?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
io_uring: ensure finish_wait() is always called in __io_uring_task_cancel()
[linux/fpc-iii.git] / net / mptcp / protocol.c
blob09b19aa2f2051a96c3d4a373922b91fb6dd4a0bc
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 (amount <= sk->sk_forward_alloc)
881 goto reserve;
882
883 /* under memory pressure try to reserve at most a single page
884 * otherwise try to reserve the full estimate and fallback
885 * to a single page before entering the error path
886 */
887 if ((tcp_under_memory_pressure(sk) && amount > PAGE_SIZE) ||
888 !sk_wmem_schedule(sk, amount)) {
889 if (amount <= PAGE_SIZE)
890 goto nomem;
891
892 amount = PAGE_SIZE;
893 if (!sk_wmem_schedule(sk, amount))
894 goto nomem;
895 }
896
897 reserve:
898 msk->wmem_reserved = amount;
899 sk->sk_forward_alloc -= amount;
900 return;
901
902 nomem:
903 /* we will wait for memory on next allocation */
904 msk->wmem_reserved = -1;
905 }
906
907 static void __mptcp_update_wmem(struct sock *sk)
908 {
909 struct mptcp_sock *msk = mptcp_sk(sk);
910
911 if (!msk->wmem_reserved)
912 return;
913
914 if (msk->wmem_reserved < 0)
915 msk->wmem_reserved = 0;
916 if (msk->wmem_reserved > 0) {
917 sk->sk_forward_alloc += msk->wmem_reserved;
918 msk->wmem_reserved = 0;
919 }
920 }
921
922 static bool mptcp_wmem_alloc(struct sock *sk, int size)
923 {
924 struct mptcp_sock *msk = mptcp_sk(sk);
925
926 /* check for pre-existing error condition */
927 if (msk->wmem_reserved < 0)
928 return false;
929
930 if (msk->wmem_reserved >= size)
931 goto account;
932
933 mptcp_data_lock(sk);
934 if (!sk_wmem_schedule(sk, size)) {
935 mptcp_data_unlock(sk);
936 return false;
937 }
938
939 sk->sk_forward_alloc -= size;
940 msk->wmem_reserved += size;
941 mptcp_data_unlock(sk);
942
943 account:
944 msk->wmem_reserved -= size;
945 return true;
946 }
947
948 static void mptcp_wmem_uncharge(struct sock *sk, int size)
949 {
950 struct mptcp_sock *msk = mptcp_sk(sk);
951
952 if (msk->wmem_reserved < 0)
953 msk->wmem_reserved = 0;
954 msk->wmem_reserved += size;
955 }
956
957 static void mptcp_mem_reclaim_partial(struct sock *sk)
958 {
959 struct mptcp_sock *msk = mptcp_sk(sk);
960
961 /* if we are experiencing a transint allocation error,
962 * the forward allocation memory has been already
963 * released
964 */
965 if (msk->wmem_reserved < 0)
966 return;
967
968 mptcp_data_lock(sk);
969 sk->sk_forward_alloc += msk->wmem_reserved;
970 sk_mem_reclaim_partial(sk);
971 msk->wmem_reserved = sk->sk_forward_alloc;
972 sk->sk_forward_alloc = 0;
973 mptcp_data_unlock(sk);
974 }
975
976 static void dfrag_uncharge(struct sock *sk, int len)
977 {
978 sk_mem_uncharge(sk, len);
979 sk_wmem_queued_add(sk, -len);
980 }
981
982 static void dfrag_clear(struct sock *sk, struct mptcp_data_frag *dfrag)
983 {
984 int len = dfrag->data_len + dfrag->overhead;
985
986 list_del(&dfrag->list);
987 dfrag_uncharge(sk, len);
988 put_page(dfrag->page);
989 }
990
991 static void __mptcp_clean_una(struct sock *sk)
992 {
993 struct mptcp_sock *msk = mptcp_sk(sk);
994 struct mptcp_data_frag *dtmp, *dfrag;
995 bool cleaned = false;
996 u64 snd_una;
997
998 /* on fallback we just need to ignore snd_una, as this is really
999 * plain TCP
1000 */
1001 if (__mptcp_check_fallback(msk))
1002 msk->snd_una = READ_ONCE(msk->snd_nxt);
1003
1004 snd_una = msk->snd_una;
1005 list_for_each_entry_safe(dfrag, dtmp, &msk->rtx_queue, list) {
1006 if (after64(dfrag->data_seq + dfrag->data_len, snd_una))
1007 break;
1008
1009 if (WARN_ON_ONCE(dfrag == msk->first_pending))
1010 break;
1011 dfrag_clear(sk, dfrag);
1012 cleaned = true;
1013 }
1014
1015 dfrag = mptcp_rtx_head(sk);
1016 if (dfrag && after64(snd_una, dfrag->data_seq)) {
1017 u64 delta = snd_una - dfrag->data_seq;
1018
1019 if (WARN_ON_ONCE(delta > dfrag->already_sent))
1020 goto out;
1021
1022 dfrag->data_seq += delta;
1023 dfrag->offset += delta;
1024 dfrag->data_len -= delta;
1025 dfrag->already_sent -= delta;
1026
1027 dfrag_uncharge(sk, delta);
1028 cleaned = true;
1029 }
1030
1031 out:
1032 if (cleaned) {
1033 if (tcp_under_memory_pressure(sk)) {
1034 __mptcp_update_wmem(sk);
1035 sk_mem_reclaim_partial(sk);
1036 }
1037
1038 if (sk_stream_is_writeable(sk)) {
1039 /* pairs with memory barrier in mptcp_poll */
1040 smp_mb();
1041 if (test_and_clear_bit(MPTCP_NOSPACE, &msk->flags))
1042 sk_stream_write_space(sk);
1043 }
1044 }
1045
1046 if (snd_una == READ_ONCE(msk->snd_nxt)) {
1047 if (msk->timer_ival)
1048 mptcp_stop_timer(sk);
1049 } else {
1050 mptcp_reset_timer(sk);
1051 }
1052 }
1053
1054 static void mptcp_enter_memory_pressure(struct sock *sk)
1055 {
1056 struct mptcp_subflow_context *subflow;
1057 struct mptcp_sock *msk = mptcp_sk(sk);
1058 bool first = true;
1059
1060 sk_stream_moderate_sndbuf(sk);
1061 mptcp_for_each_subflow(msk, subflow) {
1062 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1063
1064 if (first)
1065 tcp_enter_memory_pressure(ssk);
1066 sk_stream_moderate_sndbuf(ssk);
1067 first = false;
1068 }
1069 }
1070
1071 /* ensure we get enough memory for the frag hdr, beyond some minimal amount of
1072 * data
1073 */
1074 static bool mptcp_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1075 {
1076 if (likely(skb_page_frag_refill(32U + sizeof(struct mptcp_data_frag),
1077 pfrag, sk->sk_allocation)))
1078 return true;
1079
1080 mptcp_enter_memory_pressure(sk);
1081 return false;
1082 }
1083
1084 static struct mptcp_data_frag *
1085 mptcp_carve_data_frag(const struct mptcp_sock *msk, struct page_frag *pfrag,
1086 int orig_offset)
1087 {
1088 int offset = ALIGN(orig_offset, sizeof(long));
1089 struct mptcp_data_frag *dfrag;
1090
1091 dfrag = (struct mptcp_data_frag *)(page_to_virt(pfrag->page) + offset);
1092 dfrag->data_len = 0;
1093 dfrag->data_seq = msk->write_seq;
1094 dfrag->overhead = offset - orig_offset + sizeof(struct mptcp_data_frag);
1095 dfrag->offset = offset + sizeof(struct mptcp_data_frag);
1096 dfrag->already_sent = 0;
1097 dfrag->page = pfrag->page;
1098
1099 return dfrag;
1100 }
1101
1102 struct mptcp_sendmsg_info {
1103 int mss_now;
