1	// SPDX-License-Identifier: GPL-2.0
2	/* Multipath TCP
3	*
4	* Copyright (c) 2017 - 2019, Intel Corporation.
5	*/
6
7	#define pr_fmt(fmt) "MPTCP: " fmt
8
9	#include <linux/kernel.h>
10	#include <linux/module.h>
11	#include <linux/netdevice.h>
12	#include <linux/sched/signal.h>
13	#include <linux/atomic.h>
14	#include <net/sock.h>
15	#include <net/inet_common.h>
16	#include <net/inet_hashtables.h>
17	#include <net/protocol.h>
18	#include <net/tcp.h>
19	#include <net/tcp_states.h>
20	#if IS_ENABLED(CONFIG_MPTCP_IPV6)
21	#include <net/transp_v6.h>
22	#endif
23	#include <net/mptcp.h>
24	#include <net/xfrm.h>
25	#include <asm/ioctls.h>
26	#include "protocol.h"
27	#include "mib.h"
28
29	#define CREATE_TRACE_POINTS
30	#include <trace/events/mptcp.h>
31
32	#if IS_ENABLED(CONFIG_MPTCP_IPV6)
33	struct mptcp6_sock {
34	struct mptcp_sock msk;
35	struct ipv6_pinfo np;
36	};
37	#endif
38
39	enum {
40	MPTCP_CMSG_TS = BIT(0),
41	MPTCP_CMSG_INQ = BIT(1),
42	};
43
44	static struct percpu_counter mptcp_sockets_allocated ____cacheline_aligned_in_smp;
45
46	static void __mptcp_destroy_sock(struct sock *sk);
47	static void mptcp_check_send_data_fin(struct sock *sk);
48
49	DEFINE_PER_CPU(struct mptcp_delegated_action, mptcp_delegated_actions);
50	static struct net_device mptcp_napi_dev;
51
52	/* Returns end sequence number of the receiver's advertised window */
53	static u64 mptcp_wnd_end(const struct mptcp_sock *msk)
54	{
55	return READ_ONCE(msk->wnd_end);
56	}
57
58	static bool mptcp_is_tcpsk(struct sock *sk)
59	{
60	struct socket *sock = sk->sk_socket;
61
62	if (unlikely(sk->sk_prot == &tcp_prot)) {
63	/* we are being invoked after mptcp_accept() has
64	* accepted a non-mp-capable flow: sk is a tcp_sk,
65	* not an mptcp one.
66	*
67	* Hand the socket over to tcp so all further socket ops
68	* bypass mptcp.
69	*/
70	WRITE_ONCE(sock->ops, &inet_stream_ops);
71	return true;
72	#if IS_ENABLED(CONFIG_MPTCP_IPV6)
73	} else if (unlikely(sk->sk_prot == &tcpv6_prot)) {
74	WRITE_ONCE(sock->ops, &inet6_stream_ops);
75	return true;
76	#endif
77	}
78
79	return false;
80	}
81
82	static int __mptcp_socket_create(struct mptcp_sock *msk)
83	{
84	struct mptcp_subflow_context *subflow;
85	struct sock *sk = (struct sock *)msk;
86	struct socket *ssock;
87	int err;
88
89	err = mptcp_subflow_create_socket(sk, family: sk->sk_family, new_sock: &ssock);
90	if (err)
91	return err;
92
93	msk->scaling_ratio = tcp_sk(ssock->sk)->scaling_ratio;
94	WRITE_ONCE(msk->first, ssock->sk);
95	subflow = mptcp_subflow_ctx(sk: ssock->sk);
96	list_add(new: &subflow->node, head: &msk->conn_list);
97	sock_hold(sk: ssock->sk);
98	subflow->request_mptcp = 1;
99	subflow->subflow_id = msk->subflow_id++;
100
101	/* This is the first subflow, always with id 0 */
102	subflow->local_id_valid = 1;
103	mptcp_sock_graft(sk: msk->first, parent: sk->sk_socket);
104	iput(SOCK_INODE(socket: ssock));
105
106	return 0;
107	}
108
109	/* If the MPC handshake is not started, returns the first subflow,
110	* eventually allocating it.
111	*/
112	struct sock *__mptcp_nmpc_sk(struct mptcp_sock *msk)
113	{
114	struct sock *sk = (struct sock *)msk;
115	int ret;
116
117	if (!((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)))
118	return ERR_PTR(error: -EINVAL);
119
120	if (!msk->first) {
121	ret = __mptcp_socket_create(msk);
122	if (ret)
123	return ERR_PTR(error: ret);
124	}
125
126	return msk->first;
127	}
128
129	static void mptcp_drop(struct sock *sk, struct sk_buff *skb)
130	{
131	sk_drops_add(sk, skb);
132	__kfree_skb(skb);
133	}
134
135	static void mptcp_rmem_fwd_alloc_add(struct sock *sk, int size)
136	{
137	WRITE_ONCE(mptcp_sk(sk)->rmem_fwd_alloc,
138	mptcp_sk(sk)->rmem_fwd_alloc + size);
139	}
140
141	static void mptcp_rmem_charge(struct sock *sk, int size)
142	{
143	mptcp_rmem_fwd_alloc_add(sk, size: -size);
144	}
145
146	static bool mptcp_try_coalesce(struct sock *sk, struct sk_buff *to,
147	struct sk_buff *from)
148	{
149	bool fragstolen;
150	int delta;
151
152	if (MPTCP_SKB_CB(from)->offset ||
153	!skb_try_coalesce(to, from, fragstolen: &fragstolen, delta_truesize: &delta))
154	return false;
155
156	pr_debug("colesced seq %llx into %llx new len %d new end seq %llx",
157	MPTCP_SKB_CB(from)->map_seq, MPTCP_SKB_CB(to)->map_seq,
158	to->len, MPTCP_SKB_CB(from)->end_seq);
159	MPTCP_SKB_CB(to)->end_seq = MPTCP_SKB_CB(from)->end_seq;
160
161	/* note the fwd memory can reach a negative value after accounting
162	* for the delta, but the later skb free will restore a non
163	* negative one
164	*/
165	atomic_add(i: delta, v: &sk->sk_rmem_alloc);
166	mptcp_rmem_charge(sk, size: delta);
167	kfree_skb_partial(skb: from, head_stolen: fragstolen);
168
169	return true;
170	}
171
172	static bool mptcp_ooo_try_coalesce(struct mptcp_sock *msk, struct sk_buff *to,
173	struct sk_buff *from)
174	{
175	if (MPTCP_SKB_CB(from)->map_seq != MPTCP_SKB_CB(to)->end_seq)
176	return false;
177
178	return mptcp_try_coalesce(sk: (struct sock *)msk, to, from);
179	}
180
181	static void __mptcp_rmem_reclaim(struct sock *sk, int amount)
182	{
183	amount >>= PAGE_SHIFT;
184	mptcp_rmem_charge(sk, size: amount << PAGE_SHIFT);
185	__sk_mem_reduce_allocated(sk, amount);
186	}
187
188	static void mptcp_rmem_uncharge(struct sock *sk, int size)
189	{
190	struct mptcp_sock *msk = mptcp_sk(sk);
191	int reclaimable;
192
193	mptcp_rmem_fwd_alloc_add(sk, size);
194	reclaimable = msk->rmem_fwd_alloc - sk_unused_reserved_mem(sk);
195
196	/* see sk_mem_uncharge() for the rationale behind the following schema */
197	if (unlikely(reclaimable >= PAGE_SIZE))
198	__mptcp_rmem_reclaim(sk, amount: reclaimable);
199	}
200
201	static void mptcp_rfree(struct sk_buff *skb)
202	{
203	unsigned int len = skb->truesize;
204	struct sock *sk = skb->sk;
205
206	atomic_sub(i: len, v: &sk->sk_rmem_alloc);
207	mptcp_rmem_uncharge(sk, size: len);
208	}
209
210	void mptcp_set_owner_r(struct sk_buff *skb, struct sock *sk)
211	{
212	skb_orphan(skb);
213	skb->sk = sk;
214	skb->destructor = mptcp_rfree;
215	atomic_add(i: skb->truesize, v: &sk->sk_rmem_alloc);
216	mptcp_rmem_charge(sk, size: skb->truesize);
217	}
218
219	/* "inspired" by tcp_data_queue_ofo(), main differences:
220	* - use mptcp seqs
221	* - don't cope with sacks
222	*/
223	static void mptcp_data_queue_ofo(struct mptcp_sock *msk, struct sk_buff *skb)
224	{
225	struct sock *sk = (struct sock *)msk;
226	struct rb_node **p, *parent;
227	u64 seq, end_seq, max_seq;
228	struct sk_buff *skb1;
229
230	seq = MPTCP_SKB_CB(skb)->map_seq;
231	end_seq = MPTCP_SKB_CB(skb)->end_seq;
232	max_seq = atomic64_read(v: &msk->rcv_wnd_sent);
233
234	pr_debug("msk=%p seq=%llx limit=%llx empty=%d", msk, seq, max_seq,
235	RB_EMPTY_ROOT(&msk->out_of_order_queue));
236	if (after64(end_seq, max_seq)) {
237	/* out of window */
238	mptcp_drop(sk, skb);
239	pr_debug("oow by %lld, rcv_wnd_sent %llu\n",
240	(unsigned long long)end_seq - (unsigned long)max_seq,
241	(unsigned long long)atomic64_read(&msk->rcv_wnd_sent));
242	MPTCP_INC_STATS(net: sock_net(sk), field: MPTCP_MIB_NODSSWINDOW);
243	return;
244	}
245
246	p = &msk->out_of_order_queue.rb_node;
247	MPTCP_INC_STATS(net: sock_net(sk), field: MPTCP_MIB_OFOQUEUE);
248	if (RB_EMPTY_ROOT(&msk->out_of_order_queue)) {
249	rb_link_node(node: &skb->rbnode, NULL, rb_link: p);
250	rb_insert_color(&skb->rbnode, &msk->out_of_order_queue);
251	msk->ooo_last_skb = skb;
252	goto end;
253	}
254
255	/* with 2 subflows, adding at end of ooo queue is quite likely
256	* Use of ooo_last_skb avoids the O(Log(N)) rbtree lookup.
257	*/
258	if (mptcp_ooo_try_coalesce(msk, to: msk->ooo_last_skb, from: skb)) {
259	MPTCP_INC_STATS(net: sock_net(sk), field: MPTCP_MIB_OFOMERGE);
260	MPTCP_INC_STATS(net: sock_net(sk), field: MPTCP_MIB_OFOQUEUETAIL);
261	return;
262	}
263
264	/* Can avoid an rbtree lookup if we are adding skb after ooo_last_skb */
265	if (!before64(seq1: seq, MPTCP_SKB_CB(msk->ooo_last_skb)->end_seq)) {
266	MPTCP_INC_STATS(net: sock_net(sk), field: MPTCP_MIB_OFOQUEUETAIL);
267	parent = &msk->ooo_last_skb->rbnode;
268	p = &parent->rb_right;
269	goto insert;
270	}
271
272	/* Find place to insert this segment. Handle overlaps on the way. */
273	parent = NULL;
274	while (*p) {
275	parent = *p;
276	skb1 = rb_to_skb(parent);
277	if (before64(seq1: seq, MPTCP_SKB_CB(skb1)->map_seq)) {
278	p = &parent->rb_left;
279	continue;
280	}
281	if (before64(seq1: seq, MPTCP_SKB_CB(skb1)->end_seq)) {
282	if (!after64(end_seq, MPTCP_SKB_CB(skb1)->end_seq)) {
283	/* All the bits are present. Drop. */
284	mptcp_drop(sk, skb);
285	MPTCP_INC_STATS(net: sock_net(sk), field: MPTCP_MIB_DUPDATA);
286	return;
287	}
288	if (after64(seq, MPTCP_SKB_CB(skb1)->map_seq)) {
289	/* partial overlap:
290	* | skb |
291	* | skb1 |
292	* continue traversing
293	*/
294	} else {
295	/* skb's seq == skb1's seq and skb covers skb1.
296	* Replace skb1 with skb.
297	*/
298	rb_replace_node(victim: &skb1->rbnode, new: &skb->rbnode,
299	root: &msk->out_of_order_queue);
300	mptcp_drop(sk, skb: skb1);
301	MPTCP_INC_STATS(net: sock_net(sk), field: MPTCP_MIB_DUPDATA);
302	goto merge_right;
303	}
304	} else if (mptcp_ooo_try_coalesce(msk, to: skb1, from: skb)) {
305	MPTCP_INC_STATS(net: sock_net(sk), field: MPTCP_MIB_OFOMERGE);
306	return;
307	}
308	p = &parent->rb_right;
309	}
310
311	insert:
312	/* Insert segment into RB tree. */
313	rb_link_node(node: &skb->rbnode, parent, rb_link: p);
314	rb_insert_color(&skb->rbnode, &msk->out_of_order_queue);
315
316	merge_right:
317	/* Remove other segments covered by skb. */
318	while ((skb1 = skb_rb_next(skb)) != NULL) {
319	if (before64(seq1: end_seq, MPTCP_SKB_CB(skb1)->end_seq))
320	break;
321	rb_erase(&skb1->rbnode, &msk->out_of_order_queue);
322	mptcp_drop(sk, skb: skb1);
323	MPTCP_INC_STATS(net: sock_net(sk), field: MPTCP_MIB_DUPDATA);
324	}
325	/* If there is no skb after us, we are the last_skb ! */
326	if (!skb1)
327	msk->ooo_last_skb = skb;
328
329	end:
330	skb_condense(skb);
331	mptcp_set_owner_r(skb, sk);
332	}
333
334	static bool mptcp_rmem_schedule(struct sock *sk, struct sock *ssk, int size)
335	{
336	struct mptcp_sock *msk = mptcp_sk(sk);
337	int amt, amount;
338
339	if (size <= msk->rmem_fwd_alloc)
340	return true;
341
342	size -= msk->rmem_fwd_alloc;
343	amt = sk_mem_pages(amt: size);
344	amount = amt << PAGE_SHIFT;
345	if (!__sk_mem_raise_allocated(sk, size, amt, SK_MEM_RECV))
346	return false;
347
348	mptcp_rmem_fwd_alloc_add(sk, size: amount);
349	return true;
350	}
351
352	static bool __mptcp_move_skb(struct mptcp_sock *msk, struct sock *ssk,
353	struct sk_buff *skb, unsigned int offset,
354	size_t copy_len)
355	{
356	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk: ssk);
357	struct sock *sk = (struct sock *)msk;
358	struct sk_buff *tail;
359	bool has_rxtstamp;
360
361	__skb_unlink(skb, list: &ssk->sk_receive_queue);
362
363	skb_ext_reset(skb);
364	skb_orphan(skb);
365
366	/* try to fetch required memory from subflow */
367	if (!mptcp_rmem_schedule(sk, ssk, size: skb->truesize))
368	goto drop;
369
370	has_rxtstamp = TCP_SKB_CB(skb)->has_rxtstamp;
371
372	/* the skb map_seq accounts for the skb offset:
373	* mptcp_subflow_get_mapped_dsn() is based on the current tp->copied_seq
374	* value
375	*/
376	MPTCP_SKB_CB(skb)->map_seq = mptcp_subflow_get_mapped_dsn(subflow);
377	MPTCP_SKB_CB(skb)->end_seq = MPTCP_SKB_CB(skb)->map_seq + copy_len;
378	MPTCP_SKB_CB(skb)->offset = offset;
379	MPTCP_SKB_CB(skb)->has_rxtstamp = has_rxtstamp;
380
381	if (MPTCP_SKB_CB(skb)->map_seq == msk->ack_seq) {
382	/* in sequence */
383	msk->bytes_received += copy_len;
384	WRITE_ONCE(msk->ack_seq, msk->ack_seq + copy_len);
385	tail = skb_peek_tail(list_: &sk->sk_receive_queue);
386	if (tail && mptcp_try_coalesce(sk, to: tail, from: skb))
387	return true;
388
389	mptcp_set_owner_r(skb, sk);
390	__skb_queue_tail(list: &sk->sk_receive_queue, newsk: skb);
391	return true;
392	} else if (after64(MPTCP_SKB_CB(skb)->map_seq, msk->ack_seq)) {
393	mptcp_data_queue_ofo(msk, skb);
394	return false;
395	}
396
397	/* old data, keep it simple and drop the whole pkt, sender
398	* will retransmit as needed, if needed.
399	*/
400	MPTCP_INC_STATS(net: sock_net(sk), field: MPTCP_MIB_DUPDATA);
401	drop:
402	mptcp_drop(sk, skb);
403	return false;
404	}
405
406	static void mptcp_stop_rtx_timer(struct sock *sk)
407	{
408	struct inet_connection_sock *icsk = inet_csk(sk);
409
410	sk_stop_timer(sk, timer: &icsk->icsk_retransmit_timer);
411	mptcp_sk(sk)->timer_ival = 0;
412	}
413
414	static void mptcp_close_wake_up(struct sock *sk)
415	{
416	if (sock_flag(sk, flag: SOCK_DEAD))
417	return;
418
419	sk->sk_state_change(sk);
420	if (sk->sk_shutdown == SHUTDOWN_MASK ||
421	sk->sk_state == TCP_CLOSE)
422	sk_wake_async(sk, how: SOCK_WAKE_WAITD, POLL_HUP);
423	else
424	sk_wake_async(sk, how: SOCK_WAKE_WAITD, POLL_IN);
425	}
426
427	static bool mptcp_pending_data_fin_ack(struct sock *sk)
428	{
429	struct mptcp_sock *msk = mptcp_sk(sk);
430
431	return ((1 << sk->sk_state) &
432	(TCPF_FIN_WAIT1 | TCPF_CLOSING | TCPF_LAST_ACK)) &&
433	msk->write_seq == READ_ONCE(msk->snd_una);
434	}
435
436	static void mptcp_check_data_fin_ack(struct sock *sk)
437	{
438	struct mptcp_sock *msk = mptcp_sk(sk);
439
440	/* Look for an acknowledged DATA_FIN */
441	if (mptcp_pending_data_fin_ack(sk)) {
442	WRITE_ONCE(msk->snd_data_fin_enable, 0);
443
444	switch (sk->sk_state) {
445	case TCP_FIN_WAIT1:
446	inet_sk_state_store(sk, newstate: TCP_FIN_WAIT2);
447	break;
448	case TCP_CLOSING:
449	case TCP_LAST_ACK:
450	inet_sk_state_store(sk, newstate: TCP_CLOSE);
451	break;
452	}
453
454	mptcp_close_wake_up(sk);
455	}
456	}
457
458	static bool mptcp_pending_data_fin(struct sock *sk, u64 *seq)
459	{
460	struct mptcp_sock *msk = mptcp_sk(sk);
461
462	if (READ_ONCE(msk->rcv_data_fin) &&
463	((1 << sk->sk_state) &
464	(TCPF_ESTABLISHED | TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2))) {
465	u64 rcv_data_fin_seq = READ_ONCE(msk->rcv_data_fin_seq);
466
467	if (msk->ack_seq == rcv_data_fin_seq) {
468	if (seq)
469	*seq = rcv_data_fin_seq;
470
471	return true;
472	}
473	}
474
475	return false;
476	}
477
478	static void mptcp_set_datafin_timeout(struct sock *sk)
479	{
480	struct inet_connection_sock *icsk = inet_csk(sk);
481	u32 retransmits;
482
483	retransmits = min_t(u32, icsk->icsk_retransmits,
484	ilog2(TCP_RTO_MAX / TCP_RTO_MIN));
485
486	mptcp_sk(sk)->timer_ival = TCP_RTO_MIN << retransmits;
487	}
488
489	static void __mptcp_set_timeout(struct sock *sk, long tout)
490	{
491	mptcp_sk(sk)->timer_ival = tout > 0 ? tout : TCP_RTO_MIN;
492	}
493
494	static long mptcp_timeout_from_subflow(const struct mptcp_subflow_context *subflow)
495	{
496	const struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
497
498	return inet_csk(sk: ssk)->icsk_pending && !subflow->stale_count ?
