfrag_collapsed);
1791 
1792 		continue;
1793 
1794 wait_for_memory:
1795 		mptcp_set_nospace(sk);
1796 		__mptcp_push_pending(sk, msg->msg_flags);
1797 		ret = sk_stream_wait_memory(sk, &timeo);
1798 		if (ret)
1799 			goto out;
1800 	}
1801 
1802 	if (copied)
1803 		__mptcp_push_pending(sk, msg->msg_flags);
1804 
1805 out:
1806 	release_sock(sk);
1807 	return copied ? : ret;
1808 }
1809 
1810 static int __mptcp_recvmsg_mskq(struct mptcp_sock *msk,
1811 				struct msghdr *msg,
1812 				size_t len, int flags,
1813 				struct scm_timestamping_internal *tss,
1814 				int *cmsg_flags)
1815 {
1816 	struct sk_buff *skb, *tmp;
1817 	int copied = 0;
1818 
1819 	skb_queue_walk_safe(&msk->receive_queue, skb, tmp) {
1820 		u32 offset = MPTCP_SKB_CB(skb)->offset;
1821 		u32 data_len = skb->len - offset;
1822 		u32 count = min_t(size_t, len - copied, data_len);
1823 		int err;
1824 
1825 		if (!(flags & MSG_TRUNC)) {
1826 			err = skb_copy_datagram_msg(skb, offset, msg, count);
1827 			if (unlikely(err < 0)) {
1828 				if (!copied)
1829 					return err;
1830 				break;
1831 			}
1832 		}
1833 
1834 		if (MPTCP_SKB_CB(skb)->has_rxtstamp) {
1835 			tcp_update_recv_tstamps(skb, tss);
1836 			*cmsg_flags |= MPTCP_CMSG_TS;
1837 		}
1838 
1839 		copied += count;
1840 
1841 		if (count < data_len) {
1842 			if (!(flags & MSG_PEEK)) {
1843 				MPTCP_SKB_CB(skb)->offset += count;
1844 				MPTCP_SKB_CB(skb)->map_seq += count;
1845 			}
1846 			break;
1847 		}
1848 
1849 		if (!(flags & MSG_PEEK)) {
1850 			/* we will bulk release the skb memory later */
1851 			skb->destructor = NULL;
1852 			WRITE_ONCE(msk->rmem_released, msk->rmem_released + skb->truesize);
1853 			__skb_unlink(skb, &msk->receive_queue);
1854 			__kfree_skb(skb);
1855 		}
1856 
1857 		if (copied >= len)
1858 			break;
1859 	}
1860 
1861 	return copied;
1862 }
1863 
1864 /* receive buffer autotuning.  See tcp_rcv_space_adjust for more information.
1865  *
1866  * Only difference: Use highest rtt estimate of the subflows in use.
1867  */
1868 static void mptcp_rcv_space_adjust(struct mptcp_sock *msk, int copied)
1869 {
1870 	struct mptcp_subflow_context *subflow;
1871 	struct sock *sk = (struct sock *)msk;
1872 	u32 time, advmss = 1;
1873 	u64 rtt_us, mstamp;
1874 
1875 	sock_owned_by_me(sk);
1876 
1877 	if (copied <= 0)
1878 		return;
1879 
1880 	msk->rcvq_space.copied += copied;
1881 
1882 	mstamp = div_u64(tcp_clock_ns(), NSEC_PER_USEC);
1883 	time = tcp_stamp_us_delta(mstamp, msk->rcvq_space.time);
1884 
1885 	rtt_us = msk->rcvq_space.rtt_us;
1886 	if (rtt_us && time < (rtt_us >> 3))
1887 		return;
1888 
1889 	rtt_us = 0;
1890 	mptcp_for_each_subflow(msk, subflow) {
1891 		const struct tcp_sock *tp;
1892 		u64 sf_rtt_us;
1893 		u32 sf_advmss;
1894 
1895 		tp = tcp_sk(mptcp_subflow_tcp_sock(subflow));
1896 
1897 		sf_rtt_us = READ_ONCE(tp->rcv_rtt_est.rtt_us);
1898 		sf_advmss = READ_ONCE(tp->advmss);
1899 
1900 		rtt_us = max(sf_rtt_us, rtt_us);
1901 		advmss = max(sf_advmss, advmss);
1902 	}
1903 
1904 	msk->rcvq_space.rtt_us = rtt_us;
1905 	if (time < (rtt_us >> 3) || rtt_us == 0)
1906 		return;
1907 
1908 	if (msk->rcvq_space.copied <= msk->rcvq_space.space)
1909 		goto new_measure;
1910 
1911 	if (sock_net(sk)->ipv4.sysctl_tcp_moderate_rcvbuf &&
1912 	    !(sk->sk_userlocks & SOCK_RCVBUF_LOCK)) {
1913 		int rcvmem, rcvbuf;
1914 		u64 rcvwin, grow;
1915 
1916 		rcvwin = ((u64)msk->rcvq_space.copied << 1) + 16 * advmss;
1917 
1918 		grow = rcvwin * (msk->rcvq_space.copied - msk->rcvq_space.space);
1919 
1920 		do_div(grow, msk->rcvq_space.space);
1921 		rcvwin += (grow << 1);
1922 
1923 		rcvmem = SKB_TRUESIZE(advmss + MAX_TCP_HEADER);
1924 		while (tcp_win_from_space(sk, rcvmem) < advmss)
1925 			rcvmem += 128;
1926 
1927 		do_div(rcvwin, advmss);
1928 		rcvbuf = min_t(u64, rcvwin * rcvmem,
1929 			       sock_net(sk)->ipv4.sysctl_tcp_rmem[2]);
1930 
1931 		if (rcvbuf > sk->sk_rcvbuf) {
1932 			u32 window_clamp;
1933 
1934 			window_clamp = tcp_win_from_space(sk, rcvbuf);
1935 			WRITE_ONCE(sk->sk_rcvbuf, rcvbuf);
1936 
1937 			/* Make subflows follow along.  If we do not do this, we
1938 			 * get drops at subflow level if skbs can't be moved to
1939 			 * the mptcp rx queue fast enough (announced rcv_win can
1940 			 * exceed ssk->sk_rcvbuf).
1941 			 */
1942 			mptcp_for_each_subflow(msk, subflow) {
1943 				struct sock *ssk;
1944 				bool slow;
1945 
1946 				ssk = mptcp_subflow_tcp_sock(subflow);
1947 				slow = lock_sock_fast(ssk);
1948 				WRITE_ONCE(ssk->sk_rcvbuf, rcvbuf);
1949 				tcp_sk(ssk)->window_clamp = window_clamp;
1950 				tcp_cleanup_rbuf(ssk, 1);
1951 				unlock_sock_fast(ssk, slow);
1952 			}
1953 		}
1954 	}
1955 
1956 	msk->rcvq_space.space = msk->rcvq_space.copied;
1957 new_measure:
1958 	msk->rcvq_space.copied = 0;
1959 	msk->rcvq_space.time = mstamp;
1960 }
1961 
1962 static void __mptcp_update_rmem(struct sock *sk)
1963 {
1964 	struct mptcp_sock *msk = mptcp_sk(sk);
1965 
1966 	if (!msk->rmem_released)
1967 		return;
1968 
1969 	atomic_sub(msk->rmem_released, &sk->sk_rmem_alloc);
1970 	mptcp_rmem_uncharge(sk, msk->rmem_released);
1971 	WRITE_ONCE(msk->rmem_released, 0);
1972 }
1973 
1974 static void __mptcp_splice_receive_queue(struct sock *sk)
1975 {
1976 	struct mptcp_sock *msk = mptcp_sk(sk);
1977 
1978 	skb_queue_splice_tail_init(&sk->sk_receive_queue, &msk->receive_queue);
1979 }
1980 
1981 static bool __mptcp_move_skbs(struct mptcp_sock *msk)
1982 {
1983 	struct sock *sk = (struct sock *)msk;
1984 	unsigned int moved = 0;
1985 	bool ret, done;
1986 
1987 	do {
1988 		struct sock *ssk = mptcp_subflow_recv_lookup(msk);
1989 		bool slowpath;
1990 
1991 		/* we can have data pending in the subflows only if the msk
1992 		 * receive buffer was full at subflow_data_ready() time,
1993 		 * that is an unlikely slow path.