1104 int size_goal;
1105 u16 limit;
1106 u16 sent;
1107 unsigned int flags;
1108 };
1109
1110 static int mptcp_check_allowed_size(struct mptcp_sock *msk, u64 data_seq,
1111 int avail_size)
1112 {
1113 u64 window_end = mptcp_wnd_end(msk);
1114
1115 if (__mptcp_check_fallback(msk))
1116 return avail_size;
1117
1118 if (!before64(data_seq + avail_size, window_end)) {
1119 u64 allowed_size = window_end - data_seq;
1120
1121 return min_t(unsigned int, allowed_size, avail_size);
1122 }
1123
1124 return avail_size;
1125 }
1126
1127 static bool __mptcp_add_ext(struct sk_buff *skb, gfp_t gfp)
1128 {
1129 struct skb_ext *mpext = __skb_ext_alloc(gfp);
1130
1131 if (!mpext)
1132 return false;
1133 __skb_ext_set(skb, SKB_EXT_MPTCP, mpext);
1134 return true;
1135 }
1136
1137 static struct sk_buff *__mptcp_do_alloc_tx_skb(struct sock *sk, gfp_t gfp)
1138 {
1139 struct sk_buff *skb;
1140
1141 skb = alloc_skb_fclone(MAX_TCP_HEADER, gfp);
1142 if (likely(skb)) {
1143 if (likely(__mptcp_add_ext(skb, gfp))) {
1144 skb_reserve(skb, MAX_TCP_HEADER);
1145 skb->reserved_tailroom = skb->end - skb->tail;
1146 return skb;
1147 }
1148 __kfree_skb(skb);
1149 } else {
1150 mptcp_enter_memory_pressure(sk);
1151 }
1152 return NULL;
1153 }
1154
1155 static bool mptcp_tx_cache_refill(struct sock *sk, int size,
1156 struct sk_buff_head *skbs, int *total_ts)
1157 {
1158 struct mptcp_sock *msk = mptcp_sk(sk);
1159 struct sk_buff *skb;
1160 int space_needed;
1161
1162 if (unlikely(tcp_under_memory_pressure(sk))) {
1163 mptcp_mem_reclaim_partial(sk);
1164
1165 /* under pressure pre-allocate at most a single skb */
1166 if (msk->skb_tx_cache.qlen)
1167 return true;
1168 space_needed = msk->size_goal_cache;
1169 } else {
1170 space_needed = msk->tx_pending_data + size -
1171 msk->skb_tx_cache.qlen * msk->size_goal_cache;
1172 }
1173
1174 while (space_needed > 0) {
1175 skb = __mptcp_do_alloc_tx_skb(sk, sk->sk_allocation);
1176 if (unlikely(!skb)) {
1177 /* under memory pressure, try to pass the caller a
1178 * single skb to allow forward progress
1179 */
1180 while (skbs->qlen > 1) {
1181 skb = __skb_dequeue_tail(skbs);
1182 __kfree_skb(skb);
1183 }
1184 return skbs->qlen > 0;
1185 }
1186
1187 *total_ts += skb->truesize;
1188 __skb_queue_tail(skbs, skb);
1189 space_needed -= msk->size_goal_cache;
1190 }
1191 return true;
1192 }
1193
1194 static bool __mptcp_alloc_tx_skb(struct sock *sk, struct sock *ssk, gfp_t gfp)
1195 {
1196 struct mptcp_sock *msk = mptcp_sk(sk);
1197 struct sk_buff *skb;
1198
1199 if (ssk->sk_tx_skb_cache) {
1200 skb = ssk->sk_tx_skb_cache;
1201 if (unlikely(!skb_ext_find(skb, SKB_EXT_MPTCP) &&
1202 !__mptcp_add_ext(skb, gfp)))
1203 return false;
1204 return true;
1205 }
1206
1207 skb = skb_peek(&msk->skb_tx_cache);
1208 if (skb) {
1209 if (likely(sk_wmem_schedule(ssk, skb->truesize))) {
1210 skb = __skb_dequeue(&msk->skb_tx_cache);
1211 if (WARN_ON_ONCE(!skb))
1212 return false;
1213
1214 mptcp_wmem_uncharge(sk, skb->truesize);
1215 ssk->sk_tx_skb_cache = skb;
1216 return true;
1217 }
1218
1219 /* over memory limit, no point to try to allocate a new skb */
1220 return false;
1221 }
1222
1223 skb = __mptcp_do_alloc_tx_skb(sk, gfp);
1224 if (!skb)
1225 return false;
1226
1227 if (likely(sk_wmem_schedule(ssk, skb->truesize))) {
1228 ssk->sk_tx_skb_cache = skb;
1229 return true;
1230 }
1231 kfree_skb(skb);
1232 return false;
1233 }
1234
1235 static bool mptcp_must_reclaim_memory(struct sock *sk, struct sock *ssk)
1236 {
1237 return !ssk->sk_tx_skb_cache &&
1238 !skb_peek(&mptcp_sk(sk)->skb_tx_cache) &&
1239 tcp_under_memory_pressure(sk);
1240 }
1241
1242 static bool mptcp_alloc_tx_skb(struct sock *sk, struct sock *ssk)
1243 {
1244 if (unlikely(mptcp_must_reclaim_memory(sk, ssk)))
1245 mptcp_mem_reclaim_partial(sk);
1246 return __mptcp_alloc_tx_skb(sk, ssk, sk->sk_allocation);
1247 }
1248
1249 static int mptcp_sendmsg_frag(struct sock *sk, struct sock *ssk,
1250 struct mptcp_data_frag *dfrag,
1251 struct mptcp_sendmsg_info *info)
1252 {
1253 u64 data_seq = dfrag->data_seq + info->sent;
1254 struct mptcp_sock *msk = mptcp_sk(sk);
1255 bool zero_window_probe = false;
1256 struct mptcp_ext *mpext = NULL;
1257 struct sk_buff *skb, *tail;
1258 bool can_collapse = false;
1259 int size_bias = 0;
1260 int avail_size;
1261 size_t ret = 0;
1262
1263 pr_debug("msk=%p ssk=%p sending dfrag at seq=%lld len=%d already sent=%d",
1264 msk, ssk, dfrag->data_seq, dfrag->data_len, info->sent);
1265
1266 /* compute send limit */
1267 info->mss_now = tcp_send_mss(ssk, &info->size_goal, info->flags);
1268 avail_size = info->size_goal;
1269 msk->size_goal_cache = info->size_goal;
1270 skb = tcp_write_queue_tail(ssk);
1271 if (skb) {
1272 /* Limit the write to the size available in the
1273 * current skb, if any, so that we create at most a new skb.
1274 * Explicitly tells TCP internals to avoid collapsing on later
1275 * queue management operation, to avoid breaking the ext <->
1276 * SSN association set here
1277 */
1278 mpext = skb_ext_find(skb, SKB_EXT_MPTCP);
1279 can_collapse = (info->size_goal - skb->len > 0) &&
1280 mptcp_skb_can_collapse_to(data_seq, skb, mpext);
1281 if (!can_collapse) {
1282 TCP_SKB_CB(skb)->eor = 1;
1283 } else {
1284 size_bias = skb->len;
1285 avail_size = info->size_goal - skb->len;
1286 }
1287 }
1288
1289 /* Zero window and all data acked? Probe. */
1290 avail_size = mptcp_check_allowed_size(msk, data_seq, avail_size);
1291 if (avail_size == 0) {
1292 u64 snd_una = READ_ONCE(msk->snd_una);
1293
1294 if (skb || snd_una != msk->snd_nxt)
1295 return 0;
1296 zero_window_probe = true;
1297 data_seq = snd_una - 1;
1298 avail_size = 1;
1299 }
1300
1301 if (WARN_ON_ONCE(info->sent > info->limit ||
1302 info->limit > dfrag->data_len))
1303 return 0;
1304
1305 ret = info->limit - info->sent;
1306 tail = tcp_build_frag(ssk, avail_size + size_bias, info->flags,
1307 dfrag->page, dfrag->offset + info->sent, &ret);
1308 if (!tail) {
1309 tcp_remove_empty_skb(sk, tcp_write_queue_tail(ssk));
1310 return -ENOMEM;
1311 }
1312
1313 /* if the tail skb is still the cached one, collapsing really happened.