499	inet_csk(sk: ssk)->icsk_timeout - jiffies : 0;
500	}
501
502	static void mptcp_set_timeout(struct sock *sk)
503	{
504	struct mptcp_subflow_context *subflow;
505	long tout = 0;
506
507	mptcp_for_each_subflow(mptcp_sk(sk), subflow)
508	tout = max(tout, mptcp_timeout_from_subflow(subflow));
509	__mptcp_set_timeout(sk, tout);
510	}
511
512	static inline bool tcp_can_send_ack(const struct sock *ssk)
513	{
514	return !((1 << inet_sk_state_load(sk: ssk)) &
515	(TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_TIME_WAIT | TCPF_CLOSE | TCPF_LISTEN));
516	}
517
518	void __mptcp_subflow_send_ack(struct sock *ssk)
519	{
520	if (tcp_can_send_ack(ssk))
521	tcp_send_ack(sk: ssk);
522	}
523
524	static void mptcp_subflow_send_ack(struct sock *ssk)
525	{
526	bool slow;
527
528	slow = lock_sock_fast(sk: ssk);
529	__mptcp_subflow_send_ack(ssk);
530	unlock_sock_fast(sk: ssk, slow);
531	}
532
533	static void mptcp_send_ack(struct mptcp_sock *msk)
534	{
535	struct mptcp_subflow_context *subflow;
536
537	mptcp_for_each_subflow(msk, subflow)
538	mptcp_subflow_send_ack(ssk: mptcp_subflow_tcp_sock(subflow));
539	}
540
541	static void mptcp_subflow_cleanup_rbuf(struct sock *ssk)
542	{
543	bool slow;
544
545	slow = lock_sock_fast(sk: ssk);
546	if (tcp_can_send_ack(ssk))
547	tcp_cleanup_rbuf(sk: ssk, copied: 1);
548	unlock_sock_fast(sk: ssk, slow);
549	}
550
551	static bool mptcp_subflow_could_cleanup(const struct sock *ssk, bool rx_empty)
552	{
553	const struct inet_connection_sock *icsk = inet_csk(sk: ssk);
554	u8 ack_pending = READ_ONCE(icsk->icsk_ack.pending);
555	const struct tcp_sock *tp = tcp_sk(ssk);
556
557	return (ack_pending & ICSK_ACK_SCHED) &&
558	((READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->rcv_wup) >
559	READ_ONCE(icsk->icsk_ack.rcv_mss)) ||
560	(rx_empty && ack_pending &
561	(ICSK_ACK_PUSHED2 | ICSK_ACK_PUSHED)));
562	}
563
564	static void mptcp_cleanup_rbuf(struct mptcp_sock *msk)
565	{
566	int old_space = READ_ONCE(msk->old_wspace);
567	struct mptcp_subflow_context *subflow;
568	struct sock *sk = (struct sock *)msk;
569	int space = __mptcp_space(sk);
570	bool cleanup, rx_empty;
571
572	cleanup = (space > 0) && (space >= (old_space << 1));
573	rx_empty = !__mptcp_rmem(sk);
574
575	mptcp_for_each_subflow(msk, subflow) {
576	struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
577
578	if (cleanup || mptcp_subflow_could_cleanup(ssk, rx_empty))
579	mptcp_subflow_cleanup_rbuf(ssk);
580	}
581	}
582
583	static bool mptcp_check_data_fin(struct sock *sk)
584	{
585	struct mptcp_sock *msk = mptcp_sk(sk);
586	u64 rcv_data_fin_seq;
587	bool ret = false;
588
589	/* Need to ack a DATA_FIN received from a peer while this side
590	* of the connection is in ESTABLISHED, FIN_WAIT1, or FIN_WAIT2.
591	* msk->rcv_data_fin was set when parsing the incoming options
592	* at the subflow level and the msk lock was not held, so this
593	* is the first opportunity to act on the DATA_FIN and change
594	* the msk state.
595	*
596	* If we are caught up to the sequence number of the incoming
597	* DATA_FIN, send the DATA_ACK now and do state transition. If
598	* not caught up, do nothing and let the recv code send DATA_ACK
599	* when catching up.
600	*/
601
602	if (mptcp_pending_data_fin(sk, seq: &rcv_data_fin_seq)) {
603	WRITE_ONCE(msk->ack_seq, msk->ack_seq + 1);
604	WRITE_ONCE(msk->rcv_data_fin, 0);
605
606	WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | RCV_SHUTDOWN);
607	smp_mb__before_atomic(); /* SHUTDOWN must be visible first */
608
609	switch (sk->sk_state) {
610	case TCP_ESTABLISHED:
611	inet_sk_state_store(sk, newstate: TCP_CLOSE_WAIT);
612	break;
613	case TCP_FIN_WAIT1:
614	inet_sk_state_store(sk, newstate: TCP_CLOSING);
615	break;
616	case TCP_FIN_WAIT2:
617	inet_sk_state_store(sk, newstate: TCP_CLOSE);
618	break;
619	default:
620	/* Other states not expected */
621	WARN_ON_ONCE(1);
622	break;
623	}
624
625	ret = true;
626	if (!__mptcp_check_fallback(msk))
627	mptcp_send_ack(msk);
628	mptcp_close_wake_up(sk);
629	}
630	return ret;
631	}
632
633	static bool __mptcp_move_skbs_from_subflow(struct mptcp_sock *msk,
634	struct sock *ssk,
635	unsigned int *bytes)
636	{
637	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk: ssk);
638	struct sock *sk = (struct sock *)msk;
639	unsigned int moved = 0;
640	bool more_data_avail;
641	struct tcp_sock *tp;
642	bool done = false;
643	int sk_rbuf;
644
645	sk_rbuf = READ_ONCE(sk->sk_rcvbuf);
646
647	if (!(sk->sk_userlocks & SOCK_RCVBUF_LOCK)) {
648	int ssk_rbuf = READ_ONCE(ssk->sk_rcvbuf);
649
650	if (unlikely(ssk_rbuf > sk_rbuf)) {
651	WRITE_ONCE(sk->sk_rcvbuf, ssk_rbuf);
652	sk_rbuf = ssk_rbuf;
653	}
654	}
655
656	pr_debug("msk=%p ssk=%p", msk, ssk);
657	tp = tcp_sk(ssk);
658	do {
659	u32 map_remaining, offset;
660	u32 seq = tp->copied_seq;
661	struct sk_buff *skb;
662	bool fin;
663
664	/* try to move as much data as available */
665	map_remaining = subflow->map_data_len -
666	mptcp_subflow_get_map_offset(subflow);
667
668	skb = skb_peek(list_: &ssk->sk_receive_queue);
669	if (!skb) {
670	/* With racing move_skbs_to_msk() and __mptcp_move_skbs(),
671	* a different CPU can have already processed the pending
672	* data, stop here or we can enter an infinite loop
673	*/
674	if (!moved)
675	done = true;
676	break;
677	}
678
679	if (__mptcp_check_fallback(msk)) {
680	/* Under fallback skbs have no MPTCP extension and TCP could
681	* collapse them between the dummy map creation and the
682	* current dequeue. Be sure to adjust the map size.
683	*/
684	map_remaining = skb->len;
685	subflow->map_data_len = skb->len;
686	}
687
688	offset = seq - TCP_SKB_CB(skb)->seq;
689	fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
690	if (fin) {
691	done = true;
692	seq++;
693	}
694
695	if (offset < skb->len) {
696	size_t len = skb->len - offset;
697
698	if (tp->urg_data)
699	done = true;
700
701	if (__mptcp_move_skb(msk, ssk, skb, offset, copy_len: len))
702	moved += len;
703	seq += len;
704
705	if (WARN_ON_ONCE(map_remaining < len))
706	break;
707	} else {
708	WARN_ON_ONCE(!fin);
709	sk_eat_skb(sk: ssk, skb);
710	done = true;
711	}
712
713	WRITE_ONCE(tp->copied_seq, seq);
714	more_data_avail = mptcp_subflow_data_available(sk: ssk);
715
716	if (atomic_read(v: &sk->sk_rmem_alloc) > sk_rbuf) {
717	done = true;
718	break;
719	}
720	} while (more_data_avail);
721
722	*bytes += moved;
723	return done;
724	}
725
726	static bool __mptcp_ofo_queue(struct mptcp_sock *msk)
727	{
728	struct sock *sk = (struct sock *)msk;
729	struct sk_buff *skb, *tail;
730	bool moved = false;
731	struct rb_node *p;
732	u64 end_seq;
733
734	p = rb_first(&msk->out_of_order_queue);
735	pr_debug("msk=%p empty=%d", msk, RB_EMPTY_ROOT(&msk->out_of_order_queue));
736	while (p) {
737	skb = rb_to_skb(p);
738	if (after64(MPTCP_SKB_CB(skb)->map_seq, msk->ack_seq))
739	break;
740
741	p = rb_next(p);
742	rb_erase(&skb->rbnode, &msk->out_of_order_queue);
743
744	if (unlikely(!after64(MPTCP_SKB_CB(skb)->end_seq,
745	msk->ack_seq))) {
746	mptcp_drop(sk, skb);
747	MPTCP_INC_STATS(net: sock_net(sk), field: MPTCP_MIB_DUPDATA);
748	continue;
749	}
750
751	end_seq = MPTCP_SKB_CB(skb)->end_seq;
752	tail = skb_peek_tail(list_: &sk->sk_receive_queue);
753	if (!tail || !mptcp_ooo_try_coalesce(msk, to: tail, from: skb)) {
754	int delta = msk->ack_seq - MPTCP_SKB_CB(skb)->map_seq;
755
756	/* skip overlapping data, if any */
757	pr_debug("uncoalesced seq=%llx ack seq=%llx delta=%d",
758	MPTCP_SKB_CB(skb)->map_seq, msk->ack_seq,
759	delta);
760	MPTCP_SKB_CB(skb)->offset += delta;
761	MPTCP_SKB_CB(skb)->map_seq += delta;
762	__skb_queue_tail(list: &sk->sk_receive_queue, newsk: skb);
763	}
764	msk->bytes_received += end_seq - msk->ack_seq;
765	msk->ack_seq = end_seq;
766	moved = true;
767	}
768	return moved;
769	}
770
771	static bool __mptcp_subflow_error_report(struct sock *sk, struct sock *ssk)
772	{
773	int err = sock_error(sk: ssk);
774	int ssk_state;
775
776	if (!err)
777	return false;
778
779	/* only propagate errors on fallen-back sockets or
780	* on MPC connect
781	*/
782	if (sk->sk_state != TCP_SYN_SENT && !__mptcp_check_fallback(mptcp_sk(sk)))
783	return false;
784
785	/* We need to propagate only transition to CLOSE state.
786	* Orphaned socket will see such state change via
787	* subflow_sched_work_if_closed() and that path will properly
788	* destroy the msk as needed.
789	*/
790	ssk_state = inet_sk_state_load(sk: ssk);
791	if (ssk_state == TCP_CLOSE && !sock_flag(sk, flag: SOCK_DEAD))
792	inet_sk_state_store(sk, newstate: ssk_state);
793	WRITE_ONCE(sk->sk_err, -err);
794
795	/* This barrier is coupled with smp_rmb() in mptcp_poll() */
796	smp_wmb();
797	sk_error_report(sk);
798	return true;
799	}
800
801	void __mptcp_error_report(struct sock *sk)
802	{
803	struct mptcp_subflow_context *subflow;
804	struct mptcp_sock *msk = mptcp_sk(sk);
805
806	mptcp_for_each_subflow(msk, subflow)
807	if (__mptcp_subflow_error_report(sk, ssk: mptcp_subflow_tcp_sock(subflow)))
808	break;
809	}
810
811	/* In most cases we will be able to lock the mptcp socket. If its already
812	* owned, we need to defer to the work queue to avoid ABBA deadlock.
813	*/
814	static bool move_skbs_to_msk(struct mptcp_sock *msk, struct sock *ssk)
815	{
816	struct sock *sk = (struct sock *)msk;
817	unsigned int moved = 0;
818
819	__mptcp_move_skbs_from_subflow(msk, ssk, bytes: &moved);
820	__mptcp_ofo_queue(msk);
821	if (unlikely(ssk->sk_err)) {
822	if (!sock_owned_by_user(sk))
823	__mptcp_error_report(sk);
824	else
825	__set_bit(MPTCP_ERROR_REPORT, &msk->cb_flags);
826	}
827
828	/* If the moves have caught up with the DATA_FIN sequence number
829	* it's time to ack the DATA_FIN and change socket state, but
830	* this is not a good place to change state. Let the workqueue
831	* do it.
832	*/
833	if (mptcp_pending_data_fin(sk, NULL))
834	mptcp_schedule_work(sk);
835	return moved > 0;
836	}
837
838	void mptcp_data_ready(struct sock *sk, struct sock *ssk)
839	{
840	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk: ssk);
841	struct mptcp_sock *msk = mptcp_sk(sk);
842	int sk_rbuf, ssk_rbuf;
843
844	/* The peer can send data while we are shutting down this
845	* subflow at msk destruction time, but we must avoid enqueuing
846	* more data to the msk receive queue
847	*/
848	if (unlikely(subflow->disposable))
849	return;
850
851	ssk_rbuf = READ_ONCE(ssk->sk_rcvbuf);
852	sk_rbuf = READ_ONCE(sk->sk_rcvbuf);
853	if (unlikely(ssk_rbuf > sk_rbuf))
854	sk_rbuf = ssk_rbuf;
855
856	/* over limit? can't append more skbs to msk, Also, no need to wake-up*/
857	if (__mptcp_rmem(sk) > sk_rbuf) {
858	MPTCP_INC_STATS(net: sock_net(sk), field: MPTCP_MIB_RCVPRUNED);
859	return;
860	}
861
862	/* Wake-up the reader only for in-sequence data */
863	mptcp_data_lock(sk);
864	if (move_skbs_to_msk(msk, ssk) && mptcp_epollin_ready(sk))
865	sk->sk_data_ready(sk);
866	mptcp_data_unlock(sk);
867	}
868
869	static void mptcp_subflow_joined(struct mptcp_sock *msk, struct sock *ssk)
870	{
871	mptcp_subflow_ctx(sk: ssk)->map_seq = READ_ONCE(msk->ack_seq);
872	WRITE_ONCE(msk->allow_infinite_fallback, false);
873	mptcp_event(type: MPTCP_EVENT_SUB_ESTABLISHED, msk, ssk, GFP_ATOMIC);
874	}
875
876	static bool __mptcp_finish_join(struct mptcp_sock *msk, struct sock *ssk)
877	{
878	struct sock *sk = (struct sock *)msk;
879
880	if (sk->sk_state != TCP_ESTABLISHED)
881	return false;
882
883	/* attach to msk socket only after we are sure we will deal with it
884	* at close time
885	*/
886	if (sk->sk_socket && !ssk->sk_socket)
887	mptcp_sock_graft(sk: ssk, parent: sk->sk_socket);
888
889	mptcp_subflow_ctx(sk: ssk)->subflow_id = msk->subflow_id++;
890	mptcp_sockopt_sync_locked(msk, ssk);
891	mptcp_subflow_joined(msk, ssk);
892	mptcp_stop_tout_timer(sk);
893	__mptcp_propagate_sndbuf(sk, ssk);
894	return true;
895	}
896
897	static void __mptcp_flush_join_list(struct sock *sk, struct list_head *join_list)
898	{
899	struct mptcp_subflow_context *tmp, *subflow;
900	struct mptcp_sock *msk = mptcp_sk(sk);
901
902	list_for_each_entry_safe(subflow, tmp, join_list, node) {
903	struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
904	bool slow = lock_sock_fast(sk: ssk);
905
906	list_move_tail(list: &subflow->node, head: &msk->conn_list);
907	if (!__mptcp_finish_join(msk, ssk))
908	mptcp_subflow_reset(ssk);
909	unlock_sock_fast(sk: ssk, slow);
910	}
911	}
912
913	static bool mptcp_rtx_timer_pending(struct sock *sk)
914	{
915	return timer_pending(timer: &inet_csk(sk)->icsk_retransmit_timer);
916	}
917
918	static void mptcp_reset_rtx_timer(struct sock *sk)
919	{
920	struct inet_connection_sock *icsk = inet_csk(sk);
921	unsigned long tout;
922
923	/* prevent rescheduling on close */
924	if (unlikely(inet_sk_state_load(sk) == TCP_CLOSE))
925	return;
926
927	tout = mptcp_sk(sk)->timer_ival;
928	sk_reset_timer(sk, timer: &icsk->icsk_retransmit_timer, expires: jiffies + tout);
929	}
930
931	bool mptcp_schedule_work(struct sock *sk)
932	{
933	if (inet_sk_state_load(sk) != TCP_CLOSE &&
934	schedule_work(work: &mptcp_sk(sk)->work)) {
935	/* each subflow already holds a reference to the sk, and the
936	* workqueue is invoked by a subflow, so sk can't go away here.