1994 		 */
1995 		if (likely(!ssk))
1996 			break;
1997 
1998 		slowpath = lock_sock_fast(ssk);
1999 		mptcp_data_lock(sk);
2000 		__mptcp_update_rmem(sk);
2001 		done = __mptcp_move_skbs_from_subflow(msk, ssk, &moved);
2002 		mptcp_data_unlock(sk);
2003 
2004 		if (unlikely(ssk->sk_err))
2005 			__mptcp_error_report(sk);
2006 		unlock_sock_fast(ssk, slowpath);
2007 	} while (!done);
2008 
2009 	/* acquire the data lock only if some input data is pending */
2010 	ret = moved > 0;
2011 	if (!RB_EMPTY_ROOT(&msk->out_of_order_queue) ||
2012 	    !skb_queue_empty_lockless(&sk->sk_receive_queue)) {
2013 		mptcp_data_lock(sk);
2014 		__mptcp_update_rmem(sk);
2015 		ret |= __mptcp_ofo_queue(msk);
2016 		__mptcp_splice_receive_queue(sk);
2017 		mptcp_data_unlock(sk);
2018 	}
2019 	if (ret)
2020 		mptcp_check_data_fin((struct sock *)msk);
2021 	return !skb_queue_empty(&msk->receive_queue);
2022 }
2023 
2024 static unsigned int mptcp_inq_hint(const struct sock *sk)
2025 {
2026 	const struct mptcp_sock *msk = mptcp_sk(sk);
2027 	const struct sk_buff *skb;
2028 
2029 	skb = skb_peek(&msk->receive_queue);
2030 	if (skb) {
2031 		u64 hint_val = msk->ack_seq - MPTCP_SKB_CB(skb)->map_seq;
2032 
2033 		if (hint_val >= INT_MAX)
2034 			return INT_MAX;
2035 
2036 		return (unsigned int)hint_val;
2037 	}
2038 
2039 	if (sk->sk_state == TCP_CLOSE || (sk->sk_shutdown & RCV_SHUTDOWN))
2040 		return 1;
2041 
2042 	return 0;
2043 }
2044 
2045 static int mptcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
2046 			 int flags, int *addr_len)
2047 {
2048 	struct mptcp_sock *msk = mptcp_sk(sk);
2049 	struct scm_timestamping_internal tss;
2050 	int copied = 0, cmsg_flags = 0;
2051 	int target;
2052 	long timeo;
2053 
2054 	/* MSG_ERRQUEUE is really a no-op till we support IP_RECVERR */
2055 	if (unlikely(flags & MSG_ERRQUEUE))
2056 		return inet_recv_error(sk, msg, len, addr_len);
2057 
2058 	lock_sock(sk);
2059 	if (unlikely(sk->sk_state == TCP_LISTEN)) {
2060 		copied = -ENOTCONN;
2061 		goto out_err;
2062 	}
2063 
2064 	timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
2065 
2066 	len = min_t(size_t, len, INT_MAX);
2067 	target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
2068 
2069 	if (unlikely(msk->recvmsg_inq))
2070 		cmsg_flags = MPTCP_CMSG_INQ;
2071 
2072 	while (copied < len) {
2073 		int bytes_read;
2074 
2075 		bytes_read = __mptcp_recvmsg_mskq(msk, msg, len - copied, flags, &tss, &cmsg_flags);
2076 		if (unlikely(bytes_read < 0)) {
2077 			if (!copied)
2078 				copied = bytes_read;
2079 			goto out_err;
2080 		}
2081 
2082 		copied += bytes_read;
2083 
2084 		/* be sure to advertise window change */
2085 		mptcp_cleanup_rbuf(msk);
2086 
2087 		if (skb_queue_empty(&msk->receive_queue) && __mptcp_move_skbs(msk))
2088 			continue;
2089 
2090 		/* only the master socket status is relevant here. The exit
2091 		 * conditions mirror closely tcp_recvmsg()
2092 		 */
2093 		if (copied >= target)
2094 			break;
2095 
2096 		if (copied) {
2097 			if (sk->sk_err ||
2098 			    sk->sk_state == TCP_CLOSE ||
2099 			    (sk->sk_shutdown & RCV_SHUTDOWN) ||
2100 			    !timeo ||
2101 			    signal_pending(current))
2102 				break;
2103 		} else {
2104 			if (sk->sk_err) {
2105 				copied = sock_error(sk);
2106 				break;
2107 			}
2108 
2109 			if (test_and_clear_bit(MPTCP_WORK_EOF, &msk->flags))
2110 				mptcp_check_for_eof(msk);
2111 
2112 			if (sk->sk_shutdown & RCV_SHUTDOWN) {
2113 				/* race breaker: the shutdown could be after the
2114 				 * previous receive queue check
2115 				 */
2116 				if (__mptcp_move_skbs(msk))
2117 					continue;
2118 				break;
2119 			}
2120 
2121 			if (sk->sk_state == TCP_CLOSE) {
2122 				copied = -ENOTCONN;
2123 				break;
2124 			}
2125 
2126 			if (!timeo) {
2127 				copied = -EAGAIN;
2128 				break;
2129 			}
2130 
2131 			if (signal_pending(current)) {
2132 				copied = sock_intr_errno(timeo);
2133 				break;
2134 			}
2135 		}
2136 
2137 		pr_debug("block timeout %ld", timeo);
2138 		sk_wait_data(sk, &timeo, NULL);
2139 	}
2140 
2141 out_err:
2142 	if (cmsg_flags && copied >= 0) {
2143 		if (cmsg_flags & MPTCP_CMSG_TS)
2144 			tcp_recv_timestamp(msg, sk, &tss);
2145 
2146 		if (cmsg_flags & MPTCP_CMSG_INQ) {
2147 			unsigned int inq = mptcp_inq_hint(sk);
2148 
2149 			put_cmsg(msg, SOL_TCP, TCP_CM_INQ, sizeof(inq), &inq);
2150 		}
2151 	}
2152 
2153 	pr_debug("msk=%p rx queue empty=%d:%d copied=%d",
2154 		 msk, skb_queue_empty_lockless(&sk->sk_receive_queue),
2155 		 skb_queue_empty(&msk->receive_queue), copied);
2156 	if (!(flags & MSG_PEEK))
2157 		mptcp_rcv_space_adjust(msk, copied);
2158 
2159 	release_sock(sk);
2160 	return copied;
2161 }
2162 
2163 static void mptcp_retransmit_timer(struct timer_list *t)
2164 {
2165 	struct inet_connection_sock *icsk = from_timer(icsk, t,
2166 						       icsk_retransmit_timer);
2167 	struct sock *sk = &icsk->icsk_inet.sk;
2168 	struct mptcp_sock *msk = mptcp_sk(sk);
2169 
2170 	bh_lock_sock(sk);
2171 	if (!sock_owned_by_user(sk)) {
2172 		/* we need a process context to retransmit */
2173 		if (!test_and_set_bit(MPTCP_WORK_RTX, &msk->flags))
2174 			mptcp_schedule_work(sk);
2175 	} else {
2176 		/* delegate our work to tcp_release_cb() */
2177 		__set_bit(MPTCP_RETRANSMIT, &msk->cb_flags);
2178 	}
2179 	bh_unlock_sock(sk);
2180 	sock_put(sk);
2181 }
2182 
2183 static void mptcp_timeout_timer(struct timer_list *t)
2184 {
2185 	struct sock *sk = from_timer(sk, t, sk_timer);
2186 
2187 	mptcp_schedule_work(sk);
2188 	sock_put(sk);
2189 }
2190 
2191 /* Find an idle subflow.  Return NULL if there is unacked data at tcp
2192  * level.
2193  *
2194  * A backup subflow is returned only if that is the only kind available.
2195  */
2196 static struct sock *mptcp_subflow_get_retrans(struct mptcp_sock *msk)
2197 {
2198 	struct sock *backup = NULL, *pick = NULL;
2199 	struct mptcp_subflow_context *subflow;
2200 	int min_stale_count = INT_MAX;
2201 
2202 	sock_owned_by_me((const struct sock *)msk);
2203 
2204 	if (__mptcp_check_fallback(msk))
2205 		return NULL;
2206 
2207 	mptcp_for_each_subflow(msk, subflow) {
2208 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2209 
2210 		if (!__mptcp_subflow_active(subflow))
2211 			continue;
2212 
2213 		/* still data outstanding at TCP level? skip this */
2214 		if (!tcp_rtx_and_write_queues_empty(ssk)) {
2215 			mptcp_pm_subflow_chk_stale(msk, ssk);
2216 			min_stale_count = min_t(int, min_stale_count, subflow->stale_count);
2217 			continue;
2218 		}
2219 
2220 		if (subflow->backup) {
2221 			if (!backup)
2222 				backup = ssk;
2223 			continue;
2224 		}
2225 
2226 		if (!pick)
2227 			pick = ssk;
2228 	}
2229 
2230 	if (pick)
2231 		return pick;
2232 
2233 	/* use backup only if there are no progresses anywhere */
2234 	return min_stale_count > 1 ? backup : NULL;
2235 }
2236 
2237 static void mptcp_dispose_initial_subflow(struct mptcp_sock *msk)
2238 {
2239 	if (msk->subflow) {
2240 		iput(SOCK_INODE(msk->subflow));
2241 		msk->subflow = NULL;
2242 	}
2243 }
2244 
2245 bool __mptcp_retransmit_pending_data(struct sock *sk)
2246 {
2247 	struct mptcp_data_frag *cur, *rtx_head;
2248 	struct mptcp_sock *msk = mptcp_sk(sk);
2249 
2250 	if (__mptcp_check_fallback(mptcp_sk(sk)))
2251 		return false;
2252 
2253 	if (tcp_rtx_and_write_queues_empty(sk))
2254 		return false;
2255 
2256 	/* the closing socket has some data untransmitted and/or unacked:
2257 	 * some data in the mptcp rtx queue has not really xmitted yet.
2258 	 * keep it simple and re-inject the whole mptcp level rtx queue
2259 	 */
2260 	mptcp_data_lock(sk);
2261 	__mptcp_clean_una_wakeup(sk);
2262 	rtx_head = mptcp_rtx_head(sk);
2263 	if (!rtx_head) {
2264 		mptcp_data_unlock(sk);
2265 		return false;
2266 	}
2267 
2268 	msk->recovery_snd_nxt = msk->snd_nxt;
2269 	msk->recovery = true;
2270 	mptcp_data_unlock(sk);
2271 
2272 	msk->first_pending = rtx_head;
2273 	msk->snd_burst = 0;
2274 
2275 	/* be sure to clear the "sent status" on all re-injected fragments */
2276 	list_for_each_entry(cur, &msk->rtx_queue, list) {
2277 		if (!cur->already_sent)
2278 			break;
2279 		cur->already_sent = 0;
2280 	}
2281 
2282 	return true;
2283 }
2284 
2285 /* flags for __mptcp_close_ssk() */
2286 #define MPTCP_CF_PUSH		BIT(1)
2287 #define MPTCP_CF_FASTCLOSE	BIT(2)
2288 
2289 /* subflow sockets can be either outgoing (connect) or incoming
2290  * (accept).
2291  *
2292  * Outgoing subflows use in-kernel sockets.
2293  * Incoming subflows do not have their own 'struct socket' allocated,
2294  * so we need to use tcp_close() after detaching them from the mptcp
2295  * parent socket.