1314 */
1315 if (skb == tail) {
1316 TCP_SKB_CB(tail)->tcp_flags &= ~TCPHDR_PSH;
1317 mpext->data_len += ret;
1318 WARN_ON_ONCE(!can_collapse);
1319 WARN_ON_ONCE(zero_window_probe);
1320 goto out;
1321 }
1322
1323 mpext = skb_ext_find(tail, SKB_EXT_MPTCP);
1324 if (WARN_ON_ONCE(!mpext)) {
1325 /* should never reach here, stream corrupted */
1326 return -EINVAL;
1327 }
1328
1329 memset(mpext, 0, sizeof(*mpext));
1330 mpext->data_seq = data_seq;
1331 mpext->subflow_seq = mptcp_subflow_ctx(ssk)->rel_write_seq;
1332 mpext->data_len = ret;
1333 mpext->use_map = 1;
1334 mpext->dsn64 = 1;
1335
1336 pr_debug("data_seq=%llu subflow_seq=%u data_len=%u dsn64=%d",
1337 mpext->data_seq, mpext->subflow_seq, mpext->data_len,
1338 mpext->dsn64);
1339
1340 if (zero_window_probe) {
1341 mptcp_subflow_ctx(ssk)->rel_write_seq += ret;
1342 mpext->frozen = 1;
1343 ret = 0;
1344 tcp_push_pending_frames(ssk);
1345 }
1346 out:
1347 mptcp_subflow_ctx(ssk)->rel_write_seq += ret;
1348 return ret;
1349 }
1350
1351 #define MPTCP_SEND_BURST_SIZE ((1 << 16) - \
1352 sizeof(struct tcphdr) - \
1353 MAX_TCP_OPTION_SPACE - \
1354 sizeof(struct ipv6hdr) - \
1355 sizeof(struct frag_hdr))
1356
1357 struct subflow_send_info {
1358 struct sock *ssk;
1359 u64 ratio;
1360 };
1361
1362 static struct sock *mptcp_subflow_get_send(struct mptcp_sock *msk,
1363 u32 *sndbuf)
1364 {
1365 struct subflow_send_info send_info[2];
1366 struct mptcp_subflow_context *subflow;
1367 int i, nr_active = 0;
1368 struct sock *ssk;
1369 u64 ratio;
1370 u32 pace;
1371
1372 sock_owned_by_me((struct sock *)msk);
1373
1374 *sndbuf = 0;
1375 if (__mptcp_check_fallback(msk)) {
1376 if (!msk->first)
1377 return NULL;
1378 *sndbuf = msk->first->sk_sndbuf;
1379 return sk_stream_memory_free(msk->first) ? msk->first : NULL;
1380 }
1381
1382 /* re-use last subflow, if the burst allow that */
1383 if (msk->last_snd && msk->snd_burst > 0 &&
1384 sk_stream_memory_free(msk->last_snd) &&
1385 mptcp_subflow_active(mptcp_subflow_ctx(msk->last_snd))) {
1386 mptcp_for_each_subflow(msk, subflow) {
1387 ssk = mptcp_subflow_tcp_sock(subflow);
1388 *sndbuf = max(tcp_sk(ssk)->snd_wnd, *sndbuf);
1389 }
1390 return msk->last_snd;
1391 }
1392
1393 /* pick the subflow with the lower wmem/wspace ratio */
1394 for (i = 0; i < 2; ++i) {
1395 send_info[i].ssk = NULL;
1396 send_info[i].ratio = -1;
1397 }
1398 mptcp_for_each_subflow(msk, subflow) {
1399 ssk = mptcp_subflow_tcp_sock(subflow);
1400 if (!mptcp_subflow_active(subflow))
1401 continue;
1402
1403 nr_active += !subflow->backup;
1404 *sndbuf = max(tcp_sk(ssk)->snd_wnd, *sndbuf);
1405 if (!sk_stream_memory_free(subflow->tcp_sock))
1406 continue;
1407
1408 pace = READ_ONCE(ssk->sk_pacing_rate);
1409 if (!pace)
1410 continue;
1411
1412 ratio = div_u64((u64)READ_ONCE(ssk->sk_wmem_queued) << 32,
1413 pace);
1414 if (ratio < send_info[subflow->backup].ratio) {
1415 send_info[subflow->backup].ssk = ssk;
1416 send_info[subflow->backup].ratio = ratio;
1417 }
1418 }
1419
1420 pr_debug("msk=%p nr_active=%d ssk=%p:%lld backup=%p:%lld",
1421 msk, nr_active, send_info[0].ssk, send_info[0].ratio,
1422 send_info[1].ssk, send_info[1].ratio);
1423
1424 /* pick the best backup if no other subflow is active */
1425 if (!nr_active)
1426 send_info[0].ssk = send_info[1].ssk;
1427
1428 if (send_info[0].ssk) {
1429 msk->last_snd = send_info[0].ssk;
1430 msk->snd_burst = min_t(int, MPTCP_SEND_BURST_SIZE,
1431 sk_stream_wspace(msk->last_snd));
1432 return msk->last_snd;
1433 }
1434 return NULL;
1435 }
1436
1437 static void mptcp_push_release(struct sock *sk, struct sock *ssk,
1438 struct mptcp_sendmsg_info *info)
1439 {
1440 mptcp_set_timeout(sk, ssk);
1441 tcp_push(ssk, 0, info->mss_now, tcp_sk(ssk)->nonagle, info->size_goal);
1442 release_sock(ssk);
1443 }
1444
1445 static void mptcp_push_pending(struct sock *sk, unsigned int flags)
1446 {
1447 struct sock *prev_ssk = NULL, *ssk = NULL;
1448 struct mptcp_sock *msk = mptcp_sk(sk);
1449 struct mptcp_sendmsg_info info = {
1450 .flags = flags,
1451 };
1452 struct mptcp_data_frag *dfrag;
1453 int len, copied = 0;
1454 u32 sndbuf;
1455
1456 while ((dfrag = mptcp_send_head(sk))) {
1457 info.sent = dfrag->already_sent;
1458 info.limit = dfrag->data_len;
1459 len = dfrag->data_len - dfrag->already_sent;
1460 while (len > 0) {
1461 int ret = 0;
1462
1463 prev_ssk = ssk;
1464 __mptcp_flush_join_list(msk);
1465 ssk = mptcp_subflow_get_send(msk, &sndbuf);
1466
1467 /* do auto tuning */
1468 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK) &&
1469 sndbuf > READ_ONCE(sk->sk_sndbuf))
1470 WRITE_ONCE(sk->sk_sndbuf, sndbuf);
1471
1472 /* try to keep the subflow socket lock across
1473 * consecutive xmit on the same socket
1474 */
1475 if (ssk != prev_ssk && prev_ssk)
1476 mptcp_push_release(sk, prev_ssk, &info);
1477 if (!ssk)
1478 goto out;
1479
1480 if (ssk != prev_ssk || !prev_ssk)
1481 lock_sock(ssk);
1482
1483 /* keep it simple and always provide a new skb for the
1484 * subflow, even if we will not use it when collapsing
1485 * on the pending one
1486 */
1487 if (!mptcp_alloc_tx_skb(sk, ssk)) {
1488 mptcp_push_release(sk, ssk, &info);
1489 goto out;
1490 }
1491
1492 ret = mptcp_sendmsg_frag(sk, ssk, dfrag, &info);
1493 if (ret <= 0) {
1494 mptcp_push_release(sk, ssk, &info);
1495 goto out;
1496 }
1497
1498 info.sent += ret;
1499 dfrag->already_sent += ret;
1500 msk->snd_nxt += ret;
1501 msk->snd_burst -= ret;
1502 msk->tx_pending_data -= ret;
1503 copied += ret;
1504 len -= ret;
1505 }
1506 WRITE_ONCE(msk->first_pending, mptcp_send_next(sk));
1507 }
1508
1509 /* at this point we held the socket lock for the last subflow we used */
1510 if (ssk)
1511 mptcp_push_release(sk, ssk, &info);
1512
1513 out:
1514 if (copied) {
1515 /* start the timer, if it's not pending */
1516 if (!mptcp_timer_pending(sk))
1517 mptcp_reset_timer(sk);
1518 __mptcp_check_send_data_fin(sk);
1519 }
1520 }
1521
1522 static void __mptcp_subflow_push_pending(struct sock *sk, struct sock *ssk)
1523 {
1524 struct mptcp_sock *msk = mptcp_sk(sk);
1525 struct mptcp_sendmsg_info info;
1526 struct mptcp_data_frag *dfrag;
1527 int len, copied = 0;
1528
1529 info.flags = 0;
1530 while ((dfrag = mptcp_send_head(sk))) {
1531 info.sent = dfrag->already_sent;
1532 info.limit = dfrag->data_len;
1533 len = dfrag->data_len - dfrag->already_sent;
1534 while (len > 0) {
1535 int ret = 0;
1536
1537 /* do auto tuning */
1538 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK) &&
1539 ssk->sk_sndbuf > READ_ONCE(sk->sk_sndbuf))
1540 WRITE_ONCE(sk->sk_sndbuf, ssk->sk_sndbuf);
1541
1542 if (unlikely(mptcp_must_reclaim_memory(sk, ssk))) {
1543 __mptcp_update_wmem(sk);
1544 sk_mem_reclaim_partial(sk);
1545 }
1546 if (!__mptcp_alloc_tx_skb(sk, ssk, GFP_ATOMIC))
1547 goto out;
1548
1549 ret = mptcp_sendmsg_frag(sk, ssk, dfrag, &info);
1550 if (ret <= 0)
1551 goto out;
1552
1553 info.sent += ret;
1554 dfrag->already_sent += ret;
1555 msk->snd_nxt += ret;
1556 msk->snd_burst -= ret;
1557 msk->tx_pending_data -= ret;
1558 copied += ret;
1559 len -= ret;
1560 }
1561 WRITE_ONCE(msk->first_pending, mptcp_send_next(sk));
1562 }
1563
1564 out:
1565 /* __mptcp_alloc_tx_skb could have released some wmem and we are
1566 * not going to flush it via release_sock()
1567 */
1568 __mptcp_update_wmem(sk);
1569 if (copied) {
1570 mptcp_set_timeout(sk, ssk);
1571 tcp_push(ssk, 0, info.mss_now, tcp_sk(ssk)->nonagle,
1572 info.size_goal);
1573 if (msk->snd_data_fin_enable &&
1574 msk->snd_nxt + 1 == msk->write_seq)
1575 mptcp_schedule_work(sk);
1576 }
1577 }
1578
1579 static int mptcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1580 {
1581 struct mptcp_sock *msk = mptcp_sk(sk);
1582 struct page_frag *pfrag;
1583 size_t copied = 0;
1584 int ret = 0;
1585 long timeo;
1586
1587 if (msg->msg_flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL))
1588 return -EOPNOTSUPP;
1589
1590 mptcp_lock_sock(sk, __mptcp_wmem_reserve(sk, len));
1591
1592 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1593
1594 if ((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) {
1595 ret = sk_stream_wait_connect(sk, &timeo);
1596 if (ret)
1597 goto out;
1598 }
1599
1600 pfrag = sk_page_frag(sk);
1601
1602 while (msg_data_left(msg)) {
1603 int total_ts, frag_truesize = 0;
1604 struct mptcp_data_frag *dfrag;
1605 struct sk_buff_head skbs;
1606 bool dfrag_collapsed;
1607 size_t psize, offset;
1608
1609 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN)) {
1610 ret = -EPIPE;
1611 goto out;
1612 }
1613
1614 /* reuse tail pfrag, if possible, or carve a new one from the
1615 * page allocator
1616 */
1617 dfrag = mptcp_pending_tail(sk);
1618 dfrag_collapsed = mptcp_frag_can_collapse_to(msk, pfrag, dfrag);
1619 if (!dfrag_collapsed) {
1620 if (!sk_stream_memory_free(sk))
1621 goto wait_for_memory;
1622
1623 if (!mptcp_page_frag_refill(sk, pfrag))
1624 goto wait_for_memory;
1625
1626 dfrag = mptcp_carve_data_frag(msk, pfrag, pfrag->offset);
1627 frag_truesize = dfrag->overhead;
1628 }
1629
1630 /* we do not bound vs wspace, to allow a single packet.