937	*/
938	sock_hold(sk);
939	return true;
940	}
941	return false;
942	}
943
944	static struct sock *mptcp_subflow_recv_lookup(const struct mptcp_sock *msk)
945	{
946	struct mptcp_subflow_context *subflow;
947
948	msk_owned_by_me(msk);
949
950	mptcp_for_each_subflow(msk, subflow) {
951	if (READ_ONCE(subflow->data_avail))
952	return mptcp_subflow_tcp_sock(subflow);
953	}
954
955	return NULL;
956	}
957
958	static bool mptcp_skb_can_collapse_to(u64 write_seq,
959	const struct sk_buff *skb,
960	const struct mptcp_ext *mpext)
961	{
962	if (!tcp_skb_can_collapse_to(skb))
963	return false;
964
965	/* can collapse only if MPTCP level sequence is in order and this
966	* mapping has not been xmitted yet
967	*/
968	return mpext && mpext->data_seq + mpext->data_len == write_seq &&
969	!mpext->frozen;
970	}
971
972	/* we can append data to the given data frag if:
973	* - there is space available in the backing page_frag
974	* - the data frag tail matches the current page_frag free offset
975	* - the data frag end sequence number matches the current write seq
976	*/
977	static bool mptcp_frag_can_collapse_to(const struct mptcp_sock *msk,
978	const struct page_frag *pfrag,
979	const struct mptcp_data_frag *df)
980	{
981	return df && pfrag->page == df->page &&
982	pfrag->size - pfrag->offset > 0 &&
983	pfrag->offset == (df->offset + df->data_len) &&
984	df->data_seq + df->data_len == msk->write_seq;
985	}
986
987	static void dfrag_uncharge(struct sock *sk, int len)
988	{
989	sk_mem_uncharge(sk, size: len);
990	sk_wmem_queued_add(sk, val: -len);
991	}
992
993	static void dfrag_clear(struct sock *sk, struct mptcp_data_frag *dfrag)
994	{
995	int len = dfrag->data_len + dfrag->overhead;
996
997	list_del(entry: &dfrag->list);
998	dfrag_uncharge(sk, len);
999	put_page(page: dfrag->page);
1000	}
1001
1002	static void __mptcp_clean_una(struct sock *sk)
1003	{
1004	struct mptcp_sock *msk = mptcp_sk(sk);
1005	struct mptcp_data_frag *dtmp, *dfrag;
1006	u64 snd_una;
1007
1008	snd_una = msk->snd_una;
1009	list_for_each_entry_safe(dfrag, dtmp, &msk->rtx_queue, list) {
1010	if (after64(dfrag->data_seq + dfrag->data_len, snd_una))
1011	break;
1012
1013	if (unlikely(dfrag == msk->first_pending)) {
1014	/* in recovery mode can see ack after the current snd head */
1015	if (WARN_ON_ONCE(!msk->recovery))
1016	break;
1017
1018	WRITE_ONCE(msk->first_pending, mptcp_send_next(sk));
1019	}
1020
1021	dfrag_clear(sk, dfrag);
1022	}
1023
1024	dfrag = mptcp_rtx_head(sk);
1025	if (dfrag && after64(snd_una, dfrag->data_seq)) {
1026	u64 delta = snd_una - dfrag->data_seq;
1027
1028	/* prevent wrap around in recovery mode */
1029	if (unlikely(delta > dfrag->already_sent)) {
1030	if (WARN_ON_ONCE(!msk->recovery))
1031	goto out;
1032	if (WARN_ON_ONCE(delta > dfrag->data_len))
1033	goto out;
1034	dfrag->already_sent += delta - dfrag->already_sent;
1035	}
1036
1037	dfrag->data_seq += delta;
1038	dfrag->offset += delta;
1039	dfrag->data_len -= delta;
1040	dfrag->already_sent -= delta;
1041
1042	dfrag_uncharge(sk, len: delta);
1043	}
1044
1045	/* all retransmitted data acked, recovery completed */
1046	if (unlikely(msk->recovery) && after64(msk->snd_una, msk->recovery_snd_nxt))
1047	msk->recovery = false;
1048
1049	out:
1050	if (snd_una == READ_ONCE(msk->snd_nxt) &&
1051	snd_una == READ_ONCE(msk->write_seq)) {
1052	if (mptcp_rtx_timer_pending(sk) && !mptcp_data_fin_enabled(msk))
1053	mptcp_stop_rtx_timer(sk);
1054	} else {
1055	mptcp_reset_rtx_timer(sk);
1056	}
1057	}
1058
1059	static void __mptcp_clean_una_wakeup(struct sock *sk)
1060	{
1061	lockdep_assert_held_once(&sk->sk_lock.slock);
1062
1063	__mptcp_clean_una(sk);
1064	mptcp_write_space(sk);
1065	}
1066
1067	static void mptcp_clean_una_wakeup(struct sock *sk)
1068	{
1069	mptcp_data_lock(sk);
1070	__mptcp_clean_una_wakeup(sk);
1071	mptcp_data_unlock(sk);
1072	}
1073
1074	static void mptcp_enter_memory_pressure(struct sock *sk)
1075	{
1076	struct mptcp_subflow_context *subflow;
1077	struct mptcp_sock *msk = mptcp_sk(sk);
1078	bool first = true;
1079
1080	mptcp_for_each_subflow(msk, subflow) {
1081	struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1082
1083	if (first)
1084	tcp_enter_memory_pressure(sk: ssk);
1085	sk_stream_moderate_sndbuf(sk: ssk);
1086
1087	first = false;
1088	}
1089	__mptcp_sync_sndbuf(sk);
1090	}
1091
1092	/* ensure we get enough memory for the frag hdr, beyond some minimal amount of
1093	* data
1094	*/
1095	static bool mptcp_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1096	{
1097	if (likely(skb_page_frag_refill(32U + sizeof(struct mptcp_data_frag),
1098	pfrag, sk->sk_allocation)))
1099	return true;
1100
1101	mptcp_enter_memory_pressure(sk);
1102	return false;
1103	}
1104
1105	static struct mptcp_data_frag *
1106	mptcp_carve_data_frag(const struct mptcp_sock *msk, struct page_frag *pfrag,
1107	int orig_offset)
1108	{
1109	int offset = ALIGN(orig_offset, sizeof(long));
1110	struct mptcp_data_frag *dfrag;
1111
1112	dfrag = (struct mptcp_data_frag *)(page_to_virt(pfrag->page) + offset);
1113	dfrag->data_len = 0;
1114	dfrag->data_seq = msk->write_seq;
1115	dfrag->overhead = offset - orig_offset + sizeof(struct mptcp_data_frag);
1116	dfrag->offset = offset + sizeof(struct mptcp_data_frag);
1117	dfrag->already_sent = 0;
1118	dfrag->page = pfrag->page;
1119
1120	return dfrag;
1121	}
1122
1123	struct mptcp_sendmsg_info {
1124	int mss_now;
1125	int size_goal;
1126	u16 limit;
1127	u16 sent;
1128	unsigned int flags;
1129	bool data_lock_held;
1130	};
1131
1132	static int mptcp_check_allowed_size(const struct mptcp_sock *msk, struct sock *ssk,
1133	u64 data_seq, int avail_size)
1134	{
1135	u64 window_end = mptcp_wnd_end(msk);
1136	u64 mptcp_snd_wnd;
1137
1138	if (__mptcp_check_fallback(msk))
1139	return avail_size;
1140
1141	mptcp_snd_wnd = window_end - data_seq;
1142	avail_size = min_t(unsigned int, mptcp_snd_wnd, avail_size);
1143
1144	if (unlikely(tcp_sk(ssk)->snd_wnd < mptcp_snd_wnd)) {
1145	tcp_sk(ssk)->snd_wnd = min_t(u64, U32_MAX, mptcp_snd_wnd);
1146	MPTCP_INC_STATS(net: sock_net(sk: ssk), field: MPTCP_MIB_SNDWNDSHARED);
1147	}
1148
1149	return avail_size;
1150	}
1151
1152	static bool __mptcp_add_ext(struct sk_buff *skb, gfp_t gfp)
1153	{
1154	struct skb_ext *mpext = __skb_ext_alloc(flags: gfp);
1155
1156	if (!mpext)
1157	return false;
1158	__skb_ext_set(skb, id: SKB_EXT_MPTCP, ext: mpext);
1159	return true;
1160	}
1161
1162	static struct sk_buff *__mptcp_do_alloc_tx_skb(struct sock *sk, gfp_t gfp)
1163	{
1164	struct sk_buff *skb;
1165
1166	skb = alloc_skb_fclone(MAX_TCP_HEADER, priority: gfp);
1167	if (likely(skb)) {
1168	if (likely(__mptcp_add_ext(skb, gfp))) {
1169	skb_reserve(skb, MAX_TCP_HEADER);
1170	skb->ip_summed = CHECKSUM_PARTIAL;
1171	INIT_LIST_HEAD(list: &skb->tcp_tsorted_anchor);
1172	return skb;
1173	}
1174	__kfree_skb(skb);
1175	} else {
1176	mptcp_enter_memory_pressure(sk);
1177	}
1178	return NULL;
1179	}
1180
1181	static struct sk_buff *__mptcp_alloc_tx_skb(struct sock *sk, struct sock *ssk, gfp_t gfp)
1182	{
1183	struct sk_buff *skb;
1184
1185	skb = __mptcp_do_alloc_tx_skb(sk, gfp);
1186	if (!skb)
1187	return NULL;
1188
1189	if (likely(sk_wmem_schedule(ssk, skb->truesize))) {
1190	tcp_skb_entail(sk: ssk, skb);
1191	return skb;
1192	}
1193	tcp_skb_tsorted_anchor_cleanup(skb);
1194	kfree_skb(skb);
1195	return NULL;
1196	}
1197
1198	static struct sk_buff *mptcp_alloc_tx_skb(struct sock *sk, struct sock *ssk, bool data_lock_held)
1199	{
1200	gfp_t gfp = data_lock_held ? GFP_ATOMIC : sk->sk_allocation;
1201
1202	return __mptcp_alloc_tx_skb(sk, ssk, gfp);
1203	}
1204
1205	/* note: this always recompute the csum on the whole skb, even
1206	* if we just appended a single frag. More status info needed
1207	*/
1208	static void mptcp_update_data_checksum(struct sk_buff *skb, int added)
1209	{
1210	struct mptcp_ext *mpext = mptcp_get_ext(skb);
1211	__wsum csum = ~csum_unfold(n: mpext->csum);
1212	int offset = skb->len - added;
1213
1214	mpext->csum = csum_fold(sum: csum_block_add(csum, csum2: skb_checksum(skb, offset, len: added, csum: 0), offset));
1215	}
1216
1217	static void mptcp_update_infinite_map(struct mptcp_sock *msk,
1218	struct sock *ssk,
1219	struct mptcp_ext *mpext)
1220	{
1221	if (!mpext)
1222	return;
1223
1224	mpext->infinite_map = 1;
1225	mpext->data_len = 0;
1226
1227	MPTCP_INC_STATS(net: sock_net(sk: ssk), field: MPTCP_MIB_INFINITEMAPTX);
1228	mptcp_subflow_ctx(sk: ssk)->send_infinite_map = 0;
1229	pr_fallback(msk);
1230	mptcp_do_fallback(ssk);
1231	}
1232
1233	static int mptcp_sendmsg_frag(struct sock *sk, struct sock *ssk,
1234	struct mptcp_data_frag *dfrag,
1235	struct mptcp_sendmsg_info *info)
1236	{
1237	u64 data_seq = dfrag->data_seq + info->sent;
1238	int offset = dfrag->offset + info->sent;
1239	struct mptcp_sock *msk = mptcp_sk(sk);
1240	bool zero_window_probe = false;
1241	struct mptcp_ext *mpext = NULL;
1242	bool can_coalesce = false;
1243	bool reuse_skb = true;
1244	struct sk_buff *skb;
1245	size_t copy;
1246	int i;
1247
1248	pr_debug("msk=%p ssk=%p sending dfrag at seq=%llu len=%u already sent=%u",
1249	msk, ssk, dfrag->data_seq, dfrag->data_len, info->sent);
1250
1251	if (WARN_ON_ONCE(info->sent > info->limit ||
1252	info->limit > dfrag->data_len))
1253	return 0;
1254
1255	if (unlikely(!__tcp_can_send(ssk)))
1256	return -EAGAIN;
1257
1258	/* compute send limit */
1259	info->mss_now = tcp_send_mss(sk: ssk, size_goal: &info->size_goal, flags: info->flags);
1260	copy = info->size_goal;
1261
1262	skb = tcp_write_queue_tail(sk: ssk);
1263	if (skb && copy > skb->len) {
1264	/* Limit the write to the size available in the
1265	* current skb, if any, so that we create at most a new skb.
1266	* Explicitly tells TCP internals to avoid collapsing on later
1267	* queue management operation, to avoid breaking the ext <->
1268	* SSN association set here
1269	*/
1270	mpext = mptcp_get_ext(skb);
1271	if (!mptcp_skb_can_collapse_to(write_seq: data_seq, skb, mpext)) {
1272	TCP_SKB_CB(skb)->eor = 1;
1273	goto alloc_skb;
1274	}
1275
1276	i = skb_shinfo(skb)->nr_frags;
1277	can_coalesce = skb_can_coalesce(skb, i, page: dfrag->page, off: offset);
1278	if (!can_coalesce && i >= READ_ONCE(sysctl_max_skb_frags)) {
1279	tcp_mark_push(tcp_sk(ssk), skb);
1280	goto alloc_skb;
1281	}
1282
1283	copy -= skb->len;
1284	} else {
1285	alloc_skb:
1286	skb = mptcp_alloc_tx_skb(sk, ssk, data_lock_held: info->data_lock_held);
1287	if (!skb)
1288	return -ENOMEM;
1289
1290	i = skb_shinfo(skb)->nr_frags;
1291	reuse_skb = false;
1292	mpext = mptcp_get_ext(skb);
1293	}
1294
1295	/* Zero window and all data acked? Probe. */
1296	copy = mptcp_check_allowed_size(msk, ssk, data_seq, avail_size: copy);
1297	if (copy == 0) {
1298	u64 snd_una = READ_ONCE(msk->snd_una);
1299
1300	if (snd_una != msk->snd_nxt || tcp_write_queue_tail(sk: ssk)) {
1301	tcp_remove_empty_skb(sk: ssk);
1302	return 0;
1303	}
1304
1305	zero_window_probe = true;
1306	data_seq = snd_una - 1;
1307	copy = 1;
1308	}
1309
1310	copy = min_t(size_t, copy, info->limit - info->sent);
1311	if (!sk_wmem_schedule(sk: ssk, size: copy)) {
1312	tcp_remove_empty_skb(sk: ssk);
1313	return -ENOMEM;
1314	}
1315
1316	if (can_coalesce) {
1317	skb_frag_size_add(frag: &skb_shinfo(skb)->frags[i - 1], delta: copy);
1318	} else {
1319	get_page(page: dfrag->page);
1320	skb_fill_page_desc(skb, i, page: dfrag->page, off: offset, size: copy);
1321	}
1322
1323	skb->len += copy;
1324	skb->data_len += copy;
1325	skb->truesize += copy;
1326	sk_wmem_queued_add(sk: ssk, val: copy);
1327	sk_mem_charge(sk: ssk, size: copy);
1328	WRITE_ONCE(tcp_sk(ssk)->write_seq, tcp_sk(ssk)->write_seq + copy);
1329	TCP_SKB_CB(skb)->end_seq += copy;
1330	tcp_skb_pcount_set(skb, segs: 0);
1331
1332	/* on skb reuse we just need to update the DSS len */
1333	if (reuse_skb) {
1334	TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH;
1335	mpext->data_len += copy;
1336	goto out;
1337	}
1338
1339	memset(mpext, 0, sizeof(*mpext));
1340	mpext->data_seq = data_seq;
1341	mpext->subflow_seq = mptcp_subflow_ctx(sk: ssk)->rel_write_seq;
1342	mpext->data_len = copy;
1343	mpext->use_map = 1;
1344	mpext->dsn64 = 1;
1345
1346	pr_debug("data_seq=%llu subflow_seq=%u data_len=%u dsn64=%d",
1347	mpext->data_seq, mpext->subflow_seq, mpext->data_len,
1348	mpext->dsn64);
1349
1350	if (zero_window_probe) {
1351	mptcp_subflow_ctx(sk: ssk)->rel_write_seq += copy;
1352	mpext->frozen = 1;
1353	if (READ_ONCE(msk->csum_enabled))
1354	mptcp_update_data_checksum(skb, added: copy);
1355	tcp_push_pending_frames(sk: ssk);
1356	return 0;
1357	}
1358	out:
1359	if (READ_ONCE(msk->csum_enabled))
1360	mptcp_update_data_checksum(skb, added: copy);
1361	if (mptcp_subflow_ctx(sk: ssk)->send_infinite_map)
1362	mptcp_update_infinite_map(msk, ssk, mpext);
1363	trace_mptcp_sendmsg_frag(mpext);
1364	mptcp_subflow_ctx(sk: ssk)->rel_write_seq += copy;
1365	return copy;
1366	}
1367
1368	#define MPTCP_SEND_BURST_SIZE ((1 << 16) - \
1369	sizeof(struct tcphdr) - \
1370	MAX_TCP_OPTION_SPACE - \
1371	sizeof(struct ipv6hdr) - \
1372	sizeof(struct frag_hdr))
1373
1374	struct subflow_send_info {
1375	struct sock *ssk;
1376	u64 linger_time;
1377	};
1378
1379	void mptcp_subflow_set_active(struct mptcp_subflow_context *subflow)
1380	{
1381	if (!subflow->stale)
1382	return;
1383
1384	subflow->stale = 0;
1385	MPTCP_INC_STATS(net: sock_net(sk: mptcp_subflow_tcp_sock(subflow)), field: MPTCP_MIB_SUBFLOWRECOVER);
1386	}
1387
1388	bool mptcp_subflow_active(struct mptcp_subflow_context *subflow)
1389	{
1390	if (unlikely(subflow->stale)) {
1391	u32 rcv_tstamp = READ_ONCE(tcp_sk(mptcp_subflow_tcp_sock(subflow))->rcv_tstamp);
1392
1393	if (subflow->stale_rcv_tstamp == rcv_tstamp)
1394	return false;
1395
1396	mptcp_subflow_set_active(subflow);
1397	}
1398	return __mptcp_subflow_active(subflow);
1399	}
1400
1401	#define SSK_MODE_ACTIVE 0
1402	#define SSK_MODE_BACKUP 1
1403	#define SSK_MODE_MAX 2
1404
1405	/* implement the mptcp packet scheduler;
1406	* returns the subflow that will transmit the next DSS
1407	* additionally updates the rtx timeout
1408	*/
1409	struct sock *mptcp_subflow_get_send(struct mptcp_sock *msk)
1410	{
1411	struct subflow_send_info send_info[SSK_MODE_MAX];
1412	struct mptcp_subflow_context *subflow;
1413	struct sock *sk = (struct sock *)msk;
1414	u32 pace, burst, wmem;
1415	int i, nr_active = 0;
1416	struct sock *ssk;
1417	u64 linger_time;
1418	long tout = 0;
1419
1420	/* pick the subflow with the lower wmem/wspace ratio */
1421	for (i = 0; i < SSK_MODE_MAX; ++i) {
1422	send_info[i].ssk = NULL;
1423	send_info[i].linger_time = -1;
1424	}
1425
1426	mptcp_for_each_subflow(msk, subflow) {
1427	trace_mptcp_subflow_get_send(subflow);
1428	ssk = mptcp_subflow_tcp_sock(subflow);
1429	if (!mptcp_subflow_active(subflow))
1430	continue;
1431
1432	tout = max(tout, mptcp_timeout_from_subflow(subflow));
1433	nr_active += !subflow->backup;
1434	pace = subflow->avg_pacing_rate;
1435	if (unlikely(!pace)) {
1436	/* init pacing rate from socket */
1437	subflow->avg_pacing_rate = READ_ONCE(ssk->sk_pacing_rate);
1438	pace = subflow->avg_pacing_rate;
1439	if (!pace)
1440	continue;
1441	}
1442
1443	linger_time = div_u64(dividend: (u64)READ_ONCE(ssk->sk_wmem_queued) << 32, divisor: pace);
1444	if (linger_time < send_info[subflow->backup].linger_time) {
1445	send_info[subflow->backup].ssk = ssk;
1446	send_info[subflow->backup].linger_time = linger_time;
1447	}
1448	}
1449	__mptcp_set_timeout(sk, tout);
1450
1451	/* pick the best backup if no other subflow is active */
1452	if (!nr_active)
1453	send_info[SSK_MODE_ACTIVE].ssk = send_info[SSK_MODE_BACKUP].ssk;
1454
1455	/* According to the blest algorithm, to avoid HoL blocking for the
1456	* faster flow, we need to:
1457	* - estimate the faster flow linger time
1458	* - use the above to estimate the amount of byte transferred
1459	* by the faster flow
1460	* - check that the amount of queued data is greter than the above,
1461	* otherwise do not use the picked, slower, subflow
1462	* We select the subflow with the shorter estimated time to flush
1463	* the queued mem, which basically ensure the above. We just need
1464	* to check that subflow has a non empty cwin.
1465	*/
1466	ssk = send_info[SSK_MODE_ACTIVE].ssk;
1467	if (!ssk || !sk_stream_memory_free(sk: ssk))
1468	return NULL;
1469
1470	burst = min_t(int, MPTCP_SEND_BURST_SIZE, mptcp_wnd_end(msk) - msk->snd_nxt);
1471	wmem = READ_ONCE(ssk->sk_wmem_queued);
1472	if (!burst)
1473	return ssk;
1474
1475	subflow = mptcp_subflow_ctx(sk: ssk);
1476	subflow->avg_pacing_rate = div_u64(dividend: (u64)subflow->avg_pacing_rate * wmem +
1477	READ_ONCE(ssk->sk_pacing_rate) * burst,
1478	divisor: burst + wmem);
1479	msk->snd_burst = burst;
1480	return ssk;
1481	}
1482
1483	static void mptcp_push_release(struct sock *ssk, struct mptcp_sendmsg_info *info)
1484	{
1485	tcp_push(sk: ssk, flags: 0, mss_now: info->mss_now, tcp_sk(ssk)->nonagle, size_goal: info->size_goal);
1486	release_sock(sk: ssk);
1487	}
1488
1489	static void mptcp_update_post_push(struct mptcp_sock *msk,
1490	struct mptcp_data_frag *dfrag,
1491	u32 sent)
1492	{
1493	u64 snd_nxt_new = dfrag->data_seq;
1494
1495	dfrag->already_sent += sent;
1496
1497	msk->snd_burst -= sent;
1498
1499	snd_nxt_new += dfrag->already_sent;
1500
1501	/* snd_nxt_new can be smaller than snd_nxt in case mptcp
1502	* is recovering after a failover. In that event, this re-sends
1503	* old segments.
1504	*
1505	* Thus compute snd_nxt_new candidate based on
1506	* the dfrag->data_seq that was sent and the data
1507	* that has been handed to the subflow for transmission
1508	* and skip update in case it was old dfrag.