2296  */
2297 static void __mptcp_close_ssk(struct sock *sk, struct sock *ssk,
2298 			      struct mptcp_subflow_context *subflow,
2299 			      unsigned int flags)
2300 {
2301 	struct mptcp_sock *msk = mptcp_sk(sk);
2302 	bool need_push, dispose_it;
2303 
2304 	dispose_it = !msk->subflow || ssk != msk->subflow->sk;
2305 	if (dispose_it)
2306 		list_del(&subflow->node);
2307 
2308 	lock_sock_nested(ssk, SINGLE_DEPTH_NESTING);
2309 
2310 	if (flags & MPTCP_CF_FASTCLOSE)
2311 		subflow->send_fastclose = 1;
2312 
2313 	need_push = (flags & MPTCP_CF_PUSH) && __mptcp_retransmit_pending_data(sk);
2314 	if (!dispose_it) {
2315 		tcp_disconnect(ssk, 0);
2316 		msk->subflow->state = SS_UNCONNECTED;
2317 		mptcp_subflow_ctx_reset(subflow);
2318 		release_sock(ssk);
2319 
2320 		goto out;
2321 	}
2322 
2323 	/* if we are invoked by the msk cleanup code, the subflow is
2324 	 * already orphaned
2325 	 */
2326 	if (ssk->sk_socket)
2327 		sock_orphan(ssk);
2328 
2329 	subflow->disposable = 1;
2330 
2331 	/* if ssk hit tcp_done(), tcp_cleanup_ulp() cleared the related ops
2332 	 * the ssk has been already destroyed, we just need to release the
2333 	 * reference owned by msk;
2334 	 */
2335 	if (!inet_csk(ssk)->icsk_ulp_ops) {
2336 		kfree_rcu(subflow, rcu);
2337 	} else {
2338 		/* otherwise tcp will dispose of the ssk and subflow ctx */
2339 		if (ssk->sk_state == TCP_LISTEN) {
2340 			tcp_set_state(ssk, TCP_CLOSE);
2341 			mptcp_subflow_queue_clean(ssk);
2342 			inet_csk_listen_stop(ssk);
2343 		}
2344 		__tcp_close(ssk, 0);
2345 
2346 		/* close acquired an extra ref */
2347 		__sock_put(ssk);
2348 	}
2349 	release_sock(ssk);
2350 
2351 	sock_put(ssk);
2352 
2353 	if (ssk == msk->first)
2354 		msk->first = NULL;
2355 
2356 out:
2357 	if (ssk == msk->last_snd)
2358 		msk->last_snd = NULL;
2359 
2360 	if (need_push)
2361 		__mptcp_push_pending(sk, 0);
2362 }
2363 
2364 void mptcp_close_ssk(struct sock *sk, struct sock *ssk,
2365 		     struct mptcp_subflow_context *subflow)
2366 {
2367 	if (sk->sk_state == TCP_ESTABLISHED)
2368 		mptcp_event(MPTCP_EVENT_SUB_CLOSED, mptcp_sk(sk), ssk, GFP_KERNEL);
2369 
2370 	/* subflow aborted before reaching the fully_established status
2371 	 * attempt the creation of the next subflow
2372 	 */
2373 	mptcp_pm_subflow_check_next(mptcp_sk(sk), ssk, subflow);
2374 
2375 	__mptcp_close_ssk(sk, ssk, subflow, MPTCP_CF_PUSH);
2376 }
2377 
2378 static unsigned int mptcp_sync_mss(struct sock *sk, u32 pmtu)
2379 {
2380 	return 0;
2381 }
2382 
2383 static void __mptcp_close_subflow(struct mptcp_sock *msk)
2384 {
2385 	struct mptcp_subflow_context *subflow, *tmp;
2386 
2387 	might_sleep();
2388 
2389 	list_for_each_entry_safe(subflow, tmp, &msk->conn_list, node) {
2390 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2391 
2392 		if (inet_sk_state_load(ssk) != TCP_CLOSE)
2393 			continue;
2394 
2395 		/* 'subflow_data_ready' will re-sched once rx queue is empty */
2396 		if (!skb_queue_empty_lockless(&ssk->sk_receive_queue))
2397 			continue;
2398 
2399 		mptcp_close_ssk((struct sock *)msk, ssk, subflow);
2400 	}
2401 }
2402 
2403 static bool mptcp_check_close_timeout(const struct sock *sk)
2404 {
2405 	s32 delta = tcp_jiffies32 - inet_csk(sk)->icsk_mtup.probe_timestamp;
2406 	struct mptcp_subflow_context *subflow;
2407 
2408 	if (delta >= TCP_TIMEWAIT_LEN)
2409 		return true;
2410 
2411 	/* if all subflows are in closed status don't bother with additional
2412 	 * timeout
2413 	 */
2414 	mptcp_for_each_subflow(mptcp_sk(sk), subflow) {
2415 		if (inet_sk_state_load(mptcp_subflow_tcp_sock(subflow)) !=
2416 		    TCP_CLOSE)
2417 			return false;
2418 	}
2419 	return true;
2420 }
2421 
2422 static void mptcp_check_fastclose(struct mptcp_sock *msk)
2423 {
2424 	struct mptcp_subflow_context *subflow, *tmp;
2425 	struct sock *sk = &msk->sk.icsk_inet.sk;
2426 
2427 	if (likely(!READ_ONCE(msk->rcv_fastclose)))
2428 		return;
2429 
2430 	mptcp_token_destroy(msk);
2431 
2432 	list_for_each_entry_safe(subflow, tmp, &msk->conn_list, node) {
2433 		struct sock *tcp_sk = mptcp_subflow_tcp_sock(subflow);
2434 		bool slow;
2435 
2436 		slow = lock_sock_fast(tcp_sk);
2437 		if (tcp_sk->sk_state != TCP_CLOSE) {
2438 			tcp_send_active_reset(tcp_sk, GFP_ATOMIC);
2439 			tcp_set_state(tcp_sk, TCP_CLOSE);
2440 		}
2441 		unlock_sock_fast(tcp_sk, slow);
2442 	}
2443 
2444 	inet_sk_state_store(sk, TCP_CLOSE);
2445 	sk->sk_shutdown = SHUTDOWN_MASK;
2446 	smp_mb__before_atomic(); /* SHUTDOWN must be visible first */
2447 	set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags);
2448 
2449 	mptcp_close_wake_up(sk);
2450 }
2451 
2452 static void __mptcp_retrans(struct sock *sk)
2453 {
2454 	struct mptcp_sock *msk = mptcp_sk(sk);
2455 	struct mptcp_sendmsg_info info = {};
2456 	struct mptcp_data_frag *dfrag;
2457 	size_t copied = 0;
2458 	struct sock *ssk;
2459 	int ret;
2460 
2461 	mptcp_clean_una_wakeup(sk);
2462 
2463 	/* first check ssk: need to kick "stale" logic */
2464 	ssk = mptcp_subflow_get_retrans(msk);
2465 	dfrag = mptcp_rtx_head(sk);
2466 	if (!dfrag) {
2467 		if (mptcp_data_fin_enabled(msk)) {
2468 			struct inet_connection_sock *icsk = inet_csk(sk);
2469 
2470 			icsk->icsk_retransmits++;
2471 			mptcp_set_datafin_timeout(sk);
2472 			mptcp_send_ack(msk);
2473 
2474 			goto reset_timer;
2475 		}
2476 
2477 		if (!mptcp_send_head(sk))
2478 			return;
2479 
2480 		goto reset_timer;
2481 	}
2482 
2483 	if (!ssk)
2484 		goto reset_timer;
2485 
2486 	lock_sock(ssk);
2487 
2488 	/* limit retransmission to the bytes already sent on some subflows */
2489 	info.sent = 0;
2490 	info.limit = READ_ONCE(msk->csum_enabled) ? dfrag->data_len : dfrag->already_sent;
2491 	while (info.sent < info.limit) {
2492 		ret = mptcp_sendmsg_frag(sk, ssk, dfrag, &info);
2493 		if (ret <= 0)
2494 			break;
2495 
2496 		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_RETRANSSEGS);
2497 		copied += ret;
2498 		info.sent += ret;
2499 	}
2500 	if (copied) {
2501 		dfrag->already_sent = max(dfrag->already_sent, info.sent);
2502 		tcp_push(ssk, 0, info.mss_now, tcp_sk(ssk)->nonagle,
2503 			 info.size_goal);
2504 		WRITE_ONCE(msk->allow_infinite_fallback, false);
2505 	}
2506 
2507 	release_sock(ssk);
2508 
2509 reset_timer:
2510 	mptcp_check_and_set_pending(sk);
2511 
2512 	if (!mptcp_timer_pending(sk))
2513 		mptcp_reset_timer(sk);
2514 }
2515 
2516 /* schedule the timeout timer for the relevant event: either close timeout
2517  * or mp_fail timeout. The close timeout takes precedence on the mp_fail one
2518  */
2519 void mptcp_reset_timeout(struct mptcp_sock *msk, unsigned long fail_tout)
2520 {
2521 	struct sock *sk = (struct sock *)msk;
2522 	unsigned long timeout, close_timeout;
2523 
2524 	if (!fail_tout && !sock_flag(sk, SOCK_DEAD))
2525 		return;
2526 
2527 	close_timeout = inet_csk(sk)->icsk_mtup.probe_timestamp - tcp_jiffies32 + jiffies + TCP_TIMEWAIT_LEN;
2528 
2529 	/* the close timeout takes precedence on the fail one, and here at least one of
2530 	 * them is active
2531 	 */
2532 	timeout = sock_flag(sk, SOCK_DEAD) ? close_timeout : fail_tout;
2533 
2534 	sk_reset_timer(sk, &sk->sk_timer, timeout);
2535 }
2536 
2537 static void mptcp_mp_fail_no_response(struct mptcp_sock *msk)
2538 {
2539 	struct sock *ssk = msk->first;
2540 	bool slow;
2541 
2542 	if (!ssk)
2543 		return;
2544 
2545 	pr_debug("MP_FAIL doesn't respond, reset the subflow");
2546 
2547 	slow = lock_sock_fast(ssk);
2548 	mptcp_subflow_reset(ssk);
2549 	WRITE_ONCE(mptcp_subflow_ctx(ssk)->fail_tout, 0);
2550 	unlock_sock_fast(ssk, slow);
2551 
2552 	mptcp_reset_timeout(msk, 0);
2553 }
2554 
2555 static void mptcp_worker(struct work_struct *work)
2556 {
2557 	struct mptcp_sock *msk = container_of(work, struct mptcp_sock, work);
2558 	struct sock *sk = &msk->sk.icsk_inet.sk;
2559 	unsigned long fail_tout;
2560 	int state;
2561 
2562 	lock_sock(sk);
2563 	state = sk->sk_state;
2564 	if (unlikely(state == TCP_CLOSE))
2565 		goto unlock;
2566 
2567 	mptcp_check_data_fin_ack(sk);
2568 
2569 	mptcp_check_fastclose(msk);
2570 
2571 	mptcp_pm_nl_work(msk);
2572 
2573 	if (test_and_clear_bit(MPTCP_WORK_EOF, &msk->flags))