1631 * memory accounting will prevent execessive memory usage
1632 * anyway
1633 */
1634 offset = dfrag->offset + dfrag->data_len;
1635 psize = pfrag->size - offset;
1636 psize = min_t(size_t, psize, msg_data_left(msg));
1637 total_ts = psize + frag_truesize;
1638 __skb_queue_head_init(&skbs);
1639 if (!mptcp_tx_cache_refill(sk, psize, &skbs, &total_ts))
1640 goto wait_for_memory;
1641
1642 if (!mptcp_wmem_alloc(sk, total_ts)) {
1643 __skb_queue_purge(&skbs);
1644 goto wait_for_memory;
1645 }
1646
1647 skb_queue_splice_tail(&skbs, &msk->skb_tx_cache);
1648 if (copy_page_from_iter(dfrag->page, offset, psize,
1649 &msg->msg_iter) != psize) {
1650 mptcp_wmem_uncharge(sk, psize + frag_truesize);
1651 ret = -EFAULT;
1652 goto out;
1653 }
1654
1655 /* data successfully copied into the write queue */
1656 copied += psize;
1657 dfrag->data_len += psize;
1658 frag_truesize += psize;
1659 pfrag->offset += frag_truesize;
1660 WRITE_ONCE(msk->write_seq, msk->write_seq + psize);
1661 msk->tx_pending_data += psize;
1662
1663 /* charge data on mptcp pending queue to the msk socket
1664 * Note: we charge such data both to sk and ssk
1665 */
1666 sk_wmem_queued_add(sk, frag_truesize);
1667 if (!dfrag_collapsed) {
1668 get_page(dfrag->page);
1669 list_add_tail(&dfrag->list, &msk->rtx_queue);
1670 if (!msk->first_pending)
1671 WRITE_ONCE(msk->first_pending, dfrag);
1672 }
1673 pr_debug("msk=%p dfrag at seq=%lld len=%d sent=%d new=%d", msk,
1674 dfrag->data_seq, dfrag->data_len, dfrag->already_sent,
1675 !dfrag_collapsed);
1676
1677 continue;
1678
1679 wait_for_memory:
1680 set_bit(MPTCP_NOSPACE, &msk->flags);
1681 mptcp_push_pending(sk, msg->msg_flags);
1682 ret = sk_stream_wait_memory(sk, &timeo);
1683 if (ret)
1684 goto out;
1685 }
1686
1687 if (copied)
1688 mptcp_push_pending(sk, msg->msg_flags);
1689
1690 out:
1691 release_sock(sk);
1692 return copied ? : ret;
1693 }
1694
1695 static void mptcp_wait_data(struct sock *sk, long *timeo)
1696 {
1697 DEFINE_WAIT_FUNC(wait, woken_wake_function);
1698 struct mptcp_sock *msk = mptcp_sk(sk);
1699
1700 add_wait_queue(sk_sleep(sk), &wait);
1701 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1702
1703 sk_wait_event(sk, timeo,
1704 test_and_clear_bit(MPTCP_DATA_READY, &msk->flags), &wait);
1705
1706 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1707 remove_wait_queue(sk_sleep(sk), &wait);
1708 }
1709
1710 static int __mptcp_recvmsg_mskq(struct mptcp_sock *msk,
1711 struct msghdr *msg,
1712 size_t len)
1713 {
1714 struct sk_buff *skb;
1715 int copied = 0;
1716
1717 while ((skb = skb_peek(&msk->receive_queue)) != NULL) {
1718 u32 offset = MPTCP_SKB_CB(skb)->offset;
1719 u32 data_len = skb->len - offset;
1720 u32 count = min_t(size_t, len - copied, data_len);
1721 int err;
1722
1723 err = skb_copy_datagram_msg(skb, offset, msg, count);
1724 if (unlikely(err < 0)) {
1725 if (!copied)
1726 return err;
1727 break;
1728 }
1729
1730 copied += count;
1731
1732 if (count < data_len) {
1733 MPTCP_SKB_CB(skb)->offset += count;
1734 break;
1735 }
1736
1737 /* we will bulk release the skb memory later */
1738 skb->destructor = NULL;
1739 msk->rmem_released += skb->truesize;
1740 __skb_unlink(skb, &msk->receive_queue);
1741 __kfree_skb(skb);
1742
1743 if (copied >= len)
1744 break;
1745 }
1746
1747 return copied;
1748 }
1749
1750 /* receive buffer autotuning. See tcp_rcv_space_adjust for more information.
1751 *
1752 * Only difference: Use highest rtt estimate of the subflows in use.
1753 */
1754 static void mptcp_rcv_space_adjust(struct mptcp_sock *msk, int copied)
1755 {
1756 struct mptcp_subflow_context *subflow;
1757 struct sock *sk = (struct sock *)msk;
1758 u32 time, advmss = 1;
1759 u64 rtt_us, mstamp;
1760
1761 sock_owned_by_me(sk);
1762
1763 if (copied <= 0)
1764 return;
1765
1766 msk->rcvq_space.copied += copied;
1767
1768 mstamp = div_u64(tcp_clock_ns(), NSEC_PER_USEC);
1769 time = tcp_stamp_us_delta(mstamp, msk->rcvq_space.time);
1770
1771 rtt_us = msk->rcvq_space.rtt_us;
1772 if (rtt_us && time < (rtt_us >> 3))
1773 return;
1774
1775 rtt_us = 0;
1776 mptcp_for_each_subflow(msk, subflow) {
1777 const struct tcp_sock *tp;
1778 u64 sf_rtt_us;
1779 u32 sf_advmss;
1780
1781 tp = tcp_sk(mptcp_subflow_tcp_sock(subflow));
1782
1783 sf_rtt_us = READ_ONCE(tp->rcv_rtt_est.rtt_us);
1784 sf_advmss = READ_ONCE(tp->advmss);
1785
1786 rtt_us = max(sf_rtt_us, rtt_us);
1787 advmss = max(sf_advmss, advmss);
1788 }
1789
1790 msk->rcvq_space.rtt_us = rtt_us;
1791 if (time < (rtt_us >> 3) || rtt_us == 0)
1792 return;
1793
1794 if (msk->rcvq_space.copied <= msk->rcvq_space.space)
1795 goto new_measure;
1796
1797 if (sock_net(sk)->ipv4.sysctl_tcp_moderate_rcvbuf &&
1798 !(sk->sk_userlocks & SOCK_RCVBUF_LOCK)) {
1799 int rcvmem, rcvbuf;
1800 u64 rcvwin, grow;
1801
1802 rcvwin = ((u64)msk->rcvq_space.copied << 1) + 16 * advmss;
1803
1804 grow = rcvwin * (msk->rcvq_space.copied - msk->rcvq_space.space);
1805
1806 do_div(grow, msk->rcvq_space.space);
1807 rcvwin += (grow << 1);
1808
1809 rcvmem = SKB_TRUESIZE(advmss + MAX_TCP_HEADER);
1810 while (tcp_win_from_space(sk, rcvmem) < advmss)
1811 rcvmem += 128;
1812
1813 do_div(rcvwin, advmss);
1814 rcvbuf = min_t(u64, rcvwin * rcvmem,
1815 sock_net(sk)->ipv4.sysctl_tcp_rmem[2]);
1816
1817 if (rcvbuf > sk->sk_rcvbuf) {
1818 u32 window_clamp;
1819
1820 window_clamp = tcp_win_from_space(sk, rcvbuf);
1821 WRITE_ONCE(sk->sk_rcvbuf, rcvbuf);
1822
1823 /* Make subflows follow along. If we do not do this, we
1824 * get drops at subflow level if skbs can't be moved to
1825 * the mptcp rx queue fast enough (announced rcv_win can
1826 * exceed ssk->sk_rcvbuf).
