?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge tag 'block-5.11-2021-01-10' of git://git.kernel.dk/linux-block
[linux/fpc-iii.git] / net / mptcp / protocol.c
blob6628d8d742030736f7ede56fdfca1cc5b0b4f152
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 "protocol.h"
26 #include "mib.h"
27
28 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
29 struct mptcp6_sock {
30 struct mptcp_sock msk;
31 struct ipv6_pinfo np;
32 };
33 #endif
34
35 struct mptcp_skb_cb {
36 u64 map_seq;
37 u64 end_seq;
38 u32 offset;
39 };
40
41 #define MPTCP_SKB_CB(__skb) ((struct mptcp_skb_cb *)&((__skb)->cb[0]))
42
43 static struct percpu_counter mptcp_sockets_allocated;
44
45 static void __mptcp_destroy_sock(struct sock *sk);
46 static void __mptcp_check_send_data_fin(struct sock *sk);
47
48 /* If msk has an initial subflow socket, and the MP_CAPABLE handshake has not
49 * completed yet or has failed, return the subflow socket.
50 * Otherwise return NULL.
51 */
52 static struct socket *__mptcp_nmpc_socket(const struct mptcp_sock *msk)
53 {
54 if (!msk->subflow || READ_ONCE(msk->can_ack))
55 return NULL;
56
57 return msk->subflow;
58 }
59
60 /* Returns end sequence number of the receiver's advertised window */
61 static u64 mptcp_wnd_end(const struct mptcp_sock *msk)
62 {
63 return READ_ONCE(msk->wnd_end);
64 }
65
66 static bool mptcp_is_tcpsk(struct sock *sk)
67 {
68 struct socket *sock = sk->sk_socket;
69
70 if (unlikely(sk->sk_prot == &tcp_prot)) {
71 /* we are being invoked after mptcp_accept() has
72 * accepted a non-mp-capable flow: sk is a tcp_sk,
73 * not an mptcp one.
74 *
75 * Hand the socket over to tcp so all further socket ops
76 * bypass mptcp.
77 */
78 sock->ops = &inet_stream_ops;
79 return true;
80 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
81 } else if (unlikely(sk->sk_prot == &tcpv6_prot)) {
82 sock->ops = &inet6_stream_ops;
83 return true;
84 #endif
85 }
86
87 return false;
88 }
89
90 static struct sock *__mptcp_tcp_fallback(struct mptcp_sock *msk)
91 {
92 sock_owned_by_me((const struct sock *)msk);
93
94 if (likely(!__mptcp_check_fallback(msk)))
95 return NULL;
96
97 return msk->first;
98 }
99
100 static int __mptcp_socket_create(struct mptcp_sock *msk)
101 {
102 struct mptcp_subflow_context *subflow;
103 struct sock *sk = (struct sock *)msk;
104 struct socket *ssock;
105 int err;
106
107 err = mptcp_subflow_create_socket(sk, &ssock);
108 if (err)
109 return err;
110
111 msk->first = ssock->sk;
112 msk->subflow = ssock;
113 subflow = mptcp_subflow_ctx(ssock->sk);
114 list_add(&subflow->node, &msk->conn_list);
115 sock_hold(ssock->sk);
116 subflow->request_mptcp = 1;
117
118 /* accept() will wait on first subflow sk_wq, and we always wakes up
119 * via msk->sk_socket
120 */
121 RCU_INIT_POINTER(msk->first->sk_wq, &sk->sk_socket->wq);
122
123 return 0;
124 }
125
126 static void mptcp_drop(struct sock *sk, struct sk_buff *skb)
127 {
128 sk_drops_add(sk, skb);
129 __kfree_skb(skb);
130 }
131
132 static bool mptcp_try_coalesce(struct sock *sk, struct sk_buff *to,
133 struct sk_buff *from)
134 {
135 bool fragstolen;
136 int delta;
137
138 if (MPTCP_SKB_CB(from)->offset ||
139 !skb_try_coalesce(to, from, &fragstolen, &delta))
140 return false;
141
142 pr_debug("colesced seq %llx into %llx new len %d new end seq %llx",
143 MPTCP_SKB_CB(from)->map_seq, MPTCP_SKB_CB(to)->map_seq,
144 to->len, MPTCP_SKB_CB(from)->end_seq);
145 MPTCP_SKB_CB(to)->end_seq = MPTCP_SKB_CB(from)->end_seq;
146 kfree_skb_partial(from, fragstolen);
147 atomic_add(delta, &sk->sk_rmem_alloc);
148 sk_mem_charge(sk, delta);
149 return true;
150 }
151
152 static bool mptcp_ooo_try_coalesce(struct mptcp_sock *msk, struct sk_buff *to,
153 struct sk_buff *from)
154 {
155 if (MPTCP_SKB_CB(from)->map_seq != MPTCP_SKB_CB(to)->end_seq)
156 return false;
157
158 return mptcp_try_coalesce((struct sock *)msk, to, from);
159 }
160
161 /* "inspired" by tcp_data_queue_ofo(), main differences:
162 * - use mptcp seqs
163 * - don't cope with sacks
164 */
165 static void mptcp_data_queue_ofo(struct mptcp_sock *msk, struct sk_buff *skb)
166 {
167 struct sock *sk = (struct sock *)msk;
168 struct rb_node **p, *parent;
169 u64 seq, end_seq, max_seq;
170 struct sk_buff *skb1;
171
172 seq = MPTCP_SKB_CB(skb)->map_seq;
173 end_seq = MPTCP_SKB_CB(skb)->end_seq;
174 max_seq = READ_ONCE(msk->rcv_wnd_sent);
175
176 pr_debug("msk=%p seq=%llx limit=%llx empty=%d", msk, seq, max_seq,
177 RB_EMPTY_ROOT(&msk->out_of_order_queue));
178 if (after64(end_seq, max_seq)) {
179 /* out of window */
180 mptcp_drop(sk, skb);
181 pr_debug("oow by %lld, rcv_wnd_sent %llu\n",
182 (unsigned long long)end_seq - (unsigned long)max_seq,
183 (unsigned long long)msk->rcv_wnd_sent);
184 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_NODSSWINDOW);
185 return;
186 }
187
188 p = &msk->out_of_order_queue.rb_node;
189 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOQUEUE);
190 if (RB_EMPTY_ROOT(&msk->out_of_order_queue)) {
191 rb_link_node(&skb->rbnode, NULL, p);
192 rb_insert_color(&skb->rbnode, &msk->out_of_order_queue);
193 msk->ooo_last_skb = skb;
194 goto end;
195 }
196
197 /* with 2 subflows, adding at end of ooo queue is quite likely
198 * Use of ooo_last_skb avoids the O(Log(N)) rbtree lookup.
199 */
200 if (mptcp_ooo_try_coalesce(msk, msk->ooo_last_skb, skb)) {
201 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOMERGE);
202 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOQUEUETAIL);
203 return;
204 }
205
206 /* Can avoid an rbtree lookup if we are adding skb after ooo_last_skb */
207 if (!before64(seq, MPTCP_SKB_CB(msk->ooo_last_skb)->end_seq)) {
208 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOQUEUETAIL);
209 parent = &msk->ooo_last_skb->rbnode;
210 p = &parent->rb_right;
211 goto insert;
212 }
213
214 /* Find place to insert this segment. Handle overlaps on the way. */
215 parent = NULL;
216 while (*p) {
217 parent = *p;
218 skb1 = rb_to_skb(parent);
219 if (before64(seq, MPTCP_SKB_CB(skb1)->map_seq)) {
220 p = &parent->rb_left;
221 continue;
222 }
223 if (before64(seq, MPTCP_SKB_CB(skb1)->end_seq)) {
224 if (!after64(end_seq, MPTCP_SKB_CB(skb1)->end_seq)) {
225 /* All the bits are present. Drop. */
226 mptcp_drop(sk, skb);
227 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
228 return;
229 }
230 if (after64(seq, MPTCP_SKB_CB(skb1)->map_seq)) {
231 /* partial overlap:
232 * | skb |
233 * | skb1 |
234 * continue traversing
235 */
236 } else {
237 /* skb's seq == skb1's seq and skb covers skb1.
238 * Replace skb1 with skb.
239 */
240 rb_replace_node(&skb1->rbnode, &skb->rbnode,
241 &msk->out_of_order_queue);
242 mptcp_drop(sk, skb1);
243 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
244 goto merge_right;
245 }
246 } else if (mptcp_ooo_try_coalesce(msk, skb1, skb)) {
247 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOMERGE);
248 return;
249 }
250 p = &parent->rb_right;
251 }
252
253 insert:
254 /* Insert segment into RB tree. */
255 rb_link_node(&skb->rbnode, parent, p);
256 rb_insert_color(&skb->rbnode, &msk->out_of_order_queue);
257
258 merge_right:
259 /* Remove other segments covered by skb. */
260 while ((skb1 = skb_rb_next(skb)) != NULL) {
261 if (before64(end_seq, MPTCP_SKB_CB(skb1)->end_seq))
262 break;
263 rb_erase(&skb1->rbnode, &msk->out_of_order_queue);
264 mptcp_drop(sk, skb1);
265 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
266 }
267 /* If there is no skb after us, we are the last_skb ! */
268 if (!skb1)
269 msk->ooo_last_skb = skb;
270
271 end:
272 skb_condense(skb);
273 skb_set_owner_r(skb, sk);
274 }
275
276 static bool __mptcp_move_skb(struct mptcp_sock *msk, struct sock *ssk,
277 struct sk_buff *skb, unsigned int offset,
278 size_t copy_len)
279 {
280 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
281 struct sock *sk = (struct sock *)msk;
282 struct sk_buff *tail;
283
284 __skb_unlink(skb, &ssk->sk_receive_queue);
285
286 skb_ext_reset(skb);
287 skb_orphan(skb);
288
289 /* try to fetch required memory from subflow */
290 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
291 if (ssk->sk_forward_alloc < skb->truesize)
292 goto drop;
293 __sk_mem_reclaim(ssk, skb->truesize);
294 if (!sk_rmem_schedule(sk, skb, skb->truesize))
295 goto drop;
296 }
297
298 /* the skb map_seq accounts for the skb offset:
299 * mptcp_subflow_get_mapped_dsn() is based on the current tp->copied_seq
300 * value
301 */
302 MPTCP_SKB_CB(skb)->map_seq = mptcp_subflow_get_mapped_dsn(subflow);
303 MPTCP_SKB_CB(skb)->end_seq = MPTCP_SKB_CB(skb)->map_seq + copy_len;
304 MPTCP_SKB_CB(skb)->offset = offset;
305
306 if (MPTCP_SKB_CB(skb)->map_seq == msk->ack_seq) {
307 /* in sequence */
308 WRITE_ONCE(msk->ack_seq, msk->ack_seq + copy_len);
309 tail = skb_peek_tail(&sk->sk_receive_queue);
310 if (tail && mptcp_try_coalesce(sk, tail, skb))
311 return true;
312
313 skb_set_owner_r(skb, sk);
314 __skb_queue_tail(&sk->sk_receive_queue, skb);
315 return true;
316 } else if (after64(MPTCP_SKB_CB(skb)->map_seq, msk->ack_seq)) {
317 mptcp_data_queue_ofo(msk, skb);
318 return false;
319 }
320
321 /* old data, keep it simple and drop the whole pkt, sender
322 * will retransmit as needed, if needed.
323 */
324 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
325 drop:
326 mptcp_drop(sk, skb);
327 return false;
328 }
329
330 static void mptcp_stop_timer(struct sock *sk)
331 {
332 struct inet_connection_sock *icsk = inet_csk(sk);
333
334 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
335 mptcp_sk(sk)->timer_ival = 0;
336 }
337
338 static void mptcp_close_wake_up(struct sock *sk)
339 {
340 if (sock_flag(sk, SOCK_DEAD))
341 return;
342
343 sk->sk_state_change(sk);
344 if (sk->sk_shutdown == SHUTDOWN_MASK ||
345 sk->sk_state == TCP_CLOSE)
346 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_HUP);
347 else
348 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
349 }
350
351 static bool mptcp_pending_data_fin_ack(struct sock *sk)
352 {
353 struct mptcp_sock *msk = mptcp_sk(sk);
354
355 return !__mptcp_check_fallback(msk) &&
356 ((1 << sk->sk_state) &
357 (TCPF_FIN_WAIT1 | TCPF_CLOSING | TCPF_LAST_ACK)) &&
358 msk->write_seq == READ_ONCE(msk->snd_una);
359 }
360
361 static void mptcp_check_data_fin_ack(struct sock *sk)
362 {
363 struct mptcp_sock *msk = mptcp_sk(sk);
364
365 /* Look for an acknowledged DATA_FIN */
366 if (mptcp_pending_data_fin_ack(sk)) {
367 mptcp_stop_timer(sk);
368
369 WRITE_ONCE(msk->snd_data_fin_enable, 0);
370
371 switch (sk->sk_state) {
372 case TCP_FIN_WAIT1:
373 inet_sk_state_store(sk, TCP_FIN_WAIT2);
374 break;
375 case TCP_CLOSING:
376 case TCP_LAST_ACK:
377 inet_sk_state_store(sk, TCP_CLOSE);
378 break;
379 }
380
381 mptcp_close_wake_up(sk);
382 }
383 }
384
385 static bool mptcp_pending_data_fin(struct sock *sk, u64 *seq)
386 {
387 struct mptcp_sock *msk = mptcp_sk(sk);
388
389 if (READ_ONCE(msk->rcv_data_fin) &&
390 ((1 << sk->sk_state) &
391 (TCPF_ESTABLISHED | TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2))) {
392 u64 rcv_data_fin_seq = READ_ONCE(msk->rcv_data_fin_seq);
393
394 if (msk->ack_seq == rcv_data_fin_seq) {
395 if (seq)
396 *seq = rcv_data_fin_seq;
397
398 return true;
399 }
400 }
401
402 return false;
403 }
404
405 static void mptcp_set_timeout(const struct sock *sk, const struct sock *ssk)
406 {
407 long tout = ssk && inet_csk(ssk)->icsk_pending ?
408 inet_csk(ssk)->icsk_timeout - jiffies : 0;
409
410 if (tout <= 0)
411 tout = mptcp_sk(sk)->timer_ival;
412 mptcp_sk(sk)->timer_ival = tout > 0 ? tout : TCP_RTO_MIN;
413 }
414
415 static bool mptcp_subflow_active(struct mptcp_subflow_context *subflow)
416 {
417 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
418
419 /* can't send if JOIN hasn't completed yet (i.e. is usable for mptcp) */
420 if (subflow->request_join && !subflow->fully_established)
421 return false;
422
423 /* only send if our side has not closed yet */
424 return ((1 << ssk->sk_state) & (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT));
425 }
426
427 static bool tcp_can_send_ack(const struct sock *ssk)
428 {
429 return !((1 << inet_sk_state_load(ssk)) &
430 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_TIME_WAIT | TCPF_CLOSE));
431 }
432
433 static void mptcp_send_ack(struct mptcp_sock *msk)
434 {
435 struct mptcp_subflow_context *subflow;
436
437 mptcp_for_each_subflow(msk, subflow) {
438 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
439
440 lock_sock(ssk);
441 if (tcp_can_send_ack(ssk))
442 tcp_send_ack(ssk);
443 release_sock(ssk);
444 }
445 }
446
447 static bool mptcp_subflow_cleanup_rbuf(struct sock *ssk)
448 {
449 int ret;
450
451 lock_sock(ssk);
452 ret = tcp_can_send_ack(ssk);
453 if (ret)
454 tcp_cleanup_rbuf(ssk, 1);
455 release_sock(ssk);
456 return ret;
457 }
458
459 static void mptcp_cleanup_rbuf(struct mptcp_sock *msk)
460 {
461 struct sock *ack_hint = READ_ONCE(msk->ack_hint);
462 struct mptcp_subflow_context *subflow;
463
464 /* if the hinted ssk is still active, try to use it */
465 if (likely(ack_hint)) {
466 mptcp_for_each_subflow(msk, subflow) {
467 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
468
469 if (ack_hint == ssk && mptcp_subflow_cleanup_rbuf(ssk))
470 return;
471 }
472 }
473
474 /* otherwise pick the first active subflow */
475 mptcp_for_each_subflow(msk, subflow)
476 if (mptcp_subflow_cleanup_rbuf(mptcp_subflow_tcp_sock(subflow)))
477 return;
478 }
479
480 static bool mptcp_check_data_fin(struct sock *sk)
481 {
482 struct mptcp_sock *msk = mptcp_sk(sk);
483 u64 rcv_data_fin_seq;
484 bool ret = false;
485
486 if (__mptcp_check_fallback(msk) || !msk->first)
487 return ret;
488
489 /* Need to ack a DATA_FIN received from a peer while this side
490 * of the connection is in ESTABLISHED, FIN_WAIT1, or FIN_WAIT2.