2574 		mptcp_check_for_eof(msk);
2575 
2576 	__mptcp_check_send_data_fin(sk);
2577 	mptcp_check_data_fin(sk);
2578 
2579 	/* There is no point in keeping around an orphaned sk timedout or
2580 	 * closed, but we need the msk around to reply to incoming DATA_FIN,
2581 	 * even if it is orphaned and in FIN_WAIT2 state
2582 	 */
2583 	if (sock_flag(sk, SOCK_DEAD) &&
2584 	    (mptcp_check_close_timeout(sk) || sk->sk_state == TCP_CLOSE)) {
2585 		inet_sk_state_store(sk, TCP_CLOSE);
2586 		__mptcp_destroy_sock(sk);
2587 		goto unlock;
2588 	}
2589 
2590 	if (test_and_clear_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags))
2591 		__mptcp_close_subflow(msk);
2592 
2593 	if (test_and_clear_bit(MPTCP_WORK_RTX, &msk->flags))
2594 		__mptcp_retrans(sk);
2595 
2596 	fail_tout = msk->first ? READ_ONCE(mptcp_subflow_ctx(msk->first)->fail_tout) : 0;
2597 	if (fail_tout && time_after(jiffies, fail_tout))
2598 		mptcp_mp_fail_no_response(msk);
2599 
2600 unlock:
2601 	release_sock(sk);
2602 	sock_put(sk);
2603 }
2604 
2605 static int __mptcp_init_sock(struct sock *sk)
2606 {
2607 	struct mptcp_sock *msk = mptcp_sk(sk);
2608 
2609 	INIT_LIST_HEAD(&msk->conn_list);
2610 	INIT_LIST_HEAD(&msk->join_list);
2611 	INIT_LIST_HEAD(&msk->rtx_queue);
2612 	INIT_WORK(&msk->work, mptcp_worker);
2613 	__skb_queue_head_init(&msk->receive_queue);
2614 	msk->out_of_order_queue = RB_ROOT;
2615 	msk->first_pending = NULL;
2616 	msk->rmem_fwd_alloc = 0;
2617 	WRITE_ONCE(msk->rmem_released, 0);
2618 	msk->timer_ival = TCP_RTO_MIN;
2619 
2620 	msk->first = NULL;
2621 	inet_csk(sk)->icsk_sync_mss = mptcp_sync_mss;
2622 	WRITE_ONCE(msk->csum_enabled, mptcp_is_checksum_enabled(sock_net(sk)));
2623 	WRITE_ONCE(msk->allow_infinite_fallback, true);
2624 	msk->recovery = false;
2625 
2626 	mptcp_pm_data_init(msk);
2627 
2628 	/* re-use the csk retrans timer for MPTCP-level retrans */
2629 	timer_setup(&msk->sk.icsk_retransmit_timer, mptcp_retransmit_timer, 0);
2630 	timer_setup(&sk->sk_timer, mptcp_timeout_timer, 0);
2631 
2632 	return 0;
2633 }
2634 
2635 static void mptcp_ca_reset(struct sock *sk)
2636 {
2637 	struct inet_connection_sock *icsk = inet_csk(sk);
2638 
2639 	tcp_assign_congestion_control(sk);
2640 	strcpy(mptcp_sk(sk)->ca_name, icsk->icsk_ca_ops->name);
2641 
2642 	/* no need to keep a reference to the ops, the name will suffice */
2643 	tcp_cleanup_congestion_control(sk);
2644 	icsk->icsk_ca_ops = NULL;
2645 }
2646 
2647 static int mptcp_init_sock(struct sock *sk)
2648 {
2649 	struct net *net = sock_net(sk);
2650 	int ret;
2651 
2652 	ret = __mptcp_init_sock(sk);
2653 	if (ret)
2654 		return ret;
2655 
2656 	if (!mptcp_is_enabled(net))
2657 		return -ENOPROTOOPT;
2658 
2659 	if (unlikely(!net->mib.mptcp_statistics) && !mptcp_mib_alloc(net))
2660 		return -ENOMEM;
2661 
2662 	ret = __mptcp_socket_create(mptcp_sk(sk));
2663 	if (ret)
2664 		return ret;
2665 
2666 	/* fetch the ca name; do it outside __mptcp_init_sock(), so that clone will
2667 	 * propagate the correct value
2668 	 */
2669 	mptcp_ca_reset(sk);
2670 
2671 	sk_sockets_allocated_inc(sk);
2672 	sk->sk_rcvbuf = sock_net(sk)->ipv4.sysctl_tcp_rmem[1];
2673 	sk->sk_sndbuf = sock_net(sk)->ipv4.sysctl_tcp_wmem[1];
2674 
2675 	return 0;
2676 }
2677 
2678 static void __mptcp_clear_xmit(struct sock *sk)
2679 {
2680 	struct mptcp_sock *msk = mptcp_sk(sk);
2681 	struct mptcp_data_frag *dtmp, *dfrag;
2682 
2683 	WRITE_ONCE(msk->first_pending, NULL);
2684 	list_for_each_entry_safe(dfrag, dtmp, &msk->rtx_queue, list)
2685 		dfrag_clear(sk, dfrag);
2686 }
2687 
2688 static void mptcp_cancel_work(struct sock *sk)
2689 {
2690 	struct mptcp_sock *msk = mptcp_sk(sk);
2691 
2692 	if (cancel_work_sync(&msk->work))
2693 		__sock_put(sk);
2694 }
2695 
2696 void mptcp_subflow_shutdown(struct sock *sk, struct sock *ssk, int how)
2697 {
2698 	lock_sock(ssk);
2699 
2700 	switch (ssk->sk_state) {
2701 	case TCP_LISTEN:
2702 		if (!(how & RCV_SHUTDOWN))
2703 			break;
2704 		fallthrough;
2705 	case TCP_SYN_SENT:
2706 		tcp_disconnect(ssk, O_NONBLOCK);
2707 		break;
2708 	default:
2709 		if (__mptcp_check_fallback(mptcp_sk(sk))) {
2710 			pr_debug("Fallback");
2711 			ssk->sk_shutdown |= how;
2712 			tcp_shutdown(ssk, how);
2713 		} else {
2714 			pr_debug("Sending DATA_FIN on subflow %p", ssk);
2715 			tcp_send_ack(ssk);
2716 			if (!mptcp_timer_pending(sk))
2717 				mptcp_reset_timer(sk);
2718 		}
2719 		break;
2720 	}
2721 
2722 	release_sock(ssk);
2723 }
2724 
2725 static const unsigned char new_state[16] = {
2726 	/* current state:     new state:      action:	*/
2727 	[0 /* (Invalid) */] = TCP_CLOSE,
2728 	[TCP_ESTABLISHED]   = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2729 	[TCP_SYN_SENT]      = TCP_CLOSE,
2730 	[TCP_SYN_RECV]      = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2731 	[TCP_FIN_WAIT1]     = TCP_FIN_WAIT1,
2732 	[TCP_FIN_WAIT2]     = TCP_FIN_WAIT2,
2733 	[TCP_TIME_WAIT]     = TCP_CLOSE,	/* should not happen ! */
2734 	[TCP_CLOSE]         = TCP_CLOSE,
2735 	[TCP_CLOSE_WAIT]    = TCP_LAST_ACK  | TCP_ACTION_FIN,
2736 	[TCP_LAST_ACK]      = TCP_LAST_ACK,
2737 	[TCP_LISTEN]        = TCP_CLOSE,
2738 	[TCP_CLOSING]       = TCP_CLOSING,
2739 	[TCP_NEW_SYN_RECV]  = TCP_CLOSE,	/* should not happen ! */
2740 };
2741 
2742 static int mptcp_close_state(struct sock *sk)
2743 {
2744 	int next = (int)new_state[sk->sk_state];
2745 	int ns = next & TCP_STATE_MASK;
2746 
2747 	inet_sk_state_store(sk, ns);
2748 
2749 	return next & TCP_ACTION_FIN;
2750 }
2751 
2752 static void __mptcp_check_send_data_fin(struct sock *sk)
2753 {
2754 	struct mptcp_subflow_context *subflow;
2755 	struct mptcp_sock *msk = mptcp_sk(sk);
2756 
2757 	pr_debug("msk=%p snd_data_fin_enable=%d pending=%d snd_nxt=%llu write_seq=%llu",
2758 		 msk, msk->snd_data_fin_enable, !!mptcp_send_head(sk),
2759 		 msk->snd_nxt, msk->write_seq);
2760 
2761 	/* we still need to enqueue subflows or not really shutting down,
2762 	 * skip this
2763 	 */
2764 	if (!msk->snd_data_fin_enable || msk->snd_nxt + 1 != msk->write_seq ||
2765 	    mptcp_send_head(sk))
2766 		return;
2767 
2768 	WRITE_ONCE(msk->snd_nxt, msk->write_seq);
2769 
2770 	/* fallback socket will not get data_fin/ack, can move to the next
2771 	 * state now
2772 	 */
2773 	if (__mptcp_check_fallback(msk)) {
2774 		WRITE_ONCE(msk->snd_una, msk->write_seq);
2775 		if ((1 << sk->sk_state) & (TCPF_CLOSING | TCPF_LAST_ACK)) {
2776 			inet_sk_state_store(sk, TCP_CLOSE);
2777 			mptcp_close_wake_up(sk);
2778 		} else if (sk->sk_state == TCP_FIN_WAIT1) {
2779 			inet_sk_state_store(sk, TCP_FIN_WAIT2);
2780 		}
2781 	}
2782 
2783 	mptcp_for_each_subflow(msk, subflow) {
2784 		struct sock *tcp_sk = mptcp_subflow_tcp_sock(subflow);
2785 
2786 		mptcp_subflow_shutdown(sk, tcp_sk, SEND_SHUTDOWN);
2787 	}
2788 }
2789 
2790 static void __mptcp_wr_shutdown(struct sock *sk)
2791 {
2792 	struct mptcp_sock *msk = mptcp_sk(sk);
2793 
2794 	pr_debug("msk=%p snd_data_fin_enable=%d shutdown=%x state=%d pending=%d",
2795 		 msk, msk->snd_data_fin_enable, sk->sk_shutdown, sk->sk_state,
2796 		 !!mptcp_send_head(sk));
2797 
2798 	/* will be ignored by fallback sockets */
2799 	WRITE_ONCE(msk->write_seq, msk->write_seq + 1);
2800 	WRITE_ONCE(msk->snd_data_fin_enable, 1);
2801 
2802 	__mptcp_check_send_data_fin(sk);
2803 }
2804 
2805 static void __mptcp_destroy_sock(struct sock *sk)
2806 {
2807 	struct mptcp_subflow_context *subflow, *tmp;
2808 	struct mptcp_sock *msk = mptcp_sk(sk);
2809 	LIST_HEAD(conn_list);
2810 
2811 	pr_debug("msk=%p", msk);
2812 
2813 	might_sleep();
2814 
2815 	/* join list will be eventually flushed (with rst) at sock lock release time*/
2816 	list_splice_init(&msk->conn_list, &conn_list);
2817 
2818 	mptcp_stop_timer(sk);
2819 	sk_stop_timer(sk, &sk->sk_timer);
2820 	msk->pm.status = 0;
2821 
2822 	/* clears msk->subflow, allowing the following loop to close
2823 	 * even the initial subflow
2824 	 */
2825 	mptcp_dispose_initial_subflow(msk);
2826 	list_for_each_entry_safe(subflow, tmp, &conn_list, node) {
2827 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2828 		__mptcp_close_ssk(sk, ssk, subflow, 0);