1827 */
1828 mptcp_for_each_subflow(msk, subflow) {
1829 struct sock *ssk;
1830 bool slow;
1831
1832 ssk = mptcp_subflow_tcp_sock(subflow);
1833 slow = lock_sock_fast(ssk);
1834 WRITE_ONCE(ssk->sk_rcvbuf, rcvbuf);
1835 tcp_sk(ssk)->window_clamp = window_clamp;
1836 tcp_cleanup_rbuf(ssk, 1);
1837 unlock_sock_fast(ssk, slow);
1838 }
1839 }
1840 }
1841
1842 msk->rcvq_space.space = msk->rcvq_space.copied;
1843 new_measure:
1844 msk->rcvq_space.copied = 0;
1845 msk->rcvq_space.time = mstamp;
1846 }
1847
1848 static void __mptcp_update_rmem(struct sock *sk)
1849 {
1850 struct mptcp_sock *msk = mptcp_sk(sk);
1851
1852 if (!msk->rmem_released)
1853 return;
1854
1855 atomic_sub(msk->rmem_released, &sk->sk_rmem_alloc);
1856 sk_mem_uncharge(sk, msk->rmem_released);
1857 msk->rmem_released = 0;
1858 }
1859
1860 static void __mptcp_splice_receive_queue(struct sock *sk)
1861 {
1862 struct mptcp_sock *msk = mptcp_sk(sk);
1863
1864 skb_queue_splice_tail_init(&sk->sk_receive_queue, &msk->receive_queue);
1865 }
1866
1867 static bool __mptcp_move_skbs(struct mptcp_sock *msk, unsigned int rcv)
1868 {
1869 struct sock *sk = (struct sock *)msk;
1870 unsigned int moved = 0;
1871 bool ret, done;
1872
1873 __mptcp_flush_join_list(msk);
1874 do {
1875 struct sock *ssk = mptcp_subflow_recv_lookup(msk);
1876 bool slowpath;
1877
1878 /* we can have data pending in the subflows only if the msk
1879 * receive buffer was full at subflow_data_ready() time,
1880 * that is an unlikely slow path.
1881 */
1882 if (likely(!ssk))
1883 break;
1884
1885 slowpath = lock_sock_fast(ssk);
1886 mptcp_data_lock(sk);
1887 done = __mptcp_move_skbs_from_subflow(msk, ssk, &moved);
1888 mptcp_data_unlock(sk);
1889 if (moved && rcv) {
1890 WRITE_ONCE(msk->rmem_pending, min(rcv, moved));
1891 tcp_cleanup_rbuf(ssk, 1);
1892 WRITE_ONCE(msk->rmem_pending, 0);
1893 }
1894 unlock_sock_fast(ssk, slowpath);
1895 } while (!done);
1896
1897 /* acquire the data lock only if some input data is pending */
1898 ret = moved > 0;
1899 if (!RB_EMPTY_ROOT(&msk->out_of_order_queue) ||
1900 !skb_queue_empty_lockless(&sk->sk_receive_queue)) {
1901 mptcp_data_lock(sk);
1902 __mptcp_update_rmem(sk);
1903 ret |= __mptcp_ofo_queue(msk);
1904 __mptcp_splice_receive_queue(sk);
1905 mptcp_data_unlock(sk);
1906 }
1907 if (ret)
1908 mptcp_check_data_fin((struct sock *)msk);
1909 return !skb_queue_empty(&msk->receive_queue);
1910 }
1911
1912 static int mptcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
1913 int nonblock, int flags, int *addr_len)
1914 {
1915 struct mptcp_sock *msk = mptcp_sk(sk);
1916 int copied = 0;
1917 int target;
1918 long timeo;
1919
1920 if (msg->msg_flags & ~(MSG_WAITALL | MSG_DONTWAIT))
1921 return -EOPNOTSUPP;
1922
1923 mptcp_lock_sock(sk, __mptcp_splice_receive_queue(sk));
1924 if (unlikely(sk->sk_state == TCP_LISTEN)) {
1925 copied = -ENOTCONN;
1926 goto out_err;
1927 }
1928
1929 timeo = sock_rcvtimeo(sk, nonblock);
1930
1931 len = min_t(size_t, len, INT_MAX);
1932 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1933
1934 while (copied < len) {
1935 int bytes_read, old_space;
1936
1937 bytes_read = __mptcp_recvmsg_mskq(msk, msg, len - copied);
1938 if (unlikely(bytes_read < 0)) {
1939 if (!copied)
1940 copied = bytes_read;
1941 goto out_err;
1942 }
1943
1944 copied += bytes_read;
1945
1946 if (skb_queue_empty(&msk->receive_queue) &&
1947 __mptcp_move_skbs(msk, len - copied))
1948 continue;
1949
1950 /* be sure to advertise window change */
1951 old_space = READ_ONCE(msk->old_wspace);
1952 if ((tcp_space(sk) - old_space) >= old_space)
1953 mptcp_cleanup_rbuf(msk);
1954
1955 /* only the master socket status is relevant here. The exit
1956 * conditions mirror closely tcp_recvmsg()
1957 */
1958 if (copied >= target)
1959 break;
1960
1961 if (copied) {
1962 if (sk->sk_err ||
1963 sk->sk_state == TCP_CLOSE ||
1964 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1965 !timeo ||
1966 signal_pending(current))
1967 break;
1968 } else {
1969 if (sk->sk_err) {
1970 copied = sock_error(sk);
1971 break;
1972 }
1973
1974 if (test_and_clear_bit(MPTCP_WORK_EOF, &msk->flags))
1975 mptcp_check_for_eof(msk);
1976
1977 if (sk->sk_shutdown & RCV_SHUTDOWN) {
1978 /* race breaker: the shutdown could be after the
1979 * previous receive queue check
1980 */
1981 if (__mptcp_move_skbs(msk, len - copied))
1982 continue;
1983 break;
1984 }
1985
1986 if (sk->sk_state == TCP_CLOSE) {
1987 copied = -ENOTCONN;
1988 break;
1989 }
1990
1991 if (!timeo) {
1992 copied = -EAGAIN;
1993 break;
1994 }
1995
1996 if (signal_pending(current)) {
1997 copied = sock_intr_errno(timeo);
1998 break;
1999 }
2000 }
2001
2002 pr_debug("block timeout %ld", timeo);
2003 mptcp_wait_data(sk, &timeo);
2004 }
2005
2006 if (skb_queue_empty_lockless(&sk->sk_receive_queue) &&
2007 skb_queue_empty(&msk->receive_queue)) {
2008 /* entire backlog drained, clear DATA_READY. */
2009 clear_bit(MPTCP_DATA_READY, &msk->flags);
2010
2011 /* .. race-breaker: ssk might have gotten new data
2012 * after last __mptcp_move_skbs() returned false.
2013 */
2014 if (unlikely(__mptcp_move_skbs(msk, 0)))
2015 set_bit(MPTCP_DATA_READY, &msk->flags);
2016 } else if (unlikely(!test_bit(MPTCP_DATA_READY, &msk->flags))) {
2017 /* data to read but mptcp_wait_data() cleared DATA_READY */
2018 set_bit(MPTCP_DATA_READY, &msk->flags);
2019 }
2020 out_err:
2021 pr_debug("msk=%p data_ready=%d rx queue empty=%d copied=%d",
2022 msk, test_bit(MPTCP_DATA_READY, &msk->flags),
2023 skb_queue_empty_lockless(&sk->sk_receive_queue), copied);
2024 mptcp_rcv_space_adjust(msk, copied);
2025
2026 release_sock(sk);
2027 return copied;
2028 }
2029
2030 static void mptcp_retransmit_handler(struct sock *sk)
2031 {
2032 struct mptcp_sock *msk = mptcp_sk(sk);
2033
2034 set_bit(MPTCP_WORK_RTX, &msk->flags);
2035 mptcp_schedule_work(sk);
2036 }
2037
2038 static void mptcp_retransmit_timer(struct timer_list *t)
2039 {
2040 struct inet_connection_sock *icsk = from_timer(icsk, t,
2041 icsk_retransmit_timer);
2042 struct sock *sk = &icsk->icsk_inet.sk;
2043
2044 bh_lock_sock(sk);
2045 if (!sock_owned_by_user(sk)) {
2046 mptcp_retransmit_handler(sk);
2047 } else {
2048 /* delegate our work to tcp_release_cb() */
2049 if (!test_and_set_bit(TCP_WRITE_TIMER_DEFERRED,
2050 &sk->sk_tsq_flags))
2051 sock_hold(sk);
2052 }
2053 bh_unlock_sock(sk);
2054 sock_put(sk);
2055 }
2056
2057 static void mptcp_timeout_timer(struct timer_list *t)
2058 {
2059 struct sock *sk = from_timer(sk, t, sk_timer);
2060
2061 mptcp_schedule_work(sk);
2062 sock_put(sk);
2063 }
2064
2065 /* Find an idle subflow. Return NULL if there is unacked data at tcp
2066 * level.
2067 *
2068 * A backup subflow is returned only if that is the only kind available.
2069 */
2070 static struct sock *mptcp_subflow_get_retrans(const struct mptcp_sock *msk)
2071 {
2072 struct mptcp_subflow_context *subflow;
2073 struct sock *backup = NULL;
2074
2075 sock_owned_by_me((const struct sock *)msk);
2076
2077 if (__mptcp_check_fallback(msk))
2078 return NULL;
2079
2080 mptcp_for_each_subflow(msk, subflow) {
2081 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2082
2083 if (!mptcp_subflow_active(subflow))
2084 continue;
2085
2086 /* still data outstanding at TCP level? Don't retransmit. */
2087 if (!tcp_write_queue_empty(ssk)) {
2088 if (inet_csk(ssk)->icsk_ca_state >= TCP_CA_Loss)
2089 continue;
2090 return NULL;
2091 }
2092
2093 if (subflow->backup) {
2094 if (!backup)
2095 backup = ssk;
2096 continue;
2097 }
2098
2099 return ssk;
2100 }
2101
2102 return backup;
2103 }
2104
2105 /* subflow sockets can be either outgoing (connect) or incoming
2106 * (accept).