1509	*/
1510	if (likely(after64(snd_nxt_new, msk->snd_nxt))) {
1511	msk->bytes_sent += snd_nxt_new - msk->snd_nxt;
1512	msk->snd_nxt = snd_nxt_new;
1513	}
1514	}
1515
1516	void mptcp_check_and_set_pending(struct sock *sk)
1517	{
1518	if (mptcp_send_head(sk))
1519	mptcp_sk(sk)->push_pending |= BIT(MPTCP_PUSH_PENDING);
1520	}
1521
1522	static int __subflow_push_pending(struct sock *sk, struct sock *ssk,
1523	struct mptcp_sendmsg_info *info)
1524	{
1525	struct mptcp_sock *msk = mptcp_sk(sk);
1526	struct mptcp_data_frag *dfrag;
1527	int len, copied = 0, err = 0;
1528
1529	while ((dfrag = mptcp_send_head(sk))) {
1530	info->sent = dfrag->already_sent;
1531	info->limit = dfrag->data_len;
1532	len = dfrag->data_len - dfrag->already_sent;
1533	while (len > 0) {
1534	int ret = 0;
1535
1536	ret = mptcp_sendmsg_frag(sk, ssk, dfrag, info);
1537	if (ret <= 0) {
1538	err = copied ? : ret;
1539	goto out;
1540	}
1541
1542	info->sent += ret;
1543	copied += ret;
1544	len -= ret;
1545
1546	mptcp_update_post_push(msk, dfrag, sent: ret);
1547	}
1548	WRITE_ONCE(msk->first_pending, mptcp_send_next(sk));
1549
1550	if (msk->snd_burst <= 0 ||
1551	!sk_stream_memory_free(sk: ssk) ||
1552	!mptcp_subflow_active(subflow: mptcp_subflow_ctx(sk: ssk))) {
1553	err = copied;
1554	goto out;
1555	}
1556	mptcp_set_timeout(sk);
1557	}
1558	err = copied;
1559
1560	out:
1561	return err;
1562	}
1563
1564	void __mptcp_push_pending(struct sock *sk, unsigned int flags)
1565	{
1566	struct sock *prev_ssk = NULL, *ssk = NULL;
1567	struct mptcp_sock *msk = mptcp_sk(sk);
1568	struct mptcp_sendmsg_info info = {
1569	.flags = flags,
1570	};
1571	bool do_check_data_fin = false;
1572	int push_count = 1;
1573
1574	while (mptcp_send_head(sk) && (push_count > 0)) {
1575	struct mptcp_subflow_context *subflow;
1576	int ret = 0;
1577
1578	if (mptcp_sched_get_send(msk))
1579	break;
1580
1581	push_count = 0;
1582
1583	mptcp_for_each_subflow(msk, subflow) {
1584	if (READ_ONCE(subflow->scheduled)) {
1585	mptcp_subflow_set_scheduled(subflow, scheduled: false);
1586
1587	prev_ssk = ssk;
1588	ssk = mptcp_subflow_tcp_sock(subflow);
1589	if (ssk != prev_ssk) {
1590	/* First check. If the ssk has changed since
1591	* the last round, release prev_ssk
1592	*/
1593	if (prev_ssk)
1594	mptcp_push_release(ssk: prev_ssk, info: &info);
1595
1596	/* Need to lock the new subflow only if different
1597	* from the previous one, otherwise we are still
1598	* helding the relevant lock
1599	*/
1600	lock_sock(sk: ssk);
1601	}
1602
1603	push_count++;
1604
1605	ret = __subflow_push_pending(sk, ssk, info: &info);
1606	if (ret <= 0) {
1607	if (ret != -EAGAIN ||
1608	(1 << ssk->sk_state) &
1609	(TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2 | TCPF_CLOSE))
1610	push_count--;
1611	continue;
1612	}
1613	do_check_data_fin = true;
1614	}
1615	}
1616	}
1617
1618	/* at this point we held the socket lock for the last subflow we used */
1619	if (ssk)
1620	mptcp_push_release(ssk, info: &info);
1621
1622	/* ensure the rtx timer is running */
1623	if (!mptcp_rtx_timer_pending(sk))
1624	mptcp_reset_rtx_timer(sk);
1625	if (do_check_data_fin)
1626	mptcp_check_send_data_fin(sk);
1627	}
1628
1629	static void __mptcp_subflow_push_pending(struct sock *sk, struct sock *ssk, bool first)
1630	{
1631	struct mptcp_sock *msk = mptcp_sk(sk);
1632	struct mptcp_sendmsg_info info = {
1633	.data_lock_held = true,
1634	};
1635	bool keep_pushing = true;
1636	struct sock *xmit_ssk;
1637	int copied = 0;
1638
1639	info.flags = 0;
1640	while (mptcp_send_head(sk) && keep_pushing) {
1641	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk: ssk);
1642	int ret = 0;
1643
1644	/* check for a different subflow usage only after
1645	* spooling the first chunk of data
1646	*/
1647	if (first) {
1648	mptcp_subflow_set_scheduled(subflow, scheduled: false);
1649	ret = __subflow_push_pending(sk, ssk, info: &info);
1650	first = false;
1651	if (ret <= 0)
1652	break;
1653	copied += ret;
1654	continue;
1655	}
1656
1657	if (mptcp_sched_get_send(msk))
1658	goto out;
1659
1660	if (READ_ONCE(subflow->scheduled)) {
1661	mptcp_subflow_set_scheduled(subflow, scheduled: false);
1662	ret = __subflow_push_pending(sk, ssk, info: &info);
1663	if (ret <= 0)
1664	keep_pushing = false;
1665	copied += ret;
1666	}
1667
1668	mptcp_for_each_subflow(msk, subflow) {
1669	if (READ_ONCE(subflow->scheduled)) {
1670	xmit_ssk = mptcp_subflow_tcp_sock(subflow);
1671	if (xmit_ssk != ssk) {
1672	mptcp_subflow_delegate(subflow,
1673	MPTCP_DELEGATE_SEND);
1674	keep_pushing = false;
1675	}
1676	}
1677	}
1678	}
1679
1680	out:
1681	/* __mptcp_alloc_tx_skb could have released some wmem and we are
1682	* not going to flush it via release_sock()
1683	*/
1684	if (copied) {
1685	tcp_push(sk: ssk, flags: 0, mss_now: info.mss_now, tcp_sk(ssk)->nonagle,
1686	size_goal: info.size_goal);
1687	if (!mptcp_rtx_timer_pending(sk))
1688	mptcp_reset_rtx_timer(sk);
1689
1690	if (msk->snd_data_fin_enable &&
1691	msk->snd_nxt + 1 == msk->write_seq)
1692	mptcp_schedule_work(sk);
1693	}
1694	}
1695
1696	static void mptcp_set_nospace(struct sock *sk)
1697	{
1698	/* enable autotune */
1699	set_bit(SOCK_NOSPACE, addr: &sk->sk_socket->flags);
1700
1701	/* will be cleared on avail space */
1702	set_bit(MPTCP_NOSPACE, addr: &mptcp_sk(sk)->flags);
1703	}
1704
1705	static int mptcp_disconnect(struct sock *sk, int flags);
1706
1707	static int mptcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg,
1708	size_t len, int *copied_syn)
1709	{
1710	unsigned int saved_flags = msg->msg_flags;
1711	struct mptcp_sock *msk = mptcp_sk(sk);
1712	struct sock *ssk;
1713	int ret;
1714
1715	/* on flags based fastopen the mptcp is supposed to create the
1716	* first subflow right now. Otherwise we are in the defer_connect
1717	* path, and the first subflow must be already present.
1718	* Since the defer_connect flag is cleared after the first succsful
1719	* fastopen attempt, no need to check for additional subflow status.
1720	*/
1721	if (msg->msg_flags & MSG_FASTOPEN) {
1722	ssk = __mptcp_nmpc_sk(msk);
1723	if (IS_ERR(ptr: ssk))
1724	return PTR_ERR(ptr: ssk);
1725	}
1726	if (!msk->first)
1727	return -EINVAL;
1728
1729	ssk = msk->first;
1730
1731	lock_sock(sk: ssk);
1732	msg->msg_flags |= MSG_DONTWAIT;
1733	msk->fastopening = 1;
1734	ret = tcp_sendmsg_fastopen(sk: ssk, msg, copied: copied_syn, size: len, NULL);
1735	msk->fastopening = 0;
1736	msg->msg_flags = saved_flags;
1737	release_sock(sk: ssk);
1738
1739	/* do the blocking bits of inet_stream_connect outside the ssk socket lock */
1740	if (ret == -EINPROGRESS && !(msg->msg_flags & MSG_DONTWAIT)) {
1741	ret = __inet_stream_connect(sock: sk->sk_socket, uaddr: msg->msg_name,
1742	addr_len: msg->msg_namelen, flags: msg->msg_flags, is_sendmsg: 1);
1743
1744	/* Keep the same behaviour of plain TCP: zero the copied bytes in
1745	* case of any error, except timeout or signal
1746	*/
1747	if (ret && ret != -EINPROGRESS && ret != -ERESTARTSYS && ret != -EINTR)
1748	*copied_syn = 0;
1749	} else if (ret && ret != -EINPROGRESS) {
1750	/* The disconnect() op called by tcp_sendmsg_fastopen()/
1751	* __inet_stream_connect() can fail, due to looking check,
1752	* see mptcp_disconnect().
1753	* Attempt it again outside the problematic scope.
1754	*/
1755	if (!mptcp_disconnect(sk, flags: 0))
1756	sk->sk_socket->state = SS_UNCONNECTED;
1757	}
1758	inet_clear_bit(DEFER_CONNECT, sk);
1759
1760	return ret;
1761	}
1762
1763	static int do_copy_data_nocache(struct sock *sk, int copy,
1764	struct iov_iter *from, char *to)
1765	{
1766	if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1767	if (!copy_from_iter_full_nocache(addr: to, bytes: copy, i: from))
1768	return -EFAULT;
1769	} else if (!copy_from_iter_full(addr: to, bytes: copy, i: from)) {
1770	return -EFAULT;
1771	}
1772	return 0;
1773	}
1774
1775	static int mptcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1776	{
1777	struct mptcp_sock *msk = mptcp_sk(sk);
1778	struct page_frag *pfrag;
1779	size_t copied = 0;
1780	int ret = 0;
1781	long timeo;
1782
1783	/* silently ignore everything else */
1784	msg->msg_flags &= MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL | MSG_FASTOPEN;
1785
1786	lock_sock(sk);
1787
1788	if (unlikely(inet_test_bit(DEFER_CONNECT, sk) ||
1789	msg->msg_flags & MSG_FASTOPEN)) {
1790	int copied_syn = 0;
1791
1792	ret = mptcp_sendmsg_fastopen(sk, msg, len, copied_syn: &copied_syn);
1793	copied += copied_syn;
1794	if (ret == -EINPROGRESS && copied_syn > 0)
1795	goto out;
1796	else if (ret)
1797	goto do_error;
1798	}
1799
1800	timeo = sock_sndtimeo(sk, noblock: msg->msg_flags & MSG_DONTWAIT);
1801
1802	if ((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) {
1803	ret = sk_stream_wait_connect(sk, timeo_p: &timeo);
1804	if (ret)
1805	goto do_error;
1806	}
1807
1808	ret = -EPIPE;
1809	if (unlikely(sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN)))
1810	goto do_error;
1811
1812	pfrag = sk_page_frag(sk);
1813
1814	while (msg_data_left(msg)) {
1815	int total_ts, frag_truesize = 0;
1816	struct mptcp_data_frag *dfrag;
1817	bool dfrag_collapsed;
1818	size_t psize, offset;
1819
1820	/* reuse tail pfrag, if possible, or carve a new one from the
1821	* page allocator
1822	*/
1823	dfrag = mptcp_pending_tail(sk);
1824	dfrag_collapsed = mptcp_frag_can_collapse_to(msk, pfrag, df: dfrag);
1825	if (!dfrag_collapsed) {
1826	if (!sk_stream_memory_free(sk))
1827	goto wait_for_memory;
1828
1829	if (!mptcp_page_frag_refill(sk, pfrag))
1830	goto wait_for_memory;
1831
1832	dfrag = mptcp_carve_data_frag(msk, pfrag, orig_offset: pfrag->offset);
1833	frag_truesize = dfrag->overhead;
1834	}
1835
1836	/* we do not bound vs wspace, to allow a single packet.
1837	* memory accounting will prevent execessive memory usage
1838	* anyway
1839	*/
1840	offset = dfrag->offset + dfrag->data_len;
1841	psize = pfrag->size - offset;
1842	psize = min_t(size_t, psize, msg_data_left(msg));
1843	total_ts = psize + frag_truesize;
1844
1845	if (!sk_wmem_schedule(sk, size: total_ts))
1846	goto wait_for_memory;
1847
1848	ret = do_copy_data_nocache(sk, copy: psize, from: &msg->msg_iter,
1849	page_address(dfrag->page) + offset);
1850	if (ret)
1851	goto do_error;
1852
1853	/* data successfully copied into the write queue */
1854	sk_forward_alloc_add(sk, val: -total_ts);
1855	copied += psize;
1856	dfrag->data_len += psize;
1857	frag_truesize += psize;
1858	pfrag->offset += frag_truesize;
1859	WRITE_ONCE(msk->write_seq, msk->write_seq + psize);
1860
1861	/* charge data on mptcp pending queue to the msk socket
1862	* Note: we charge such data both to sk and ssk
1863	*/
1864	sk_wmem_queued_add(sk, val: frag_truesize);
1865	if (!dfrag_collapsed) {
1866	get_page(page: dfrag->page);
1867	list_add_tail(new: &dfrag->list, head: &msk->rtx_queue);
1868	if (!msk->first_pending)
1869	WRITE_ONCE(msk->first_pending, dfrag);
1870	}
1871	pr_debug("msk=%p dfrag at seq=%llu len=%u sent=%u new=%d", msk,
1872	dfrag->data_seq, dfrag->data_len, dfrag->already_sent,
1873	!dfrag_collapsed);
1874
1875	continue;
1876
1877	wait_for_memory:
1878	mptcp_set_nospace(sk);
1879	__mptcp_push_pending(sk, flags: msg->msg_flags);
1880	ret = sk_stream_wait_memory(sk, timeo_p: &timeo);
1881	if (ret)
1882	goto do_error;
1883	}
1884
1885	if (copied)
1886	__mptcp_push_pending(sk, flags: msg->msg_flags);
1887
1888	out:
1889	release_sock(sk);
1890	return copied;
1891
1892	do_error:
1893	if (copied)
1894	goto out;
1895
1896	copied = sk_stream_error(sk, flags: msg->msg_flags, err: ret);
1897	goto out;
1898	}
1899
1900	static int __mptcp_recvmsg_mskq(struct mptcp_sock *msk,
1901	struct msghdr *msg,
1902	size_t len, int flags,
1903	struct scm_timestamping_internal *tss,
1904	int *cmsg_flags)
1905	{
1906	struct sk_buff *skb, *tmp;
1907	int copied = 0;
1908
1909	skb_queue_walk_safe(&msk->receive_queue, skb, tmp) {
1910	u32 offset = MPTCP_SKB_CB(skb)->offset;
1911	u32 data_len = skb->len - offset;
1912	u32 count = min_t(size_t, len - copied, data_len);
1913	int err;
1914
1915	if (!(flags & MSG_TRUNC)) {
1916	err = skb_copy_datagram_msg(from: skb, offset, msg, size: count);
1917	if (unlikely(err < 0)) {
1918	if (!copied)
1919	return err;
1920	break;
1921	}
1922	}
1923
1924	if (MPTCP_SKB_CB(skb)->has_rxtstamp) {
1925	tcp_update_recv_tstamps(skb, tss);
1926	*cmsg_flags |= MPTCP_CMSG_TS;
1927	}
1928
1929	copied += count;
1930
1931	if (count < data_len) {
1932	if (!(flags & MSG_PEEK)) {
1933	MPTCP_SKB_CB(skb)->offset += count;
1934	MPTCP_SKB_CB(skb)->map_seq += count;
1935	msk->bytes_consumed += count;
1936	}
1937	break;
1938	}
1939
1940	if (!(flags & MSG_PEEK)) {
1941	/* we will bulk release the skb memory later */
1942	skb->destructor = NULL;
1943	WRITE_ONCE(msk->rmem_released, msk->rmem_released + skb->truesize);
1944	__skb_unlink(skb, list: &msk->receive_queue);
1945	__kfree_skb(skb);
1946	msk->bytes_consumed += count;
1947	}
1948
1949	if (copied >= len)
1950	break;
1951	}
1952
1953	return copied;
1954	}
1955
1956	/* receive buffer autotuning. See tcp_rcv_space_adjust for more information.
1957	*
1958	* Only difference: Use highest rtt estimate of the subflows in use.
1959	*/
1960	static void mptcp_rcv_space_adjust(struct mptcp_sock *msk, int copied)
1961	{
1962	struct mptcp_subflow_context *subflow;
1963	struct sock *sk = (struct sock *)msk;
1964	u8 scaling_ratio = U8_MAX;
1965	u32 time, advmss = 1;
1966	u64 rtt_us, mstamp;
1967
1968	msk_owned_by_me(msk);
1969
1970	if (copied <= 0)
1971	return;
1972
1973	msk->rcvq_space.copied += copied;
1974
1975	mstamp = div_u64(dividend: tcp_clock_ns(), NSEC_PER_USEC);
1976	time = tcp_stamp_us_delta(t1: mstamp, t0: msk->rcvq_space.time);
1977
1978	rtt_us = msk->rcvq_space.rtt_us;
1979	if (rtt_us && time < (rtt_us >> 3))
1980	return;
1981
1982	rtt_us = 0;
1983	mptcp_for_each_subflow(msk, subflow) {
1984	const struct tcp_sock *tp;
1985	u64 sf_rtt_us;
1986	u32 sf_advmss;
1987
1988	tp = tcp_sk(mptcp_subflow_tcp_sock(subflow));
1989
1990	sf_rtt_us = READ_ONCE(tp->rcv_rtt_est.rtt_us);
1991	sf_advmss = READ_ONCE(tp->advmss);
1992
1993	rtt_us = max(sf_rtt_us, rtt_us);
1994	advmss = max(sf_advmss, advmss);
1995	scaling_ratio = min(tp->scaling_ratio, scaling_ratio);
1996	}
1997
1998	msk->rcvq_space.rtt_us = rtt_us;
1999	msk->scaling_ratio = scaling_ratio;
2000	if (time < (rtt_us >> 3) || rtt_us == 0)
2001	return;
2002
2003	if (msk->rcvq_space.copied <= msk->rcvq_space.space)
2004	goto new_measure;
2005
2006	if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_moderate_rcvbuf) &&
2007	!(sk->sk_userlocks & SOCK_RCVBUF_LOCK)) {
2008	u64 rcvwin, grow;
2009	int rcvbuf;
2010
2011	rcvwin = ((u64)msk->rcvq_space.copied << 1) + 16 * advmss;
2012
2013	grow = rcvwin * (msk->rcvq_space.copied - msk->rcvq_space.space);
2014
2015	do_div(grow, msk->rcvq_space.space);
2016	rcvwin += (grow << 1);
2017
2018	rcvbuf = min_t(u64, __tcp_space_from_win(scaling_ratio, rcvwin),
2019	READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_rmem[2]));
2020
2021	if (rcvbuf > sk->sk_rcvbuf) {
2022	u32 window_clamp;
2023
2024	window_clamp = __tcp_win_from_space(scaling_ratio, space: rcvbuf);
2025	WRITE_ONCE(sk->sk_rcvbuf, rcvbuf);
2026
2027	/* Make subflows follow along. If we do not do this, we
2028	* get drops at subflow level if skbs can't be moved to
2029	* the mptcp rx queue fast enough (announced rcv_win can
2030	* exceed ssk->sk_rcvbuf).
2031	*/
2032	mptcp_for_each_subflow(msk, subflow) {
2033	struct sock *ssk;
2034	bool slow;
2035
2036	ssk = mptcp_subflow_tcp_sock(subflow);
2037	slow = lock_sock_fast(sk: ssk);
2038	WRITE_ONCE(ssk->sk_rcvbuf, rcvbuf);
2039	tcp_sk(ssk)->window_clamp = window_clamp;
2040	tcp_cleanup_rbuf(sk: ssk, copied: 1);
2041	unlock_sock_fast(sk: ssk, slow);
2042	}
2043	}
2044	}
2045
2046	msk->rcvq_space.space = msk->rcvq_space.copied;
2047	new_measure:
2048	msk->rcvq_space.copied = 0;
2049	msk->rcvq_space.time = mstamp;
2050	}
2051
2052	static void __mptcp_update_rmem(struct sock *sk)
2053	{
2054	struct mptcp_sock *msk = mptcp_sk(sk);
2055
2056	if (!msk->rmem_released)
2057	return;
2058
2059	atomic_sub(i: msk->rmem_released, v: &sk->sk_rmem_alloc);
2060	mptcp_rmem_uncharge(sk, size: msk->rmem_released);
2061	WRITE_ONCE(msk->rmem_released, 0);
2062	}
2063
2064	static void __mptcp_splice_receive_queue(struct sock *sk)
2065	{
2066	struct mptcp_sock *msk = mptcp_sk(sk);
2067
2068	skb_queue_splice_tail_init(list: &sk->sk_receive_queue, head: &msk->receive_queue);
2069	}
2070
2071	static bool __mptcp_move_skbs(struct mptcp_sock *msk)
2072	{
2073	struct sock *sk = (struct sock *)msk;
2074	unsigned int moved = 0;
2075	bool ret, done;
2076
2077	do {
2078	struct sock *ssk = mptcp_subflow_recv_lookup(msk);
2079	bool slowpath;
2080
2081	/* we can have data pending in the subflows only if the msk
2082	* receive buffer was full at subflow_data_ready() time,
2083	* that is an unlikely slow path.