491 * msk->rcv_data_fin was set when parsing the incoming options
492 * at the subflow level and the msk lock was not held, so this
493 * is the first opportunity to act on the DATA_FIN and change
494 * the msk state.
495 *
496 * If we are caught up to the sequence number of the incoming
497 * DATA_FIN, send the DATA_ACK now and do state transition. If
498 * not caught up, do nothing and let the recv code send DATA_ACK
499 * when catching up.
500 */
501
502 if (mptcp_pending_data_fin(sk, &rcv_data_fin_seq)) {
503 WRITE_ONCE(msk->ack_seq, msk->ack_seq + 1);
504 WRITE_ONCE(msk->rcv_data_fin, 0);
505
506 sk->sk_shutdown |= RCV_SHUTDOWN;
507 smp_mb__before_atomic(); /* SHUTDOWN must be visible first */
508 set_bit(MPTCP_DATA_READY, &msk->flags);
509
510 switch (sk->sk_state) {
511 case TCP_ESTABLISHED:
512 inet_sk_state_store(sk, TCP_CLOSE_WAIT);
513 break;
514 case TCP_FIN_WAIT1:
515 inet_sk_state_store(sk, TCP_CLOSING);
516 break;
517 case TCP_FIN_WAIT2:
518 inet_sk_state_store(sk, TCP_CLOSE);
519 break;
520 default:
521 /* Other states not expected */
522 WARN_ON_ONCE(1);
523 break;
524 }
525
526 ret = true;
527 mptcp_set_timeout(sk, NULL);
528 mptcp_send_ack(msk);
529 mptcp_close_wake_up(sk);
530 }
531 return ret;
532 }
533
534 static bool __mptcp_move_skbs_from_subflow(struct mptcp_sock *msk,
535 struct sock *ssk,
536 unsigned int *bytes)
537 {
538 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
539 struct sock *sk = (struct sock *)msk;
540 unsigned int moved = 0;
541 bool more_data_avail;
542 struct tcp_sock *tp;
543 bool done = false;
544 int sk_rbuf;
545
546 sk_rbuf = READ_ONCE(sk->sk_rcvbuf);
547
548 if (!(sk->sk_userlocks & SOCK_RCVBUF_LOCK)) {
549 int ssk_rbuf = READ_ONCE(ssk->sk_rcvbuf);
550
551 if (unlikely(ssk_rbuf > sk_rbuf)) {
552 WRITE_ONCE(sk->sk_rcvbuf, ssk_rbuf);
553 sk_rbuf = ssk_rbuf;
554 }
555 }
556
557 pr_debug("msk=%p ssk=%p", msk, ssk);
558 tp = tcp_sk(ssk);
559 do {
560 u32 map_remaining, offset;
561 u32 seq = tp->copied_seq;
562 struct sk_buff *skb;
563 bool fin;
564
565 /* try to move as much data as available */
566 map_remaining = subflow->map_data_len -
567 mptcp_subflow_get_map_offset(subflow);
568
569 skb = skb_peek(&ssk->sk_receive_queue);
570 if (!skb) {
571 /* if no data is found, a racing workqueue/recvmsg
572 * already processed the new data, stop here or we
573 * can enter an infinite loop
574 */
575 if (!moved)
576 done = true;
577 break;
578 }
579
580 if (__mptcp_check_fallback(msk)) {
581 /* if we are running under the workqueue, TCP could have
582 * collapsed skbs between dummy map creation and now
583 * be sure to adjust the size
584 */
585 map_remaining = skb->len;
586 subflow->map_data_len = skb->len;
587 }
588
589 offset = seq - TCP_SKB_CB(skb)->seq;
590 fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
591 if (fin) {
592 done = true;
593 seq++;
594 }
595
596 if (offset < skb->len) {
597 size_t len = skb->len - offset;
598
599 if (tp->urg_data)
600 done = true;
601
602 if (__mptcp_move_skb(msk, ssk, skb, offset, len))
603 moved += len;
604 seq += len;
605
606 if (WARN_ON_ONCE(map_remaining < len))
607 break;
608 } else {
609 WARN_ON_ONCE(!fin);
610 sk_eat_skb(ssk, skb);
611 done = true;
612 }
613
614 WRITE_ONCE(tp->copied_seq, seq);
615 more_data_avail = mptcp_subflow_data_available(ssk);
616
617 if (atomic_read(&sk->sk_rmem_alloc) > sk_rbuf) {
618 done = true;
619 break;
620 }
621 } while (more_data_avail);
622 WRITE_ONCE(msk->ack_hint, ssk);
623
624 *bytes += moved;
625 return done;
626 }
627
628 static bool __mptcp_ofo_queue(struct mptcp_sock *msk)
629 {
630 struct sock *sk = (struct sock *)msk;
631 struct sk_buff *skb, *tail;
632 bool moved = false;
633 struct rb_node *p;
634 u64 end_seq;
635
636 p = rb_first(&msk->out_of_order_queue);
637 pr_debug("msk=%p empty=%d", msk, RB_EMPTY_ROOT(&msk->out_of_order_queue));
638 while (p) {
639 skb = rb_to_skb(p);
640 if (after64(MPTCP_SKB_CB(skb)->map_seq, msk->ack_seq))
641 break;
642
643 p = rb_next(p);
644 rb_erase(&skb->rbnode, &msk->out_of_order_queue);
645
646 if (unlikely(!after64(MPTCP_SKB_CB(skb)->end_seq,
647 msk->ack_seq))) {
648 mptcp_drop(sk, skb);
649 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
650 continue;
651 }
652
653 end_seq = MPTCP_SKB_CB(skb)->end_seq;
654 tail = skb_peek_tail(&sk->sk_receive_queue);
655 if (!tail || !mptcp_ooo_try_coalesce(msk, tail, skb)) {
656 int delta = msk->ack_seq - MPTCP_SKB_CB(skb)->map_seq;
657
658 /* skip overlapping data, if any */
659 pr_debug("uncoalesced seq=%llx ack seq=%llx delta=%d",
660 MPTCP_SKB_CB(skb)->map_seq, msk->ack_seq,
661 delta);
662 MPTCP_SKB_CB(skb)->offset += delta;
663 __skb_queue_tail(&sk->sk_receive_queue, skb);
664 }
665 msk->ack_seq = end_seq;
666 moved = true;
667 }
668 return moved;
669 }
670
671 /* In most cases we will be able to lock the mptcp socket. If its already
672 * owned, we need to defer to the work queue to avoid ABBA deadlock.
673 */
674 static void move_skbs_to_msk(struct mptcp_sock *msk, struct sock *ssk)
675 {
676 struct sock *sk = (struct sock *)msk;
677 unsigned int moved = 0;
678
679 if (inet_sk_state_load(sk) == TCP_CLOSE)
680 return;
681
682 mptcp_data_lock(sk);
683
684 __mptcp_move_skbs_from_subflow(msk, ssk, &moved);
685 __mptcp_ofo_queue(msk);
686
687 /* If the moves have caught up with the DATA_FIN sequence number
688 * it's time to ack the DATA_FIN and change socket state, but
689 * this is not a good place to change state. Let the workqueue
690 * do it.
691 */
692 if (mptcp_pending_data_fin(sk, NULL))
693 mptcp_schedule_work(sk);
694 mptcp_data_unlock(sk);
695 }
696
697 void mptcp_data_ready(struct sock *sk, struct sock *ssk)
698 {
699 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
700 struct mptcp_sock *msk = mptcp_sk(sk);
701 int sk_rbuf, ssk_rbuf;
702 bool wake;
703
704 /* The peer can send data while we are shutting down this
705 * subflow at msk destruction time, but we must avoid enqueuing
706 * more data to the msk receive queue
707 */
708 if (unlikely(subflow->disposable))
709 return;
710
711 /* move_skbs_to_msk below can legitly clear the data_avail flag,
712 * but we will need later to properly woke the reader, cache its
713 * value
714 */
715 wake = subflow->data_avail == MPTCP_SUBFLOW_DATA_AVAIL;
716 if (wake)
717 set_bit(MPTCP_DATA_READY, &msk->flags);
718
719 ssk_rbuf = READ_ONCE(ssk->sk_rcvbuf);
720 sk_rbuf = READ_ONCE(sk->sk_rcvbuf);
721 if (unlikely(ssk_rbuf > sk_rbuf))
722 sk_rbuf = ssk_rbuf;
723
724 /* over limit? can't append more skbs to msk */
725 if (atomic_read(&sk->sk_rmem_alloc) > sk_rbuf)
726 goto wake;
727
728 move_skbs_to_msk(msk, ssk);
729
730 wake:
731 if (wake)
732 sk->sk_data_ready(sk);
733 }
734
735 void __mptcp_flush_join_list(struct mptcp_sock *msk)
736 {
737 if (likely(list_empty(&msk->join_list)))
738 return;
739
740 spin_lock_bh(&msk->join_list_lock);
741 list_splice_tail_init(&msk->join_list, &msk->conn_list);
742 spin_unlock_bh(&msk->join_list_lock);
743 }
744
745 static bool mptcp_timer_pending(struct sock *sk)
746 {
747 return timer_pending(&inet_csk(sk)->icsk_retransmit_timer);
748 }
749
750 static void mptcp_reset_timer(struct sock *sk)
751 {
752 struct inet_connection_sock *icsk = inet_csk(sk);
753 unsigned long tout;
754
755 /* prevent rescheduling on close */
756 if (unlikely(inet_sk_state_load(sk) == TCP_CLOSE))
757 return;
758
759 /* should never be called with mptcp level timer cleared */
760 tout = READ_ONCE(mptcp_sk(sk)->timer_ival);
761 if (WARN_ON_ONCE(!tout))
762 tout = TCP_RTO_MIN;
763 sk_reset_timer(sk, &icsk->icsk_retransmit_timer, jiffies + tout);
764 }
765
766 bool mptcp_schedule_work(struct sock *sk)
767 {
768 if (inet_sk_state_load(sk) != TCP_CLOSE &&
769 schedule_work(&mptcp_sk(sk)->work)) {
770 /* each subflow already holds a reference to the sk, and the
771 * workqueue is invoked by a subflow, so sk can't go away here.
772 */
773 sock_hold(sk);
774 return true;
775 }
776 return false;
777 }
778
779 void mptcp_subflow_eof(struct sock *sk)
780 {
781 if (!test_and_set_bit(MPTCP_WORK_EOF, &mptcp_sk(sk)->flags))
782 mptcp_schedule_work(sk);
783 }
784
785 static void mptcp_check_for_eof(struct mptcp_sock *msk)
786 {
787 struct mptcp_subflow_context *subflow;
788 struct sock *sk = (struct sock *)msk;
789 int receivers = 0;
790
791 mptcp_for_each_subflow(msk, subflow)
792 receivers += !subflow->rx_eof;
793 if (receivers)
794 return;
795
796 if (!(sk->sk_shutdown & RCV_SHUTDOWN)) {
797 /* hopefully temporary hack: propagate shutdown status
798 * to msk, when all subflows agree on it
799 */
800 sk->sk_shutdown |= RCV_SHUTDOWN;
801
802 smp_mb__before_atomic(); /* SHUTDOWN must be visible first */
803 set_bit(MPTCP_DATA_READY, &msk->flags);
804 sk->sk_data_ready(sk);
805 }
806
807 switch (sk->sk_state) {
808 case TCP_ESTABLISHED:
809 inet_sk_state_store(sk, TCP_CLOSE_WAIT);
810 break;
811 case TCP_FIN_WAIT1:
812 inet_sk_state_store(sk, TCP_CLOSING);
813 break;
814 case TCP_FIN_WAIT2:
815 inet_sk_state_store(sk, TCP_CLOSE);
816 break;
817 default:
818 return;
819 }
820 mptcp_close_wake_up(sk);
821 }
822
823 static struct sock *mptcp_subflow_recv_lookup(const struct mptcp_sock *msk)
824 {
825 struct mptcp_subflow_context *subflow;
826 struct sock *sk = (struct sock *)msk;
827
828 sock_owned_by_me(sk);
829
830 mptcp_for_each_subflow(msk, subflow) {
831 if (subflow->data_avail)
832 return mptcp_subflow_tcp_sock(subflow);
833 }
834
835 return NULL;
836 }
837
838 static bool mptcp_skb_can_collapse_to(u64 write_seq,
839 const struct sk_buff *skb,
840 const struct mptcp_ext *mpext)
841 {
842 if (!tcp_skb_can_collapse_to(skb))
843 return false;
844
845 /* can collapse only if MPTCP level sequence is in order and this
846 * mapping has not been xmitted yet
847 */
848 return mpext && mpext->data_seq + mpext->data_len == write_seq &&
849 !mpext->frozen;
850 }
851
852 static bool mptcp_frag_can_collapse_to(const struct mptcp_sock *msk,
853 const struct page_frag *pfrag,
854 const struct mptcp_data_frag *df)
855 {
856 return df && pfrag->page == df->page &&
857 pfrag->size - pfrag->offset > 0 &&
858 df->data_seq + df->data_len == msk->write_seq;
859 }
860
861 static int mptcp_wmem_with_overhead(struct sock *sk, int size)
862 {
863 struct mptcp_sock *msk = mptcp_sk(sk);
864 int ret, skbs;
865
866 ret = size + ((sizeof(struct mptcp_data_frag) * size) >> PAGE_SHIFT);
867 skbs = (msk->tx_pending_data + size) / msk->size_goal_cache;
868 if (skbs < msk->skb_tx_cache.qlen)
869 return ret;
870
871 return ret + (skbs - msk->skb_tx_cache.qlen) * SKB_TRUESIZE(MAX_TCP_HEADER);
872 }
873
874 static void __mptcp_wmem_reserve(struct sock *sk, int size)
875 {
876 int amount = mptcp_wmem_with_overhead(sk, size);
877 struct mptcp_sock *msk = mptcp_sk(sk);
878
879 WARN_ON_ONCE(msk->wmem_reserved);
880 if (WARN_ON_ONCE(amount < 0))
881 amount = 0;
882
883 if (amount <= sk->sk_forward_alloc)
884 goto reserve;
885
886 /* under memory pressure try to reserve at most a single page
887 * otherwise try to reserve the full estimate and fallback
888 * to a single page before entering the error path
889 */
890 if ((tcp_under_memory_pressure(sk) && amount > PAGE_SIZE) ||
891 !sk_wmem_schedule(sk, amount)) {
892 if (amount <= PAGE_SIZE)
893 goto nomem;
894
895 amount = PAGE_SIZE;
896 if (!sk_wmem_schedule(sk, amount))
897 goto nomem;
898 }
899
900 reserve:
901 msk->wmem_reserved = amount;