2107 *
2108 * Outgoing subflows use in-kernel sockets.
2109 * Incoming subflows do not have their own 'struct socket' allocated,
2110 * so we need to use tcp_close() after detaching them from the mptcp
2111 * parent socket.
2112 */
2113 void __mptcp_close_ssk(struct sock *sk, struct sock *ssk,
2114 struct mptcp_subflow_context *subflow)
2115 {
2116 bool dispose_socket = false;
2117 struct socket *sock;
2118
2119 list_del(&subflow->node);
2120
2121 lock_sock_nested(ssk, SINGLE_DEPTH_NESTING);
2122
2123 /* if we are invoked by the msk cleanup code, the subflow is
2124 * already orphaned
2125 */
2126 sock = ssk->sk_socket;
2127 if (sock) {
2128 dispose_socket = sock != sk->sk_socket;
2129 sock_orphan(ssk);
2130 }
2131
2132 subflow->disposable = 1;
2133
2134 /* if ssk hit tcp_done(), tcp_cleanup_ulp() cleared the related ops
2135 * the ssk has been already destroyed, we just need to release the
2136 * reference owned by msk;
2137 */
2138 if (!inet_csk(ssk)->icsk_ulp_ops) {
2139 kfree_rcu(subflow, rcu);
2140 } else {
2141 /* otherwise tcp will dispose of the ssk and subflow ctx */
2142 __tcp_close(ssk, 0);
2143
2144 /* close acquired an extra ref */
2145 __sock_put(ssk);
2146 }
2147 release_sock(ssk);
2148 if (dispose_socket)
2149 iput(SOCK_INODE(sock));
2150
2151 sock_put(ssk);
2152 }
2153
2154 static unsigned int mptcp_sync_mss(struct sock *sk, u32 pmtu)
2155 {
2156 return 0;
2157 }
2158
2159 static void pm_work(struct mptcp_sock *msk)
2160 {
2161 struct mptcp_pm_data *pm = &msk->pm;
2162
2163 spin_lock_bh(&msk->pm.lock);
2164
2165 pr_debug("msk=%p status=%x", msk, pm->status);
2166 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) {
2167 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED);
2168 mptcp_pm_nl_add_addr_received(msk);
2169 }
2170 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) {
2171 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK);
2172 mptcp_pm_nl_add_addr_send_ack(msk);
2173 }
2174 if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) {
2175 pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED);
2176 mptcp_pm_nl_rm_addr_received(msk);
2177 }
2178 if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) {
2179 pm->status &= ~BIT(MPTCP_PM_ESTABLISHED);
2180 mptcp_pm_nl_fully_established(msk);
2181 }
2182 if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) {
2183 pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED);
2184 mptcp_pm_nl_subflow_established(msk);
2185 }
2186
2187 spin_unlock_bh(&msk->pm.lock);
2188 }
2189
2190 static void __mptcp_close_subflow(struct mptcp_sock *msk)
2191 {
2192 struct mptcp_subflow_context *subflow, *tmp;
2193
2194 list_for_each_entry_safe(subflow, tmp, &msk->conn_list, node) {
2195 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2196
2197 if (inet_sk_state_load(ssk) != TCP_CLOSE)
2198 continue;
2199
2200 __mptcp_close_ssk((struct sock *)msk, ssk, subflow);
2201 }
2202 }
2203
2204 static bool mptcp_check_close_timeout(const struct sock *sk)
2205 {
2206 s32 delta = tcp_jiffies32 - inet_csk(sk)->icsk_mtup.probe_timestamp;
2207 struct mptcp_subflow_context *subflow;
2208
2209 if (delta >= TCP_TIMEWAIT_LEN)
2210 return true;
2211
2212 /* if all subflows are in closed status don't bother with additional
2213 * timeout
2214 */
2215 mptcp_for_each_subflow(mptcp_sk(sk), subflow) {
2216 if (inet_sk_state_load(mptcp_subflow_tcp_sock(subflow)) !=
2217 TCP_CLOSE)
2218 return false;
2219 }
2220 return true;
2221 }
2222
2223 static void mptcp_check_fastclose(struct mptcp_sock *msk)
2224 {
2225 struct mptcp_subflow_context *subflow, *tmp;
2226 struct sock *sk = &msk->sk.icsk_inet.sk;
2227
2228 if (likely(!READ_ONCE(msk->rcv_fastclose)))
2229 return;
2230
2231 mptcp_token_destroy(msk);
2232
2233 list_for_each_entry_safe(subflow, tmp, &msk->conn_list, node) {
2234 struct sock *tcp_sk = mptcp_subflow_tcp_sock(subflow);
2235
2236 lock_sock(tcp_sk);
2237 if (tcp_sk->sk_state != TCP_CLOSE) {
2238 tcp_send_active_reset(tcp_sk, GFP_ATOMIC);
2239 tcp_set_state(tcp_sk, TCP_CLOSE);
2240 }
2241 release_sock(tcp_sk);
2242 }
2243
2244 inet_sk_state_store(sk, TCP_CLOSE);
2245 sk->sk_shutdown = SHUTDOWN_MASK;
2246 smp_mb__before_atomic(); /* SHUTDOWN must be visible first */
2247 set_bit(MPTCP_DATA_READY, &msk->flags);
2248 set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags);
2249
2250 mptcp_close_wake_up(sk);
2251 }
2252
2253 static void mptcp_worker(struct work_struct *work)
2254 {
2255 struct mptcp_sock *msk = container_of(work, struct mptcp_sock, work);
2256 struct sock *ssk, *sk = &msk->sk.icsk_inet.sk;
2257 struct mptcp_sendmsg_info info = {};
2258 struct mptcp_data_frag *dfrag;
2259 size_t copied = 0;
2260 int state, ret;
2261
2262 lock_sock(sk);
2263 state = sk->sk_state;
2264 if (unlikely(state == TCP_CLOSE))
2265 goto unlock;
2266
2267 mptcp_check_data_fin_ack(sk);
2268 __mptcp_flush_join_list(msk);
2269
2270 mptcp_check_fastclose(msk);
2271
2272 if (test_and_clear_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags))
2273 __mptcp_close_subflow(msk);
2274
2275 if (msk->pm.status)
2276 pm_work(msk);
2277
2278 if (test_and_clear_bit(MPTCP_WORK_EOF, &msk->flags))
2279 mptcp_check_for_eof(msk);
2280
2281 __mptcp_check_send_data_fin(sk);
2282 mptcp_check_data_fin(sk);
2283
2284 /* if the msk data is completely acked, or the socket timedout,
2285 * there is no point in keeping around an orphaned sk
2286 */
2287 if (sock_flag(sk, SOCK_DEAD) &&
2288 (mptcp_check_close_timeout(sk) ||
2289 (state != sk->sk_state &&
2290 ((1 << inet_sk_state_load(sk)) & (TCPF_CLOSE | TCPF_FIN_WAIT2))))) {
2291 inet_sk_state_store(sk, TCP_CLOSE);
2292 __mptcp_destroy_sock(sk);
2293 goto unlock;
2294 }
2295
2296 if (!test_and_clear_bit(MPTCP_WORK_RTX, &msk->flags))
2297 goto unlock;
2298
2299 dfrag = mptcp_rtx_head(sk);
2300 if (!dfrag)
2301 goto unlock;
2302
2303 ssk = mptcp_subflow_get_retrans(msk);
2304 if (!ssk)
2305 goto reset_unlock;
2306
2307 lock_sock(ssk);
2308
2309 /* limit retransmission to the bytes already sent on some subflows */
2310 info.sent = 0;
2311 info.limit = dfrag->already_sent;
2312 while (info.sent < dfrag->already_sent) {
2313 if (!mptcp_alloc_tx_skb(sk, ssk))
2314 break;
2315
2316 ret = mptcp_sendmsg_frag(sk, ssk, dfrag, &info);
2317 if (ret <= 0)
2318 break;
2319