2084	*/
2085	if (likely(!ssk))
2086	break;
2087
2088	slowpath = lock_sock_fast(sk: ssk);
2089	mptcp_data_lock(sk);
2090	__mptcp_update_rmem(sk);
2091	done = __mptcp_move_skbs_from_subflow(msk, ssk, bytes: &moved);
2092	mptcp_data_unlock(sk);
2093
2094	if (unlikely(ssk->sk_err))
2095	__mptcp_error_report(sk);
2096	unlock_sock_fast(sk: ssk, slow: slowpath);
2097	} while (!done);
2098
2099	/* acquire the data lock only if some input data is pending */
2100	ret = moved > 0;
2101	if (!RB_EMPTY_ROOT(&msk->out_of_order_queue) ||
2102	!skb_queue_empty_lockless(list: &sk->sk_receive_queue)) {
2103	mptcp_data_lock(sk);
2104	__mptcp_update_rmem(sk);
2105	ret |= __mptcp_ofo_queue(msk);
2106	__mptcp_splice_receive_queue(sk);
2107	mptcp_data_unlock(sk);
2108	}
2109	if (ret)
2110	mptcp_check_data_fin(sk: (struct sock *)msk);
2111	return !skb_queue_empty(list: &msk->receive_queue);
2112	}
2113
2114	static unsigned int mptcp_inq_hint(const struct sock *sk)
2115	{
2116	const struct mptcp_sock *msk = mptcp_sk(sk);
2117	const struct sk_buff *skb;
2118
2119	skb = skb_peek(list_: &msk->receive_queue);
2120	if (skb) {
2121	u64 hint_val = msk->ack_seq - MPTCP_SKB_CB(skb)->map_seq;
2122
2123	if (hint_val >= INT_MAX)
2124	return INT_MAX;
2125
2126	return (unsigned int)hint_val;
2127	}
2128
2129	if (sk->sk_state == TCP_CLOSE || (sk->sk_shutdown & RCV_SHUTDOWN))
2130	return 1;
2131
2132	return 0;
2133	}
2134
2135	static int mptcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
2136	int flags, int *addr_len)
2137	{
2138	struct mptcp_sock *msk = mptcp_sk(sk);
2139	struct scm_timestamping_internal tss;
2140	int copied = 0, cmsg_flags = 0;
2141	int target;
2142	long timeo;
2143
2144	/* MSG_ERRQUEUE is really a no-op till we support IP_RECVERR */
2145	if (unlikely(flags & MSG_ERRQUEUE))
2146	return inet_recv_error(sk, msg, len, addr_len);
2147
2148	lock_sock(sk);
2149	if (unlikely(sk->sk_state == TCP_LISTEN)) {
2150	copied = -ENOTCONN;
2151	goto out_err;
2152	}
2153
2154	timeo = sock_rcvtimeo(sk, noblock: flags & MSG_DONTWAIT);
2155
2156	len = min_t(size_t, len, INT_MAX);
2157	target = sock_rcvlowat(sk, waitall: flags & MSG_WAITALL, len);
2158
2159	if (unlikely(msk->recvmsg_inq))
2160	cmsg_flags = MPTCP_CMSG_INQ;
2161
2162	while (copied < len) {
2163	int bytes_read;
2164
2165	bytes_read = __mptcp_recvmsg_mskq(msk, msg, len: len - copied, flags, tss: &tss, cmsg_flags: &cmsg_flags);
2166	if (unlikely(bytes_read < 0)) {
2167	if (!copied)
2168	copied = bytes_read;
2169	goto out_err;
2170	}
2171
2172	copied += bytes_read;
2173
2174	/* be sure to advertise window change */
2175	mptcp_cleanup_rbuf(msk);
2176
2177	if (skb_queue_empty(list: &msk->receive_queue) && __mptcp_move_skbs(msk))
2178	continue;
2179
2180	/* only the master socket status is relevant here. The exit
2181	* conditions mirror closely tcp_recvmsg()
2182	*/
2183	if (copied >= target)
2184	break;
2185
2186	if (copied) {
2187	if (sk->sk_err ||
2188	sk->sk_state == TCP_CLOSE ||
2189	(sk->sk_shutdown & RCV_SHUTDOWN) ||
2190	!timeo ||
2191	signal_pending(current))
2192	break;
2193	} else {
2194	if (sk->sk_err) {
2195	copied = sock_error(sk);
2196	break;
2197	}
2198
2199	if (sk->sk_shutdown & RCV_SHUTDOWN) {
2200	/* race breaker: the shutdown could be after the
2201	* previous receive queue check
2202	*/
2203	if (__mptcp_move_skbs(msk))
2204	continue;
2205	break;
2206	}
2207
2208	if (sk->sk_state == TCP_CLOSE) {
2209	copied = -ENOTCONN;
2210	break;
2211	}
2212
2213	if (!timeo) {
2214	copied = -EAGAIN;
2215	break;
2216	}
2217
2218	if (signal_pending(current)) {
2219	copied = sock_intr_errno(timeo);
2220	break;
2221	}
2222	}
2223
2224	pr_debug("block timeout %ld", timeo);
2225	sk_wait_data(sk, timeo: &timeo, NULL);
2226	}
2227
2228	out_err:
2229	if (cmsg_flags && copied >= 0) {
2230	if (cmsg_flags & MPTCP_CMSG_TS)
2231	tcp_recv_timestamp(msg, sk, tss: &tss);
2232
2233	if (cmsg_flags & MPTCP_CMSG_INQ) {
2234	unsigned int inq = mptcp_inq_hint(sk);
2235
2236	put_cmsg(msg, SOL_TCP, TCP_CM_INQ, len: sizeof(inq), data: &inq);
2237	}
2238	}
2239
2240	pr_debug("msk=%p rx queue empty=%d:%d copied=%d",
2241	msk, skb_queue_empty_lockless(&sk->sk_receive_queue),
2242	skb_queue_empty(&msk->receive_queue), copied);
2243	if (!(flags & MSG_PEEK))
2244	mptcp_rcv_space_adjust(msk, copied);
2245
2246	release_sock(sk);
2247	return copied;
2248	}
2249
2250	static void mptcp_retransmit_timer(struct timer_list *t)
2251	{
2252	struct inet_connection_sock *icsk = from_timer(icsk, t,
2253	icsk_retransmit_timer);
2254	struct sock *sk = &icsk->icsk_inet.sk;
2255	struct mptcp_sock *msk = mptcp_sk(sk);
2256
2257	bh_lock_sock(sk);
2258	if (!sock_owned_by_user(sk)) {
2259	/* we need a process context to retransmit */
2260	if (!test_and_set_bit(MPTCP_WORK_RTX, addr: &msk->flags))
2261	mptcp_schedule_work(sk);
2262	} else {
2263	/* delegate our work to tcp_release_cb() */
2264	__set_bit(MPTCP_RETRANSMIT, &msk->cb_flags);
2265	}
2266	bh_unlock_sock(sk);
2267	sock_put(sk);
2268	}
2269
2270	static void mptcp_tout_timer(struct timer_list *t)
2271	{
2272	struct sock *sk = from_timer(sk, t, sk_timer);
2273
2274	mptcp_schedule_work(sk);
2275	sock_put(sk);
2276	}
2277
2278	/* Find an idle subflow. Return NULL if there is unacked data at tcp
2279	* level.
2280	*
2281	* A backup subflow is returned only if that is the only kind available.
2282	*/
2283	struct sock *mptcp_subflow_get_retrans(struct mptcp_sock *msk)
2284	{
2285	struct sock *backup = NULL, *pick = NULL;
2286	struct mptcp_subflow_context *subflow;
2287	int min_stale_count = INT_MAX;
2288
2289	mptcp_for_each_subflow(msk, subflow) {
2290	struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2291
2292	if (!__mptcp_subflow_active(subflow))
2293	continue;
2294
2295	/* still data outstanding at TCP level? skip this */
2296	if (!tcp_rtx_and_write_queues_empty(sk: ssk)) {
2297	mptcp_pm_subflow_chk_stale(msk, ssk);
2298	min_stale_count = min_t(int, min_stale_count, subflow->stale_count);
2299	continue;
2300	}
2301
2302	if (subflow->backup) {
2303	if (!backup)
2304	backup = ssk;
2305	continue;
2306	}
2307
2308	if (!pick)
2309	pick = ssk;
2310	}
2311
2312	if (pick)
2313	return pick;
2314
2315	/* use backup only if there are no progresses anywhere */
2316	return min_stale_count > 1 ? backup : NULL;
2317	}
2318
2319	bool __mptcp_retransmit_pending_data(struct sock *sk)
2320	{
2321	struct mptcp_data_frag *cur, *rtx_head;
2322	struct mptcp_sock *msk = mptcp_sk(sk);
2323
2324	if (__mptcp_check_fallback(msk))
2325	return false;
2326
2327	if (tcp_rtx_and_write_queues_empty(sk))
2328	return false;
2329
2330	/* the closing socket has some data untransmitted and/or unacked:
2331	* some data in the mptcp rtx queue has not really xmitted yet.
2332	* keep it simple and re-inject the whole mptcp level rtx queue
2333	*/
2334	mptcp_data_lock(sk);
2335	__mptcp_clean_una_wakeup(sk);
2336	rtx_head = mptcp_rtx_head(sk);
2337	if (!rtx_head) {
2338	mptcp_data_unlock(sk);
2339	return false;
2340	}
2341
2342	msk->recovery_snd_nxt = msk->snd_nxt;
2343	msk->recovery = true;
2344	mptcp_data_unlock(sk);
2345
2346	msk->first_pending = rtx_head;
2347	msk->snd_burst = 0;
2348
2349	/* be sure to clear the "sent status" on all re-injected fragments */
2350	list_for_each_entry(cur, &msk->rtx_queue, list) {
2351	if (!cur->already_sent)
2352	break;
2353	cur->already_sent = 0;
2354	}
2355
2356	return true;
2357	}
2358
2359	/* flags for __mptcp_close_ssk() */
2360	#define MPTCP_CF_PUSH BIT(1)
2361	#define MPTCP_CF_FASTCLOSE BIT(2)
2362
2363	/* be sure to send a reset only if the caller asked for it, also
2364	* clean completely the subflow status when the subflow reaches
2365	* TCP_CLOSE state
2366	*/
2367	static void __mptcp_subflow_disconnect(struct sock *ssk,
2368	struct mptcp_subflow_context *subflow,
2369	unsigned int flags)
2370	{
2371	if (((1 << ssk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)) ||
2372	(flags & MPTCP_CF_FASTCLOSE)) {
2373	/* The MPTCP code never wait on the subflow sockets, TCP-level
2374	* disconnect should never fail
2375	*/
2376	WARN_ON_ONCE(tcp_disconnect(ssk, 0));
2377	mptcp_subflow_ctx_reset(subflow);
2378	} else {
2379	tcp_shutdown(sk: ssk, SEND_SHUTDOWN);
2380	}
2381	}
2382
2383	/* subflow sockets can be either outgoing (connect) or incoming
2384	* (accept).
2385	*
2386	* Outgoing subflows use in-kernel sockets.
2387	* Incoming subflows do not have their own 'struct socket' allocated,
2388	* so we need to use tcp_close() after detaching them from the mptcp
2389	* parent socket.
2390	*/
2391	static void __mptcp_close_ssk(struct sock *sk, struct sock *ssk,
2392	struct mptcp_subflow_context *subflow,
2393	unsigned int flags)
2394	{
2395	struct mptcp_sock *msk = mptcp_sk(sk);
2396	bool dispose_it, need_push = false;
2397
2398	/* If the first subflow moved to a close state before accept, e.g. due
2399	* to an incoming reset or listener shutdown, the subflow socket is
2400	* already deleted by inet_child_forget() and the mptcp socket can't
2401	* survive too.
2402	*/
2403	if (msk->in_accept_queue && msk->first == ssk &&
2404	(sock_flag(sk, flag: SOCK_DEAD) || sock_flag(sk: ssk, flag: SOCK_DEAD))) {
2405	/* ensure later check in mptcp_worker() will dispose the msk */
2406	sock_set_flag(sk, flag: SOCK_DEAD);
2407	mptcp_set_close_tout(sk, tcp_jiffies32 - (mptcp_close_timeout(sk) + 1));
2408	lock_sock_nested(sk: ssk, SINGLE_DEPTH_NESTING);
2409	mptcp_subflow_drop_ctx(ssk);
2410	goto out_release;
2411	}
2412
2413	dispose_it = msk->free_first || ssk != msk->first;
2414	if (dispose_it)
2415	list_del(entry: &subflow->node);
2416
2417	lock_sock_nested(sk: ssk, SINGLE_DEPTH_NESTING);
2418
2419	if ((flags & MPTCP_CF_FASTCLOSE) && !__mptcp_check_fallback(msk)) {
2420	/* be sure to force the tcp_close path
2421	* to generate the egress reset
2422	*/
2423	ssk->sk_lingertime = 0;
2424	sock_set_flag(sk: ssk, flag: SOCK_LINGER);
2425	subflow->send_fastclose = 1;
2426	}
2427
2428	need_push = (flags & MPTCP_CF_PUSH) && __mptcp_retransmit_pending_data(sk);
2429	if (!dispose_it) {
2430	__mptcp_subflow_disconnect(ssk, subflow, flags);
2431	release_sock(sk: ssk);
2432
2433	goto out;
2434	}
2435
2436	subflow->disposable = 1;
2437
2438	/* if ssk hit tcp_done(), tcp_cleanup_ulp() cleared the related ops
2439	* the ssk has been already destroyed, we just need to release the
2440	* reference owned by msk;
2441	*/
2442	if (!inet_csk(sk: ssk)->icsk_ulp_ops) {
2443	WARN_ON_ONCE(!sock_flag(ssk, SOCK_DEAD));
2444	kfree_rcu(subflow, rcu);
2445	} else {
2446	/* otherwise tcp will dispose of the ssk and subflow ctx */
2447	__tcp_close(sk: ssk, timeout: 0);
2448
2449	/* close acquired an extra ref */
2450	__sock_put(sk: ssk);
2451	}
2452
2453	out_release:
2454	__mptcp_subflow_error_report(sk, ssk);
2455	release_sock(sk: ssk);
2456
2457	sock_put(sk: ssk);
2458
2459	if (ssk == msk->first)
2460	WRITE_ONCE(msk->first, NULL);
2461
2462	out:
2463	__mptcp_sync_sndbuf(sk);
2464	if (need_push)
2465	__mptcp_push_pending(sk, flags: 0);
2466
2467	/* Catch every 'all subflows closed' scenario, including peers silently
2468	* closing them, e.g. due to timeout.
2469	* For established sockets, allow an additional timeout before closing,
2470	* as the protocol can still create more subflows.