902 sk->sk_forward_alloc -= amount;
903 return;
904
905 nomem:
906 /* we will wait for memory on next allocation */
907 msk->wmem_reserved = -1;
908 }
909
910 static void __mptcp_update_wmem(struct sock *sk)
911 {
912 struct mptcp_sock *msk = mptcp_sk(sk);
913
914 if (!msk->wmem_reserved)
915 return;
916
917 if (msk->wmem_reserved < 0)
918 msk->wmem_reserved = 0;
919 if (msk->wmem_reserved > 0) {
920 sk->sk_forward_alloc += msk->wmem_reserved;
921 msk->wmem_reserved = 0;
922 }
923 }
924
925 static bool mptcp_wmem_alloc(struct sock *sk, int size)
926 {
927 struct mptcp_sock *msk = mptcp_sk(sk);
928
929 /* check for pre-existing error condition */
930 if (msk->wmem_reserved < 0)
931 return false;
932
933 if (msk->wmem_reserved >= size)
934 goto account;
935
936 mptcp_data_lock(sk);
937 if (!sk_wmem_schedule(sk, size)) {
938 mptcp_data_unlock(sk);
939 return false;
940 }
941
942 sk->sk_forward_alloc -= size;
943 msk->wmem_reserved += size;
944 mptcp_data_unlock(sk);
945
946 account:
947 msk->wmem_reserved -= size;
948 return true;
949 }
950
951 static void mptcp_wmem_uncharge(struct sock *sk, int size)
952 {
953 struct mptcp_sock *msk = mptcp_sk(sk);
954
955 if (msk->wmem_reserved < 0)
956 msk->wmem_reserved = 0;
957 msk->wmem_reserved += size;
958 }
959
960 static void mptcp_mem_reclaim_partial(struct sock *sk)
961 {
962 struct mptcp_sock *msk = mptcp_sk(sk);
963
964 /* if we are experiencing a transint allocation error,
965 * the forward allocation memory has been already
966 * released
967 */
968 if (msk->wmem_reserved < 0)
969 return;
970
971 mptcp_data_lock(sk);
972 sk->sk_forward_alloc += msk->wmem_reserved;
973 sk_mem_reclaim_partial(sk);
974 msk->wmem_reserved = sk->sk_forward_alloc;
975 sk->sk_forward_alloc = 0;
976 mptcp_data_unlock(sk);
977 }
978
979 static void dfrag_uncharge(struct sock *sk, int len)
980 {
981 sk_mem_uncharge(sk, len);
982 sk_wmem_queued_add(sk, -len);
983 }
984
985 static void dfrag_clear(struct sock *sk, struct mptcp_data_frag *dfrag)
986 {
987 int len = dfrag->data_len + dfrag->overhead;
988
989 list_del(&dfrag->list);
990 dfrag_uncharge(sk, len);
991 put_page(dfrag->page);
992 }
993
994 static void __mptcp_clean_una(struct sock *sk)
995 {
996 struct mptcp_sock *msk = mptcp_sk(sk);
997 struct mptcp_data_frag *dtmp, *dfrag;
998 bool cleaned = false;
999 u64 snd_una;
1000
1001 /* on fallback we just need to ignore snd_una, as this is really
1002 * plain TCP
1003 */
1004 if (__mptcp_check_fallback(msk))
1005 msk->snd_una = READ_ONCE(msk->snd_nxt);
1006
1007 snd_una = msk->snd_una;
1008 list_for_each_entry_safe(dfrag, dtmp, &msk->rtx_queue, list) {
1009 if (after64(dfrag->data_seq + dfrag->data_len, snd_una))
1010 break;
1011
1012 if (WARN_ON_ONCE(dfrag == msk->first_pending))
1013 break;
1014 dfrag_clear(sk, dfrag);
1015 cleaned = true;
1016 }
1017
1018 dfrag = mptcp_rtx_head(sk);
1019 if (dfrag && after64(snd_una, dfrag->data_seq)) {
1020 u64 delta = snd_una - dfrag->data_seq;
1021
1022 if (WARN_ON_ONCE(delta > dfrag->already_sent))
1023 goto out;
1024
1025 dfrag->data_seq += delta;
1026 dfrag->offset += delta;
1027 dfrag->data_len -= delta;
1028 dfrag->already_sent -= delta;
1029
1030 dfrag_uncharge(sk, delta);
1031 cleaned = true;
1032 }
1033
1034 out:
1035 if (cleaned) {
1036 if (tcp_under_memory_pressure(sk)) {
1037 __mptcp_update_wmem(sk);
1038 sk_mem_reclaim_partial(sk);
1039 }
1040
1041 if (sk_stream_is_writeable(sk)) {
1042 /* pairs with memory barrier in mptcp_poll */
1043 smp_mb();
1044 if (test_and_clear_bit(MPTCP_NOSPACE, &msk->flags))
1045 sk_stream_write_space(sk);
1046 }
1047 }
1048
1049 if (snd_una == READ_ONCE(msk->snd_nxt)) {
1050 if (msk->timer_ival)
1051 mptcp_stop_timer(sk);
1052 } else {
1053 mptcp_reset_timer(sk);
1054 }
1055 }
1056
1057 static void mptcp_enter_memory_pressure(struct sock *sk)
1058 {
1059 struct mptcp_subflow_context *subflow;
1060 struct mptcp_sock *msk = mptcp_sk(sk);
1061 bool first = true;
1062
1063 sk_stream_moderate_sndbuf(sk);
1064 mptcp_for_each_subflow(msk, subflow) {
1065 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1066
1067 if (first)
1068 tcp_enter_memory_pressure(ssk);
1069 sk_stream_moderate_sndbuf(ssk);
1070 first = false;
1071 }
1072 }
1073
1074 /* ensure we get enough memory for the frag hdr, beyond some minimal amount of
1075 * data
1076 */
1077 static bool mptcp_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1078 {
1079 if (likely(skb_page_frag_refill(32U + sizeof(struct mptcp_data_frag),
1080 pfrag, sk->sk_allocation)))
1081 return true;
1082
1083 mptcp_enter_memory_pressure(sk);
1084 return false;
1085 }
1086
1087 static struct mptcp_data_frag *
1088 mptcp_carve_data_frag(const struct mptcp_sock *msk, struct page_frag *pfrag,
1089 int orig_offset)
1090 {
1091 int offset = ALIGN(orig_offset, sizeof(long));
1092 struct mptcp_data_frag *dfrag;
1093
1094 dfrag = (struct mptcp_data_frag *)(page_to_virt(pfrag->page) + offset);
1095 dfrag->data_len = 0;
1096 dfrag->data_seq = msk->write_seq;
1097 dfrag->overhead = offset - orig_offset + sizeof(struct mptcp_data_frag);
1098 dfrag->offset = offset + sizeof(struct mptcp_data_frag);
1099 dfrag->already_sent = 0;
1100 dfrag->page = pfrag->page;
1101
1102 return dfrag;
1103 }
1104
1105 struct mptcp_sendmsg_info {
1106 int mss_now;
1107 int size_goal;
1108 u16 limit;
1109 u16 sent;
1110 unsigned int flags;
1111 };
1112
1113 static int mptcp_check_allowed_size(struct mptcp_sock *msk, u64 data_seq,
1114 int avail_size)
1115 {
1116 u64 window_end = mptcp_wnd_end(msk);
1117
1118 if (__mptcp_check_fallback(msk))
1119 return avail_size;
1120
1121 if (!before64(data_seq + avail_size, window_end)) {
1122 u64 allowed_size = window_end - data_seq;
1123
1124 return min_t(unsigned int, allowed_size, avail_size);
1125 }
1126
1127 return avail_size;
1128 }
1129
1130 static bool __mptcp_add_ext(struct sk_buff *skb, gfp_t gfp)
1131 {
1132 struct skb_ext *mpext = __skb_ext_alloc(gfp);
1133
1134 if (!mpext)
1135 return false;
1136 __skb_ext_set(skb, SKB_EXT_MPTCP, mpext);
1137 return true;
1138 }
1139
1140 static struct sk_buff *__mptcp_do_alloc_tx_skb(struct sock *sk, gfp_t gfp)
1141 {
1142 struct sk_buff *skb;
1143
1144 skb = alloc_skb_fclone(MAX_TCP_HEADER, gfp);
1145 if (likely(skb)) {
1146 if (likely(__mptcp_add_ext(skb, gfp))) {
1147 skb_reserve(skb, MAX_TCP_HEADER);
1148 skb->reserved_tailroom = skb->end - skb->tail;
1149 return skb;
1150 }
1151 __kfree_skb(skb);
1152 } else {
1153 mptcp_enter_memory_pressure(sk);
1154 }
1155 return NULL;
1156 }
1157
1158 static bool mptcp_tx_cache_refill(struct sock *sk, int size,
1159 struct sk_buff_head *skbs, int *total_ts)
1160 {
1161 struct mptcp_sock *msk = mptcp_sk(sk);
1162 struct sk_buff *skb;
1163 int space_needed;
1164
1165 if (unlikely(tcp_under_memory_pressure(sk))) {
1166 mptcp_mem_reclaim_partial(sk);
1167
1168 /* under pressure pre-allocate at most a single skb */
1169 if (msk->skb_tx_cache.qlen)
1170 return true;
1171 space_needed = msk->size_goal_cache;
1172 } else {
1173 space_needed = msk->tx_pending_data + size -
1174 msk->skb_tx_cache.qlen * msk->size_goal_cache;
1175 }
1176
1177 while (space_needed > 0) {
1178 skb = __mptcp_do_alloc_tx_skb(sk, sk->sk_allocation);
1179 if (unlikely(!skb)) {
1180 /* under memory pressure, try to pass the caller a
1181 * single skb to allow forward progress
1182 */
1183 while (skbs->qlen > 1) {
1184 skb = __skb_dequeue_tail(skbs);
1185 __kfree_skb(skb);
1186 }
1187 return skbs->qlen > 0;
1188 }
1189
1190 *total_ts += skb->truesize;
1191 __skb_queue_tail(skbs, skb);
1192 space_needed -= msk->size_goal_cache;
1193 }
1194 return true;
1195 }
1196
1197 static bool __mptcp_alloc_tx_skb(struct sock *sk, struct sock *ssk, gfp_t gfp)
1198 {
1199 struct mptcp_sock *msk = mptcp_sk(sk);
1200 struct sk_buff *skb;
1201
1202 if (ssk->sk_tx_skb_cache) {
1203 skb = ssk->sk_tx_skb_cache;
1204 if (unlikely(!skb_ext_find(skb, SKB_EXT_MPTCP) &&
1205 !__mptcp_add_ext(skb, gfp)))
1206 return false;
1207 return true;
1208 }
1209
1210 skb = skb_peek(&msk->skb_tx_cache);
1211 if (skb) {
1212 if (likely(sk_wmem_schedule(ssk, skb->truesize))) {
1213 skb = __skb_dequeue(&msk->skb_tx_cache);
1214 if (WARN_ON_ONCE(!skb))
1215 return false;
1216
1217 mptcp_wmem_uncharge(sk, skb->truesize);
1218 ssk->sk_tx_skb_cache = skb;
1219 return true;
1220 }
1221
1222 /* over memory limit, no point to try to allocate a new skb */
1223 return false;
1224 }
1225
1226 skb = __mptcp_do_alloc_tx_skb(sk, gfp);
1227 if (!skb)
1228 return false;
1229
1230 if (likely(sk_wmem_schedule(ssk, skb->truesize))) {
1231 ssk->sk_tx_skb_cache = skb;
1232 return true;
1233 }
1234 kfree_skb(skb);
1235 return false;
1236 }
1237
1238 static bool mptcp_must_reclaim_memory(struct sock *sk, struct sock *ssk)
1239 {
1240 return !ssk->sk_tx_skb_cache &&
1241 !skb_peek(&mptcp_sk(sk)->skb_tx_cache) &&
1242 tcp_under_memory_pressure(sk);
1243 }
1244
1245 static bool mptcp_alloc_tx_skb(struct sock *sk, struct sock *ssk)
1246 {
1247 if (unlikely(mptcp_must_reclaim_memory(sk, ssk)))
1248 mptcp_mem_reclaim_partial(sk);
1249 return __mptcp_alloc_tx_skb(sk, ssk, sk->sk_allocation);
1250 }
1251
1252 static int mptcp_sendmsg_frag(struct sock *sk, struct sock *ssk,
1253 struct mptcp_data_frag *dfrag,
1254 struct mptcp_sendmsg_info *info)
1255 {
1256 u64 data_seq = dfrag->data_seq + info->sent;
1257 struct mptcp_sock *msk = mptcp_sk(sk);
1258 bool zero_window_probe = false;
1259 struct mptcp_ext *mpext = NULL;
1260 struct sk_buff *skb, *tail;
1261 bool can_collapse = false;
1262 int size_bias = 0;
1263 int avail_size;
1264 size_t ret = 0;
1265
1266 pr_debug("msk=%p ssk=%p sending dfrag at seq=%lld len=%d already sent=%d",
1267 msk, ssk, dfrag->data_seq, dfrag->data_len, info->sent);
1268
1269 /* compute send limit */
1270 info->mss_now = tcp_send_mss(ssk, &info->size_goal, info->flags);
1271 avail_size = info->size_goal;
1272 msk->size_goal_cache = info->size_goal;
1273 skb = tcp_write_queue_tail(ssk);
1274 if (skb) {
1275 /* Limit the write to the size available in the
1276 * current skb, if any, so that we create at most a new skb.
1277 * Explicitly tells TCP internals to avoid collapsing on later
1278 * queue management operation, to avoid breaking the ext <->
1279 * SSN association set here
1280 */
1281 mpext = skb_ext_find(skb, SKB_EXT_MPTCP);
1282 can_collapse = (info->size_goal - skb->len > 0) &&
1283 mptcp_skb_can_collapse_to(data_seq, skb, mpext);
1284 if (!can_collapse) {
1285 TCP_SKB_CB(skb)->eor = 1;
1286 } else {
1287 size_bias = skb->len;
1288 avail_size = info->size_goal - skb->len;
1289 }
1290 }
1291
1292 /* Zero window and all data acked? Probe. */
1293 avail_size = mptcp_check_allowed_size(msk, data_seq, avail_size);
1294 if (avail_size == 0) {
1295 u64 snd_una = READ_ONCE(msk->snd_una);
1296
1297 if (skb || snd_una != msk->snd_nxt)
1298 return 0;
1299 zero_window_probe = true;
1300 data_seq = snd_una - 1;
1301 avail_size = 1;
1302 }
1303
1304 if (WARN_ON_ONCE(info->sent > info->limit ||
1305 info->limit > dfrag->data_len))
1306 return 0;
1307
1308 ret = info->limit - info->sent;
1309 tail = tcp_build_frag(ssk, avail_size + size_bias, info->flags,
1310 dfrag->page, dfrag->offset + info->sent, &ret);
1311 if (!tail) {
1312 tcp_remove_empty_skb(sk, tcp_write_queue_tail(ssk));
1313 return -ENOMEM;
1314 }
1315
1316 /* if the tail skb is still the cached one, collapsing really happened.