2320 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_RETRANSSEGS);
2321 copied += ret;
2322 info.sent += ret;
2323 }
2324 if (copied)
2325 tcp_push(ssk, 0, info.mss_now, tcp_sk(ssk)->nonagle,
2326 info.size_goal);
2327
2328 mptcp_set_timeout(sk, ssk);
2329 release_sock(ssk);
2330
2331 reset_unlock:
2332 if (!mptcp_timer_pending(sk))
2333 mptcp_reset_timer(sk);
2334
2335 unlock:
2336 release_sock(sk);
2337 sock_put(sk);
2338 }
2339
2340 static int __mptcp_init_sock(struct sock *sk)
2341 {
2342 struct mptcp_sock *msk = mptcp_sk(sk);
2343
2344 spin_lock_init(&msk->join_list_lock);
2345
2346 INIT_LIST_HEAD(&msk->conn_list);
2347 INIT_LIST_HEAD(&msk->join_list);
2348 INIT_LIST_HEAD(&msk->rtx_queue);
2349 INIT_WORK(&msk->work, mptcp_worker);
2350 __skb_queue_head_init(&msk->receive_queue);
2351 __skb_queue_head_init(&msk->skb_tx_cache);
2352 msk->out_of_order_queue = RB_ROOT;
2353 msk->first_pending = NULL;
2354 msk->wmem_reserved = 0;
2355 msk->rmem_released = 0;
2356 msk->tx_pending_data = 0;
2357 msk->size_goal_cache = TCP_BASE_MSS;
2358
2359 msk->ack_hint = NULL;
2360 msk->first = NULL;
2361 inet_csk(sk)->icsk_sync_mss = mptcp_sync_mss;
2362
2363 mptcp_pm_data_init(msk);
2364
2365 /* re-use the csk retrans timer for MPTCP-level retrans */
2366 timer_setup(&msk->sk.icsk_retransmit_timer, mptcp_retransmit_timer, 0);
2367 timer_setup(&sk->sk_timer, mptcp_timeout_timer, 0);
2368 return 0;
2369 }
2370
2371 static int mptcp_init_sock(struct sock *sk)
2372 {
2373 struct net *net = sock_net(sk);
2374 int ret;
2375
2376 ret = __mptcp_init_sock(sk);
2377 if (ret)
2378 return ret;
2379
2380 if (!mptcp_is_enabled(net))
2381 return -ENOPROTOOPT;
2382
2383 if (unlikely(!net->mib.mptcp_statistics) && !mptcp_mib_alloc(net))
2384 return -ENOMEM;
2385
2386 ret = __mptcp_socket_create(mptcp_sk(sk));
2387 if (ret)
2388 return ret;
2389
2390 sk_sockets_allocated_inc(sk);
2391 sk->sk_rcvbuf = sock_net(sk)->ipv4.sysctl_tcp_rmem[1];
2392 sk->sk_sndbuf = sock_net(sk)->ipv4.sysctl_tcp_wmem[1];
2393
2394 return 0;
2395 }
2396
2397 static void __mptcp_clear_xmit(struct sock *sk)
2398 {
2399 struct mptcp_sock *msk = mptcp_sk(sk);
2400 struct mptcp_data_frag *dtmp, *dfrag;
2401 struct sk_buff *skb;
2402
2403 WRITE_ONCE(msk->first_pending, NULL);
2404 list_for_each_entry_safe(dfrag, dtmp, &msk->rtx_queue, list)
2405 dfrag_clear(sk, dfrag);
2406 while ((skb = __skb_dequeue(&msk->skb_tx_cache)) != NULL) {
2407 sk->sk_forward_alloc += skb->truesize;
2408 kfree_skb(skb);
2409 }
2410 }
2411
2412 static void mptcp_cancel_work(struct sock *sk)
2413 {
2414 struct mptcp_sock *msk = mptcp_sk(sk);
2415
2416 if (cancel_work_sync(&msk->work))
2417 __sock_put(sk);
2418 }
2419
2420 void mptcp_subflow_shutdown(struct sock *sk, struct sock *ssk, int how)
2421 {
2422 lock_sock(ssk);
2423
2424 switch (ssk->sk_state) {
2425 case TCP_LISTEN:
2426 if (!(how & RCV_SHUTDOWN))
2427 break;
2428 fallthrough;
2429 case TCP_SYN_SENT:
2430 tcp_disconnect(ssk, O_NONBLOCK);
2431 break;
2432 default:
2433 if (__mptcp_check_fallback(mptcp_sk(sk))) {
2434 pr_debug("Fallback");
2435 ssk->sk_shutdown |= how;
2436 tcp_shutdown(ssk, how);
2437 } else {
2438 pr_debug("Sending DATA_FIN on subflow %p", ssk);
2439 mptcp_set_timeout(sk, ssk);
2440 tcp_send_ack(ssk);
2441 }
2442 break;
2443 }
2444
2445 release_sock(ssk);
2446 }
2447
2448 static const unsigned char new_state[16] = {
2449 /* current state: new state: action: */
2450 [0 /* (Invalid) */] = TCP_CLOSE,
2451 [TCP_ESTABLISHED] = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2452 [TCP_SYN_SENT] = TCP_CLOSE,
2453 [TCP_SYN_RECV] = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2454 [TCP_FIN_WAIT1] = TCP_FIN_WAIT1,
2455 [TCP_FIN_WAIT2] = TCP_FIN_WAIT2,
2456 [TCP_TIME_WAIT] = TCP_CLOSE, /* should not happen ! */
2457 [TCP_CLOSE] = TCP_CLOSE,
2458 [TCP_CLOSE_WAIT] = TCP_LAST_ACK | TCP_ACTION_FIN,
2459 [TCP_LAST_ACK] = TCP_LAST_ACK,
2460 [TCP_LISTEN] = TCP_CLOSE,
2461 [TCP_CLOSING] = TCP_CLOSING,
2462 [TCP_NEW_SYN_RECV] = TCP_CLOSE, /* should not happen ! */
2463 };
2464
2465 static int mptcp_close_state(struct sock *sk)
2466 {
2467 int next = (int)new_state[sk->sk_state];
2468 int ns = next & TCP_STATE_MASK;
2469
2470 inet_sk_state_store(sk, ns);
2471
2472 return next & TCP_ACTION_FIN;
2473 }
2474
2475 static void __mptcp_check_send_data_fin(struct sock *sk)
2476 {
2477 struct mptcp_subflow_context *subflow;
2478 struct mptcp_sock *msk = mptcp_sk(sk);
2479
2480 pr_debug("msk=%p snd_data_fin_enable=%d pending=%d snd_nxt=%llu write_seq=%llu",
2481 msk, msk->snd_data_fin_enable, !!mptcp_send_head(sk),
2482 msk->snd_nxt, msk->write_seq);
2483
2484 /* we still need to enqueue subflows or not really shutting down,
2485 * skip this
2486 */
2487 if (!msk->snd_data_fin_enable || msk->snd_nxt + 1 != msk->write_seq ||
2488 mptcp_send_head(sk))
2489 return;
2490
2491 WRITE_ONCE(msk->snd_nxt, msk->write_seq);
2492
2493 /* fallback socket will not get data_fin/ack, can move to the next
2494 * state now
2495 */
2496 if (__mptcp_check_fallback(msk)) {
2497 if ((1 << sk->sk_state) & (TCPF_CLOSING | TCPF_LAST_ACK)) {
2498 inet_sk_state_store(sk, TCP_CLOSE);
2499 mptcp_close_wake_up(sk);
2500 } else if (sk->sk_state == TCP_FIN_WAIT1) {
2501 inet_sk_state_store(sk, TCP_FIN_WAIT2);
2502 }
2503 }
2504
2505 __mptcp_flush_join_list(msk);
2506 mptcp_for_each_subflow(msk, subflow) {
2507 struct sock *tcp_sk = mptcp_subflow_tcp_sock(subflow);
2508
2509 mptcp_subflow_shutdown(sk, tcp_sk, SEND_SHUTDOWN);
2510 }
2511 }
2512
2513 static void __mptcp_wr_shutdown(struct sock *sk)
2514 {
2515 struct mptcp_sock *msk = mptcp_sk(sk);
2516
2517 pr_debug("msk=%p snd_data_fin_enable=%d shutdown=%x state=%d pending=%d",
2518 msk, msk->snd_data_fin_enable, sk->sk_shutdown, sk->sk_state,
2519 !!mptcp_send_head(sk));
2520
2521 /* will be ignored by fallback sockets */
2522 WRITE_ONCE(msk->write_seq, msk->write_seq + 1);
2523 WRITE_ONCE(msk->snd_data_fin_enable, 1);
2524
2525 __mptcp_check_send_data_fin(sk);
2526 }
2527
2528 static void __mptcp_destroy_sock(struct sock *sk)
2529 {
2530 struct mptcp_subflow_context *subflow, *tmp;
2531 struct mptcp_sock *msk = mptcp_sk(sk);
2532 LIST_HEAD(conn_list);
2533
2534 pr_debug("msk=%p", msk);
2535
2536 /* be sure to always acquire the join list lock, to sync vs
2537 * mptcp_finish_join().