2471	*/
2472	if (list_is_singular(head: &msk->conn_list) && msk->first &&
2473	inet_sk_state_load(sk: msk->first) == TCP_CLOSE) {
2474	if (sk->sk_state != TCP_ESTABLISHED ||
2475	msk->in_accept_queue || sock_flag(sk, flag: SOCK_DEAD)) {
2476	inet_sk_state_store(sk, newstate: TCP_CLOSE);
2477	mptcp_close_wake_up(sk);
2478	} else {
2479	mptcp_start_tout_timer(sk);
2480	}
2481	}
2482	}
2483
2484	void mptcp_close_ssk(struct sock *sk, struct sock *ssk,
2485	struct mptcp_subflow_context *subflow)
2486	{
2487	if (sk->sk_state == TCP_ESTABLISHED)
2488	mptcp_event(type: MPTCP_EVENT_SUB_CLOSED, mptcp_sk(sk), ssk, GFP_KERNEL);
2489
2490	/* subflow aborted before reaching the fully_established status
2491	* attempt the creation of the next subflow
2492	*/
2493	mptcp_pm_subflow_check_next(mptcp_sk(sk), subflow);
2494
2495	__mptcp_close_ssk(sk, ssk, subflow, MPTCP_CF_PUSH);
2496	}
2497
2498	static unsigned int mptcp_sync_mss(struct sock *sk, u32 pmtu)
2499	{
2500	return 0;
2501	}
2502
2503	static void __mptcp_close_subflow(struct sock *sk)
2504	{
2505	struct mptcp_subflow_context *subflow, *tmp;
2506	struct mptcp_sock *msk = mptcp_sk(sk);
2507
2508	might_sleep();
2509
2510	mptcp_for_each_subflow_safe(msk, subflow, tmp) {
2511	struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2512
2513	if (inet_sk_state_load(sk: ssk) != TCP_CLOSE)
2514	continue;
2515
2516	/* 'subflow_data_ready' will re-sched once rx queue is empty */
2517	if (!skb_queue_empty_lockless(list: &ssk->sk_receive_queue))
2518	continue;
2519
2520	mptcp_close_ssk(sk, ssk, subflow);
2521	}
2522
2523	}
2524
2525	static bool mptcp_close_tout_expired(const struct sock *sk)
2526	{
2527	if (!inet_csk(sk)->icsk_mtup.probe_timestamp ||
2528	sk->sk_state == TCP_CLOSE)
2529	return false;
2530
2531	return time_after32(tcp_jiffies32,
2532	inet_csk(sk)->icsk_mtup.probe_timestamp + mptcp_close_timeout(sk));
2533	}
2534
2535	static void mptcp_check_fastclose(struct mptcp_sock *msk)
2536	{
2537	struct mptcp_subflow_context *subflow, *tmp;
2538	struct sock *sk = (struct sock *)msk;
2539
2540	if (likely(!READ_ONCE(msk->rcv_fastclose)))
2541	return;
2542
2543	mptcp_token_destroy(msk);
2544
2545	mptcp_for_each_subflow_safe(msk, subflow, tmp) {
2546	struct sock *tcp_sk = mptcp_subflow_tcp_sock(subflow);
2547	bool slow;
2548
2549	slow = lock_sock_fast(sk: tcp_sk);
2550	if (tcp_sk->sk_state != TCP_CLOSE) {
2551	tcp_send_active_reset(sk: tcp_sk, GFP_ATOMIC);
2552	tcp_set_state(sk: tcp_sk, state: TCP_CLOSE);
2553	}
2554	unlock_sock_fast(sk: tcp_sk, slow);
2555	}
2556
2557	/* Mirror the tcp_reset() error propagation */
2558	switch (sk->sk_state) {
2559	case TCP_SYN_SENT:
2560	WRITE_ONCE(sk->sk_err, ECONNREFUSED);
2561	break;
2562	case TCP_CLOSE_WAIT:
2563	WRITE_ONCE(sk->sk_err, EPIPE);
2564	break;
2565	case TCP_CLOSE:
2566	return;
2567	default:
2568	WRITE_ONCE(sk->sk_err, ECONNRESET);
2569	}
2570
2571	inet_sk_state_store(sk, newstate: TCP_CLOSE);
2572	WRITE_ONCE(sk->sk_shutdown, SHUTDOWN_MASK);
2573	smp_mb__before_atomic(); /* SHUTDOWN must be visible first */
2574	set_bit(MPTCP_WORK_CLOSE_SUBFLOW, addr: &msk->flags);
2575
2576	/* the calling mptcp_worker will properly destroy the socket */
2577	if (sock_flag(sk, flag: SOCK_DEAD))
2578	return;
2579
2580	sk->sk_state_change(sk);
2581	sk_error_report(sk);
2582	}
2583
2584	static void __mptcp_retrans(struct sock *sk)
2585	{
2586	struct mptcp_sock *msk = mptcp_sk(sk);
2587	struct mptcp_subflow_context *subflow;
2588	struct mptcp_sendmsg_info info = {};
2589	struct mptcp_data_frag *dfrag;
2590	struct sock *ssk;
2591	int ret, err;
2592	u16 len = 0;
2593
2594	mptcp_clean_una_wakeup(sk);
2595
2596	/* first check ssk: need to kick "stale" logic */
2597	err = mptcp_sched_get_retrans(msk);
2598	dfrag = mptcp_rtx_head(sk);
2599	if (!dfrag) {
2600	if (mptcp_data_fin_enabled(msk)) {
2601	struct inet_connection_sock *icsk = inet_csk(sk);
2602
2603	icsk->icsk_retransmits++;
2604	mptcp_set_datafin_timeout(sk);
2605	mptcp_send_ack(msk);
2606
2607	goto reset_timer;
2608	}
2609
2610	if (!mptcp_send_head(sk))
2611	return;
2612
2613	goto reset_timer;
2614	}
2615
2616	if (err)
2617	goto reset_timer;
2618
2619	mptcp_for_each_subflow(msk, subflow) {
2620	if (READ_ONCE(subflow->scheduled)) {
2621	u16 copied = 0;
2622
2623	mptcp_subflow_set_scheduled(subflow, scheduled: false);
2624
2625	ssk = mptcp_subflow_tcp_sock(subflow);
2626
2627	lock_sock(sk: ssk);
2628
2629	/* limit retransmission to the bytes already sent on some subflows */
2630	info.sent = 0;
2631	info.limit = READ_ONCE(msk->csum_enabled) ? dfrag->data_len :
2632	dfrag->already_sent;
2633	while (info.sent < info.limit) {
2634	ret = mptcp_sendmsg_frag(sk, ssk, dfrag, info: &info);
2635	if (ret <= 0)
2636	break;
2637
2638	MPTCP_INC_STATS(net: sock_net(sk), field: MPTCP_MIB_RETRANSSEGS);
2639	copied += ret;
2640	info.sent += ret;
2641	}
2642	if (copied) {
2643	len = max(copied, len);
2644	tcp_push(sk: ssk, flags: 0, mss_now: info.mss_now, tcp_sk(ssk)->nonagle,
2645	size_goal: info.size_goal);
2646	WRITE_ONCE(msk->allow_infinite_fallback, false);
2647	}
2648
2649	release_sock(sk: ssk);
2650	}
2651	}
2652
2653	msk->bytes_retrans += len;
2654	dfrag->already_sent = max(dfrag->already_sent, len);
2655
2656	reset_timer:
2657	mptcp_check_and_set_pending(sk);
2658
2659	if (!mptcp_rtx_timer_pending(sk))
2660	mptcp_reset_rtx_timer(sk);
2661	}
2662
2663	/* schedule the timeout timer for the relevant event: either close timeout
2664	* or mp_fail timeout. The close timeout takes precedence on the mp_fail one
2665	*/
2666	void mptcp_reset_tout_timer(struct mptcp_sock *msk, unsigned long fail_tout)
2667	{
2668	struct sock *sk = (struct sock *)msk;
2669	unsigned long timeout, close_timeout;
2670
2671	if (!fail_tout && !inet_csk(sk)->icsk_mtup.probe_timestamp)
2672	return;
2673
2674	close_timeout = inet_csk(sk)->icsk_mtup.probe_timestamp - tcp_jiffies32 + jiffies +
2675	mptcp_close_timeout(sk);
2676
2677	/* the close timeout takes precedence on the fail one, and here at least one of
2678	* them is active
2679	*/
2680	timeout = inet_csk(sk)->icsk_mtup.probe_timestamp ? close_timeout : fail_tout;
2681
2682	sk_reset_timer(sk, timer: &sk->sk_timer, expires: timeout);
2683	}
2684
2685	static void mptcp_mp_fail_no_response(struct mptcp_sock *msk)
2686	{
2687	struct sock *ssk = msk->first;
2688	bool slow;
2689
2690	if (!ssk)
2691	return;
2692
2693	pr_debug("MP_FAIL doesn't respond, reset the subflow");
2694
2695	slow = lock_sock_fast(sk: ssk);
2696	mptcp_subflow_reset(ssk);
2697	WRITE_ONCE(mptcp_subflow_ctx(ssk)->fail_tout, 0);
2698	unlock_sock_fast(sk: ssk, slow);
2699	}
2700
2701	static void mptcp_do_fastclose(struct sock *sk)
2702	{
2703	struct mptcp_subflow_context *subflow, *tmp;
2704	struct mptcp_sock *msk = mptcp_sk(sk);
2705
2706	inet_sk_state_store(sk, newstate: TCP_CLOSE);
2707	mptcp_for_each_subflow_safe(msk, subflow, tmp)
2708	__mptcp_close_ssk(sk, ssk: mptcp_subflow_tcp_sock(subflow),
2709	subflow, MPTCP_CF_FASTCLOSE);
2710	}
2711
2712	static void mptcp_worker(struct work_struct *work)
2713	{
2714	struct mptcp_sock *msk = container_of(work, struct mptcp_sock, work);
2715	struct sock *sk = (struct sock *)msk;
2716	unsigned long fail_tout;
2717	int state;
2718
2719	lock_sock(sk);
2720	state = sk->sk_state;
2721	if (unlikely((1 << state) & (TCPF_CLOSE | TCPF_LISTEN)))
2722	goto unlock;
2723
2724	mptcp_check_fastclose(msk);
2725
2726	mptcp_pm_nl_work(msk);
2727
2728	mptcp_check_send_data_fin(sk);
2729	mptcp_check_data_fin_ack(sk);
2730	mptcp_check_data_fin(sk);
2731
2732	if (test_and_clear_bit(MPTCP_WORK_CLOSE_SUBFLOW, addr: &msk->flags))
2733	__mptcp_close_subflow(sk);
2734
2735	if (mptcp_close_tout_expired(sk)) {
2736	mptcp_do_fastclose(sk);
2737	mptcp_close_wake_up(sk);
2738	}
2739
2740	if (sock_flag(sk, flag: SOCK_DEAD) && sk->sk_state == TCP_CLOSE) {
2741	__mptcp_destroy_sock(sk);
2742	goto unlock;
2743	}
2744
2745	if (test_and_clear_bit(MPTCP_WORK_RTX, addr: &msk->flags))
2746	__mptcp_retrans(sk);
2747
2748	fail_tout = msk->first ? READ_ONCE(mptcp_subflow_ctx(msk->first)->fail_tout) : 0;
2749	if (fail_tout && time_after(jiffies, fail_tout))
2750	mptcp_mp_fail_no_response(msk);
2751
2752	unlock:
2753	release_sock(sk);
2754	sock_put(sk);
2755	}
2756
2757	static void __mptcp_init_sock(struct sock *sk)
2758	{
2759	struct mptcp_sock *msk = mptcp_sk(sk);
2760
2761	INIT_LIST_HEAD(list: &msk->conn_list);
2762	INIT_LIST_HEAD(list: &msk->join_list);
2763	INIT_LIST_HEAD(list: &msk->rtx_queue);
2764	INIT_WORK(&msk->work, mptcp_worker);
2765	__skb_queue_head_init(list: &msk->receive_queue);
2766	msk->out_of_order_queue = RB_ROOT;
2767	msk->first_pending = NULL;
2768	msk->rmem_fwd_alloc = 0;
2769	WRITE_ONCE(msk->rmem_released, 0);
2770	msk->timer_ival = TCP_RTO_MIN;
2771	msk->scaling_ratio = TCP_DEFAULT_SCALING_RATIO;
2772
2773	WRITE_ONCE(msk->first, NULL);
2774	inet_csk(sk)->icsk_sync_mss = mptcp_sync_mss;
2775	WRITE_ONCE(msk->csum_enabled, mptcp_is_checksum_enabled(sock_net(sk)));
2776	WRITE_ONCE(msk->allow_infinite_fallback, true);
2777	msk->recovery = false;
2778	msk->subflow_id = 1;
2779
2780	mptcp_pm_data_init(msk);
2781
2782	/* re-use the csk retrans timer for MPTCP-level retrans */
2783	timer_setup(&msk->sk.icsk_retransmit_timer, mptcp_retransmit_timer, 0);
2784	timer_setup(&sk->sk_timer, mptcp_tout_timer, 0);
2785	}
2786
2787	static void mptcp_ca_reset(struct sock *sk)
2788	{
2789	struct inet_connection_sock *icsk = inet_csk(sk);
2790
2791	tcp_assign_congestion_control(sk);
2792	strcpy(mptcp_sk(sk)->ca_name, q: icsk->icsk_ca_ops->name);
2793
2794	/* no need to keep a reference to the ops, the name will suffice */
2795	tcp_cleanup_congestion_control(sk);
2796	icsk->icsk_ca_ops = NULL;
2797	}
2798
2799	static int mptcp_init_sock(struct sock *sk)
2800	{
2801	struct net *net = sock_net(sk);
2802	int ret;
2803
2804	__mptcp_init_sock(sk);
2805
2806	if (!mptcp_is_enabled(net))
2807	return -ENOPROTOOPT;
2808
2809	if (unlikely(!net->mib.mptcp_statistics) && !mptcp_mib_alloc(net))
2810	return -ENOMEM;
2811
2812	ret = mptcp_init_sched(mptcp_sk(sk),
2813	sched: mptcp_sched_find(name: mptcp_get_scheduler(net)));
2814	if (ret)
2815	return ret;
2816
2817	set_bit(SOCK_CUSTOM_SOCKOPT, addr: &sk->sk_socket->flags);
2818
2819	/* fetch the ca name; do it outside __mptcp_init_sock(), so that clone will
2820	* propagate the correct value
2821	*/
2822	mptcp_ca_reset(sk);
2823
2824	sk_sockets_allocated_inc(sk);
2825	sk->sk_rcvbuf = READ_ONCE(net->ipv4.sysctl_tcp_rmem[1]);
2826	sk->sk_sndbuf = READ_ONCE(net->ipv4.sysctl_tcp_wmem[1]);
2827
2828	return 0;
2829	}
2830
2831	static void __mptcp_clear_xmit(struct sock *sk)
2832	{
2833	struct mptcp_sock *msk = mptcp_sk(sk);
2834	struct mptcp_data_frag *dtmp, *dfrag;
2835
2836	WRITE_ONCE(msk->first_pending, NULL);
2837	list_for_each_entry_safe(dfrag, dtmp, &msk->rtx_queue, list)
2838	dfrag_clear(sk, dfrag);
2839	}
2840
2841	void mptcp_cancel_work(struct sock *sk)
2842	{
2843	struct mptcp_sock *msk = mptcp_sk(sk);
2844
2845	if (cancel_work_sync(work: &msk->work))
2846	__sock_put(sk);
2847	}
2848
2849	void mptcp_subflow_shutdown(struct sock *sk, struct sock *ssk, int how)
2850	{
2851	lock_sock(sk: ssk);
2852
2853	switch (ssk->sk_state) {
2854	case TCP_LISTEN:
2855	if (!(how & RCV_SHUTDOWN))
2856	break;
2857	fallthrough;
2858	case TCP_SYN_SENT:
2859	WARN_ON_ONCE(tcp_disconnect(ssk, O_NONBLOCK));
2860	break;
2861	default:
2862	if (__mptcp_check_fallback(mptcp_sk(sk))) {
2863	pr_debug("Fallback");
2864	ssk->sk_shutdown |= how;
2865	tcp_shutdown(sk: ssk, how);
2866
2867	/* simulate the data_fin ack reception to let the state
2868	* machine move forward
2869	*/
2870	WRITE_ONCE(mptcp_sk(sk)->snd_una, mptcp_sk(sk)->snd_nxt);
2871	mptcp_schedule_work(sk);
2872	} else {
2873	pr_debug("Sending DATA_FIN on subflow %p", ssk);
2874	tcp_send_ack(sk: ssk);
2875	if (!mptcp_rtx_timer_pending(sk))
2876	mptcp_reset_rtx_timer(sk);
2877	}
2878	break;
2879	}
2880
2881	release_sock(sk: ssk);
2882	}
2883
2884	static const unsigned char new_state[16] = {
2885	/* current state: new state: action: */
2886	[0 /* (Invalid) */] = TCP_CLOSE,
2887	[TCP_ESTABLISHED] = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2888	[TCP_SYN_SENT] = TCP_CLOSE,
2889	[TCP_SYN_RECV] = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2890	[TCP_FIN_WAIT1] = TCP_FIN_WAIT1,
2891	[TCP_FIN_WAIT2] = TCP_FIN_WAIT2,
2892	[TCP_TIME_WAIT] = TCP_CLOSE, /* should not happen ! */
2893	[TCP_CLOSE] = TCP_CLOSE,
2894	[TCP_CLOSE_WAIT] = TCP_LAST_ACK | TCP_ACTION_FIN,
2895	[TCP_LAST_ACK] = TCP_LAST_ACK,
2896	[TCP_LISTEN] = TCP_CLOSE,
2897	[TCP_CLOSING] = TCP_CLOSING,
2898	[TCP_NEW_SYN_RECV] = TCP_CLOSE, /* should not happen ! */
2899	};
2900
2901	static int mptcp_close_state(struct sock *sk)
2902	{
2903	int next = (int)new_state[sk->sk_state];
2904	int ns = next & TCP_STATE_MASK;
2905
2906	inet_sk_state_store(sk, newstate: ns);
2907
2908	return next & TCP_ACTION_FIN;
2909	}
2910
2911	static void mptcp_check_send_data_fin(struct sock *sk)
2912	{
2913	struct mptcp_subflow_context *subflow;
2914	struct mptcp_sock *msk = mptcp_sk(sk);
2915
2916	pr_debug("msk=%p snd_data_fin_enable=%d pending=%d snd_nxt=%llu write_seq=%llu",
2917	msk, msk->snd_data_fin_enable, !!mptcp_send_head(sk),
2918	msk->snd_nxt, msk->write_seq);
2919
2920	/* we still need to enqueue subflows or not really shutting down,
2921	* skip this
2922	*/
2923	if (!msk->snd_data_fin_enable || msk->snd_nxt + 1 != msk->write_seq ||
2924	mptcp_send_head(sk))
2925	return;
2926
2927	WRITE_ONCE(msk->snd_nxt, msk->write_seq);
2928
2929	mptcp_for_each_subflow(msk, subflow) {
2930	struct sock *tcp_sk = mptcp_subflow_tcp_sock(subflow);
2931
2932	mptcp_subflow_shutdown(sk, ssk: tcp_sk, SEND_SHUTDOWN);
2933	}
2934	}
2935
2936	static void __mptcp_wr_shutdown(struct sock *sk)
2937	{
2938	struct mptcp_sock *msk = mptcp_sk(sk);
2939
2940	pr_debug("msk=%p snd_data_fin_enable=%d shutdown=%x state=%d pending=%d",
2941	msk, msk->snd_data_fin_enable, sk->sk_shutdown, sk->sk_state,
2942	!!mptcp_send_head(sk));
2943
2944	/* will be ignored by fallback sockets */
2945	WRITE_ONCE(msk->write_seq, msk->write_seq + 1);
2946	WRITE_ONCE(msk->snd_data_fin_enable, 1);
2947
2948	mptcp_check_send_data_fin(sk);
2949	}
2950
2951	static void __mptcp_destroy_sock(struct sock *sk)
2952	{
2953	struct mptcp_sock *msk = mptcp_sk(sk);
2954
2955	pr_debug("msk=%p", msk);
2956
2957	might_sleep();
2958
2959	mptcp_stop_rtx_timer(sk);
2960	sk_stop_timer(sk, timer: &sk->sk_timer);
2961	msk->pm.status = 0;
2962	mptcp_release_sched(msk);
2963
2964	sk->sk_prot->destroy(sk);
2965
2966	WARN_ON_ONCE(msk->rmem_fwd_alloc);
2967	WARN_ON_ONCE(msk->rmem_released);
2968	sk_stream_kill_queues(sk);
2969	xfrm_sk_free_policy(sk);
2970
2971	sock_put(sk);
2972	}
2973
2974	void __mptcp_unaccepted_force_close(struct sock *sk)
2975	{
2976	sock_set_flag(sk, flag: SOCK_DEAD);
2977	mptcp_do_fastclose(sk);
2978	__mptcp_destroy_sock(sk);
2979	}
2980
2981	static __poll_t mptcp_check_readable(struct sock *sk)
2982	{
2983	return mptcp_epollin_ready(sk) ? EPOLLIN | EPOLLRDNORM : 0;
2984	}
2985
2986	static void mptcp_check_listen_stop(struct sock *sk)
2987	{
2988	struct sock *ssk;
2989
2990	if (inet_sk_state_load(sk) != TCP_LISTEN)
2991	return;
2992
2993	sock_prot_inuse_add(net: sock_net(sk), prot: sk->sk_prot, val: -1);
2994	ssk = mptcp_sk(sk)->first;
2995	if (WARN_ON_ONCE(!ssk || inet_sk_state_load(ssk) != TCP_LISTEN))
2996	return;
2997
2998	lock_sock_nested(sk: ssk, SINGLE_DEPTH_NESTING);
2999	tcp_set_state(sk: ssk, state: TCP_CLOSE);
3000	mptcp_subflow_queue_clean(sk, ssk);
3001	inet_csk_listen_stop(sk: ssk);
3002	mptcp_event_pm_listener(ssk, event: MPTCP_EVENT_LISTENER_CLOSED);
3003	release_sock(sk: ssk);
3004	}
3005
3006	bool __mptcp_close(struct sock *sk, long timeout)
3007	{
3008	struct mptcp_subflow_context *subflow;
3009	struct mptcp_sock *msk = mptcp_sk(sk);
3010	bool do_cancel_work = false;
3011	int subflows_alive = 0;
3012
3013	WRITE_ONCE(sk->sk_shutdown, SHUTDOWN_MASK);
3014
3015	if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE)) {
3016	mptcp_check_listen_stop(sk);
3017	inet_sk_state_store(sk, newstate: TCP_CLOSE);
3018	goto cleanup;
3019	}
3020
3021	if (mptcp_data_avail(msk) || timeout < 0) {
3022	/* If the msk has read data, or the caller explicitly ask it,
3023	* do the MPTCP equivalent of TCP reset, aka MPTCP fastclose
3024	*/
3025	mptcp_do_fastclose(sk);
3026	timeout = 0;
3027	} else if (mptcp_close_state(sk)) {
3028	__mptcp_wr_shutdown(sk);
3029	}
3030
3031	sk_stream_wait_close(sk, timeo_p: timeout);
3032
3033	cleanup:
3034	/* orphan all the subflows */
3035	mptcp_for_each_subflow(msk, subflow) {
3036	struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
3037	bool slow = lock_sock_fast_nested(sk: ssk);
3038
3039	subflows_alive += ssk->sk_state != TCP_CLOSE;
3040
3041	/* since the close timeout takes precedence on the fail one,
3042	* cancel the latter
3043	*/
3044	if (ssk == msk->first)
3045	subflow->fail_tout = 0;
3046
3047	/* detach from the parent socket, but allow data_ready to
3048	* push incoming data into the mptcp stack, to properly ack it
3049	*/
3050	ssk->sk_socket = NULL;
3051	ssk->sk_wq = NULL;
3052	unlock_sock_fast(sk: ssk, slow);
3053	}
3054	sock_orphan(sk);
3055
3056	/* all the subflows are closed, only timeout can change the msk
3057	* state, let's not keep resources busy for no reasons
3058	*/
3059	if (subflows_alive == 0)
3060	inet_sk_state_store(sk, newstate: TCP_CLOSE);
3061
3062	sock_hold(sk);
3063	pr_debug("msk=%p state=%d", sk, sk->sk_state);
3064	if (msk->token)
3065	mptcp_event(type: MPTCP_EVENT_CLOSED, msk, NULL, GFP_KERNEL);
3066
3067	if (sk->sk_state == TCP_CLOSE) {
3068	__mptcp_destroy_sock(sk);
3069	do_cancel_work = true;
3070	} else {
3071	mptcp_start_tout_timer(sk);
3072	}
3073
3074	return do_cancel_work;
3075	}
3076
3077	static void mptcp_close(struct sock *sk, long timeout)
3078	{
3079	bool do_cancel_work;
3080
3081	lock_sock(sk);
3082
3083	do_cancel_work = __mptcp_close(sk, timeout);
3084	release_sock(sk);
3085	if (do_cancel_work)
3086	mptcp_cancel_work(sk);
3087
3088	sock_put(sk);
3089	}
3090
3091	static void mptcp_copy_inaddrs(struct sock *msk, const struct sock *ssk)
3092	{
3093	#if IS_ENABLED(CONFIG_MPTCP_IPV6)
3094	const struct ipv6_pinfo *ssk6 = inet6_sk(sk: ssk);
3095	struct ipv6_pinfo *msk6 = inet6_sk(sk: msk);
3096
3097	msk->sk_v6_daddr = ssk->sk_v6_daddr;
3098	msk->sk_v6_rcv_saddr = ssk->sk_v6_rcv_saddr;
3099
3100	if (msk6 && ssk6) {
3101	msk6->saddr = ssk6->saddr;
3102	msk6->flow_label = ssk6->flow_label;
3103	}
3104	#endif
3105
3106	inet_sk(msk)->inet_num = inet_sk(ssk)->inet_num;
3107	inet_sk(msk)->inet_dport = inet_sk(ssk)->inet_dport;
3108	inet_sk(msk)->inet_sport = inet_sk(ssk)->inet_sport;
3109	inet_sk(msk)->inet_daddr = inet_sk(ssk)->inet_daddr;
3110	inet_sk(msk)->inet_saddr = inet_sk(ssk)->inet_saddr;
3111	inet_sk(msk)->inet_rcv_saddr = inet_sk(ssk)->inet_rcv_saddr;
3112	}
3113
3114	static int mptcp_disconnect(struct sock *sk, int flags)
3115	{
3116	struct mptcp_sock *msk = mptcp_sk(sk);
3117
3118	/* We are on the fastopen error path. We can't call straight into the
3119	* subflows cleanup code due to lock nesting (we are already under
3120	* msk->firstsocket lock).