1317 */
1318 if (skb == tail) {
1319 TCP_SKB_CB(tail)->tcp_flags &= ~TCPHDR_PSH;
1320 mpext->data_len += ret;
1321 WARN_ON_ONCE(!can_collapse);
1322 WARN_ON_ONCE(zero_window_probe);
1323 goto out;
1324 }
1325
1326 mpext = skb_ext_find(tail, SKB_EXT_MPTCP);
1327 if (WARN_ON_ONCE(!mpext)) {
1328 /* should never reach here, stream corrupted */
1329 return -EINVAL;
1330 }
1331
1332 memset(mpext, 0, sizeof(*mpext));
1333 mpext->data_seq = data_seq;
1334 mpext->subflow_seq = mptcp_subflow_ctx(ssk)->rel_write_seq;
1335 mpext->data_len = ret;
1336 mpext->use_map = 1;
1337 mpext->dsn64 = 1;
1338
1339 pr_debug("data_seq=%llu subflow_seq=%u data_len=%u dsn64=%d",
1340 mpext->data_seq, mpext->subflow_seq, mpext->data_len,
1341 mpext->dsn64);
1342
1343 if (zero_window_probe) {
1344 mptcp_subflow_ctx(ssk)->rel_write_seq += ret;
1345 mpext->frozen = 1;
1346 ret = 0;
1347 tcp_push_pending_frames(ssk);
1348 }
1349 out:
1350 mptcp_subflow_ctx(ssk)->rel_write_seq += ret;
1351 return ret;
1352 }
1353
1354 #define MPTCP_SEND_BURST_SIZE ((1 << 16) - \
1355 sizeof(struct tcphdr) - \
1356 MAX_TCP_OPTION_SPACE - \
1357 sizeof(struct ipv6hdr) - \
1358 sizeof(struct frag_hdr))
1359
1360 struct subflow_send_info {
1361 struct sock *ssk;
1362 u64 ratio;
1363 };
1364
1365 static struct sock *mptcp_subflow_get_send(struct mptcp_sock *msk,
1366 u32 *sndbuf)
1367 {
1368 struct subflow_send_info send_info[2];
1369 struct mptcp_subflow_context *subflow;
1370 int i, nr_active = 0;
1371 struct sock *ssk;
1372 u64 ratio;
1373 u32 pace;
1374
1375 sock_owned_by_me((struct sock *)msk);
1376
1377 *sndbuf = 0;
1378 if (__mptcp_check_fallback(msk)) {
1379 if (!msk->first)
1380 return NULL;
1381 *sndbuf = msk->first->sk_sndbuf;
1382 return sk_stream_memory_free(msk->first) ? msk->first : NULL;
1383 }
1384
1385 /* re-use last subflow, if the burst allow that */
1386 if (msk->last_snd && msk->snd_burst > 0 &&
1387 sk_stream_memory_free(msk->last_snd) &&
1388 mptcp_subflow_active(mptcp_subflow_ctx(msk->last_snd))) {
1389 mptcp_for_each_subflow(msk, subflow) {
1390 ssk = mptcp_subflow_tcp_sock(subflow);
1391 *sndbuf = max(tcp_sk(ssk)->snd_wnd, *sndbuf);
1392 }
1393 return msk->last_snd;
1394 }
1395
1396 /* pick the subflow with the lower wmem/wspace ratio */
1397 for (i = 0; i < 2; ++i) {
1398 send_info[i].ssk = NULL;
1399 send_info[i].ratio = -1;
1400 }
1401 mptcp_for_each_subflow(msk, subflow) {
1402 ssk = mptcp_subflow_tcp_sock(subflow);
1403 if (!mptcp_subflow_active(subflow))
1404 continue;
1405
1406 nr_active += !subflow->backup;
1407 *sndbuf = max(tcp_sk(ssk)->snd_wnd, *sndbuf);
1408 if (!sk_stream_memory_free(subflow->tcp_sock))
1409 continue;
1410
1411 pace = READ_ONCE(ssk->sk_pacing_rate);
1412 if (!pace)
1413 continue;
1414
1415 ratio = div_u64((u64)READ_ONCE(ssk->sk_wmem_queued) << 32,
1416 pace);
1417 if (ratio < send_info[subflow->backup].ratio) {
1418 send_info[subflow->backup].ssk = ssk;
1419 send_info[subflow->backup].ratio = ratio;
1420 }
1421 }
1422
1423 pr_debug("msk=%p nr_active=%d ssk=%p:%lld backup=%p:%lld",
1424 msk, nr_active, send_info[0].ssk, send_info[0].ratio,
1425 send_info[1].ssk, send_info[1].ratio);
1426
1427 /* pick the best backup if no other subflow is active */
1428 if (!nr_active)
1429 send_info[0].ssk = send_info[1].ssk;
1430
1431 if (send_info[0].ssk) {
1432 msk->last_snd = send_info[0].ssk;
1433 msk->snd_burst = min_t(int, MPTCP_SEND_BURST_SIZE,
1434 sk_stream_wspace(msk->last_snd));
1435 return msk->last_snd;
1436 }
1437 return NULL;
1438 }
1439
1440 static void mptcp_push_release(struct sock *sk, struct sock *ssk,
1441 struct mptcp_sendmsg_info *info)
1442 {
1443 mptcp_set_timeout(sk, ssk);
1444 tcp_push(ssk, 0, info->mss_now, tcp_sk(ssk)->nonagle, info->size_goal);
1445 release_sock(ssk);
1446 }
1447
1448 static void mptcp_push_pending(struct sock *sk, unsigned int flags)
1449 {
1450 struct sock *prev_ssk = NULL, *ssk = NULL;
1451 struct mptcp_sock *msk = mptcp_sk(sk);
1452 struct mptcp_sendmsg_info info = {
1453 .flags = flags,
1454 };
1455 struct mptcp_data_frag *dfrag;
1456 int len, copied = 0;
1457 u32 sndbuf;
1458
1459 while ((dfrag = mptcp_send_head(sk))) {
1460 info.sent = dfrag->already_sent;
1461 info.limit = dfrag->data_len;
1462 len = dfrag->data_len - dfrag->already_sent;
1463 while (len > 0) {
1464 int ret = 0;
1465
1466 prev_ssk = ssk;
1467 __mptcp_flush_join_list(msk);
1468 ssk = mptcp_subflow_get_send(msk, &sndbuf);
1469
1470 /* do auto tuning */
1471 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK) &&
1472 sndbuf > READ_ONCE(sk->sk_sndbuf))
1473 WRITE_ONCE(sk->sk_sndbuf, sndbuf);
1474
1475 /* try to keep the subflow socket lock across
1476 * consecutive xmit on the same socket
1477 */
1478 if (ssk != prev_ssk && prev_ssk)
1479 mptcp_push_release(sk, prev_ssk, &info);
1480 if (!ssk)
1481 goto out;
1482
1483 if (ssk != prev_ssk || !prev_ssk)
1484 lock_sock(ssk);
1485
1486 /* keep it simple and always provide a new skb for the
1487 * subflow, even if we will not use it when collapsing
1488 * on the pending one
1489 */
1490 if (!mptcp_alloc_tx_skb(sk, ssk)) {
1491 mptcp_push_release(sk, ssk, &info);
1492 goto out;
1493 }
1494
1495 ret = mptcp_sendmsg_frag(sk, ssk, dfrag, &info);
1496 if (ret <= 0) {
1497 mptcp_push_release(sk, ssk, &info);
1498 goto out;
1499 }
1500
1501 info.sent += ret;
1502 dfrag->already_sent += ret;
1503 msk->snd_nxt += ret;
1504 msk->snd_burst -= ret;
1505 msk->tx_pending_data -= ret;
1506 copied += ret;
1507 len -= ret;
1508 }
1509 WRITE_ONCE(msk->first_pending, mptcp_send_next(sk));
1510 }
1511
1512 /* at this point we held the socket lock for the last subflow we used */
1513 if (ssk)
1514 mptcp_push_release(sk, ssk, &info);
1515
1516 out:
1517 if (copied) {
1518 /* start the timer, if it's not pending */
1519 if (!mptcp_timer_pending(sk))
1520 mptcp_reset_timer(sk);
1521 __mptcp_check_send_data_fin(sk);
1522 }
1523 }
1524
1525 static void __mptcp_subflow_push_pending(struct sock *sk, struct sock *ssk)
1526 {
1527 struct mptcp_sock *msk = mptcp_sk(sk);
1528 struct mptcp_sendmsg_info info;
1529 struct mptcp_data_frag *dfrag;
1530 int len, copied = 0;
1531
1532 info.flags = 0;
1533 while ((dfrag = mptcp_send_head(sk))) {
1534 info.sent = dfrag->already_sent;
1535 info.limit = dfrag->data_len;
1536 len = dfrag->data_len - dfrag->already_sent;
1537 while (len > 0) {
1538 int ret = 0;
1539
1540 /* do auto tuning */
1541 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK) &&
1542 ssk->sk_sndbuf > READ_ONCE(sk->sk_sndbuf))
1543 WRITE_ONCE(sk->sk_sndbuf, ssk->sk_sndbuf);
1544
1545 if (unlikely(mptcp_must_reclaim_memory(sk, ssk))) {
1546 __mptcp_update_wmem(sk);
1547 sk_mem_reclaim_partial(sk);
1548 }
1549 if (!__mptcp_alloc_tx_skb(sk, ssk, GFP_ATOMIC))
1550 goto out;
1551
1552 ret = mptcp_sendmsg_frag(sk, ssk, dfrag, &info);
1553 if (ret <= 0)
1554 goto out;
1555
1556 info.sent += ret;
1557 dfrag->already_sent += ret;
1558 msk->snd_nxt += ret;
1559 msk->snd_burst -= ret;
1560 msk->tx_pending_data -= ret;
1561 copied += ret;
1562 len -= ret;
1563 }
1564 WRITE_ONCE(msk->first_pending, mptcp_send_next(sk));
1565 }
1566
1567 out:
1568 /* __mptcp_alloc_tx_skb could have released some wmem and we are
1569 * not going to flush it via release_sock()
1570 */
1571 __mptcp_update_wmem(sk);
1572 if (copied) {
1573 mptcp_set_timeout(sk, ssk);
1574 tcp_push(ssk, 0, info.mss_now, tcp_sk(ssk)->nonagle,
1575 info.size_goal);
1576 if (msk->snd_data_fin_enable &&
1577 msk->snd_nxt + 1 == msk->write_seq)
1578 mptcp_schedule_work(sk);
1579 }
1580 }
1581
1582 static int mptcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1583 {
1584 struct mptcp_sock *msk = mptcp_sk(sk);
1585 struct page_frag *pfrag;
1586 size_t copied = 0;
1587 int ret = 0;
1588 long timeo;
1589
1590 if (msg->msg_flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL))
1591 return -EOPNOTSUPP;
1592
1593 mptcp_lock_sock(sk, __mptcp_wmem_reserve(sk, min_t(size_t, 1 << 20, len)));
1594
1595 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1596
1597 if ((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) {
1598 ret = sk_stream_wait_connect(sk, &timeo);
1599 if (ret)
1600 goto out;
1601 }
1602
1603 pfrag = sk_page_frag(sk);
1604
1605 while (msg_data_left(msg)) {
1606 int total_ts, frag_truesize = 0;
1607 struct mptcp_data_frag *dfrag;
1608 struct sk_buff_head skbs;
1609 bool dfrag_collapsed;
1610 size_t psize, offset;
1611
1612 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN)) {
1613 ret = -EPIPE;
1614 goto out;
1615 }
1616
1617 /* reuse tail pfrag, if possible, or carve a new one from the
1618 * page allocator
1619 */
1620 dfrag = mptcp_pending_tail(sk);
1621 dfrag_collapsed = mptcp_frag_can_collapse_to(msk, pfrag, dfrag);
1622 if (!dfrag_collapsed) {
1623 if (!sk_stream_memory_free(sk))
1624 goto wait_for_memory;
1625
1626 if (!mptcp_page_frag_refill(sk, pfrag))
1627 goto wait_for_memory;
1628
1629 dfrag = mptcp_carve_data_frag(msk, pfrag, pfrag->offset);
1630 frag_truesize = dfrag->overhead;
1631 }
1632
1633 /* we do not bound vs wspace, to allow a single packet.
1634 * memory accounting will prevent execessive memory usage
1635 * anyway
1636 */
1637 offset = dfrag->offset + dfrag->data_len;
1638 psize = pfrag->size - offset;
1639 psize = min_t(size_t, psize, msg_data_left(msg));
1640 total_ts = psize + frag_truesize;
1641 __skb_queue_head_init(&skbs);
1642 if (!mptcp_tx_cache_refill(sk, psize, &skbs, &total_ts))
1643 goto wait_for_memory;
1644
1645 if (!mptcp_wmem_alloc(sk, total_ts)) {
1646 __skb_queue_purge(&skbs);
1647 goto wait_for_memory;
1648 }
1649
1650 skb_queue_splice_tail(&skbs, &msk->skb_tx_cache);
1651 if (copy_page_from_iter(dfrag->page, offset, psize,
1652 &msg->msg_iter) != psize) {
1653 mptcp_wmem_uncharge(sk, psize + frag_truesize);
1654 ret = -EFAULT;
1655 goto out;
1656 }
1657
1658 /* data successfully copied into the write queue */
1659 copied += psize;
1660 dfrag->data_len += psize;
1661 frag_truesize += psize;
1662 pfrag->offset += frag_truesize;
1663 WRITE_ONCE(msk->write_seq, msk->write_seq + psize);
1664 msk->tx_pending_data += psize;
1665
1666 /* charge data on mptcp pending queue to the msk socket
1667 * Note: we charge such data both to sk and ssk
1668 */
1669 sk_wmem_queued_add(sk, frag_truesize);
1670 if (!dfrag_collapsed) {
1671 get_page(dfrag->page);
1672 list_add_tail(&dfrag->list, &msk->rtx_queue);
1673 if (!msk->first_pending)
1674 WRITE_ONCE(msk->first_pending, dfrag);
1675 }
1676 pr_debug("msk=%p dfrag at seq=%lld len=%d sent=%d new=%d", msk,
1677 dfrag->data_seq, dfrag->data_len, dfrag->already_sent,
1678 !dfrag_collapsed);
1679
1680 continue;
1681
1682 wait_for_memory:
1683 set_bit(MPTCP_NOSPACE, &msk->flags);
1684 mptcp_push_pending(sk, msg->msg_flags);
1685 ret = sk_stream_wait_memory(sk, &timeo);
1686 if (ret)
1687 goto out;
1688 }
1689
1690 if (copied)
1691 mptcp_push_pending(sk, msg->msg_flags);
1692
1693 out:
1694 release_sock(sk);
1695 return copied ? : ret;
1696 }
1697
1698 static void mptcp_wait_data(struct sock *sk, long *timeo)
1699 {
1700 DEFINE_WAIT_FUNC(wait, woken_wake_function);
1701 struct mptcp_sock *msk = mptcp_sk(sk);
1702
1703 add_wait_queue(sk_sleep(sk), &wait);
1704 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1705
1706 sk_wait_event(sk, timeo,
1707 test_and_clear_bit(MPTCP_DATA_READY, &msk->flags), &wait);
1708
1709 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1710 remove_wait_queue(sk_sleep(sk), &wait);
1711 }
1712
1713 static int __mptcp_recvmsg_mskq(struct mptcp_sock *msk,
1714 struct msghdr *msg,
1715 size_t len)
1716 {
1717 struct sk_buff *skb;
1718 int copied = 0;
1719
1720 while ((skb = skb_peek(&msk->receive_queue)) != NULL) {
1721 u32 offset = MPTCP_SKB_CB(skb)->offset;
1722 u32 data_len = skb->len - offset;
1723 u32 count = min_t(size_t, len - copied, data_len);
1724 int err;
1725
1726 err = skb_copy_datagram_msg(skb, offset, msg, count);
1727 if (unlikely(err < 0)) {
1728 if (!copied)
1729 return err;
1730 break;
1731 }
1732
1733 copied += count;
1734
1735 if (count < data_len) {
1736 MPTCP_SKB_CB(skb)->offset += count;
1737 break;
1738 }
1739
1740 /* we will bulk release the skb memory later */
1741 skb->destructor = NULL;
1742 msk->rmem_released += skb->truesize;
1743 __skb_unlink(skb, &msk->receive_queue);
1744 __kfree_skb(skb);
1745
1746 if (copied >= len)
1747 break;
1748 }
1749
1750 return copied;
1751 }
1752
1753 /* receive buffer autotuning. See tcp_rcv_space_adjust for more information.
1754 *
1755 * Only difference: Use highest rtt estimate of the subflows in use.
1756 */
1757 static void mptcp_rcv_space_adjust(struct mptcp_sock *msk, int copied)
1758 {
1759 struct mptcp_subflow_context *subflow;
1760 struct sock *sk = (struct sock *)msk;
1761 u32 time, advmss = 1;
1762 u64 rtt_us, mstamp;
1763
1764 sock_owned_by_me(sk);
1765
1766 if (copied <= 0)
1767 return;
1768
1769 msk->rcvq_space.copied += copied;
1770
1771 mstamp = div_u64(tcp_clock_ns(), NSEC_PER_USEC);
1772 time = tcp_stamp_us_delta(mstamp, msk->rcvq_space.time);
1773
1774 rtt_us = msk->rcvq_space.rtt_us;
1775 if (rtt_us && time < (rtt_us >> 3))
1776 return;
1777
1778 rtt_us = 0;
1779 mptcp_for_each_subflow(msk, subflow) {
1780 const struct tcp_sock *tp;
1781 u64 sf_rtt_us;
1782 u32 sf_advmss;
1783
1784 tp = tcp_sk(mptcp_subflow_tcp_sock(subflow));
1785
1786 sf_rtt_us = READ_ONCE(tp->rcv_rtt_est.rtt_us);
1787 sf_advmss = READ_ONCE(tp->advmss);
1788
1789 rtt_us = max(sf_rtt_us, rtt_us);
1790 advmss = max(sf_advmss, advmss);
1791 }
1792
1793 msk->rcvq_space.rtt_us = rtt_us;
1794 if (time < (rtt_us >> 3) || rtt_us == 0)
1795 return;
1796
1797 if (msk->rcvq_space.copied <= msk->rcvq_space.space)
1798 goto new_measure;
1799
1800 if (sock_net(sk)->ipv4.sysctl_tcp_moderate_rcvbuf &&
1801 !(sk->sk_userlocks & SOCK_RCVBUF_LOCK)) {
1802 int rcvmem, rcvbuf;
1803 u64 rcvwin, grow;
1804
1805 rcvwin = ((u64)msk->rcvq_space.copied << 1) + 16 * advmss;
1806
1807 grow = rcvwin * (msk->rcvq_space.copied - msk->rcvq_space.space);
1808
1809 do_div(grow, msk->rcvq_space.space);
1810 rcvwin += (grow << 1);
1811
1812 rcvmem = SKB_TRUESIZE(advmss + MAX_TCP_HEADER);
1813 while (tcp_win_from_space(sk, rcvmem) < advmss)
1814 rcvmem += 128;
1815
1816 do_div(rcvwin, advmss);
1817 rcvbuf = min_t(u64, rcvwin * rcvmem,
1818 sock_net(sk)->ipv4.sysctl_tcp_rmem[2]);
1819
1820 if (rcvbuf > sk->sk_rcvbuf) {
1821 u32 window_clamp;
1822
1823 window_clamp = tcp_win_from_space(sk, rcvbuf);
1824 WRITE_ONCE(sk->sk_rcvbuf, rcvbuf);
1825
1826 /* Make subflows follow along. If we do not do this, we
1827 * get drops at subflow level if skbs can't be moved to
1828 * the mptcp rx queue fast enough (announced rcv_win can
1829 * exceed ssk->sk_rcvbuf).