2829 	}
2830 
2831 	sk->sk_prot->destroy(sk);
2832 
2833 	WARN_ON_ONCE(msk->rmem_fwd_alloc);
2834 	WARN_ON_ONCE(msk->rmem_released);
2835 	sk_stream_kill_queues(sk);
2836 	xfrm_sk_free_policy(sk);
2837 
2838 	sk_refcnt_debug_release(sk);
2839 	sock_put(sk);
2840 }
2841 
2842 static void mptcp_close(struct sock *sk, long timeout)
2843 {
2844 	struct mptcp_subflow_context *subflow;
2845 	struct mptcp_sock *msk = mptcp_sk(sk);
2846 	bool do_cancel_work = false;
2847 
2848 	lock_sock(sk);
2849 	sk->sk_shutdown = SHUTDOWN_MASK;
2850 
2851 	if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE)) {
2852 		inet_sk_state_store(sk, TCP_CLOSE);
2853 		goto cleanup;
2854 	}
2855 
2856 	if (mptcp_close_state(sk))
2857 		__mptcp_wr_shutdown(sk);
2858 
2859 	sk_stream_wait_close(sk, timeout);
2860 
2861 cleanup:
2862 	/* orphan all the subflows */
2863 	inet_csk(sk)->icsk_mtup.probe_timestamp = tcp_jiffies32;
2864 	mptcp_for_each_subflow(msk, subflow) {
2865 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2866 		bool slow = lock_sock_fast_nested(ssk);
2867 
2868 		/* since the close timeout takes precedence on the fail one,
2869 		 * cancel the latter
2870 		 */
2871 		if (ssk == msk->first)
2872 			subflow->fail_tout = 0;
2873 
2874 		sock_orphan(ssk);
2875 		unlock_sock_fast(ssk, slow);
2876 	}
2877 	sock_orphan(sk);
2878 
2879 	sock_hold(sk);
2880 	pr_debug("msk=%p state=%d", sk, sk->sk_state);
2881 	if (mptcp_sk(sk)->token)
2882 		mptcp_event(MPTCP_EVENT_CLOSED, msk, NULL, GFP_KERNEL);
2883 
2884 	if (sk->sk_state == TCP_CLOSE) {
2885 		__mptcp_destroy_sock(sk);
2886 		do_cancel_work = true;
2887 	} else {
2888 		mptcp_reset_timeout(msk, 0);
2889 	}
2890 	release_sock(sk);
2891 	if (do_cancel_work)
2892 		mptcp_cancel_work(sk);
2893 
2894 	sock_put(sk);
2895 }
2896 
2897 static void mptcp_copy_inaddrs(struct sock *msk, const struct sock *ssk)
2898 {
2899 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2900 	const struct ipv6_pinfo *ssk6 = inet6_sk(ssk);
2901 	struct ipv6_pinfo *msk6 = inet6_sk(msk);
2902 
2903 	msk->sk_v6_daddr = ssk->sk_v6_daddr;
2904 	msk->sk_v6_rcv_saddr = ssk->sk_v6_rcv_saddr;
2905 
2906 	if (msk6 && ssk6) {
2907 		msk6->saddr = ssk6->saddr;
2908 		msk6->flow_label = ssk6->flow_label;
2909 	}
2910 #endif
2911 
2912 	inet_sk(msk)->inet_num = inet_sk(ssk)->inet_num;
2913 	inet_sk(msk)->inet_dport = inet_sk(ssk)->inet_dport;
2914 	inet_sk(msk)->inet_sport = inet_sk(ssk)->inet_sport;
2915 	inet_sk(msk)->inet_daddr = inet_sk(ssk)->inet_daddr;
2916 	inet_sk(msk)->inet_saddr = inet_sk(ssk)->inet_saddr;
2917 	inet_sk(msk)->inet_rcv_saddr = inet_sk(ssk)->inet_rcv_saddr;
2918 }
2919 
2920 static int mptcp_disconnect(struct sock *sk, int flags)
2921 {
2922 	struct mptcp_subflow_context *subflow, *tmp;
2923 	struct mptcp_sock *msk = mptcp_sk(sk);
2924 
2925 	inet_sk_state_store(sk, TCP_CLOSE);
2926 
2927 	list_for_each_entry_safe(subflow, tmp, &msk->conn_list, node) {
2928 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2929 
2930 		__mptcp_close_ssk(sk, ssk, subflow, MPTCP_CF_FASTCLOSE);
2931 	}
2932 
2933 	mptcp_stop_timer(sk);
2934 	sk_stop_timer(sk, &sk->sk_timer);
2935 
2936 	if (mptcp_sk(sk)->token)
2937 		mptcp_event(MPTCP_EVENT_CLOSED, mptcp_sk(sk), NULL, GFP_KERNEL);
2938 
2939 	mptcp_destroy_common(msk);
2940 	msk->last_snd = NULL;
2941 	WRITE_ONCE(msk->flags, 0);
2942 	msk->cb_flags = 0;
2943 	msk->push_pending = 0;
2944 	msk->recovery = false;
2945 	msk->can_ack = false;
2946 	msk->fully_established = false;
2947 	msk->rcv_data_fin = false;
2948 	msk->snd_data_fin_enable = false;
2949 	msk->rcv_fastclose = false;
2950 	msk->use_64bit_ack = false;
2951 	WRITE_ONCE(msk->csum_enabled, mptcp_is_checksum_enabled(sock_net(sk)));
2952 	mptcp_pm_data_reset(msk);
2953 	mptcp_ca_reset(sk);
2954 
2955 	sk->sk_shutdown = 0;
2956 	sk_error_report(sk);
2957 	return 0;
2958 }
2959 
2960 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2961 static struct ipv6_pinfo *mptcp_inet6_sk(const struct sock *sk)
2962 {
2963 	unsigned int offset = sizeof(struct mptcp6_sock) - sizeof(struct ipv6_pinfo);
2964 
2965 	return (struct ipv6_pinfo *)(((u8 *)sk) + offset);
2966 }
2967 #endif
2968 
2969 struct sock *mptcp_sk_clone(const struct sock *sk,
2970 			    const struct mptcp_options_received *mp_opt,
2971 			    struct request_sock *req)
2972 {
2973 	struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
2974 	struct sock *nsk = sk_clone_lock(sk, GFP_ATOMIC);
2975 	struct mptcp_sock *msk;
2976 	u64 ack_seq;
2977 
2978 	if (!nsk)
2979 		return NULL;
2980 
2981 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2982 	if (nsk->sk_family == AF_INET6)
2983 		inet_sk(nsk)->pinet6 = mptcp_inet6_sk(nsk);
2984 #endif
2985 
2986 	__mptcp_init_sock(nsk);
2987 
2988 	msk = mptcp_sk(nsk);
2989 	msk->local_key = subflow_req->local_key;
2990 	msk->token = subflow_req->token;
2991 	msk->subflow = NULL;
2992 	WRITE_ONCE(msk->fully_established, false);
2993 	if (mp_opt->suboptions & OPTION_MPTCP_CSUMREQD)
2994 		WRITE_ONCE(msk->csum_enabled, true);
2995 
2996 	msk->write_seq = subflow_req->idsn + 1;
2997 	msk->snd_nxt = msk->write_seq;
2998 	msk->snd_una = msk->write_seq;
2999 	msk->wnd_end = msk->snd_nxt + req->rsk_rcv_wnd;
3000 	msk->setsockopt_seq = mptcp_sk(sk)->setsockopt_seq;
3001 
3002 	if (mp_opt->suboptions & OPTIONS_MPTCP_MPC) {
3003 		msk->can_ack = true;
3004 		msk->remote_key = mp_opt->sndr_key;
3005 		mptcp_crypto_key_sha(msk->remote_key, NULL, &ack_seq);
3006 		ack_seq++;
3007 		WRITE_ONCE(msk->ack_seq, ack_seq);
3008 		atomic64_set(&msk->rcv_wnd_sent, ack_seq);
3009 	}
3010 
3011 	sock_reset_flag(nsk, SOCK_RCU_FREE);
3012 	/* will be fully established after successful MPC subflow creation */
3013 	inet_sk_state_store(nsk, TCP_SYN_RECV);
3014 
3015 	security_inet_csk_clone(nsk, req);
3016 	bh_unlock_sock(nsk);
3017 
3018 	/* keep a single reference */
3019 	__sock_put(nsk);
3020 	return nsk;
3021 }
3022 
3023 void mptcp_rcv_space_init(struct mptcp_sock *msk, const struct sock *ssk)
3024 {
3025 	const struct tcp_sock *tp = tcp_sk(ssk);
3026 
3027 	msk->rcvq_space.copied = 0;
3028 	msk->rcvq_space.rtt_us = 0;
3029 
3030 	msk->rcvq_space.time = tp->tcp_mstamp;
3031 
3032 	/* initial rcv_space offering made to peer */
3033 	msk->rcvq_space.space = min_t(u32, tp->rcv_wnd,
3034 				      TCP_INIT_CWND * tp->advmss);
3035 	if (msk->rcvq_space.space == 0)
3036 		msk->rcvq_space.space = TCP_INIT_CWND * TCP_MSS_DEFAULT;
3037 
3038 	WRITE_ONCE(msk->wnd_end, msk->snd_nxt + tcp_sk(ssk)->snd_wnd);
3039 }
3040 
3041 static struct sock *mptcp_accept(struct sock *sk, int flags, int *err,
3042 				 bool kern)
3043 {
3044 	struct mptcp_sock *msk = mptcp_sk(sk);
3045 	struct socket *listener;
3046 	struct sock *newsk;
3047 
3048 	listener = __mptcp_nmpc_socket(msk);
3049 	if (WARN_ON_ONCE(!listener)) {
3050 		*err = -EINVAL;
3051 		return NULL;
3052 	}
3053 
3054 	pr_debug("msk=%p, listener=%p", msk, mptcp_subflow_ctx(listener->sk));
3055 	newsk = inet_csk_accept(listener->sk, flags, err, kern);
3056 	if (!newsk)
3057 		return NULL;
3058 
3059 	pr_debug("msk=%p, subflow is mptcp=%d", msk, sk_is_mptcp(newsk));
3060 	if (sk_is_mptcp(newsk)) {
3061 		struct mptcp_subflow_context *subflow;
3062 		struct sock *new_mptcp_sock;
3063 
3064 		subflow = mptcp_subflow_ctx(newsk);
3065 		new_mptcp_sock = subflow->conn;
3066 
3067 		/* is_mptcp should be false if subflow->conn is missing, see
3068 		 * subflow_syn_recv_sock()
3069 		 */
3070 		if (WARN_ON_ONCE(!new_mptcp_sock)) {
3071 			tcp_sk(newsk)->is_mptcp = 0;
3072 			goto out;
3073 		}
3074 
3075 		/* acquire the 2nd reference for the owning socket */
3076 		sock_hold(new_mptcp_sock);
3077 		newsk = new_mptcp_sock;
3078 		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPCAPABLEPASSIVEACK);
3079 	} else {
3080 		MPTCP_INC_STATS(sock_net(sk),
3081 				MPTCP_MIB_MPCAPABLEPASSIVEFALLBACK);
3082 	}
3083 
3084 out:
3085 	newsk->sk_kern_sock = kern;
3086 	return newsk;
3087 }
3088 
3089 void mptcp_destroy_common(struct mptcp_sock *msk)
3090 {
3091 	struct sock *sk = (struct sock *)msk;