3121	*/
3122	if (msk->fastopening)
3123	return -EBUSY;
3124
3125	mptcp_check_listen_stop(sk);
3126	inet_sk_state_store(sk, newstate: TCP_CLOSE);
3127
3128	mptcp_stop_rtx_timer(sk);
3129	mptcp_stop_tout_timer(sk);
3130
3131	if (msk->token)
3132	mptcp_event(type: MPTCP_EVENT_CLOSED, msk, NULL, GFP_KERNEL);
3133
3134	/* msk->subflow is still intact, the following will not free the first
3135	* subflow
3136	*/
3137	mptcp_destroy_common(msk, MPTCP_CF_FASTCLOSE);
3138	WRITE_ONCE(msk->flags, 0);
3139	msk->cb_flags = 0;
3140	msk->push_pending = 0;
3141	msk->recovery = false;
3142	msk->can_ack = false;
3143	msk->fully_established = false;
3144	msk->rcv_data_fin = false;
3145	msk->snd_data_fin_enable = false;
3146	msk->rcv_fastclose = false;
3147	msk->use_64bit_ack = false;
3148	msk->bytes_consumed = 0;
3149	WRITE_ONCE(msk->csum_enabled, mptcp_is_checksum_enabled(sock_net(sk)));
3150	mptcp_pm_data_reset(msk);
3151	mptcp_ca_reset(sk);
3152	msk->bytes_acked = 0;
3153	msk->bytes_received = 0;
3154	msk->bytes_sent = 0;
3155	msk->bytes_retrans = 0;
3156
3157	WRITE_ONCE(sk->sk_shutdown, 0);
3158	sk_error_report(sk);
3159	return 0;
3160	}
3161
3162	#if IS_ENABLED(CONFIG_MPTCP_IPV6)
3163	static struct ipv6_pinfo *mptcp_inet6_sk(const struct sock *sk)
3164	{
3165	unsigned int offset = sizeof(struct mptcp6_sock) - sizeof(struct ipv6_pinfo);
3166
3167	return (struct ipv6_pinfo *)(((u8 *)sk) + offset);
3168	}
3169	#endif
3170
3171	struct sock *mptcp_sk_clone_init(const struct sock *sk,
3172	const struct mptcp_options_received *mp_opt,
3173	struct sock *ssk,
3174	struct request_sock *req)
3175	{
3176	struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(rsk: req);
3177	struct sock *nsk = sk_clone_lock(sk, GFP_ATOMIC);
3178	struct mptcp_sock *msk;
3179
3180	if (!nsk)
3181	return NULL;
3182
3183	#if IS_ENABLED(CONFIG_MPTCP_IPV6)
3184	if (nsk->sk_family == AF_INET6)
3185	inet_sk(nsk)->pinet6 = mptcp_inet6_sk(sk: nsk);
3186	#endif
3187
3188	__mptcp_init_sock(sk: nsk);
3189
3190	msk = mptcp_sk(nsk);
3191	msk->local_key = subflow_req->local_key;
3192	msk->token = subflow_req->token;
3193	msk->in_accept_queue = 1;
3194	WRITE_ONCE(msk->fully_established, false);
3195	if (mp_opt->suboptions & OPTION_MPTCP_CSUMREQD)
3196	WRITE_ONCE(msk->csum_enabled, true);
3197
3198	msk->write_seq = subflow_req->idsn + 1;
3199	msk->snd_nxt = msk->write_seq;
3200	msk->snd_una = msk->write_seq;
3201	msk->wnd_end = msk->snd_nxt + req->rsk_rcv_wnd;
3202	msk->setsockopt_seq = mptcp_sk(sk)->setsockopt_seq;
3203	mptcp_init_sched(msk, mptcp_sk(sk)->sched);
3204
3205	/* passive msk is created after the first/MPC subflow */
3206	msk->subflow_id = 2;
3207
3208	sock_reset_flag(sk: nsk, flag: SOCK_RCU_FREE);
3209	security_inet_csk_clone(newsk: nsk, req);
3210
3211	/* this can't race with mptcp_close(), as the msk is
3212	* not yet exposted to user-space
3213	*/
3214	inet_sk_state_store(sk: nsk, newstate: TCP_ESTABLISHED);
3215
3216	/* The msk maintain a ref to each subflow in the connections list */
3217	WRITE_ONCE(msk->first, ssk);
3218	list_add(new: &mptcp_subflow_ctx(sk: ssk)->node, head: &msk->conn_list);
3219	sock_hold(sk: ssk);
3220
3221	/* new mpc subflow takes ownership of the newly
3222	* created mptcp socket
3223	*/
3224	mptcp_token_accept(r: subflow_req, msk);
3225
3226	/* set msk addresses early to ensure mptcp_pm_get_local_id()
3227	* uses the correct data
3228	*/
3229	mptcp_copy_inaddrs(msk: nsk, ssk);
3230	__mptcp_propagate_sndbuf(sk: nsk, ssk);
3231
3232	mptcp_rcv_space_init(msk, ssk);
3233	bh_unlock_sock(nsk);
3234
3235	/* note: the newly allocated socket refcount is 2 now */
3236	return nsk;
3237	}
3238
3239	void mptcp_rcv_space_init(struct mptcp_sock *msk, const struct sock *ssk)
3240	{
3241	const struct tcp_sock *tp = tcp_sk(ssk);
3242
3243	msk->rcvq_space.copied = 0;
3244	msk->rcvq_space.rtt_us = 0;
3245
3246	msk->rcvq_space.time = tp->tcp_mstamp;
3247
3248	/* initial rcv_space offering made to peer */
3249	msk->rcvq_space.space = min_t(u32, tp->rcv_wnd,
3250	TCP_INIT_CWND * tp->advmss);
3251	if (msk->rcvq_space.space == 0)
3252	msk->rcvq_space.space = TCP_INIT_CWND * TCP_MSS_DEFAULT;
3253
3254	WRITE_ONCE(msk->wnd_end, msk->snd_nxt + tcp_sk(ssk)->snd_wnd);
3255	}
3256
3257	static struct sock *mptcp_accept(struct sock *ssk, int flags, int *err,
3258	bool kern)
3259	{
3260	struct sock *newsk;
3261
3262	pr_debug("ssk=%p, listener=%p", ssk, mptcp_subflow_ctx(ssk));
3263	newsk = inet_csk_accept(sk: ssk, flags, err, kern);
3264	if (!newsk)
3265	return NULL;
3266
3267	pr_debug("newsk=%p, subflow is mptcp=%d", newsk, sk_is_mptcp(newsk));
3268	if (sk_is_mptcp(sk: newsk)) {
3269	struct mptcp_subflow_context *subflow;
3270	struct sock *new_mptcp_sock;
3271
3272	subflow = mptcp_subflow_ctx(sk: newsk);
3273	new_mptcp_sock = subflow->conn;
3274
3275	/* is_mptcp should be false if subflow->conn is missing, see
3276	* subflow_syn_recv_sock()
3277	*/
3278	if (WARN_ON_ONCE(!new_mptcp_sock)) {
3279	tcp_sk(newsk)->is_mptcp = 0;
3280	goto out;
3281	}
3282
3283	newsk = new_mptcp_sock;
3284	MPTCP_INC_STATS(net: sock_net(sk: ssk), field: MPTCP_MIB_MPCAPABLEPASSIVEACK);
3285	} else {
3286	MPTCP_INC_STATS(net: sock_net(sk: ssk),
3287	field: MPTCP_MIB_MPCAPABLEPASSIVEFALLBACK);
3288	}
3289
3290	out:
3291	newsk->sk_kern_sock = kern;
3292	return newsk;
3293	}
3294
3295	void mptcp_destroy_common(struct mptcp_sock *msk, unsigned int flags)
3296	{
3297	struct mptcp_subflow_context *subflow, *tmp;
3298	struct sock *sk = (struct sock *)msk;
3299
3300	__mptcp_clear_xmit(sk);
3301
3302	/* join list will be eventually flushed (with rst) at sock lock release time */
3303	mptcp_for_each_subflow_safe(msk, subflow, tmp)
3304	__mptcp_close_ssk(sk, ssk: mptcp_subflow_tcp_sock(subflow), subflow, flags);
3305
3306	/* move to sk_receive_queue, sk_stream_kill_queues will purge it */
3307	mptcp_data_lock(sk);
3308	skb_queue_splice_tail_init(list: &msk->receive_queue, head: &sk->sk_receive_queue);
3309	__skb_queue_purge(list: &sk->sk_receive_queue);
3310	skb_rbtree_purge(root: &msk->out_of_order_queue);
3311	mptcp_data_unlock(sk);
3312
3313	/* move all the rx fwd alloc into the sk_mem_reclaim_final in
3314	* inet_sock_destruct() will dispose it
3315	*/
3316	sk_forward_alloc_add(sk, val: msk->rmem_fwd_alloc);
3317	WRITE_ONCE(msk->rmem_fwd_alloc, 0);
3318	mptcp_token_destroy(msk);
3319	mptcp_pm_free_anno_list(msk);
3320	mptcp_free_local_addr_list(msk);
3321	}
3322
3323	static void mptcp_destroy(struct sock *sk)
3324	{
3325	struct mptcp_sock *msk = mptcp_sk(sk);
3326
3327	/* allow the following to close even the initial subflow */
3328	msk->free_first = 1;
3329	mptcp_destroy_common(msk, flags: 0);
3330	sk_sockets_allocated_dec(sk);
3331	}
3332
3333	void __mptcp_data_acked(struct sock *sk)
3334	{
3335	if (!sock_owned_by_user(sk))
3336	__mptcp_clean_una(sk);
3337	else
3338	__set_bit(MPTCP_CLEAN_UNA, &mptcp_sk(sk)->cb_flags);
3339
3340	if (mptcp_pending_data_fin_ack(sk))
3341	mptcp_schedule_work(sk);
3342	}
3343
3344	void __mptcp_check_push(struct sock *sk, struct sock *ssk)
3345	{
3346	if (!mptcp_send_head(sk))
3347	return;
3348
3349	if (!sock_owned_by_user(sk))
3350	__mptcp_subflow_push_pending(sk, ssk, first: false);
3351	else
3352	__set_bit(MPTCP_PUSH_PENDING, &mptcp_sk(sk)->cb_flags);
3353	}
3354
3355	#define MPTCP_FLAGS_PROCESS_CTX_NEED (BIT(MPTCP_PUSH_PENDING) | \
3356	BIT(MPTCP_RETRANSMIT) | \
3357	BIT(MPTCP_FLUSH_JOIN_LIST))
3358
3359	/* processes deferred events and flush wmem */
3360	static void mptcp_release_cb(struct sock *sk)
3361	__must_hold(&sk->sk_lock.slock)
3362	{
3363	struct mptcp_sock *msk = mptcp_sk(sk);
3364
3365	for (;;) {
3366	unsigned long flags = (msk->cb_flags & MPTCP_FLAGS_PROCESS_CTX_NEED) |
3367	msk->push_pending;
3368	struct list_head join_list;
3369
3370	if (!flags)
3371	break;
3372
3373	INIT_LIST_HEAD(list: &join_list);
3374	list_splice_init(list: &msk->join_list, head: &join_list);
3375
3376	/* the following actions acquire the subflow socket lock
3377	*
3378	* 1) can't be invoked in atomic scope
3379	* 2) must avoid ABBA deadlock with msk socket spinlock: the RX
3380	* datapath acquires the msk socket spinlock while helding
3381	* the subflow socket lock
3382	*/
3383	msk->push_pending = 0;
3384	msk->cb_flags &= ~flags;
3385	spin_unlock_bh(lock: &sk->sk_lock.slock);
3386
3387	if (flags & BIT(MPTCP_FLUSH_JOIN_LIST))
3388	__mptcp_flush_join_list(sk, join_list: &join_list);
3389	if (flags & BIT(MPTCP_PUSH_PENDING))
3390	__mptcp_push_pending(sk, flags: 0);
3391	if (flags & BIT(MPTCP_RETRANSMIT))
3392	__mptcp_retrans(sk);
3393
3394	cond_resched();
3395	spin_lock_bh(lock: &sk->sk_lock.slock);
3396	}
3397
3398	if (__test_and_clear_bit(MPTCP_CLEAN_UNA, &msk->cb_flags))
3399	__mptcp_clean_una_wakeup(sk);
3400	if (unlikely(msk->cb_flags)) {
3401	/* be sure to set the current sk state before tacking actions
3402	* depending on sk_state, that is processing MPTCP_ERROR_REPORT
3403	*/
3404	if (__test_and_clear_bit(MPTCP_CONNECTED, &msk->cb_flags))
3405	__mptcp_set_connected(sk);
3406	if (__test_and_clear_bit(MPTCP_ERROR_REPORT, &msk->cb_flags))
3407	__mptcp_error_report(sk);
3408	if (__test_and_clear_bit(MPTCP_SYNC_SNDBUF, &msk->cb_flags))
3409	__mptcp_sync_sndbuf(sk);
3410	}
3411
3412	__mptcp_update_rmem(sk);
3413	}
3414
3415	/* MP_JOIN client subflow must wait for 4th ack before sending any data:
3416	* TCP can't schedule delack timer before the subflow is fully established.
3417	* MPTCP uses the delack timer to do 3rd ack retransmissions
3418	*/
3419	static void schedule_3rdack_retransmission(struct sock *ssk)
3420	{
3421	struct inet_connection_sock *icsk = inet_csk(sk: ssk);
3422	struct tcp_sock *tp = tcp_sk(ssk);
3423	unsigned long timeout;
3424
3425	if (mptcp_subflow_ctx(sk: ssk)->fully_established)
3426	return;
3427
3428	/* reschedule with a timeout above RTT, as we must look only for drop */
3429	if (tp->srtt_us)
3430	timeout = usecs_to_jiffies(u: tp->srtt_us >> (3 - 1));
3431	else
3432	timeout = TCP_TIMEOUT_INIT;
3433	timeout += jiffies;
3434
3435	WARN_ON_ONCE(icsk->icsk_ack.pending & ICSK_ACK_TIMER);
3436	icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
3437	icsk->icsk_ack.timeout = timeout;
3438	sk_reset_timer(sk: ssk, timer: &icsk->icsk_delack_timer, expires: timeout);
3439	}
3440
3441	void mptcp_subflow_process_delegated(struct sock *ssk, long status)
3442	{
3443	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk: ssk);
3444	struct sock *sk = subflow->conn;
3445
3446	if (status & BIT(MPTCP_DELEGATE_SEND)) {
3447	mptcp_data_lock(sk);
3448	if (!sock_owned_by_user(sk))
3449	__mptcp_subflow_push_pending(sk, ssk, first: true);
3450	else
3451	__set_bit(MPTCP_PUSH_PENDING, &mptcp_sk(sk)->cb_flags);
3452	mptcp_data_unlock(sk);
3453	}
3454	if (status & BIT(MPTCP_DELEGATE_SNDBUF)) {
3455	mptcp_data_lock(sk);
3456	if (!sock_owned_by_user(sk))
3457	__mptcp_sync_sndbuf(sk);
3458	else
3459	__set_bit(MPTCP_SYNC_SNDBUF, &mptcp_sk(sk)->cb_flags);
3460	mptcp_data_unlock(sk);
3461	}
3462	if (status & BIT(MPTCP_DELEGATE_ACK))
3463	schedule_3rdack_retransmission(ssk);
3464	}
3465
3466	static int mptcp_hash(struct sock *sk)
3467	{
3468	/* should never be called,
3469	* we hash the TCP subflows not the master socket
3470	*/
3471	WARN_ON_ONCE(1);
3472	return 0;
3473	}
3474
3475	static void mptcp_unhash(struct sock *sk)
3476	{
3477	/* called from sk_common_release(), but nothing to do here */
3478	}
3479
3480	static int mptcp_get_port(struct sock *sk, unsigned short snum)
3481	{
3482	struct mptcp_sock *msk = mptcp_sk(sk);
3483
3484	pr_debug("msk=%p, ssk=%p", msk, msk->first);
3485	if (WARN_ON_ONCE(!msk->first))
3486	return -EINVAL;
3487
3488	return inet_csk_get_port(sk: msk->first, snum);
3489	}
3490
3491	void mptcp_finish_connect(struct sock *ssk)
3492	{
3493	struct mptcp_subflow_context *subflow;
3494	struct mptcp_sock *msk;
3495	struct sock *sk;
3496
3497	subflow = mptcp_subflow_ctx(sk: ssk);
3498	sk = subflow->conn;
3499	msk = mptcp_sk(sk);
3500
3501	pr_debug("msk=%p, token=%u", sk, subflow->token);
3502
3503	subflow->map_seq = subflow->iasn;
3504	subflow->map_subflow_seq = 1;
3505
3506	/* the socket is not connected yet, no msk/subflow ops can access/race
3507	* accessing the field below
3508	*/
3509	WRITE_ONCE(msk->local_key, subflow->local_key);
3510	WRITE_ONCE(msk->write_seq, subflow->idsn + 1);
3511	WRITE_ONCE(msk->snd_nxt, msk->write_seq);
3512	WRITE_ONCE(msk->snd_una, msk->write_seq);
3513
3514	mptcp_pm_new_connection(msk, ssk, server_side: 0);
3515
3516	mptcp_rcv_space_init(msk, ssk);
3517	}
3518
3519	void mptcp_sock_graft(struct sock *sk, struct socket *parent)
3520	{
3521	write_lock_bh(&sk->sk_callback_lock);
3522	rcu_assign_pointer(sk->sk_wq, &parent->wq);
3523	sk_set_socket(sk, sock: parent);
3524	sk->sk_uid = SOCK_INODE(socket: parent)->i_uid;
3525	write_unlock_bh(&sk->sk_callback_lock);
3526	}
3527
3528	bool mptcp_finish_join(struct sock *ssk)
3529	{
3530	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk: ssk);
3531	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
3532	struct sock *parent = (void *)msk;
3533	bool ret = true;
3534
3535	pr_debug("msk=%p, subflow=%p", msk, subflow);
3536
3537	/* mptcp socket already closing? */
3538	if (!mptcp_is_fully_established(sk: parent)) {
3539	subflow->reset_reason = MPTCP_RST_EMPTCP;
3540	return false;
3541	}
3542
3543	/* active subflow, already present inside the conn_list */
3544	if (!list_empty(head: &subflow->node)) {
3545	mptcp_subflow_joined(msk, ssk);
3546	mptcp_propagate_sndbuf(sk: parent, ssk);
3547	return true;
3548	}
3549
3550	if (!mptcp_pm_allow_new_subflow(msk))
3551	goto err_prohibited;
3552
3553	/* If we can't acquire msk socket lock here, let the release callback
3554	* handle it
3555	*/
3556	mptcp_data_lock(parent);
3557	if (!sock_owned_by_user(sk: parent)) {
3558	ret = __mptcp_finish_join(msk, ssk);
3559	if (ret) {
3560	sock_hold(sk: ssk);
3561	list_add_tail(new: &subflow->node, head: &msk->conn_list);
3562	}
3563	} else {
3564	sock_hold(sk: ssk);
3565	list_add_tail(new: &subflow->node, head: &msk->join_list);
3566	__set_bit(MPTCP_FLUSH_JOIN_LIST, &msk->cb_flags);
3567	}
3568	mptcp_data_unlock(parent);
3569
3570	if (!ret) {
3571	err_prohibited:
3572	subflow->reset_reason = MPTCP_RST_EPROHIBIT;
3573	return false;
3574	}
3575
3576	return true;
3577	}
3578
3579	static void mptcp_shutdown(struct sock *sk, int how)
3580	{
3581	pr_debug("sk=%p, how=%d", sk, how);
3582
3583	if ((how & SEND_SHUTDOWN) && mptcp_close_state(sk))
3584	__mptcp_wr_shutdown(sk);
3585	}
3586
3587	static int mptcp_forward_alloc_get(const struct sock *sk)
3588	{
3589	return READ_ONCE(sk->sk_forward_alloc) +
3590	READ_ONCE(mptcp_sk(sk)->rmem_fwd_alloc);
3591	}
3592
3593	static int mptcp_ioctl_outq(const struct mptcp_sock *msk, u64 v)
3594	{
3595	const struct sock *sk = (void *)msk;
3596	u64 delta;
3597
3598	if (sk->sk_state == TCP_LISTEN)
3599	return -EINVAL;
3600
3601	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
3602	return 0;
3603
3604	delta = msk->write_seq - v;
3605	if (__mptcp_check_fallback(msk) && msk->first) {
3606	struct tcp_sock *tp = tcp_sk(msk->first);
3607
3608	/* the first subflow is disconnected after close - see
3609	* __mptcp_close_ssk(). tcp_disconnect() moves the write_seq
3610	* so ignore that status, too.