1830 */
1831 mptcp_for_each_subflow(msk, subflow) {
1832 struct sock *ssk;
1833 bool slow;
1834
1835 ssk = mptcp_subflow_tcp_sock(subflow);
1836 slow = lock_sock_fast(ssk);
1837 WRITE_ONCE(ssk->sk_rcvbuf, rcvbuf);
1838 tcp_sk(ssk)->window_clamp = window_clamp;
1839 tcp_cleanup_rbuf(ssk, 1);
1840 unlock_sock_fast(ssk, slow);
1841 }
1842 }
1843 }
1844
1845 msk->rcvq_space.space = msk->rcvq_space.copied;
1846 new_measure:
1847 msk->rcvq_space.copied = 0;
1848 msk->rcvq_space.time = mstamp;
1849 }
1850
1851 static void __mptcp_update_rmem(struct sock *sk)
1852 {
1853 struct mptcp_sock *msk = mptcp_sk(sk);
1854
1855 if (!msk->rmem_released)
1856 return;
1857
1858 atomic_sub(msk->rmem_released, &sk->sk_rmem_alloc);
1859 sk_mem_uncharge(sk, msk->rmem_released);
1860 msk->rmem_released = 0;
1861 }
1862
1863 static void __mptcp_splice_receive_queue(struct sock *sk)
1864 {
1865 struct mptcp_sock *msk = mptcp_sk(sk);
1866
1867 skb_queue_splice_tail_init(&sk->sk_receive_queue, &msk->receive_queue);
1868 }
1869
1870 static bool __mptcp_move_skbs(struct mptcp_sock *msk, unsigned int rcv)
1871 {
1872 struct sock *sk = (struct sock *)msk;
1873 unsigned int moved = 0;
1874 bool ret, done;
1875
1876 __mptcp_flush_join_list(msk);
1877 do {
1878 struct sock *ssk = mptcp_subflow_recv_lookup(msk);
1879 bool slowpath;
1880
1881 /* we can have data pending in the subflows only if the msk
1882 * receive buffer was full at subflow_data_ready() time,
1883 * that is an unlikely slow path.
1884 */
1885 if (likely(!ssk))
1886 break;
1887
1888 slowpath = lock_sock_fast(ssk);
1889 mptcp_data_lock(sk);
1890 done = __mptcp_move_skbs_from_subflow(msk, ssk, &moved);
1891 mptcp_data_unlock(sk);
1892 if (moved && rcv) {
1893 WRITE_ONCE(msk->rmem_pending, min(rcv, moved));
1894 tcp_cleanup_rbuf(ssk, 1);
1895 WRITE_ONCE(msk->rmem_pending, 0);
1896 }
1897 unlock_sock_fast(ssk, slowpath);
1898 } while (!done);
1899
1900 /* acquire the data lock only if some input data is pending */
1901 ret = moved > 0;
1902 if (!RB_EMPTY_ROOT(&msk->out_of_order_queue) ||
1903 !skb_queue_empty_lockless(&sk->sk_receive_queue)) {
1904 mptcp_data_lock(sk);
1905 __mptcp_update_rmem(sk);
1906 ret |= __mptcp_ofo_queue(msk);
1907 __mptcp_splice_receive_queue(sk);
1908 mptcp_data_unlock(sk);
1909 }
1910 if (ret)
1911 mptcp_check_data_fin((struct sock *)msk);
1912 return !skb_queue_empty(&msk->receive_queue);
1913 }
1914
1915 static int mptcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
1916 int nonblock, int flags, int *addr_len)
1917 {
1918 struct mptcp_sock *msk = mptcp_sk(sk);
1919 int copied = 0;
1920 int target;
1921 long timeo;
1922
1923 if (msg->msg_flags & ~(MSG_WAITALL | MSG_DONTWAIT))
1924 return -EOPNOTSUPP;
1925
1926 mptcp_lock_sock(sk, __mptcp_splice_receive_queue(sk));
1927 if (unlikely(sk->sk_state == TCP_LISTEN)) {
1928 copied = -ENOTCONN;
1929 goto out_err;
1930 }
1931
1932 timeo = sock_rcvtimeo(sk, nonblock);
1933
1934 len = min_t(size_t, len, INT_MAX);
1935 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1936
1937 while (copied < len) {
1938 int bytes_read, old_space;
1939
1940 bytes_read = __mptcp_recvmsg_mskq(msk, msg, len - copied);
1941 if (unlikely(bytes_read < 0)) {
1942 if (!copied)
1943 copied = bytes_read;
1944 goto out_err;
1945 }
1946
1947 copied += bytes_read;
1948
1949 if (skb_queue_empty(&msk->receive_queue) &&
1950 __mptcp_move_skbs(msk, len - copied))
1951 continue;
1952
1953 /* be sure to advertise window change */
1954 old_space = READ_ONCE(msk->old_wspace);
1955 if ((tcp_space(sk) - old_space) >= old_space)
1956 mptcp_cleanup_rbuf(msk);
1957
1958 /* only the master socket status is relevant here. The exit
1959 * conditions mirror closely tcp_recvmsg()
1960 */
1961 if (copied >= target)
1962 break;
1963
1964 if (copied) {
1965 if (sk->sk_err ||
1966 sk->sk_state == TCP_CLOSE ||
1967 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1968 !timeo ||
1969 signal_pending(current))
1970 break;
1971 } else {
1972 if (sk->sk_err) {
1973 copied = sock_error(sk);
1974 break;
1975 }
1976
1977 if (test_and_clear_bit(MPTCP_WORK_EOF, &msk->flags))
1978 mptcp_check_for_eof(msk);
1979
1980 if (sk->sk_shutdown & RCV_SHUTDOWN) {
1981 /* race breaker: the shutdown could be after the
1982 * previous receive queue check
1983 */
1984 if (__mptcp_move_skbs(msk, len - copied))
1985 continue;
1986 break;
1987 }
1988
1989 if (sk->sk_state == TCP_CLOSE) {
1990 copied = -ENOTCONN;
1991 break;
1992 }
1993
1994 if (!timeo) {
1995 copied = -EAGAIN;
1996 break;
1997 }
1998
1999 if (signal_pending(current)) {
2000 copied = sock_intr_errno(timeo);
2001 break;
2002 }
2003 }
2004
2005 pr_debug("block timeout %ld", timeo);
2006 mptcp_wait_data(sk, &timeo);
2007 }
2008
2009 if (skb_queue_empty_lockless(&sk->sk_receive_queue) &&
2010 skb_queue_empty(&msk->receive_queue)) {
2011 /* entire backlog drained, clear DATA_READY. */
2012 clear_bit(MPTCP_DATA_READY, &msk->flags);
2013
2014 /* .. race-breaker: ssk might have gotten new data
2015 * after last __mptcp_move_skbs() returned false.
2016 */
2017 if (unlikely(__mptcp_move_skbs(msk, 0)))
2018 set_bit(MPTCP_DATA_READY, &msk->flags);
2019 } else if (unlikely(!test_bit(MPTCP_DATA_READY, &msk->flags))) {
2020 /* data to read but mptcp_wait_data() cleared DATA_READY */
2021 set_bit(MPTCP_DATA_READY, &msk->flags);
2022 }
2023 out_err:
2024 pr_debug("msk=%p data_ready=%d rx queue empty=%d copied=%d",
2025 msk, test_bit(MPTCP_DATA_READY, &msk->flags),
2026 skb_queue_empty_lockless(&sk->sk_receive_queue), copied);
2027 mptcp_rcv_space_adjust(msk, copied);
2028
2029 release_sock(sk);
2030 return copied;
2031 }
2032
2033 static void mptcp_retransmit_handler(struct sock *sk)
2034 {
2035 struct mptcp_sock *msk = mptcp_sk(sk);
2036
2037 set_bit(MPTCP_WORK_RTX, &msk->flags);
2038 mptcp_schedule_work(sk);
2039 }
2040
2041 static void mptcp_retransmit_timer(struct timer_list *t)
2042 {
2043 struct inet_connection_sock *icsk = from_timer(icsk, t,
2044 icsk_retransmit_timer);
2045 struct sock *sk = &icsk->icsk_inet.sk;
2046
2047 bh_lock_sock(sk);
2048 if (!sock_owned_by_user(sk)) {
2049 mptcp_retransmit_handler(sk);
2050 } else {
2051 /* delegate our work to tcp_release_cb() */
2052 if (!test_and_set_bit(TCP_WRITE_TIMER_DEFERRED,
2053 &sk->sk_tsq_flags))
2054 sock_hold(sk);
2055 }
2056 bh_unlock_sock(sk);
2057 sock_put(sk);
2058 }
2059
2060 static void mptcp_timeout_timer(struct timer_list *t)
2061 {
2062 struct sock *sk = from_timer(sk, t, sk_timer);
2063
2064 mptcp_schedule_work(sk);
2065 sock_put(sk);
2066 }
2067
2068 /* Find an idle subflow. Return NULL if there is unacked data at tcp
2069 * level.
2070 *
2071 * A backup subflow is returned only if that is the only kind available.
2072 */
2073 static struct sock *mptcp_subflow_get_retrans(const struct mptcp_sock *msk)
2074 {
2075 struct mptcp_subflow_context *subflow;
2076 struct sock *backup = NULL;
2077
2078 sock_owned_by_me((const struct sock *)msk);
2079
2080 if (__mptcp_check_fallback(msk))
2081 return NULL;
2082
2083 mptcp_for_each_subflow(msk, subflow) {
2084 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2085
2086 if (!mptcp_subflow_active(subflow))
2087 continue;
2088
2089 /* still data outstanding at TCP level? Don't retransmit. */
2090 if (!tcp_write_queue_empty(ssk)) {
2091 if (inet_csk(ssk)->icsk_ca_state >= TCP_CA_Loss)
2092 continue;
2093 return NULL;
2094 }
2095
2096 if (subflow->backup) {
2097 if (!backup)
2098 backup = ssk;
2099 continue;
2100 }
2101
2102 return ssk;
2103 }
2104
2105 return backup;
2106 }
2107
2108 /* subflow sockets can be either outgoing (connect) or incoming
2109 * (accept).
2110 *
2111 * Outgoing subflows use in-kernel sockets.
2112 * Incoming subflows do not have their own 'struct socket' allocated,
2113 * so we need to use tcp_close() after detaching them from the mptcp
2114 * parent socket.