3092 
3093 	__mptcp_clear_xmit(sk);
3094 
3095 	/* move to sk_receive_queue, sk_stream_kill_queues will purge it */
3096 	mptcp_data_lock(sk);
3097 	skb_queue_splice_tail_init(&msk->receive_queue, &sk->sk_receive_queue);
3098 	__skb_queue_purge(&sk->sk_receive_queue);
3099 	skb_rbtree_purge(&msk->out_of_order_queue);
3100 	mptcp_data_unlock(sk);
3101 
3102 	/* move all the rx fwd alloc into the sk_mem_reclaim_final in
3103 	 * inet_sock_destruct() will dispose it
3104 	 */
3105 	sk->sk_forward_alloc += msk->rmem_fwd_alloc;
3106 	msk->rmem_fwd_alloc = 0;
3107 	mptcp_token_destroy(msk);
3108 	mptcp_pm_free_anno_list(msk);
3109 	mptcp_free_local_addr_list(msk);
3110 }
3111 
3112 static void mptcp_destroy(struct sock *sk)
3113 {
3114 	struct mptcp_sock *msk = mptcp_sk(sk);
3115 
3116 	mptcp_destroy_common(msk);
3117 	sk_sockets_allocated_dec(sk);
3118 }
3119 
3120 void __mptcp_data_acked(struct sock *sk)
3121 {
3122 	if (!sock_owned_by_user(sk))
3123 		__mptcp_clean_una(sk);
3124 	else
3125 		__set_bit(MPTCP_CLEAN_UNA, &mptcp_sk(sk)->cb_flags);
3126 
3127 	if (mptcp_pending_data_fin_ack(sk))
3128 		mptcp_schedule_work(sk);
3129 }
3130 
3131 void __mptcp_check_push(struct sock *sk, struct sock *ssk)
3132 {
3133 	if (!mptcp_send_head(sk))
3134 		return;
3135 
3136 	if (!sock_owned_by_user(sk)) {
3137 		struct sock *xmit_ssk = mptcp_subflow_get_send(mptcp_sk(sk));
3138 
3139 		if (xmit_ssk == ssk)
3140 			__mptcp_subflow_push_pending(sk, ssk);
3141 		else if (xmit_ssk)
3142 			mptcp_subflow_delegate(mptcp_subflow_ctx(xmit_ssk), MPTCP_DELEGATE_SEND);
3143 	} else {
3144 		__set_bit(MPTCP_PUSH_PENDING, &mptcp_sk(sk)->cb_flags);
3145 	}
3146 }
3147 
3148 #define MPTCP_FLAGS_PROCESS_CTX_NEED (BIT(MPTCP_PUSH_PENDING) | \
3149 				      BIT(MPTCP_RETRANSMIT) | \
3150 				      BIT(MPTCP_FLUSH_JOIN_LIST))
3151 
3152 /* processes deferred events and flush wmem */
3153 static void mptcp_release_cb(struct sock *sk)
3154 	__must_hold(&sk->sk_lock.slock)
3155 {
3156 	struct mptcp_sock *msk = mptcp_sk(sk);
3157 
3158 	for (;;) {
3159 		unsigned long flags = (msk->cb_flags & MPTCP_FLAGS_PROCESS_CTX_NEED) |
3160 				      msk->push_pending;
3161 		if (!flags)
3162 			break;
3163 
3164 		/* the following actions acquire the subflow socket lock
3165 		 *
3166 		 * 1) can't be invoked in atomic scope
3167 		 * 2) must avoid ABBA deadlock with msk socket spinlock: the RX
3168 		 *    datapath acquires the msk socket spinlock while helding
3169 		 *    the subflow socket lock
3170 		 */
3171 		msk->push_pending = 0;
3172 		msk->cb_flags &= ~flags;
3173 		spin_unlock_bh(&sk->sk_lock.slock);
3174 		if (flags & BIT(MPTCP_FLUSH_JOIN_LIST))
3175 			__mptcp_flush_join_list(sk);
3176 		if (flags & BIT(MPTCP_PUSH_PENDING))
3177 			__mptcp_push_pending(sk, 0);
3178 		if (flags & BIT(MPTCP_RETRANSMIT))
3179 			__mptcp_retrans(sk);
3180 
3181 		cond_resched();
3182 		spin_lock_bh(&sk->sk_lock.slock);
3183 	}
3184 
3185 	if (__test_and_clear_bit(MPTCP_CLEAN_UNA, &msk->cb_flags))
3186 		__mptcp_clean_una_wakeup(sk);
3187 	if (unlikely(&msk->cb_flags)) {
3188 		/* be sure to set the current sk state before tacking actions
3189 		 * depending on sk_state, that is processing MPTCP_ERROR_REPORT
3190 		 */
3191 		if (__test_and_clear_bit(MPTCP_CONNECTED, &msk->cb_flags))
3192 			__mptcp_set_connected(sk);
3193 		if (__test_and_clear_bit(MPTCP_ERROR_REPORT, &msk->cb_flags))
3194 			__mptcp_error_report(sk);
3195 		if (__test_and_clear_bit(MPTCP_RESET_SCHEDULER, &msk->cb_flags))
3196 			msk->last_snd = NULL;
3197 	}
3198 
3199 	__mptcp_update_rmem(sk);
3200 }
3201 
3202 /* MP_JOIN client subflow must wait for 4th ack before sending any data:
3203  * TCP can't schedule delack timer before the subflow is fully established.
3204  * MPTCP uses the delack timer to do 3rd ack retransmissions
3205  */
3206 static void schedule_3rdack_retransmission(struct sock *ssk)
3207 {
3208 	struct inet_connection_sock *icsk = inet_csk(ssk);
3209 	struct tcp_sock *tp = tcp_sk(ssk);
3210 	unsigned long timeout;
3211 
3212 	if (mptcp_subflow_ctx(ssk)->fully_established)
3213 		return;
3214 
3215 	/* reschedule with a timeout above RTT, as we must look only for drop */
3216 	if (tp->srtt_us)
3217 		timeout = usecs_to_jiffies(tp->srtt_us >> (3 - 1));
3218 	else
3219 		timeout = TCP_TIMEOUT_INIT;
3220 	timeout += jiffies;
3221 
3222 	WARN_ON_ONCE(icsk->icsk_ack.pending & ICSK_ACK_TIMER);
3223 	icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
3224 	icsk->icsk_ack.timeout = timeout;
3225 	sk_reset_timer(ssk, &icsk->icsk_delack_timer, timeout);
3226 }
3227 
3228 void mptcp_subflow_process_delegated(struct sock *ssk)
3229 {
3230 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
3231 	struct sock *sk = subflow->conn;
3232 
3233 	if (test_bit(MPTCP_DELEGATE_SEND, &subflow->delegated_status)) {
3234 		mptcp_data_lock(sk);
3235 		if (!sock_owned_by_user(sk))
3236 			__mptcp_subflow_push_pending(sk, ssk);
3237 		else
3238 			__set_bit(MPTCP_PUSH_PENDING, &mptcp_sk(sk)->cb_flags);
3239 		mptcp_data_unlock(sk);
3240 		mptcp_subflow_delegated_done(subflow, MPTCP_DELEGATE_SEND);
3241 	}
3242 	if (test_bit(MPTCP_DELEGATE_ACK, &subflow->delegated_status)) {
3243 		schedule_3rdack_retransmission(ssk);
3244 		mptcp_subflow_delegated_done(subflow, MPTCP_DELEGATE_ACK);
3245 	}
3246 }
3247 
3248 static int mptcp_hash(struct sock *sk)
3249 {
3250 	/* should never be called,
3251 	 * we hash the TCP subflows not the master socket
3252 	 */
3253 	WARN_ON_ONCE(1);
3254 	return 0;
3255 }
3256 
3257 static void mptcp_unhash(struct sock *sk)
3258 {
3259 	/* called from sk_common_release(), but nothing to do here */
3260 }
3261 
3262 static int mptcp_get_port(struct sock *sk, unsigned short snum)
3263 {
3264 	struct mptcp_sock *msk = mptcp_sk(sk);
3265 	struct socket *ssock;
3266 
3267 	ssock = __mptcp_nmpc_socket(msk);
3268 	pr_debug("msk=%p, subflow=%p", msk, ssock);
3269 	if (WARN_ON_ONCE(!ssock))
3270 		return -EINVAL;
3271 
3272 	return inet_csk_get_port(ssock->sk, snum);
3273 }
3274 
3275 void mptcp_finish_connect(struct sock *ssk)
3276 {
3277 	struct mptcp_subflow_context *subflow;
3278 	struct mptcp_sock *msk;
3279 	struct sock *sk;
3280 	u64 ack_seq;
3281 
3282 	subflow = mptcp_subflow_ctx(ssk);
3283 	sk = subflow->conn;
3284 	msk = mptcp_sk(sk);
3285 
3286 	pr_debug("msk=%p, token=%u", sk, subflow->token);
3287 
3288 	mptcp_crypto_key_sha(subflow->remote_key, NULL, &ack_seq);
3289 	ack_seq++;
3290 	subflow->map_seq = ack_seq;
3291 	subflow->map_subflow_seq = 1;
3292 
3293 	/* the socket is not connected yet, no msk/subflow ops can access/race
3294 	 * accessing the field below
3295 	 */
3296 	WRITE_ONCE(msk->remote_key, subflow->remote_key);
3297 	WRITE_ONCE(msk->local_key, subflow->local_key);
3298 	WRITE_ONCE(msk->write_seq, subflow->idsn + 1);
3299 	WRITE_ONCE(msk->snd_nxt, msk->write_seq);
3300 	WRITE_ONCE(msk->ack_seq, ack_seq);
3301 	WRITE_ONCE(msk->can_ack, 1);
3302 	WRITE_ONCE(msk->snd_una, msk->write_seq);
3303 	atomic64_set(&msk->rcv_wnd_sent, ack_seq);
3304 
3305 	mptcp_pm_new_connection(msk, ssk, 0);
3306 
3307 	mptcp_rcv_space_init(msk, ssk);
3308 }
3309 
3310 void mptcp_sock_graft(struct sock *sk, struct socket *parent)
3311 {
3312 	write_lock_bh(&sk->sk_callback_lock);
3313 	rcu_assign_pointer(sk->sk_wq, &parent->wq);
3314 	sk_set_socket(sk, parent);
3315 	sk->sk_uid = SOCK_INODE(parent)->i_uid;
3316 	write_unlock_bh(&sk->sk_callback_lock);
3317 }
3318 
3319 bool mptcp_finish_join(struct sock *ssk)
3320 {
3321 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
3322 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
3323 	struct sock *parent = (void *)msk;
3324 	bool ret = true;
3325 
3326 	pr_debug("msk=%p, subflow=%p", msk, subflow);
3327 
3328 	/* mptcp socket already closing? */
3329 	if (!mptcp_is_fully_established(parent)) {
3330 		subflow->reset_reason = MPTCP_RST_EMPTCP;
3331 		return false;
3332 	}
3333 
3334 	if (!list_empty(&subflow->node))
3335 		goto out;
3336 
3337 	if (!mptcp_pm_allow_new_subflow(msk))
3338 		goto err_prohibited;
3339 
3340 	/* active connections are already on conn_list.