3611	*/
3612	if (!((1 << msk->first->sk_state) &
3613	(TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE)))
3614	delta += READ_ONCE(tp->write_seq) - tp->snd_una;
3615	}
3616	if (delta > INT_MAX)
3617	delta = INT_MAX;
3618
3619	return (int)delta;
3620	}
3621
3622	static int mptcp_ioctl(struct sock *sk, int cmd, int *karg)
3623	{
3624	struct mptcp_sock *msk = mptcp_sk(sk);
3625	bool slow;
3626
3627	switch (cmd) {
3628	case SIOCINQ:
3629	if (sk->sk_state == TCP_LISTEN)
3630	return -EINVAL;
3631
3632	lock_sock(sk);
3633	__mptcp_move_skbs(msk);
3634	*karg = mptcp_inq_hint(sk);
3635	release_sock(sk);
3636	break;
3637	case SIOCOUTQ:
3638	slow = lock_sock_fast(sk);
3639	*karg = mptcp_ioctl_outq(msk, READ_ONCE(msk->snd_una));
3640	unlock_sock_fast(sk, slow);
3641	break;
3642	case SIOCOUTQNSD:
3643	slow = lock_sock_fast(sk);
3644	*karg = mptcp_ioctl_outq(msk, v: msk->snd_nxt);
3645	unlock_sock_fast(sk, slow);
3646	break;
3647	default:
3648	return -ENOIOCTLCMD;
3649	}
3650
3651	return 0;
3652	}
3653
3654	static void mptcp_subflow_early_fallback(struct mptcp_sock *msk,
3655	struct mptcp_subflow_context *subflow)
3656	{
3657	subflow->request_mptcp = 0;
3658	__mptcp_do_fallback(msk);
3659	}
3660
3661	static int mptcp_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
3662	{
3663	struct mptcp_subflow_context *subflow;
3664	struct mptcp_sock *msk = mptcp_sk(sk);
3665	int err = -EINVAL;
3666	struct sock *ssk;
3667
3668	ssk = __mptcp_nmpc_sk(msk);
3669	if (IS_ERR(ptr: ssk))
3670	return PTR_ERR(ptr: ssk);
3671
3672	inet_sk_state_store(sk, newstate: TCP_SYN_SENT);
3673	subflow = mptcp_subflow_ctx(sk: ssk);
3674	#ifdef CONFIG_TCP_MD5SIG
3675	/* no MPTCP if MD5SIG is enabled on this socket or we may run out of
3676	* TCP option space.
3677	*/
3678	if (rcu_access_pointer(tcp_sk(ssk)->md5sig_info))
3679	mptcp_subflow_early_fallback(msk, subflow);
3680	#endif
3681	if (subflow->request_mptcp && mptcp_token_new_connect(ssk)) {
3682	MPTCP_INC_STATS(net: sock_net(sk: ssk), field: MPTCP_MIB_TOKENFALLBACKINIT);
3683	mptcp_subflow_early_fallback(msk, subflow);
3684	}
3685	if (likely(!__mptcp_check_fallback(msk)))
3686	MPTCP_INC_STATS(net: sock_net(sk), field: MPTCP_MIB_MPCAPABLEACTIVE);
3687
3688	/* if reaching here via the fastopen/sendmsg path, the caller already
3689	* acquired the subflow socket lock, too.
3690	*/
3691	if (!msk->fastopening)
3692	lock_sock(sk: ssk);
3693
3694	/* the following mirrors closely a very small chunk of code from
3695	* __inet_stream_connect()
3696	*/
3697	if (ssk->sk_state != TCP_CLOSE)
3698	goto out;
3699
3700	if (BPF_CGROUP_PRE_CONNECT_ENABLED(ssk)) {
3701	err = ssk->sk_prot->pre_connect(ssk, uaddr, addr_len);
3702	if (err)
3703	goto out;
3704	}
3705
3706	err = ssk->sk_prot->connect(ssk, uaddr, addr_len);
3707	if (err < 0)
3708	goto out;
3709
3710	inet_assign_bit(DEFER_CONNECT, sk, inet_test_bit(DEFER_CONNECT, ssk));
3711
3712	out:
3713	if (!msk->fastopening)
3714	release_sock(sk: ssk);
3715
3716	/* on successful connect, the msk state will be moved to established by
3717	* subflow_finish_connect()
3718	*/
3719	if (unlikely(err)) {
3720	/* avoid leaving a dangling token in an unconnected socket */
3721	mptcp_token_destroy(msk);
3722	inet_sk_state_store(sk, newstate: TCP_CLOSE);
3723	return err;
3724	}
3725
3726	mptcp_copy_inaddrs(msk: sk, ssk);
3727	return 0;
3728	}
3729
3730	static struct proto mptcp_prot = {
3731	.name = "MPTCP",
3732	.owner = THIS_MODULE,
3733	.init = mptcp_init_sock,
3734	.connect = mptcp_connect,
3735	.disconnect = mptcp_disconnect,
3736	.close = mptcp_close,
3737	.accept = mptcp_accept,
3738	.setsockopt = mptcp_setsockopt,
3739	.getsockopt = mptcp_getsockopt,
3740	.shutdown = mptcp_shutdown,
3741	.destroy = mptcp_destroy,
3742	.sendmsg = mptcp_sendmsg,
3743	.ioctl = mptcp_ioctl,
3744	.recvmsg = mptcp_recvmsg,
3745	.release_cb = mptcp_release_cb,
3746	.hash = mptcp_hash,
3747	.unhash = mptcp_unhash,
3748	.get_port = mptcp_get_port,
3749	.forward_alloc_get = mptcp_forward_alloc_get,
3750	.sockets_allocated = &mptcp_sockets_allocated,
3751
3752	.memory_allocated = &tcp_memory_allocated,
3753	.per_cpu_fw_alloc = &tcp_memory_per_cpu_fw_alloc,
3754
3755	.memory_pressure = &tcp_memory_pressure,
3756	.sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_tcp_wmem),
3757	.sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_tcp_rmem),
3758	.sysctl_mem = sysctl_tcp_mem,
3759	.obj_size = sizeof(struct mptcp_sock),
3760	.slab_flags = SLAB_TYPESAFE_BY_RCU,
3761	.no_autobind = true,
3762	};
3763
3764	static int mptcp_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3765	{
3766	struct mptcp_sock *msk = mptcp_sk(sock->sk);
3767	struct sock *ssk, *sk = sock->sk;
3768	int err = -EINVAL;
3769
3770	lock_sock(sk);
3771	ssk = __mptcp_nmpc_sk(msk);
3772	if (IS_ERR(ptr: ssk)) {
3773	err = PTR_ERR(ptr: ssk);
3774	goto unlock;
3775	}
3776
3777	if (sk->sk_family == AF_INET)
3778	err = inet_bind_sk(sk: ssk, uaddr, addr_len);
3779	#if IS_ENABLED(CONFIG_MPTCP_IPV6)
3780	else if (sk->sk_family == AF_INET6)
3781	err = inet6_bind_sk(sk: ssk, uaddr, addr_len);
3782	#endif
3783	if (!err)
3784	mptcp_copy_inaddrs(msk: sk, ssk);
3785
3786	unlock:
3787	release_sock(sk);
3788	return err;
3789	}
3790
3791	static int mptcp_listen(struct socket *sock, int backlog)
3792	{
3793	struct mptcp_sock *msk = mptcp_sk(sock->sk);
3794	struct sock *sk = sock->sk;
3795	struct sock *ssk;
3796	int err;
3797
3798	pr_debug("msk=%p", msk);
3799
3800	lock_sock(sk);
3801
3802	err = -EINVAL;
3803	if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
3804	goto unlock;
3805
3806	ssk = __mptcp_nmpc_sk(msk);
3807	if (IS_ERR(ptr: ssk)) {
3808	err = PTR_ERR(ptr: ssk);
3809	goto unlock;
3810	}
3811
3812	inet_sk_state_store(sk, newstate: TCP_LISTEN);
3813	sock_set_flag(sk, flag: SOCK_RCU_FREE);
3814
3815	lock_sock(sk: ssk);
3816	err = __inet_listen_sk(sk: ssk, backlog);
3817	release_sock(sk: ssk);
3818	inet_sk_state_store(sk, newstate: inet_sk_state_load(sk: ssk));
3819
3820	if (!err) {
3821	sock_prot_inuse_add(net: sock_net(sk), prot: sk->sk_prot, val: 1);
3822	mptcp_copy_inaddrs(msk: sk, ssk);
3823	mptcp_event_pm_listener(ssk, event: MPTCP_EVENT_LISTENER_CREATED);
3824	}
3825
3826	unlock:
3827	release_sock(sk);
3828	return err;
3829	}
3830
3831	static int mptcp_stream_accept(struct socket *sock, struct socket *newsock,
3832	int flags, bool kern)
3833	{
3834	struct mptcp_sock *msk = mptcp_sk(sock->sk);
3835	struct sock *ssk, *newsk;
3836	int err;
3837
3838	pr_debug("msk=%p", msk);
3839
3840	/* Buggy applications can call accept on socket states other then LISTEN
3841	* but no need to allocate the first subflow just to error out.
3842	*/
3843	ssk = READ_ONCE(msk->first);
3844	if (!ssk)
3845	return -EINVAL;
3846
3847	newsk = mptcp_accept(ssk, flags, err: &err, kern);
3848	if (!newsk)
3849	return err;
3850
3851	lock_sock(sk: newsk);
3852
3853	__inet_accept(sock, newsock, newsk);
3854	if (!mptcp_is_tcpsk(sk: newsock->sk)) {
3855	struct mptcp_sock *msk = mptcp_sk(newsk);
3856	struct mptcp_subflow_context *subflow;
3857
3858	set_bit(SOCK_CUSTOM_SOCKOPT, addr: &newsock->flags);
3859	msk->in_accept_queue = 0;
3860
3861	/* set ssk->sk_socket of accept()ed flows to mptcp socket.
3862	* This is needed so NOSPACE flag can be set from tcp stack.
3863	*/
3864	mptcp_for_each_subflow(msk, subflow) {
3865	struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
3866
3867	if (!ssk->sk_socket)
3868	mptcp_sock_graft(sk: ssk, parent: newsock);
3869	}
3870
3871	/* Do late cleanup for the first subflow as necessary. Also
3872	* deal with bad peers not doing a complete shutdown.
3873	*/
3874	if (unlikely(inet_sk_state_load(msk->first) == TCP_CLOSE)) {
3875	__mptcp_close_ssk(sk: newsk, ssk: msk->first,
3876	subflow: mptcp_subflow_ctx(sk: msk->first), flags: 0);
3877	if (unlikely(list_is_singular(&msk->conn_list)))
3878	inet_sk_state_store(sk: newsk, newstate: TCP_CLOSE);
3879	}
3880	}
3881	release_sock(sk: newsk);
3882
3883	return 0;
3884	}
3885
3886	static __poll_t mptcp_check_writeable(struct mptcp_sock *msk)
3887	{
3888	struct sock *sk = (struct sock *)msk;
3889
3890	if (sk_stream_is_writeable(sk))
3891	return EPOLLOUT | EPOLLWRNORM;
3892
3893	mptcp_set_nospace(sk);
3894	smp_mb__after_atomic(); /* msk->flags is changed by write_space cb */
3895	if (sk_stream_is_writeable(sk))
3896	return EPOLLOUT | EPOLLWRNORM;
3897
3898	return 0;
3899	}
3900
3901	static __poll_t mptcp_poll(struct file *file, struct socket *sock,
3902	struct poll_table_struct *wait)
3903	{
3904	struct sock *sk = sock->sk;
3905	struct mptcp_sock *msk;
3906	__poll_t mask = 0;
3907	u8 shutdown;
3908	int state;
3909
3910	msk = mptcp_sk(sk);
3911	sock_poll_wait(filp: file, sock, p: wait);
3912
3913	state = inet_sk_state_load(sk);
3914	pr_debug("msk=%p state=%d flags=%lx", msk, state, msk->flags);
3915	if (state == TCP_LISTEN) {
3916	struct sock *ssk = READ_ONCE(msk->first);
3917
3918	if (WARN_ON_ONCE(!ssk))
3919	return 0;
3920
3921	return inet_csk_listen_poll(sk: ssk);
3922	}
3923
3924	shutdown = READ_ONCE(sk->sk_shutdown);
3925	if (shutdown == SHUTDOWN_MASK || state == TCP_CLOSE)
3926	mask |= EPOLLHUP;
3927	if (shutdown & RCV_SHUTDOWN)
3928	mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
3929
3930	if (state != TCP_SYN_SENT && state != TCP_SYN_RECV) {
3931	mask |= mptcp_check_readable(sk);
3932	if (shutdown & SEND_SHUTDOWN)
3933	mask |= EPOLLOUT | EPOLLWRNORM;
3934	else
3935	mask |= mptcp_check_writeable(msk);
3936	} else if (state == TCP_SYN_SENT &&
3937	inet_test_bit(DEFER_CONNECT, sk)) {
3938	/* cf tcp_poll() note about TFO */
3939	mask |= EPOLLOUT | EPOLLWRNORM;
3940	}
3941
3942	/* This barrier is coupled with smp_wmb() in __mptcp_error_report() */
3943	smp_rmb();
3944	if (READ_ONCE(sk->sk_err))
3945	mask |= EPOLLERR;
3946
3947	return mask;
3948	}
3949
3950	static const struct proto_ops mptcp_stream_ops = {
3951	.family = PF_INET,
3952	.owner = THIS_MODULE,
3953	.release = inet_release,
3954	.bind = mptcp_bind,
3955	.connect = inet_stream_connect,
3956	.socketpair = sock_no_socketpair,
3957	.accept = mptcp_stream_accept,
3958	.getname = inet_getname,
3959	.poll = mptcp_poll,
3960	.ioctl = inet_ioctl,
3961	.gettstamp = sock_gettstamp,
3962	.listen = mptcp_listen,
3963	.shutdown = inet_shutdown,
3964	.setsockopt = sock_common_setsockopt,
3965	.getsockopt = sock_common_getsockopt,
3966	.sendmsg = inet_sendmsg,
3967	.recvmsg = inet_recvmsg,
3968	.mmap = sock_no_mmap,
3969	.set_rcvlowat = mptcp_set_rcvlowat,
3970	};
3971
3972	static struct inet_protosw mptcp_protosw = {
3973	.type = SOCK_STREAM,
3974	.protocol = IPPROTO_MPTCP,
3975	.prot = &mptcp_prot,
3976	.ops = &mptcp_stream_ops,
3977	.flags = INET_PROTOSW_ICSK,
3978	};
3979
3980	static int mptcp_napi_poll(struct napi_struct *napi, int budget)
3981	{
3982	struct mptcp_delegated_action *delegated;
3983	struct mptcp_subflow_context *subflow;
3984	int work_done = 0;
3985
3986	delegated = container_of(napi, struct mptcp_delegated_action, napi);
3987	while ((subflow = mptcp_subflow_delegated_next(delegated)) != NULL) {
3988	struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
3989
3990	bh_lock_sock_nested(ssk);
3991	if (!sock_owned_by_user(sk: ssk)) {
3992	mptcp_subflow_process_delegated(ssk, xchg(&subflow->delegated_status, 0));
3993	} else {
3994	/* tcp_release_cb_override already processed
3995	* the action or will do at next release_sock().
3996	* In both case must dequeue the subflow here - on the same
3997	* CPU that scheduled it.
3998	*/
3999	smp_wmb();
4000	clear_bit(MPTCP_DELEGATE_SCHEDULED, addr: &subflow->delegated_status);
4001	}
4002	bh_unlock_sock(ssk);
4003	sock_put(sk: ssk);
4004
4005	if (++work_done == budget)
4006	return budget;
4007	}
4008
4009	/* always provide a 0 'work_done' argument, so that napi_complete_done
4010	* will not try accessing the NULL napi->dev ptr
4011	*/
4012	napi_complete_done(n: napi, work_done: 0);
4013	return work_done;
4014	}
4015
4016	void __init mptcp_proto_init(void)
4017	{
4018	struct mptcp_delegated_action *delegated;
4019	int cpu;
4020
4021	mptcp_prot.h.hashinfo = tcp_prot.h.hashinfo;
4022
4023	if (percpu_counter_init(&mptcp_sockets_allocated, 0, GFP_KERNEL))
4024	panic(fmt: "Failed to allocate MPTCP pcpu counter\n");
4025
4026	init_dummy_netdev(dev: &mptcp_napi_dev);
4027	for_each_possible_cpu(cpu) {
4028	delegated = per_cpu_ptr(&mptcp_delegated_actions, cpu);
4029	INIT_LIST_HEAD(list: &delegated->head);
4030	netif_napi_add_tx(dev: &mptcp_napi_dev, napi: &delegated->napi,
4031	poll: mptcp_napi_poll);
4032	napi_enable(n: &delegated->napi);
4033	}
4034
4035	mptcp_subflow_init();
4036	mptcp_pm_init();
4037	mptcp_sched_init();
4038	mptcp_token_init();
4039
4040	if (proto_register(prot: &mptcp_prot, alloc_slab: 1) != 0)
4041	panic(fmt: "Failed to register MPTCP proto.\n");
4042
4043	inet_register_protosw(p: &mptcp_protosw);
4044
4045	BUILD_BUG_ON(sizeof(struct mptcp_skb_cb) > sizeof_field(struct sk_buff, cb));
4046	}
4047
4048	#if IS_ENABLED(CONFIG_MPTCP_IPV6)
4049	static const struct proto_ops mptcp_v6_stream_ops = {
4050	.family = PF_INET6,
4051	.owner = THIS_MODULE,
4052	.release = inet6_release,
4053	.bind = mptcp_bind,
4054	.connect = inet_stream_connect,
4055	.socketpair = sock_no_socketpair,
4056	.accept = mptcp_stream_accept,
4057	.getname = inet6_getname,
4058	.poll = mptcp_poll,
4059	.ioctl = inet6_ioctl,
4060	.gettstamp = sock_gettstamp,
4061	.listen = mptcp_listen,
4062	.shutdown = inet_shutdown,
4063	.setsockopt = sock_common_setsockopt,
4064	.getsockopt = sock_common_getsockopt,
4065	.sendmsg = inet6_sendmsg,
4066	.recvmsg = inet6_recvmsg,
4067	.mmap = sock_no_mmap,
4068	#ifdef CONFIG_COMPAT
4069	.compat_ioctl = inet6_compat_ioctl,
4070	#endif
4071	.set_rcvlowat = mptcp_set_rcvlowat,
4072	};
4073
4074	static struct proto mptcp_v6_prot;
4075
4076	static struct inet_protosw mptcp_v6_protosw = {
4077	.type = SOCK_STREAM,
4078	.protocol = IPPROTO_MPTCP,
4079	.prot = &mptcp_v6_prot,
4080	.ops = &mptcp_v6_stream_ops,
4081	.flags = INET_PROTOSW_ICSK,
4082	};
4083
4084	int __init mptcp_proto_v6_init(void)
4085	{
4086	int err;
4087
4088	mptcp_v6_prot = mptcp_prot;
4089	strcpy(p: mptcp_v6_prot.name, q: "MPTCPv6");
4090	mptcp_v6_prot.slab = NULL;
4091	mptcp_v6_prot.obj_size = sizeof(struct mptcp6_sock);
4092	mptcp_v6_prot.ipv6_pinfo_offset = offsetof(struct mptcp6_sock, np);
4093
4094	err = proto_register(prot: &mptcp_v6_prot, alloc_slab: 1);
4095	if (err)
4096	return err;
4097
4098	err = inet6_register_protosw(p: &mptcp_v6_protosw);
4099	if (err)
4100	proto_unregister(prot: &mptcp_v6_prot);
4101
4102	return err;
4103	}
4104	#endif
4105