2115 */
2116 void __mptcp_close_ssk(struct sock *sk, struct sock *ssk,
2117 struct mptcp_subflow_context *subflow)
2118 {
2119 bool dispose_socket = false;
2120 struct socket *sock;
2121
2122 list_del(&subflow->node);
2123
2124 lock_sock_nested(ssk, SINGLE_DEPTH_NESTING);
2125
2126 /* if we are invoked by the msk cleanup code, the subflow is
2127 * already orphaned
2128 */
2129 sock = ssk->sk_socket;
2130 if (sock) {
2131 dispose_socket = sock != sk->sk_socket;
2132 sock_orphan(ssk);
2133 }
2134
2135 subflow->disposable = 1;
2136
2137 /* if ssk hit tcp_done(), tcp_cleanup_ulp() cleared the related ops
2138 * the ssk has been already destroyed, we just need to release the
2139 * reference owned by msk;
2140 */
2141 if (!inet_csk(ssk)->icsk_ulp_ops) {
2142 kfree_rcu(subflow, rcu);
2143 } else {
2144 /* otherwise tcp will dispose of the ssk and subflow ctx */
2145 __tcp_close(ssk, 0);
2146
2147 /* close acquired an extra ref */
2148 __sock_put(ssk);
2149 }
2150 release_sock(ssk);
2151 if (dispose_socket)
2152 iput(SOCK_INODE(sock));
2153
2154 sock_put(ssk);
2155 }
2156
2157 static unsigned int mptcp_sync_mss(struct sock *sk, u32 pmtu)
2158 {
2159 return 0;
2160 }
2161
2162 static void pm_work(struct mptcp_sock *msk)
2163 {
2164 struct mptcp_pm_data *pm = &msk->pm;
2165
2166 spin_lock_bh(&msk->pm.lock);
2167
2168 pr_debug("msk=%p status=%x", msk, pm->status);
2169 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) {
2170 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED);
2171 mptcp_pm_nl_add_addr_received(msk);
2172 }
2173 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) {
2174 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK);
2175 mptcp_pm_nl_add_addr_send_ack(msk);
2176 }
2177 if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) {
2178 pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED);
2179 mptcp_pm_nl_rm_addr_received(msk);
2180 }
2181 if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) {
2182 pm->status &= ~BIT(MPTCP_PM_ESTABLISHED);
2183 mptcp_pm_nl_fully_established(msk);
2184 }
2185 if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) {
2186 pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED);
2187 mptcp_pm_nl_subflow_established(msk);
2188 }
2189
2190 spin_unlock_bh(&msk->pm.lock);
2191 }
2192
2193 static void __mptcp_close_subflow(struct mptcp_sock *msk)
2194 {
2195 struct mptcp_subflow_context *subflow, *tmp;
2196
2197 list_for_each_entry_safe(subflow, tmp, &msk->conn_list, node) {
2198 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2199
2200 if (inet_sk_state_load(ssk) != TCP_CLOSE)
2201 continue;
2202
2203 __mptcp_close_ssk((struct sock *)msk, ssk, subflow);
2204 }
2205 }
2206
2207 static bool mptcp_check_close_timeout(const struct sock *sk)
2208 {
2209 s32 delta = tcp_jiffies32 - inet_csk(sk)->icsk_mtup.probe_timestamp;
2210 struct mptcp_subflow_context *subflow;
2211
2212 if (delta >= TCP_TIMEWAIT_LEN)
2213 return true;
2214
2215 /* if all subflows are in closed status don't bother with additional
2216 * timeout
2217 */
2218 mptcp_for_each_subflow(mptcp_sk(sk), subflow) {
2219 if (inet_sk_state_load(mptcp_subflow_tcp_sock(subflow)) !=
2220 TCP_CLOSE)
2221 return false;
2222 }
2223 return true;
2224 }
2225
2226 static void mptcp_check_fastclose(struct mptcp_sock *msk)
2227 {
2228 struct mptcp_subflow_context *subflow, *tmp;
2229 struct sock *sk = &msk->sk.icsk_inet.sk;
2230
2231 if (likely(!READ_ONCE(msk->rcv_fastclose)))
2232 return;
2233
2234 mptcp_token_destroy(msk);
2235
2236 list_for_each_entry_safe(subflow, tmp, &msk->conn_list, node) {
2237 struct sock *tcp_sk = mptcp_subflow_tcp_sock(subflow);
2238
2239 lock_sock(tcp_sk);
2240 if (tcp_sk->sk_state != TCP_CLOSE) {
2241 tcp_send_active_reset(tcp_sk, GFP_ATOMIC);
2242 tcp_set_state(tcp_sk, TCP_CLOSE);
2243 }
2244 release_sock(tcp_sk);
2245 }
2246
2247 inet_sk_state_store(sk, TCP_CLOSE);
2248 sk->sk_shutdown = SHUTDOWN_MASK;
2249 smp_mb__before_atomic(); /* SHUTDOWN must be visible first */
2250 set_bit(MPTCP_DATA_READY, &msk->flags);
2251 set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags);
2252
2253 mptcp_close_wake_up(sk);
2254 }
2255
2256 static void mptcp_worker(struct work_struct *work)
2257 {
2258 struct mptcp_sock *msk = container_of(work, struct mptcp_sock, work);
2259 struct sock *ssk, *sk = &msk->sk.icsk_inet.sk;
2260 struct mptcp_sendmsg_info info = {};
2261 struct mptcp_data_frag *dfrag;
2262 size_t copied = 0;
2263 int state, ret;
2264
2265 lock_sock(sk);
2266 state = sk->sk_state;
2267 if (unlikely(state == TCP_CLOSE))
2268 goto unlock;
2269
2270 mptcp_check_data_fin_ack(sk);
2271 __mptcp_flush_join_list(msk);
2272
2273 mptcp_check_fastclose(msk);
2274
2275 if (test_and_clear_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags))
2276 __mptcp_close_subflow(msk);
2277
2278 if (msk->pm.status)
2279 pm_work(msk);
2280
2281 if (test_and_clear_bit(MPTCP_WORK_EOF, &msk->flags))
2282 mptcp_check_for_eof(msk);
2283
2284 __mptcp_check_send_data_fin(sk);
2285 mptcp_check_data_fin(sk);
2286
2287 /* if the msk data is completely acked, or the socket timedout,
2288 * there is no point in keeping around an orphaned sk
2289 */
2290 if (sock_flag(sk, SOCK_DEAD) &&
2291 (mptcp_check_close_timeout(sk) ||
2292 (state != sk->sk_state &&
2293 ((1 << inet_sk_state_load(sk)) & (TCPF_CLOSE | TCPF_FIN_WAIT2))))) {
2294 inet_sk_state_store(sk, TCP_CLOSE);
2295 __mptcp_destroy_sock(sk);
2296 goto unlock;
2297 }
2298
2299 if (!test_and_clear_bit(MPTCP_WORK_RTX, &msk->flags))
2300 goto unlock;
2301
2302 dfrag = mptcp_rtx_head(sk);
2303 if (!dfrag)
2304 goto unlock;
2305
2306 ssk = mptcp_subflow_get_retrans(msk);
2307 if (!ssk)
2308 goto reset_unlock;
2309
2310 lock_sock(ssk);
2311
2312 /* limit retransmission to the bytes already sent on some subflows */
2313 info.sent = 0;
2314 info.limit = dfrag->already_sent;
2315 while (info.sent < dfrag->already_sent) {
2316 if (!mptcp_alloc_tx_skb(sk, ssk))
2317 break;
2318
2319 ret = mptcp_sendmsg_frag(sk, ssk, dfrag, &info);
2320 if (ret <= 0)
2321 break;
2322
2323 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_RETRANSSEGS);
2324 copied += ret;
2325 info.sent += ret;
2326 }
2327 if (copied)
2328 tcp_push(ssk, 0, info.mss_now, tcp_sk(ssk)->nonagle,
2329 info.size_goal);
2330
2331 mptcp_set_timeout(sk, ssk);
2332 release_sock(ssk);
2333
2334 reset_unlock:
2335 if (!mptcp_timer_pending(sk))
2336 mptcp_reset_timer(sk);
2337
2338 unlock:
2339 release_sock(sk);
2340 sock_put(sk);
2341 }
2342
2343 static int __mptcp_init_sock(struct sock *sk)
2344 {
2345 struct mptcp_sock *msk = mptcp_sk(sk);
2346
2347 spin_lock_init(&msk->join_list_lock);
2348
2349 INIT_LIST_HEAD(&msk->conn_list);
2350 INIT_LIST_HEAD(&msk->join_list);
2351 INIT_LIST_HEAD(&msk->rtx_queue);
2352 INIT_WORK(&msk->work, mptcp_worker);
2353 __skb_queue_head_init(&msk->receive_queue);
2354 __skb_queue_head_init(&msk->skb_tx_cache);
2355 msk->out_of_order_queue = RB_ROOT;
2356 msk->first_pending = NULL;
2357 msk->wmem_reserved = 0;
2358 msk->rmem_released = 0;
2359 msk->tx_pending_data = 0;
2360 msk->size_goal_cache = TCP_BASE_MSS;
2361
2362 msk->ack_hint = NULL;
2363 msk->first = NULL;
2364 inet_csk(sk)->icsk_sync_mss = mptcp_sync_mss;
2365
2366 mptcp_pm_data_init(msk);
2367
2368 /* re-use the csk retrans timer for MPTCP-level retrans */
2369 timer_setup(&msk->sk.icsk_retransmit_timer, mptcp_retransmit_timer, 0);
2370 timer_setup(&sk->sk_timer, mptcp_timeout_timer, 0);
2371 return 0;
2372 }
2373
2374 static int mptcp_init_sock(struct sock *sk)
2375 {
2376 struct net *net = sock_net(sk);
2377 int ret;
2378
2379 ret = __mptcp_init_sock(sk);
2380 if (ret)
2381 return ret;
2382
2383 if (!mptcp_is_enabled(net))
2384 return -ENOPROTOOPT;
2385
2386 if (unlikely(!net->mib.mptcp_statistics) && !mptcp_mib_alloc(net))
2387 return -ENOMEM;
2388
2389 ret = __mptcp_socket_create(mptcp_sk(sk));
2390 if (ret)
2391 return ret;
2392
2393 sk_sockets_allocated_inc(sk);
2394 sk->sk_rcvbuf = sock_net(sk)->ipv4.sysctl_tcp_rmem[1];
2395 sk->sk_sndbuf = sock_net(sk)->ipv4.sysctl_tcp_wmem[1];
2396
2397 return 0;
2398 }
2399
2400 static void __mptcp_clear_xmit(struct sock *sk)
2401 {
2402 struct mptcp_sock *msk = mptcp_sk(sk);
2403 struct mptcp_data_frag *dtmp, *dfrag;
2404 struct sk_buff *skb;
2405
2406 WRITE_ONCE(msk->first_pending, NULL);
2407 list_for_each_entry_safe(dfrag, dtmp, &msk->rtx_queue, list)
2408 dfrag_clear(sk, dfrag);
2409 while ((skb = __skb_dequeue(&msk->skb_tx_cache)) != NULL) {
2410 sk->sk_forward_alloc += skb->truesize;
2411 kfree_skb(skb);
2412 }
2413 }
2414
2415 static void mptcp_cancel_work(struct sock *sk)
2416 {
2417 struct mptcp_sock *msk = mptcp_sk(sk);
2418
2419 if (cancel_work_sync(&msk->work))
2420 __sock_put(sk);
2421 }
2422
2423 void mptcp_subflow_shutdown(struct sock *sk, struct sock *ssk, int how)
2424 {
2425 lock_sock(ssk);
2426
2427 switch (ssk->sk_state) {
2428 case TCP_LISTEN:
2429 if (!(how & RCV_SHUTDOWN))
2430 break;
2431 fallthrough;
2432 case TCP_SYN_SENT:
2433 tcp_disconnect(ssk, O_NONBLOCK);
2434 break;
2435 default:
2436 if (__mptcp_check_fallback(mptcp_sk(sk))) {
2437 pr_debug("Fallback");
2438 ssk->sk_shutdown |= how;
2439 tcp_shutdown(ssk, how);
2440 } else {
2441 pr_debug("Sending DATA_FIN on subflow %p", ssk);
2442 mptcp_set_timeout(sk, ssk);
2443 tcp_send_ack(ssk);
2444 }
2445 break;
2446 }
2447
2448 release_sock(ssk);
2449 }
2450
2451 static const unsigned char new_state[16] = {
2452 /* current state: new state: action: */
2453 [0 /* (Invalid) */] = TCP_CLOSE,
2454 [TCP_ESTABLISHED] = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2455 [TCP_SYN_SENT] = TCP_CLOSE,
2456 [TCP_SYN_RECV] = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2457 [TCP_FIN_WAIT1] = TCP_FIN_WAIT1,
2458 [TCP_FIN_WAIT2] = TCP_FIN_WAIT2,
2459 [TCP_TIME_WAIT] = TCP_CLOSE, /* should not happen ! */
2460 [TCP_CLOSE] = TCP_CLOSE,
2461 [TCP_CLOSE_WAIT] = TCP_LAST_ACK | TCP_ACTION_FIN,
2462 [TCP_LAST_ACK] = TCP_LAST_ACK,
2463 [TCP_LISTEN] = TCP_CLOSE,
2464 [TCP_CLOSING] = TCP_CLOSING,
2465 [TCP_NEW_SYN_RECV] = TCP_CLOSE, /* should not happen ! */
2466 };
2467
2468 static int mptcp_close_state(struct sock *sk)
2469 {
2470 int next = (int)new_state[sk->sk_state];
2471 int ns = next & TCP_STATE_MASK;
2472
2473 inet_sk_state_store(sk, ns);
2474
2475 return next & TCP_ACTION_FIN;
2476 }
2477
2478 static void __mptcp_check_send_data_fin(struct sock *sk)
2479 {
2480 struct mptcp_subflow_context *subflow;
2481 struct mptcp_sock *msk = mptcp_sk(sk);
2482
2483 pr_debug("msk=%p snd_data_fin_enable=%d pending=%d snd_nxt=%llu write_seq=%llu",
2484 msk, msk->snd_data_fin_enable, !!mptcp_send_head(sk),
2485 msk->snd_nxt, msk->write_seq);
2486
2487 /* we still need to enqueue subflows or not really shutting down,
2488 * skip this
2489 */
2490 if (!msk->snd_data_fin_enable || msk->snd_nxt + 1 != msk->write_seq ||
2491 mptcp_send_head(sk))
2492 return;
2493
2494 WRITE_ONCE(msk->snd_nxt, msk->write_seq);
2495
2496 /* fallback socket will not get data_fin/ack, can move to the next
2497 * state now
2498 */
2499 if (__mptcp_check_fallback(msk)) {
2500 if ((1 << sk->sk_state) & (TCPF_CLOSING | TCPF_LAST_ACK)) {
2501 inet_sk_state_store(sk, TCP_CLOSE);
2502 mptcp_close_wake_up(sk);
2503 } else if (sk->sk_state == TCP_FIN_WAIT1) {
2504 inet_sk_state_store(sk, TCP_FIN_WAIT2);
2505 }
2506 }
2507
2508 __mptcp_flush_join_list(msk);
2509 mptcp_for_each_subflow(msk, subflow) {
2510 struct sock *tcp_sk = mptcp_subflow_tcp_sock(subflow);
2511
2512 mptcp_subflow_shutdown(sk, tcp_sk, SEND_SHUTDOWN);
2513 }
2514 }
2515
2516 static void __mptcp_wr_shutdown(struct sock *sk)
2517 {
2518 struct mptcp_sock *msk = mptcp_sk(sk);
2519
2520 pr_debug("msk=%p snd_data_fin_enable=%d shutdown=%x state=%d pending=%d",
2521 msk, msk->snd_data_fin_enable, sk->sk_shutdown, sk->sk_state,
2522 !!mptcp_send_head(sk));
2523
2524 /* will be ignored by fallback sockets */
2525 WRITE_ONCE(msk->write_seq, msk->write_seq + 1);
2526 WRITE_ONCE(msk->snd_data_fin_enable, 1);
2527
2528 __mptcp_check_send_data_fin(sk);
2529 }
2530
2531 static void __mptcp_destroy_sock(struct sock *sk)
2532 {
2533 struct mptcp_subflow_context *subflow, *tmp;
2534 struct mptcp_sock *msk = mptcp_sk(sk);
2535 LIST_HEAD(conn_list);
2536
2537 pr_debug("msk=%p", msk);
2538
2539 /* be sure to always acquire the join list lock, to sync vs
2540 * mptcp_finish_join().
2541 */
2542 spin_lock_bh(&msk->join_list_lock);
2543 list_splice_tail_init(&msk->join_list, &msk->conn_list);
2544 spin_unlock_bh(&msk->join_list_lock);
2545 list_splice_init(&msk->conn_list, &conn_list);
2546
2547 sk_stop_timer(sk, &msk->sk.icsk_retransmit_timer);
2548 sk_stop_timer(sk, &sk->sk_timer);
2549 msk->pm.status = 0;
2550
2551 list_for_each_entry_safe(subflow, tmp, &conn_list, node) {
2552 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2553 __mptcp_close_ssk(sk, ssk, subflow);
2554 }
2555
2556 sk->sk_prot->destroy(sk);
2557
2558 WARN_ON_ONCE(msk->wmem_reserved);
2559 WARN_ON_ONCE(msk->rmem_released);
2560 sk_stream_kill_queues(sk);
2561 xfrm_sk_free_policy(sk);
2562 sk_refcnt_debug_release(sk);
2563 sock_put(sk);
2564 }
2565
2566 static void mptcp_close(struct sock *sk, long timeout)
2567 {
2568 struct mptcp_subflow_context *subflow;
2569 bool do_cancel_work = false;
2570
2571 lock_sock(sk);
2572 sk->sk_shutdown = SHUTDOWN_MASK;
2573
2574 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE)) {
2575 inet_sk_state_store(sk, TCP_CLOSE);
2576 goto cleanup;
2577 }
2578
2579 if (mptcp_close_state(sk))
2580 __mptcp_wr_shutdown(sk);
2581
2582 sk_stream_wait_close(sk, timeout);
2583
2584 cleanup:
2585 /* orphan all the subflows */
2586 inet_csk(sk)->icsk_mtup.probe_timestamp = tcp_jiffies32;
2587 list_for_each_entry(subflow, &mptcp_sk(sk)->conn_list, node) {
2588 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2589 bool slow, dispose_socket;
2590 struct socket *sock;
2591
2592 slow = lock_sock_fast(ssk);
2593 sock = ssk->sk_socket;
2594 dispose_socket = sock && sock != sk->sk_socket;
2595 sock_orphan(ssk);
2596 unlock_sock_fast(ssk, slow);
2597
2598 /* for the outgoing subflows we additionally need to free
2599 * the associated socket
2600 */
2601 if (dispose_socket)
2602 iput(SOCK_INODE(sock));
2603 }
2604 sock_orphan(sk);
2605
2606 sock_hold(sk);
2607 pr_debug("msk=%p state=%d", sk, sk->sk_state);
2608 if (sk->sk_state == TCP_CLOSE) {
2609 __mptcp_destroy_sock(sk);
2610 do_cancel_work = true;
2611 } else {
2612 sk_reset_timer(sk, &sk->sk_timer, jiffies + TCP_TIMEWAIT_LEN);
2613 }
2614 release_sock(sk);
2615 if (do_cancel_work)
2616 mptcp_cancel_work(sk);
2617 sock_put(sk);
2618 }
2619
2620 static void mptcp_copy_inaddrs(struct sock *msk, const struct sock *ssk)
2621 {
2622 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2623 const struct ipv6_pinfo *ssk6 = inet6_sk(ssk);
2624 struct ipv6_pinfo *msk6 = inet6_sk(msk);
2625
2626 msk->sk_v6_daddr = ssk->sk_v6_daddr;
2627 msk->sk_v6_rcv_saddr = ssk->sk_v6_rcv_saddr;
2628
2629 if (msk6 && ssk6) {
2630 msk6->saddr = ssk6->saddr;
2631 msk6->flow_label = ssk6->flow_label;
2632 }
2633 #endif
2634
2635 inet_sk(msk)->inet_num = inet_sk(ssk)->inet_num;
2636 inet_sk(msk)->inet_dport = inet_sk(ssk)->inet_dport;
2637 inet_sk(msk)->inet_sport = inet_sk(ssk)->inet_sport;
2638 inet_sk(msk)->inet_daddr = inet_sk(ssk)->inet_daddr;
2639 inet_sk(msk)->inet_saddr = inet_sk(ssk)->inet_saddr;
2640 inet_sk(msk)->inet_rcv_saddr = inet_sk(ssk)->inet_rcv_saddr;
2641 }
2642
2643 static int mptcp_disconnect(struct sock *sk, int flags)
2644 {
2645 /* Should never be called.
2646 * inet_stream_connect() calls ->disconnect, but that
2647 * refers to the subflow socket, not the mptcp one.