3341 	 * If we can't acquire msk socket lock here, let the release callback
3342 	 * handle it
3343 	 */
3344 	mptcp_data_lock(parent);
3345 	if (!sock_owned_by_user(parent)) {
3346 		ret = __mptcp_finish_join(msk, ssk);
3347 		if (ret) {
3348 			sock_hold(ssk);
3349 			list_add_tail(&subflow->node, &msk->conn_list);
3350 		}
3351 	} else {
3352 		sock_hold(ssk);
3353 		list_add_tail(&subflow->node, &msk->join_list);
3354 		__set_bit(MPTCP_FLUSH_JOIN_LIST, &msk->cb_flags);
3355 	}
3356 	mptcp_data_unlock(parent);
3357 
3358 	if (!ret) {
3359 err_prohibited:
3360 		subflow->reset_reason = MPTCP_RST_EPROHIBIT;
3361 		return false;
3362 	}
3363 
3364 	subflow->map_seq = READ_ONCE(msk->ack_seq);
3365 	WRITE_ONCE(msk->allow_infinite_fallback, false);
3366 
3367 out:
3368 	mptcp_event(MPTCP_EVENT_SUB_ESTABLISHED, msk, ssk, GFP_ATOMIC);
3369 	return true;
3370 }
3371 
3372 static void mptcp_shutdown(struct sock *sk, int how)
3373 {
3374 	pr_debug("sk=%p, how=%d", sk, how);
3375 
3376 	if ((how & SEND_SHUTDOWN) && mptcp_close_state(sk))
3377 		__mptcp_wr_shutdown(sk);
3378 }
3379 
3380 static int mptcp_forward_alloc_get(const struct sock *sk)
3381 {
3382 	return sk->sk_forward_alloc + mptcp_sk(sk)->rmem_fwd_alloc;
3383 }
3384 
3385 static int mptcp_ioctl_outq(const struct mptcp_sock *msk, u64 v)
3386 {
3387 	const struct sock *sk = (void *)msk;
3388 	u64 delta;
3389 
3390 	if (sk->sk_state == TCP_LISTEN)
3391 		return -EINVAL;
3392 
3393 	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
3394 		return 0;
3395 
3396 	delta = msk->write_seq - v;
3397 	if (__mptcp_check_fallback(msk) && msk->first) {
3398 		struct tcp_sock *tp = tcp_sk(msk->first);
3399 
3400 		/* the first subflow is disconnected after close - see
3401 		 * __mptcp_close_ssk(). tcp_disconnect() moves the write_seq
3402 		 * so ignore that status, too.
3403 		 */
3404 		if (!((1 << msk->first->sk_state) &
3405 		      (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE)))
3406 			delta += READ_ONCE(tp->write_seq) - tp->snd_una;
3407 	}
3408 	if (delta > INT_MAX)
3409 		delta = INT_MAX;
3410 
3411 	return (int)delta;
3412 }
3413 
3414 static int mptcp_ioctl(struct sock *sk, int cmd, unsigned long arg)
3415 {
3416 	struct mptcp_sock *msk = mptcp_sk(sk);
3417 	bool slow;
3418 	int answ;
3419 
3420 	switch (cmd) {
3421 	case SIOCINQ:
3422 		if (sk->sk_state == TCP_LISTEN)
3423 			return -EINVAL;
3424 
3425 		lock_sock(sk);
3426 		__mptcp_move_skbs(msk);
3427 		answ = mptcp_inq_hint(sk);
3428 		release_sock(sk);
3429 		break;
3430 	case SIOCOUTQ:
3431 		slow = lock_sock_fast(sk);
3432 		answ = mptcp_ioctl_outq(msk, READ_ONCE(msk->snd_una));
3433 		unlock_sock_fast(sk, slow);
3434 		break;
3435 	case SIOCOUTQNSD:
3436 		slow = lock_sock_fast(sk);
3437 		answ = mptcp_ioctl_outq(msk, msk->snd_nxt);
3438 		unlock_sock_fast(sk, slow);
3439 		break;
3440 	default:
3441 		return -ENOIOCTLCMD;
3442 	}
3443 
3444 	return put_user(answ, (int __user *)arg);
3445 }
3446 
3447 static struct proto mptcp_prot = {
3448 	.name		= "MPTCP",
3449 	.owner		= THIS_MODULE,
3450 	.init		= mptcp_init_sock,
3451 	.disconnect	= mptcp_disconnect,
3452 	.close		= mptcp_close,
3453 	.accept		= mptcp_accept,
3454 	.setsockopt	= mptcp_setsockopt,
3455 	.getsockopt	= mptcp_getsockopt,
3456 	.shutdown	= mptcp_shutdown,
3457 	.destroy	= mptcp_destroy,
3458 	.sendmsg	= mptcp_sendmsg,
3459 	.ioctl		= mptcp_ioctl,
3460 	.recvmsg	= mptcp_recvmsg,
3461 	.release_cb	= mptcp_release_cb,
3462 	.hash		= mptcp_hash,
3463 	.unhash		= mptcp_unhash,
3464 	.get_port	= mptcp_get_port,
3465 	.forward_alloc_get	= mptcp_forward_alloc_get,
3466 	.sockets_allocated	= &mptcp_sockets_allocated,
3467 	.memory_allocated	= &tcp_memory_allocated,
3468 	.memory_pressure	= &tcp_memory_pressure,
3469 	.sysctl_wmem_offset	= offsetof(struct net, ipv4.sysctl_tcp_wmem),
3470 	.sysctl_rmem_offset	= offsetof(struct net, ipv4.sysctl_tcp_rmem),
3471 	.sysctl_mem	= sysctl_tcp_mem,
3472 	.obj_size	= sizeof(struct mptcp_sock),
3473 	.slab_flags	= SLAB_TYPESAFE_BY_RCU,
3474 	.no_autobind	= true,
3475 };
3476 
3477 static int mptcp_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3478 {
3479 	struct mptcp_sock *msk = mptcp_sk(sock->sk);
3480 	struct socket *ssock;
3481 	int err;
3482 
3483 	lock_sock(sock->sk);
3484 	ssock = __mptcp_nmpc_socket(msk);
3485 	if (!ssock) {
3486 		err = -EINVAL;
3487 		goto unlock;
3488 	}
3489 
3490 	err = ssock->ops->bind(ssock, uaddr, addr_len);
3491 	if (!err)
3492 		mptcp_copy_inaddrs(sock->sk, ssock->sk);
3493 
3494 unlock:
3495 	release_sock(sock->sk);
3496 	return err;
3497 }
3498 
3499 static void mptcp_subflow_early_fallback(struct mptcp_sock *msk,
3500 					 struct mptcp_subflow_context *subflow)
3501 {
3502 	subflow->request_mptcp = 0;
3503 	__mptcp_do_fallback(msk);
3504 }
3505 
3506 static int mptcp_stream_connect(struct socket *sock, struct sockaddr *uaddr,
3507 				int addr_len, int flags)
3508 {
3509 	struct mptcp_sock *msk = mptcp_sk(sock->sk);
3510 	struct mptcp_subflow_context *subflow;
3511 	struct socket *ssock;
3512 	int err = -EINVAL;
3513 
3514 	lock_sock(sock->sk);
3515 	if (uaddr) {
3516 		if (addr_len < sizeof(uaddr->sa_family))
3517 			goto unlock;
3518 
3519 		if (uaddr->sa_family == AF_UNSPEC) {
3520 			err = mptcp_disconnect(sock->sk, flags);
3521 			sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
3522 			goto unlock;
3523 		}
3524 	}
3525 
3526 	if (sock->state != SS_UNCONNECTED && msk->subflow) {
3527 		/* pending connection or invalid state, let existing subflow
3528 		 * cope with that
3529 		 */
3530 		ssock = msk->subflow;
3531 		goto do_connect;
3532 	}
3533 
3534 	ssock = __mptcp_nmpc_socket(msk);
3535 	if (!ssock)
3536 		goto unlock;
3537 
3538 	mptcp_token_destroy(msk);
3539 	inet_sk_state_store(sock->sk, TCP_SYN_SENT);
3540 	subflow = mptcp_subflow_ctx(ssock->sk);
3541 #ifdef CONFIG_TCP_MD5SIG
3542 	/* no MPTCP if MD5SIG is enabled on this socket or we may run out of
3543 	 * TCP option space.
3544 	 */
3545 	if (rcu_access_pointer(tcp_sk(ssock->sk)->md5sig_info))
3546 		mptcp_subflow_early_fallback(msk, subflow);
3547 #endif
3548 	if (subflow->request_mptcp && mptcp_token_new_connect(ssock->sk)) {
3549 		MPTCP_INC_STATS(sock_net(ssock->sk), MPTCP_MIB_TOKENFALLBACKINIT);
3550 		mptcp_subflow_early_fallback(msk, subflow);
3551 	}
3552 	if (likely(!__mptcp_check_fallback(msk)))
3553 		MPTCP_INC_STATS(sock_net(sock->sk), MPTCP_MIB_MPCAPABLEACTIVE);
3554 
3555 do_connect:
3556 	err = ssock->ops->connect(ssock, uaddr, addr_len, flags);
3557 	sock->state = ssock->state;
3558 
3559 	/* on successful connect, the msk state will be moved to established by
3560 	 * subflow_finish_connect()
3561 	 */
3562 	if (!err || err == -EINPROGRESS)
3563 		mptcp_copy_inaddrs(sock->sk, ssock->sk);
3564 	else
3565 		inet_sk_state_store(sock->sk, inet_sk_state_load(ssock->sk));
3566 
3567 unlock:
3568 	release_sock(sock->sk);
3569 	return err;
3570 }
3571 
3572 static int mptcp_listen(struct socket *sock, int backlog)
3573 {
3574 	struct mptcp_sock *msk = mptcp_sk(sock->sk);
3575 	struct socket *ssock;
3576 	int err;
3577 
3578 	pr_debug("msk=%p", msk);
3579 
3580 	lock_sock(sock->sk);
3581 	ssock = __mptcp_nmpc_socket(msk);
3582 	if (!ssock) {
3583 		err = -EINVAL;
3584 		goto unlock;
3585 	}
3586 
3587 	mptcp_token_destroy(msk);
3588 	inet_sk_state_store(sock->sk, TCP_LISTEN);
3589 	sock_set_flag(sock->sk, SOCK_RCU_FREE);
3590 
3591 	err = ssock->ops->listen(ssock, backlog);
3592 	inet_sk_state_store(sock->sk, inet_sk_state_load(ssock->sk));
3593 	if (!err)
3594 		mptcp_copy_inaddrs(sock->sk, ssock->sk);
3595 
3596 unlock:
3597 	release_sock(sock->sk);
3598 	return err;
3599 }
3600 
3601 static int mptcp_stream_accept(struct socket *sock, struct socket *newsock,
3602 			       int flags, bool kern)
3603 {
3604 	struct mptcp_sock *msk = mptcp_sk(sock->sk);
3605 	struct socket *ssock;
3606 	int err;
3607 
3608 	pr_debug("msk=%p", msk);
3609 
3610 	ssock = __mptcp_nmpc_socket(msk);
3611 	if (!ssock)
3612 		return -EINVAL;
3613 
3614 	err = ssock->ops->accept(sock, newsock, flags, kern);
3615 	if (err == 0 && !mptcp_is_tcpsk(newsock->sk)) {
3616 		struct mptcp_sock *msk = mptcp_sk(newsock->sk);
3617 		struct mptcp_subflow_context *subflow;
3618 		struct sock *newsk = newsock->sk;
3619 
3620 		lock_sock(newsk);
3621 
3622 		/* PM/worker can now acquire the first subflow socket
3623 		 * lock without racing with listener queue cleanup,
3624 		 * we can notify it, if needed.