2648 */
2649 WARN_ON_ONCE(1);
2650 return 0;
2651 }
2652
2653 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2654 static struct ipv6_pinfo *mptcp_inet6_sk(const struct sock *sk)
2655 {
2656 unsigned int offset = sizeof(struct mptcp6_sock) - sizeof(struct ipv6_pinfo);
2657
2658 return (struct ipv6_pinfo *)(((u8 *)sk) + offset);
2659 }
2660 #endif
2661
2662 struct sock *mptcp_sk_clone(const struct sock *sk,
2663 const struct mptcp_options_received *mp_opt,
2664 struct request_sock *req)
2665 {
2666 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
2667 struct sock *nsk = sk_clone_lock(sk, GFP_ATOMIC);
2668 struct mptcp_sock *msk;
2669 u64 ack_seq;
2670
2671 if (!nsk)
2672 return NULL;
2673
2674 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2675 if (nsk->sk_family == AF_INET6)
2676 inet_sk(nsk)->pinet6 = mptcp_inet6_sk(nsk);
2677 #endif
2678
2679 __mptcp_init_sock(nsk);
2680
2681 msk = mptcp_sk(nsk);
2682 msk->local_key = subflow_req->local_key;
2683 msk->token = subflow_req->token;
2684 msk->subflow = NULL;
2685 WRITE_ONCE(msk->fully_established, false);
2686
2687 msk->write_seq = subflow_req->idsn + 1;
2688 msk->snd_nxt = msk->write_seq;
2689 msk->snd_una = msk->write_seq;
2690 msk->wnd_end = msk->snd_nxt + req->rsk_rcv_wnd;
2691
2692 if (mp_opt->mp_capable) {
2693 msk->can_ack = true;
2694 msk->remote_key = mp_opt->sndr_key;
2695 mptcp_crypto_key_sha(msk->remote_key, NULL, &ack_seq);
2696 ack_seq++;
2697 WRITE_ONCE(msk->ack_seq, ack_seq);
2698 WRITE_ONCE(msk->rcv_wnd_sent, ack_seq);
2699 }
2700
2701 sock_reset_flag(nsk, SOCK_RCU_FREE);
2702 /* will be fully established after successful MPC subflow creation */
2703 inet_sk_state_store(nsk, TCP_SYN_RECV);
2704
2705 security_inet_csk_clone(nsk, req);
2706 bh_unlock_sock(nsk);
2707
2708 /* keep a single reference */
2709 __sock_put(nsk);
2710 return nsk;
2711 }
2712
2713 void mptcp_rcv_space_init(struct mptcp_sock *msk, const struct sock *ssk)
2714 {
2715 const struct tcp_sock *tp = tcp_sk(ssk);
2716
2717 msk->rcvq_space.copied = 0;
2718 msk->rcvq_space.rtt_us = 0;
2719
2720 msk->rcvq_space.time = tp->tcp_mstamp;
2721
2722 /* initial rcv_space offering made to peer */
2723 msk->rcvq_space.space = min_t(u32, tp->rcv_wnd,
2724 TCP_INIT_CWND * tp->advmss);
2725 if (msk->rcvq_space.space == 0)
2726 msk->rcvq_space.space = TCP_INIT_CWND * TCP_MSS_DEFAULT;
2727
2728 WRITE_ONCE(msk->wnd_end, msk->snd_nxt + tcp_sk(ssk)->snd_wnd);
2729 }
2730
2731 static struct sock *mptcp_accept(struct sock *sk, int flags, int *err,
2732 bool kern)
2733 {
2734 struct mptcp_sock *msk = mptcp_sk(sk);
2735 struct socket *listener;
2736 struct sock *newsk;
2737
2738 listener = __mptcp_nmpc_socket(msk);
2739 if (WARN_ON_ONCE(!listener)) {
2740 *err = -EINVAL;
2741 return NULL;
2742 }
2743
2744 pr_debug("msk=%p, listener=%p", msk, mptcp_subflow_ctx(listener->sk));
2745 newsk = inet_csk_accept(listener->sk, flags, err, kern);
2746 if (!newsk)
2747 return NULL;
2748
2749 pr_debug("msk=%p, subflow is mptcp=%d", msk, sk_is_mptcp(newsk));
2750 if (sk_is_mptcp(newsk)) {
2751 struct mptcp_subflow_context *subflow;
2752 struct sock *new_mptcp_sock;
2753
2754 subflow = mptcp_subflow_ctx(newsk);
2755 new_mptcp_sock = subflow->conn;
2756
2757 /* is_mptcp should be false if subflow->conn is missing, see
2758 * subflow_syn_recv_sock()
2759 */
2760 if (WARN_ON_ONCE(!new_mptcp_sock)) {
2761 tcp_sk(newsk)->is_mptcp = 0;
2762 return newsk;
2763 }
2764
2765 /* acquire the 2nd reference for the owning socket */
2766 sock_hold(new_mptcp_sock);
2767 newsk = new_mptcp_sock;
2768 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPCAPABLEPASSIVEACK);
2769 } else {
2770 MPTCP_INC_STATS(sock_net(sk),
2771 MPTCP_MIB_MPCAPABLEPASSIVEFALLBACK);
2772 }
2773
2774 return newsk;
2775 }
2776
2777 void mptcp_destroy_common(struct mptcp_sock *msk)
2778 {
2779 struct sock *sk = (struct sock *)msk;
2780
2781 __mptcp_clear_xmit(sk);
2782
2783 /* move to sk_receive_queue, sk_stream_kill_queues will purge it */
2784 skb_queue_splice_tail_init(&msk->receive_queue, &sk->sk_receive_queue);
2785
2786 skb_rbtree_purge(&msk->out_of_order_queue);
2787 mptcp_token_destroy(msk);
2788 mptcp_pm_free_anno_list(msk);
2789 }
2790
2791 static void mptcp_destroy(struct sock *sk)
2792 {
2793 struct mptcp_sock *msk = mptcp_sk(sk);
2794
2795 mptcp_destroy_common(msk);
2796 sk_sockets_allocated_dec(sk);
2797 }
2798
2799 static int mptcp_setsockopt_sol_socket(struct mptcp_sock *msk, int optname,
2800 sockptr_t optval, unsigned int optlen)
2801 {
2802 struct sock *sk = (struct sock *)msk;
2803 struct socket *ssock;
2804 int ret;
2805
2806 switch (optname) {
2807 case SO_REUSEPORT:
2808 case SO_REUSEADDR:
2809 lock_sock(sk);
2810 ssock = __mptcp_nmpc_socket(msk);
2811 if (!ssock) {
2812 release_sock(sk);
2813 return -EINVAL;
2814 }
2815
2816 ret = sock_setsockopt(ssock, SOL_SOCKET, optname, optval, optlen);
2817 if (ret == 0) {
2818 if (optname == SO_REUSEPORT)
2819 sk->sk_reuseport = ssock->sk->sk_reuseport;
2820 else if (optname == SO_REUSEADDR)
2821 sk->sk_reuse = ssock->sk->sk_reuse;
2822 }
2823 release_sock(sk);
2824 return ret;
2825 }
2826
2827 return sock_setsockopt(sk->sk_socket, SOL_SOCKET, optname, optval, optlen);
2828 }
2829
2830 static int mptcp_setsockopt_v6(struct mptcp_sock *msk, int optname,
2831 sockptr_t optval, unsigned int optlen)
2832 {
2833 struct sock *sk = (struct sock *)msk;
2834 int ret = -EOPNOTSUPP;
2835 struct socket *ssock;
2836
2837 switch (optname) {
2838 case IPV6_V6ONLY:
2839 lock_sock(sk);
2840 ssock = __mptcp_nmpc_socket(msk);
2841 if (!ssock) {
2842 release_sock(sk);
2843 return -EINVAL;
2844 }
2845
2846 ret = tcp_setsockopt(ssock->sk, SOL_IPV6, optname, optval, optlen);
2847 if (ret == 0)
2848 sk->sk_ipv6only = ssock->sk->sk_ipv6only;
2849
2850 release_sock(sk);
2851 break;
2852 }
2853
2854 return ret;
2855 }
2856
2857 static int mptcp_setsockopt(struct sock *sk, int level, int optname,
2858 sockptr_t optval, unsigned int optlen)
2859 {
2860 struct mptcp_sock *msk = mptcp_sk(sk);
2861 struct sock *ssk;
2862
2863 pr_debug("msk=%p", msk);
2864
2865 if (level == SOL_SOCKET)
2866 return mptcp_setsockopt_sol_socket(msk, optname, optval, optlen);
2867
2868 /* @@ the meaning of setsockopt() when the socket is connected and
2869 * there are multiple subflows is not yet defined. It is up to the
2870 * MPTCP-level socket to configure the subflows until the subflow
2871 * is in TCP fallback, when TCP socket options are passed through
2872 * to the one remaining subflow.
2873 */
2874 lock_sock(sk);
2875 ssk = __mptcp_tcp_fallback(msk);
2876 release_sock(sk);
2877 if (ssk)
2878 return tcp_setsockopt(ssk, level, optname, optval, optlen);
2879
2880 if (level == SOL_IPV6)
2881 return mptcp_setsockopt_v6(msk, optname, optval, optlen);
2882
2883 return -EOPNOTSUPP;
2884 }
2885
2886 static int mptcp_getsockopt(struct sock *sk, int level, int optname,
2887 char __user *optval, int __user *option)
2888 {
2889 struct mptcp_sock *msk = mptcp_sk(sk);
2890 struct sock *ssk;
2891
2892 pr_debug("msk=%p", msk);
2893
2894 /* @@ the meaning of setsockopt() when the socket is connected and
2895 * there are multiple subflows is not yet defined. It is up to the
2896 * MPTCP-level socket to configure the subflows until the subflow
2897 * is in TCP fallback, when socket options are passed through
2898 * to the one remaining subflow.
2899 */
2900 lock_sock(sk);
2901 ssk = __mptcp_tcp_fallback(msk);
2902 release_sock(sk);
2903 if (ssk)
2904 return tcp_getsockopt(ssk, level, optname, optval, option);
2905
2906 return -EOPNOTSUPP;
2907 }
2908
2909 void __mptcp_data_acked(struct sock *sk)
2910 {
2911 if (!sock_owned_by_user(sk))
2912 __mptcp_clean_una(sk);
2913 else
2914 set_bit(MPTCP_CLEAN_UNA, &mptcp_sk(sk)->flags);
2915
2916 if (mptcp_pending_data_fin_ack(sk))
2917 mptcp_schedule_work(sk);
2918 }
2919
2920 void __mptcp_check_push(struct sock *sk, struct sock *ssk)
2921 {
2922 if (!mptcp_send_head(sk))
2923 return;
2924
2925 if (!sock_owned_by_user(sk))
2926 __mptcp_subflow_push_pending(sk, ssk);
2927 else
2928 set_bit(MPTCP_PUSH_PENDING, &mptcp_sk(sk)->flags);
2929 }
2930
2931 #define MPTCP_DEFERRED_ALL (TCPF_WRITE_TIMER_DEFERRED)
2932
2933 /* processes deferred events and flush wmem */
2934 static void mptcp_release_cb(struct sock *sk)
2935 {
2936 unsigned long flags, nflags;
2937
2938 /* push_pending may touch wmem_reserved, do it before the later
2939 * cleanup
2940 */
2941 if (test_and_clear_bit(MPTCP_CLEAN_UNA, &mptcp_sk(sk)->flags))
2942 __mptcp_clean_una(sk);
2943 if (test_and_clear_bit(MPTCP_PUSH_PENDING, &mptcp_sk(sk)->flags)) {
2944 /* mptcp_push_pending() acquires the subflow socket lock
2945 *
2946 * 1) can't be invoked in atomic scope
2947 * 2) must avoid ABBA deadlock with msk socket spinlock: the RX
2948 * datapath acquires the msk socket spinlock while helding
2949 * the subflow socket lock
2950 */
2951
2952 spin_unlock_bh(&sk->sk_lock.slock);
2953 mptcp_push_pending(sk, 0);
2954 spin_lock_bh(&sk->sk_lock.slock);
2955 }
2956
2957 /* clear any wmem reservation and errors */
2958 __mptcp_update_wmem(sk);
2959 __mptcp_update_rmem(sk);
2960
2961 do {
2962 flags = sk->sk_tsq_flags;
2963 if (!(flags & MPTCP_DEFERRED_ALL))
2964 return;
2965 nflags = flags & ~MPTCP_DEFERRED_ALL;
2966 } while (cmpxchg(&sk->sk_tsq_flags, flags, nflags) != flags);
2967
2968 sock_release_ownership(sk);
2969
2970 if (flags & TCPF_WRITE_TIMER_DEFERRED) {
2971 mptcp_retransmit_handler(sk);
2972 __sock_put(sk);
2973 }
2974 }
2975
2976 static int mptcp_hash(struct sock *sk)
2977 {
2978 /* should never be called,
2979 * we hash the TCP subflows not the master socket
2980 */
2981 WARN_ON_ONCE(1);
2982 return 0;
2983 }
2984
2985 static void mptcp_unhash(struct sock *sk)
2986 {
2987 /* called from sk_common_release(), but nothing to do here */
2988 }
2989
2990 static int mptcp_get_port(struct sock *sk, unsigned short snum)
2991 {
2992 struct mptcp_sock *msk = mptcp_sk(sk);
2993 struct socket *ssock;
2994
2995 ssock = __mptcp_nmpc_socket(msk);
2996 pr_debug("msk=%p, subflow=%p", msk, ssock);
2997 if (WARN_ON_ONCE(!ssock))
2998 return -EINVAL;
2999
3000 return inet_csk_get_port(ssock->sk, snum);
3001 }
3002
3003 void mptcp_finish_connect(struct sock *ssk)
3004 {
3005 struct mptcp_subflow_context *subflow;
3006 struct mptcp_sock *msk;
3007 struct sock *sk;
3008 u64 ack_seq;
3009
3010 subflow = mptcp_subflow_ctx(ssk);
3011 sk = subflow->conn;
3012 msk = mptcp_sk(sk);
3013
3014 pr_debug("msk=%p, token=%u", sk, subflow->token);
3015
3016 mptcp_crypto_key_sha(subflow->remote_key, NULL, &ack_seq);
3017 ack_seq++;
3018 subflow->map_seq = ack_seq;
3019 subflow->map_subflow_seq = 1;
3020
3021 /* the socket is not connected yet, no msk/subflow ops can access/race
3022 * accessing the field below
3023 */
3024 WRITE_ONCE(msk->remote_key, subflow->remote_key);
3025 WRITE_ONCE(msk->local_key, subflow->local_key);
3026 WRITE_ONCE(msk->write_seq, subflow->idsn + 1);
3027 WRITE_ONCE(msk->snd_nxt, msk->write_seq);
3028 WRITE_ONCE(msk->ack_seq, ack_seq);
3029 WRITE_ONCE(msk->rcv_wnd_sent, ack_seq);
3030 WRITE_ONCE(msk->can_ack, 1);
3031 WRITE_ONCE(msk->snd_una, msk->write_seq);
3032
3033 mptcp_pm_new_connection(msk, 0);
3034
3035 mptcp_rcv_space_init(msk, ssk);
3036 }
3037
3038 static void mptcp_sock_graft(struct sock *sk, struct socket *parent)
3039 {
3040 write_lock_bh(&sk->sk_callback_lock);
3041 rcu_assign_pointer(sk->sk_wq, &parent->wq);
3042 sk_set_socket(sk, parent);
3043 sk->sk_uid = SOCK_INODE(parent)->i_uid;
3044 write_unlock_bh(&sk->sk_callback_lock);
3045 }
3046
3047 bool mptcp_finish_join(struct sock *ssk)
3048 {
3049 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
3050 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
3051 struct sock *parent = (void *)msk;
3052 struct socket *parent_sock;
3053 bool ret;
3054
3055 pr_debug("msk=%p, subflow=%p", msk, subflow);
3056
3057 /* mptcp socket already closing? */
3058 if (!mptcp_is_fully_established(parent))
3059 return false;
3060
3061 if (!msk->pm.server_side)
3062 return true;
3063
3064 if (!mptcp_pm_allow_new_subflow(msk))
3065 return false;
3066
3067 /* active connections are already on conn_list, and we can't acquire
3068 * msk lock here.
3069 * use the join list lock as synchronization point and double-check
3070 * msk status to avoid racing with __mptcp_destroy_sock()
3071 */
3072 spin_lock_bh(&msk->join_list_lock);
3073 ret = inet_sk_state_load(parent) == TCP_ESTABLISHED;
3074 if (ret && !WARN_ON_ONCE(!list_empty(&subflow->node))) {
3075 list_add_tail(&subflow->node, &msk->join_list);
3076 sock_hold(ssk);
3077 }
3078 spin_unlock_bh(&msk->join_list_lock);
3079 if (!ret)
3080 return false;
3081
3082 /* attach to msk socket only after we are sure he will deal with us
3083 * at close time
3084 */
3085 parent_sock = READ_ONCE(parent->sk_socket);
3086 if (parent_sock && !ssk->sk_socket)
3087 mptcp_sock_graft(ssk, parent_sock);
3088 subflow->map_seq = READ_ONCE(msk->ack_seq);
3089 return true;
3090 }
3091
3092 static struct proto mptcp_prot = {
3093 .name = "MPTCP",
3094 .owner = THIS_MODULE,
3095 .init = mptcp_init_sock,
3096 .disconnect = mptcp_disconnect,
3097 .close = mptcp_close,
3098 .accept = mptcp_accept,
3099 .setsockopt = mptcp_setsockopt,
3100 .getsockopt = mptcp_getsockopt,
3101 .shutdown = tcp_shutdown,
3102 .destroy = mptcp_destroy,
3103 .sendmsg = mptcp_sendmsg,
3104 .recvmsg = mptcp_recvmsg,
3105 .release_cb = mptcp_release_cb,
3106 .hash = mptcp_hash,
3107 .unhash = mptcp_unhash,
3108 .get_port = mptcp_get_port,
3109 .sockets_allocated = &mptcp_sockets_allocated,
3110 .memory_allocated = &tcp_memory_allocated,
3111 .memory_pressure = &tcp_memory_pressure,
3112 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_tcp_wmem),
3113 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_tcp_rmem),
3114 .sysctl_mem = sysctl_tcp_mem,
3115 .obj_size = sizeof(struct mptcp_sock),
3116 .slab_flags = SLAB_TYPESAFE_BY_RCU,
3117 .no_autobind = true,
3118 };
3119
3120 static int mptcp_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3121 {
3122 struct mptcp_sock *msk = mptcp_sk(sock->sk);
3123 struct socket *ssock;
3124 int err;
3125
3126 lock_sock(sock->sk);
3127 ssock = __mptcp_nmpc_socket(msk);
3128 if (!ssock) {
3129 err = -EINVAL;
3130 goto unlock;
3131 }
3132
3133 err = ssock->ops->bind(ssock, uaddr, addr_len);
3134 if (!err)
3135 mptcp_copy_inaddrs(sock->sk, ssock->sk);
3136
3137 unlock:
3138 release_sock(sock->sk);
3139 return err;
3140 }
3141
3142 static void mptcp_subflow_early_fallback(struct mptcp_sock *msk,
3143 struct mptcp_subflow_context *subflow)
3144 {
3145 subflow->request_mptcp = 0;
3146 __mptcp_do_fallback(msk);
3147 }
3148
3149 static int mptcp_stream_connect(struct socket *sock, struct sockaddr *uaddr,
3150 int addr_len, int flags)
3151 {
3152 struct mptcp_sock *msk = mptcp_sk(sock->sk);
3153 struct mptcp_subflow_context *subflow;
3154 struct socket *ssock;
3155 int err;
3156
3157 lock_sock(sock->sk);
3158 if (sock->state != SS_UNCONNECTED && msk->subflow) {
3159 /* pending connection or invalid state, let existing subflow
3160 * cope with that
3161 */
3162 ssock = msk->subflow;
3163 goto do_connect;
3164 }
3165
3166 ssock = __mptcp_nmpc_socket(msk);
3167 if (!ssock) {
3168 err = -EINVAL;
3169 goto unlock;
3170 }
3171
3172 mptcp_token_destroy(msk);
3173 inet_sk_state_store(sock->sk, TCP_SYN_SENT);
3174 subflow = mptcp_subflow_ctx(ssock->sk);
3175 #ifdef CONFIG_TCP_MD5SIG
3176 /* no MPTCP if MD5SIG is enabled on this socket or we may run out of
3177 * TCP option space.
3178 */
3179 if (rcu_access_pointer(tcp_sk(ssock->sk)->md5sig_info))
3180 mptcp_subflow_early_fallback(msk, subflow);
3181 #endif
3182 if (subflow->request_mptcp && mptcp_token_new_connect(ssock->sk))
3183 mptcp_subflow_early_fallback(msk, subflow);
3184
3185 do_connect:
3186 err = ssock->ops->connect(ssock, uaddr, addr_len, flags);
3187 sock->state = ssock->state;
3188
3189 /* on successful connect, the msk state will be moved to established by
3190 * subflow_finish_connect()
3191 */
3192 if (!err || err == -EINPROGRESS)
3193 mptcp_copy_inaddrs(sock->sk, ssock->sk);
3194 else
3195 inet_sk_state_store(sock->sk, inet_sk_state_load(ssock->sk));
3196
3197 unlock:
3198 release_sock(sock->sk);
3199 return err;
3200 }
3201
3202 static int mptcp_listen(struct socket *sock, int backlog)
3203 {
3204 struct mptcp_sock *msk = mptcp_sk(sock->sk);
3205 struct socket *ssock;
3206 int err;
3207
3208 pr_debug("msk=%p", msk);
3209
3210 lock_sock(sock->sk);
3211 ssock = __mptcp_nmpc_socket(msk);
3212 if (!ssock) {
3213 err = -EINVAL;
3214 goto unlock;
3215 }
3216
3217 mptcp_token_destroy(msk);
3218 inet_sk_state_store(sock->sk, TCP_LISTEN);
3219 sock_set_flag(sock->sk, SOCK_RCU_FREE);
3220
3221 err = ssock->ops->listen(ssock, backlog);
3222 inet_sk_state_store(sock->sk, inet_sk_state_load(ssock->sk));
3223 if (!err)
3224 mptcp_copy_inaddrs(sock->sk, ssock->sk);
3225
3226 unlock:
3227 release_sock(sock->sk);
3228 return err;
3229 }
3230
3231 static int mptcp_stream_accept(struct socket *sock, struct socket *newsock,
3232 int flags, bool kern)
3233 {
3234 struct mptcp_sock *msk = mptcp_sk(sock->sk);
3235 struct socket *ssock;
3236 int err;
3237
3238 pr_debug("msk=%p", msk);
3239
3240 lock_sock(sock->sk);
3241 if (sock->sk->sk_state != TCP_LISTEN)
3242 goto unlock_fail;
3243
3244 ssock = __mptcp_nmpc_socket(msk);
3245 if (!ssock)
3246 goto unlock_fail;
3247
3248 clear_bit(MPTCP_DATA_READY, &msk->flags);
3249 sock_hold(ssock->sk);
3250 release_sock(sock->sk);
3251
3252 err = ssock->ops->accept(sock, newsock, flags, kern);
3253 if (err == 0 && !mptcp_is_tcpsk(newsock->sk)) {
3254 struct mptcp_sock *msk = mptcp_sk(newsock->sk);
3255 struct mptcp_subflow_context *subflow;
3256 struct sock *newsk = newsock->sk;
3257 bool slowpath;
3258
3259 slowpath = lock_sock_fast(newsk);
3260
3261 /* PM/worker can now acquire the first subflow socket
3262 * lock without racing with listener queue cleanup,
3263 * we can notify it, if needed.
3264 */
3265 subflow = mptcp_subflow_ctx(msk->first);
3266 list_add(&subflow->node, &msk->conn_list);
3267 sock_hold(msk->first);
3268 if (mptcp_is_fully_established(newsk))
3269 mptcp_pm_fully_established(msk);
3270
3271 mptcp_copy_inaddrs(newsk, msk->first);
3272 mptcp_rcv_space_init(msk, msk->first);
3273
3274 /* set ssk->sk_socket of accept()ed flows to mptcp socket.
3275 * This is needed so NOSPACE flag can be set from tcp stack.
3276 */
3277 __mptcp_flush_join_list(msk);
3278 mptcp_for_each_subflow(msk, subflow) {
3279 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
3280
3281 if (!ssk->sk_socket)
3282 mptcp_sock_graft(ssk, newsock);
3283 }
3284 unlock_sock_fast(newsk, slowpath);
3285 }
3286
3287 if (inet_csk_listen_poll(ssock->sk))
3288 set_bit(MPTCP_DATA_READY, &msk->flags);
3289 sock_put(ssock->sk);
3290 return err;
3291
3292 unlock_fail:
3293 release_sock(sock->sk);
3294 return -EINVAL;
3295 }
3296
3297 static __poll_t mptcp_check_readable(struct mptcp_sock *msk)
3298 {
3299 return test_bit(MPTCP_DATA_READY, &msk->flags) ? EPOLLIN | EPOLLRDNORM :
3300 0;
3301 }
3302
3303 static __poll_t mptcp_check_writeable(struct mptcp_sock *msk)
3304 {
3305 struct sock *sk = (struct sock *)msk;
3306
3307 if (unlikely(sk->sk_shutdown & SEND_SHUTDOWN))
3308 return 0;
3309
3310 if (sk_stream_is_writeable(sk))
3311 return EPOLLOUT | EPOLLWRNORM;
3312
3313 set_bit(MPTCP_NOSPACE, &msk->flags);
3314 smp_mb__after_atomic(); /* msk->flags is changed by write_space cb */
3315 if (sk_stream_is_writeable(sk))
3316 return EPOLLOUT | EPOLLWRNORM;
3317
3318 return 0;
3319 }
3320
3321 static __poll_t mptcp_poll(struct file *file, struct socket *sock,
3322 struct poll_table_struct *wait)
3323 {
3324 struct sock *sk = sock->sk;
3325 struct mptcp_sock *msk;
3326 __poll_t mask = 0;
3327 int state;
3328
3329 msk = mptcp_sk(sk);
3330 sock_poll_wait(file, sock, wait);
3331
3332 state = inet_sk_state_load(sk);
3333 pr_debug("msk=%p state=%d flags=%lx", msk, state, msk->flags);
3334 if (state == TCP_LISTEN)
3335 return mptcp_check_readable(msk);
3336
3337 if (state != TCP_SYN_SENT && state != TCP_SYN_RECV) {
3338 mask |= mptcp_check_readable(msk);
3339 mask |= mptcp_check_writeable(msk);
3340 }
3341 if (sk->sk_shutdown & RCV_SHUTDOWN)
3342 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
3343
3344 return mask;
3345 }
3346
3347 static int mptcp_shutdown(struct socket *sock, int how)
3348 {
3349 struct mptcp_sock *msk = mptcp_sk(sock->sk);
3350 struct sock *sk = sock->sk;
3351 int ret = 0;
3352
3353 pr_debug("sk=%p, how=%d", msk, how);
3354
3355 lock_sock(sk);
3356
3357 how++;
3358 if ((how & ~SHUTDOWN_MASK) || !how) {
3359 ret = -EINVAL;
3360 goto out_unlock;
3361 }
3362
3363 if (sock->state == SS_CONNECTING) {
3364 if ((1 << sk->sk_state) &
3365 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
3366 sock->state = SS_DISCONNECTING;
3367 else
3368 sock->state = SS_CONNECTED;
3369 }
3370
3371 sk->sk_shutdown |= how;
3372 if ((how & SEND_SHUTDOWN) && mptcp_close_state(sk))
3373 __mptcp_wr_shutdown(sk);
3374
3375 /* Wake up anyone sleeping in poll. */
3376 sk->sk_state_change(sk);
3377
3378 out_unlock:
3379 release_sock(sk);
3380
3381 return ret;
3382 }
3383
3384 static const struct proto_ops mptcp_stream_ops = {
3385 .family = PF_INET,
3386 .owner = THIS_MODULE,
3387 .release = inet_release,
3388 .bind = mptcp_bind,
3389 .connect = mptcp_stream_connect,
3390 .socketpair = sock_no_socketpair,
3391 .accept = mptcp_stream_accept,
3392 .getname = inet_getname,
3393 .poll = mptcp_poll,
3394 .ioctl = inet_ioctl,
3395 .gettstamp = sock_gettstamp,
3396 .listen = mptcp_listen,
3397 .shutdown = mptcp_shutdown,
3398 .setsockopt = sock_common_setsockopt,
3399 .getsockopt = sock_common_getsockopt,
3400 .sendmsg = inet_sendmsg,
3401 .recvmsg = inet_recvmsg,
3402 .mmap = sock_no_mmap,
3403 .sendpage = inet_sendpage,
3404 };
3405
3406 static struct inet_protosw mptcp_protosw = {
3407 .type = SOCK_STREAM,
3408 .protocol = IPPROTO_MPTCP,
3409 .prot = &mptcp_prot,
3410 .ops = &mptcp_stream_ops,
3411 .flags = INET_PROTOSW_ICSK,
3412 };
3413
3414 void __init mptcp_proto_init(void)
3415 {
3416 mptcp_prot.h.hashinfo = tcp_prot.h.hashinfo;
3417
3418 if (percpu_counter_init(&mptcp_sockets_allocated, 0, GFP_KERNEL))
3419 panic("Failed to allocate MPTCP pcpu counter\n");
3420
3421 mptcp_subflow_init();
3422 mptcp_pm_init();
3423 mptcp_token_init();
3424
3425 if (proto_register(&mptcp_prot, 1) != 0)
3426 panic("Failed to register MPTCP proto.\n");
3427
3428 inet_register_protosw(&mptcp_protosw);
3429
3430 BUILD_BUG_ON(sizeof(struct mptcp_skb_cb) > sizeof_field(struct sk_buff, cb));
3431 }
3432
3433 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
3434 static const struct proto_ops mptcp_v6_stream_ops = {
3435 .family = PF_INET6,
3436 .owner = THIS_MODULE,
3437 .release = inet6_release,
3438 .bind = mptcp_bind,
3439 .connect = mptcp_stream_connect,
3440 .socketpair = sock_no_socketpair,
3441 .accept = mptcp_stream_accept,
3442 .getname = inet6_getname,
3443 .poll = mptcp_poll,
3444 .ioctl = inet6_ioctl,
3445 .gettstamp = sock_gettstamp,
3446 .listen = mptcp_listen,
3447 .shutdown = mptcp_shutdown,
3448 .setsockopt = sock_common_setsockopt,
3449 .getsockopt = sock_common_getsockopt,
3450 .sendmsg = inet6_sendmsg,
3451 .recvmsg = inet6_recvmsg,
3452 .mmap = sock_no_mmap,
3453 .sendpage = inet_sendpage,
3454 #ifdef CONFIG_COMPAT
3455 .compat_ioctl = inet6_compat_ioctl,
3456 #endif
3457 };
3458
3459 static struct proto mptcp_v6_prot;
3460
3461 static void mptcp_v6_destroy(struct sock *sk)
3462 {
3463 mptcp_destroy(sk);
3464 inet6_destroy_sock(sk);
3465 }
3466
3467 static struct inet_protosw mptcp_v6_protosw = {
3468 .type = SOCK_STREAM,
3469 .protocol = IPPROTO_MPTCP,
3470 .prot = &mptcp_v6_prot,
3471 .ops = &mptcp_v6_stream_ops,
3472 .flags = INET_PROTOSW_ICSK,
3473 };
3474
3475 int __init mptcp_proto_v6_init(void)
3476 {
3477 int err;
3478
3479 mptcp_v6_prot = mptcp_prot;
3480 strcpy(mptcp_v6_prot.name, "MPTCPv6");
3481 mptcp_v6_prot.slab = NULL;
3482 mptcp_v6_prot.destroy = mptcp_v6_destroy;
3483 mptcp_v6_prot.obj_size = sizeof(struct mptcp6_sock);
3484
3485 err = proto_register(&mptcp_v6_prot, 1);
3486 if (err)
3487 return err;
3488
3489 err = inet6_register_protosw(&mptcp_v6_protosw);
3490 if (err)
3491 proto_unregister(&mptcp_v6_prot);
3492
3493 return err;
3494 }
3495 #endif
Linux with added FPC-III support