3625 		 *
3626 		 * Even if remote has reset the initial subflow by now
3627 		 * the refcnt is still at least one.
3628 		 */
3629 		subflow = mptcp_subflow_ctx(msk->first);
3630 		list_add(&subflow->node, &msk->conn_list);
3631 		sock_hold(msk->first);
3632 		if (mptcp_is_fully_established(newsk))
3633 			mptcp_pm_fully_established(msk, msk->first, GFP_KERNEL);
3634 
3635 		mptcp_copy_inaddrs(newsk, msk->first);
3636 		mptcp_rcv_space_init(msk, msk->first);
3637 		mptcp_propagate_sndbuf(newsk, msk->first);
3638 
3639 		/* set ssk->sk_socket of accept()ed flows to mptcp socket.
3640 		 * This is needed so NOSPACE flag can be set from tcp stack.
3641 		 */
3642 		mptcp_for_each_subflow(msk, subflow) {
3643 			struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
3644 
3645 			if (!ssk->sk_socket)
3646 				mptcp_sock_graft(ssk, newsock);
3647 		}
3648 		release_sock(newsk);
3649 	}
3650 
3651 	return err;
3652 }
3653 
3654 static __poll_t mptcp_check_readable(struct mptcp_sock *msk)
3655 {
3656 	/* Concurrent splices from sk_receive_queue into receive_queue will
3657 	 * always show at least one non-empty queue when checked in this order.
3658 	 */
3659 	if (skb_queue_empty_lockless(&((struct sock *)msk)->sk_receive_queue) &&
3660 	    skb_queue_empty_lockless(&msk->receive_queue))
3661 		return 0;
3662 
3663 	return EPOLLIN | EPOLLRDNORM;
3664 }
3665 
3666 static __poll_t mptcp_check_writeable(struct mptcp_sock *msk)
3667 {
3668 	struct sock *sk = (struct sock *)msk;
3669 
3670 	if (unlikely(sk->sk_shutdown & SEND_SHUTDOWN))
3671 		return EPOLLOUT | EPOLLWRNORM;
3672 
3673 	if (sk_stream_is_writeable(sk))
3674 		return EPOLLOUT | EPOLLWRNORM;
3675 
3676 	mptcp_set_nospace(sk);
3677 	smp_mb__after_atomic(); /* msk->flags is changed by write_space cb */
3678 	if (sk_stream_is_writeable(sk))
3679 		return EPOLLOUT | EPOLLWRNORM;
3680 
3681 	return 0;
3682 }
3683 
3684 static __poll_t mptcp_poll(struct file *file, struct socket *sock,
3685 			   struct poll_table_struct *wait)
3686 {
3687 	struct sock *sk = sock->sk;
3688 	struct mptcp_sock *msk;
3689 	__poll_t mask = 0;
3690 	int state;
3691 
3692 	msk = mptcp_sk(sk);
3693 	sock_poll_wait(file, sock, wait);
3694 
3695 	state = inet_sk_state_load(sk);
3696 	pr_debug("msk=%p state=%d flags=%lx", msk, state, msk->flags);
3697 	if (state == TCP_LISTEN) {
3698 		if (WARN_ON_ONCE(!msk->subflow || !msk->subflow->sk))
3699 			return 0;
3700 
3701 		return inet_csk_listen_poll(msk->subflow->sk);
3702 	}
3703 
3704 	if (state != TCP_SYN_SENT && state != TCP_SYN_RECV) {
3705 		mask |= mptcp_check_readable(msk);
3706 		mask |= mptcp_check_writeable(msk);
3707 	}
3708 	if (sk->sk_shutdown == SHUTDOWN_MASK || state == TCP_CLOSE)
3709 		mask |= EPOLLHUP;
3710 	if (sk->sk_shutdown & RCV_SHUTDOWN)
3711 		mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
3712 
3713 	/* This barrier is coupled with smp_wmb() in tcp_reset() */
3714 	smp_rmb();
3715 	if (sk->sk_err)
3716 		mask |= EPOLLERR;
3717 
3718 	return mask;
3719 }
3720 
3721 static const struct proto_ops mptcp_stream_ops = {
3722 	.family		   = PF_INET,
3723 	.owner		   = THIS_MODULE,
3724 	.release	   = inet_release,
3725 	.bind		   = mptcp_bind,
3726 	.connect	   = mptcp_stream_connect,
3727 	.socketpair	   = sock_no_socketpair,
3728 	.accept		   = mptcp_stream_accept,
3729 	.getname	   = inet_getname,
3730 	.poll		   = mptcp_poll,
3731 	.ioctl		   = inet_ioctl,
3732 	.gettstamp	   = sock_gettstamp,
3733 	.listen		   = mptcp_listen,
3734 	.shutdown	   = inet_shutdown,
3735 	.setsockopt	   = sock_common_setsockopt,
3736 	.getsockopt	   = sock_common_getsockopt,
3737 	.sendmsg	   = inet_sendmsg,
3738 	.recvmsg	   = inet_recvmsg,
3739 	.mmap		   = sock_no_mmap,
3740 	.sendpage	   = inet_sendpage,
3741 };
3742 
3743 static struct inet_protosw mptcp_protosw = {
3744 	.type		= SOCK_STREAM,
3745 	.protocol	= IPPROTO_MPTCP,
3746 	.prot		= &mptcp_prot,
3747 	.ops		= &mptcp_stream_ops,
3748 	.flags		= INET_PROTOSW_ICSK,
3749 };
3750 
3751 static int mptcp_napi_poll(struct napi_struct *napi, int budget)
3752 {
3753 	struct mptcp_delegated_action *delegated;
3754 	struct mptcp_subflow_context *subflow;
3755 	int work_done = 0;
3756 
3757 	delegated = container_of(napi, struct mptcp_delegated_action, napi);
3758 	while ((subflow = mptcp_subflow_delegated_next(delegated)) != NULL) {
3759 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
3760 
3761 		bh_lock_sock_nested(ssk);
3762 		if (!sock_owned_by_user(ssk) &&
3763 		    mptcp_subflow_has_delegated_action(subflow))
3764 			mptcp_subflow_process_delegated(ssk);
3765 		/* ... elsewhere tcp_release_cb_override already processed
3766 		 * the action or will do at next release_sock().
3767 		 * In both case must dequeue the subflow here - on the same
3768 		 * CPU that scheduled it.
3769 		 */
3770 		bh_unlock_sock(ssk);
3771 		sock_put(ssk);
3772 
3773 		if (++work_done == budget)
3774 			return budget;
3775 	}
3776 
3777 	/* always provide a 0 'work_done' argument, so that napi_complete_done
3778 	 * will not try accessing the NULL napi->dev ptr
3779 	 */
3780 	napi_complete_done(napi, 0);
3781 	return work_done;
3782 }
3783 
3784 void __init mptcp_proto_init(void)
3785 {
3786 	struct mptcp_delegated_action *delegated;
3787 	int cpu;
3788 
3789 	mptcp_prot.h.hashinfo = tcp_prot.h.hashinfo;
3790 
3791 	if (percpu_counter_init(&mptcp_sockets_allocated, 0, GFP_KERNEL))
3792 		panic("Failed to allocate MPTCP pcpu counter\n");
3793 
3794 	init_dummy_netdev(&mptcp_napi_dev);
3795 	for_each_possible_cpu(cpu) {
3796 		delegated = per_cpu_ptr(&mptcp_delegated_actions, cpu);
3797 		INIT_LIST_HEAD(&delegated->head);
3798 		netif_napi_add_tx(&mptcp_napi_dev, &delegated->napi,
3799 				  mptcp_napi_poll);
3800 		napi_enable(&delegated->napi);
3801 	}
3802 
3803 	mptcp_subflow_init();
3804 	mptcp_pm_init();
3805 	mptcp_token_init();
3806 
3807 	if (proto_register(&mptcp_prot, 1) != 0)
3808 		panic("Failed to register MPTCP proto.\n");
3809 
3810 	inet_register_protosw(&mptcp_protosw);
3811 
3812 	BUILD_BUG_ON(sizeof(struct mptcp_skb_cb) > sizeof_field(struct sk_buff, cb));
3813 }
3814 
3815 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
3816 static const struct proto_ops mptcp_v6_stream_ops = {
3817 	.family		   = PF_INET6,
3818 	.owner		   = THIS_MODULE,
3819 	.release	   = inet6_release,
3820 	.bind		   = mptcp_bind,
3821 	.connect	   = mptcp_stream_connect,
3822 	.socketpair	   = sock_no_socketpair,
3823 	.accept		   = mptcp_stream_accept,
3824 	.getname	   = inet6_getname,
3825 	.poll		   = mptcp_poll,
3826 	.ioctl		   = inet6_ioctl,
3827 	.gettstamp	   = sock_gettstamp,
3828 	.listen		   = mptcp_listen,
3829 	.shutdown	   = inet_shutdown,
3830 	.setsockopt	   = sock_common_setsockopt,
3831 	.getsockopt	   = sock_common_getsockopt,
3832 	.sendmsg	   = inet6_sendmsg,
3833 	.recvmsg	   = inet6_recvmsg,
3834 	.mmap		   = sock_no_mmap,
3835 	.sendpage	   = inet_sendpage,
3836 #ifdef CONFIG_COMPAT
3837 	.compat_ioctl	   = inet6_compat_ioctl,
3838 #endif
3839 };
3840 
3841 static struct proto mptcp_v6_prot;
3842 
3843 static void mptcp_v6_destroy(struct sock *sk)
3844 {
3845 	mptcp_destroy(sk);
3846 	inet6_destroy_sock(sk);
3847 }
3848 
3849 static struct inet_protosw mptcp_v6_protosw = {
3850 	.type		= SOCK_STREAM,
3851 	.protocol	= IPPROTO_MPTCP,
3852 	.prot		= &mptcp_v6_prot,
3853 	.ops		= &mptcp_v6_stream_ops,
3854 	.flags		= INET_PROTOSW_ICSK,
3855 };
3856 
3857 int __init mptcp_proto_v6_init(void)
3858 {
3859 	int err;
3860 
3861 	mptcp_v6_prot = mptcp_prot;
3862 	strcpy(mptcp_v6_prot.name, "MPTCPv6");
3863 	mptcp_v6_prot.slab = NULL;
3864 	mptcp_v6_prot.destroy = mptcp_v6_destroy;
3865 	mptcp_v6_prot.obj_size = sizeof(struct mptcp6_sock);
3866 
3867 	err = proto_register(&mptcp_v6_prot, 1);
3868 	if (err)
3869 		return err;
3870 
3871 	err = inet6_register_protosw(&mptcp_v6_protosw);
3872 	if (err)
3873 		proto_unregister(&mptcp_v6_prot);
3874 
3875 	return err;
3876 }
3877 #endif
3878