2538 */
2539 spin_lock_bh(&msk->join_list_lock);
2540 list_splice_tail_init(&msk->join_list, &msk->conn_list);
2541 spin_unlock_bh(&msk->join_list_lock);
2542 list_splice_init(&msk->conn_list, &conn_list);
2543
2544 sk_stop_timer(sk, &msk->sk.icsk_retransmit_timer);
2545 sk_stop_timer(sk, &sk->sk_timer);
2546 msk->pm.status = 0;
2547
2548 list_for_each_entry_safe(subflow, tmp, &conn_list, node) {
2549 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2550 __mptcp_close_ssk(sk, ssk, subflow);
2551 }
2552
2553 sk->sk_prot->destroy(sk);
2554
2555 WARN_ON_ONCE(msk->wmem_reserved);
2556 WARN_ON_ONCE(msk->rmem_released);
2557 sk_stream_kill_queues(sk);
2558 xfrm_sk_free_policy(sk);
2559 sk_refcnt_debug_release(sk);
2560 sock_put(sk);
2561 }
2562
2563 static void mptcp_close(struct sock *sk, long timeout)
2564 {
2565 struct mptcp_subflow_context *subflow;
2566 bool do_cancel_work = false;
2567
2568 lock_sock(sk);
2569 sk->sk_shutdown = SHUTDOWN_MASK;
2570
2571 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE)) {
2572 inet_sk_state_store(sk, TCP_CLOSE);
2573 goto cleanup;
2574 }
2575
2576 if (mptcp_close_state(sk))
2577 __mptcp_wr_shutdown(sk);
2578
2579 sk_stream_wait_close(sk, timeout);
2580
2581 cleanup:
2582 /* orphan all the subflows */
2583 inet_csk(sk)->icsk_mtup.probe_timestamp = tcp_jiffies32;
2584 list_for_each_entry(subflow, &mptcp_sk(sk)->conn_list, node) {
2585 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2586 bool slow, dispose_socket;
2587 struct socket *sock;
2588
2589 slow = lock_sock_fast(ssk);
2590 sock = ssk->sk_socket;
2591 dispose_socket = sock && sock != sk->sk_socket;
2592 sock_orphan(ssk);
2593 unlock_sock_fast(ssk, slow);
2594
2595 /* for the outgoing subflows we additionally need to free
2596 * the associated socket
2597 */
2598 if (dispose_socket)
2599 iput(SOCK_INODE(sock));
2600 }
2601 sock_orphan(sk);
2602
2603 sock_hold(sk);
2604 pr_debug("msk=%p state=%d", sk, sk->sk_state);
2605 if (sk->sk_state == TCP_CLOSE) {
2606 __mptcp_destroy_sock(sk);
2607 do_cancel_work = true;
2608 } else {
2609 sk_reset_timer(sk, &sk->sk_timer, jiffies + TCP_TIMEWAIT_LEN);
2610 }
2611 release_sock(sk);
2612 if (do_cancel_work)
2613 mptcp_cancel_work(sk);
2614 sock_put(sk);
2615 }
2616
2617 static void mptcp_copy_inaddrs(struct sock *msk, const struct sock *ssk)
2618 {
2619 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2620 const struct ipv6_pinfo *ssk6 = inet6_sk(ssk);
2621 struct ipv6_pinfo *msk6 = inet6_sk(msk);
2622
2623 msk->sk_v6_daddr = ssk->sk_v6_daddr;
2624 msk->sk_v6_rcv_saddr = ssk->sk_v6_rcv_saddr;
2625
2626 if (msk6 && ssk6) {
2627 msk6->saddr = ssk6->saddr;
2628 msk6->flow_label = ssk6->flow_label;
2629 }
2630 #endif
2631
2632 inet_sk(msk)->inet_num = inet_sk(ssk)->inet_num;
2633 inet_sk(msk)->inet_dport = inet_sk(ssk)->inet_dport;
2634 inet_sk(msk)->inet_sport = inet_sk(ssk)->inet_sport;
2635 inet_sk(msk)->inet_daddr = inet_sk(ssk)->inet_daddr;
2636 inet_sk(msk)->inet_saddr = inet_sk(ssk)->inet_saddr;
2637 inet_sk(msk)->inet_rcv_saddr = inet_sk(ssk)->inet_rcv_saddr;
2638 }
2639
2640 static int mptcp_disconnect(struct sock *sk, int flags)
2641 {
2642 /* Should never be called.
2643 * inet_stream_connect() calls ->disconnect, but that
2644 * refers to the subflow socket, not the mptcp one.
2645 */
2646 WARN_ON_ONCE(1);
2647 return 0;
2648 }
2649
2650 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2651 static struct ipv6_pinfo *mptcp_inet6_sk(const struct sock *sk)
2652 {
2653 unsigned int offset = sizeof(struct mptcp6_sock) - sizeof(struct ipv6_pinfo);
2654
2655 return (struct ipv6_pinfo *)(((u8 *)sk) + offset);
2656 }
2657 #endif
2658
2659 struct sock *mptcp_sk_clone(const struct sock *sk,
2660 const struct mptcp_options_received *mp_opt,
2661 struct request_sock *req)
2662 {
2663 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
2664 struct sock *nsk = sk_clone_lock(sk, GFP_ATOMIC);
2665 struct mptcp_sock *msk;
2666 u64 ack_seq;
2667
2668 if (!nsk)
2669 return NULL;
2670
2671 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2672 if (nsk->sk_family == AF_INET6)
2673 inet_sk(nsk)->pinet6 = mptcp_inet6_sk(nsk);
2674 #endif
2675
2676 __mptcp_init_sock(nsk);
2677
2678 msk = mptcp_sk(nsk);
2679 msk->local_key = subflow_req->local_key;
2680 msk->token = subflow_req->token;
2681 msk->subflow = NULL;
2682 WRITE_ONCE(msk->fully_established, false);
2683
2684 msk->write_seq = subflow_req->idsn + 1;
2685 msk->snd_nxt = msk->write_seq;
2686 msk->snd_una = msk->write_seq;
2687 msk->wnd_end = msk->snd_nxt + req->rsk_rcv_wnd;
2688
2689 if (mp_opt->mp_capable) {
2690 msk->can_ack = true;
2691 msk->remote_key = mp_opt->sndr_key;
2692 mptcp_crypto_key_sha(msk->remote_key, NULL, &ack_seq);
2693 ack_seq++;
2694 WRITE_ONCE(msk->ack_seq, ack_seq);
2695 WRITE_ONCE(msk->rcv_wnd_sent, ack_seq);
2696 }
2697
2698 sock_reset_flag(nsk, SOCK_RCU_FREE);
2699 /* will be fully established after successful MPC subflow creation */
2700 inet_sk_state_store(nsk, TCP_SYN_RECV);
2701
2702 security_inet_csk_clone(nsk, req);
2703 bh_unlock_sock(nsk);
2704
2705 /* keep a single reference */
2706 __sock_put(nsk);
2707 return nsk;
2708 }
2709
2710 void mptcp_rcv_space_init(struct mptcp_sock *msk, const struct sock *ssk)
2711 {
2712 const struct tcp_sock *tp = tcp_sk(ssk);
2713
2714 msk->rcvq_space.copied = 0;
2715 msk->rcvq_space.rtt_us = 0;
2716
2717 msk->rcvq_space.time = tp->tcp_mstamp;
2718
2719 /* initial rcv_space offering made to peer */
2720 msk->rcvq_space.space = min_t(u32, tp->rcv_wnd,
2721 TCP_INIT_CWND * tp->advmss);
2722 if (msk->rcvq_space.space == 0)
2723 msk->rcvq_space.space = TCP_INIT_CWND * TCP_MSS_DEFAULT;
2724
2725 WRITE_ONCE(msk->wnd_end, msk->snd_nxt + tcp_sk(ssk)->snd_wnd);
2726 }
2727
2728 static struct sock *mptcp_accept(struct sock *sk, int flags, int *err,
2729 bool kern)
2730 {
2731 struct mptcp_sock *msk = mptcp_sk(sk);
2732 struct socket *listener;
2733 struct sock *newsk;
2734
2735 listener = __mptcp_nmpc_socket(msk);
2736 if (WARN_ON_ONCE(!listener)) {
2737 *err = -EINVAL;
2738 return NULL;
2739 }
2740
2741 pr_debug("msk=%p, listener=%p", msk, mptcp_subflow_ctx(listener->sk));
2742 newsk = inet_csk_accept(listener->sk, flags, err, kern);
2743 if (!newsk)
2744 return NULL;
2745
2746 pr_debug("msk=%p, subflow is mptcp=%d", msk, sk_is_mptcp(newsk));
2747 if (sk_is_mptcp(newsk)) {
2748 struct mptcp_subflow_context *subflow;
2749 struct sock *new_mptcp_sock;
2750
2751 subflow = mptcp_subflow_ctx(newsk);
2752 new_mptcp_sock = subflow->conn;
2753
2754 /* is_mptcp should be false if subflow->conn is missing, see
2755 * subflow_syn_recv_sock()
2756 */
2757 if (WARN_ON_ONCE(!new_mptcp_sock)) {
2758 tcp_sk(newsk)->is_mptcp = 0;
2759 return newsk;
2760 }
2761
2762 /* acquire the 2nd reference for the owning socket */
2763 sock_hold(new_mptcp_sock);
2764 newsk = new_mptcp_sock;
2765 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPCAPABLEPASSIVEACK);
2766 } else {
2767 MPTCP_INC_STATS(sock_net(sk),
2768 MPTCP_MIB_MPCAPABLEPASSIVEFALLBACK);
2769 }
2770
2771 return newsk;
2772 }
2773
2774 void mptcp_destroy_common(struct mptcp_sock *msk)
2775 {
2776 struct sock *sk = (struct sock *)msk;
2777
2778 __mptcp_clear_xmit(sk);
2779
2780 /* move to sk_receive_queue, sk_stream_kill_queues will purge it */
2781 skb_queue_splice_tail_init(&msk->receive_queue, &sk->sk_receive_queue);
2782
2783 skb_rbtree_purge(&msk->out_of_order_queue);
2784 mptcp_token_destroy(msk);
2785 mptcp_pm_free_anno_list(msk);
2786 }
2787
2788 static void mptcp_destroy(struct sock *sk)
2789 {
2790 struct mptcp_sock *msk = mptcp_sk(sk);
2791
2792 mptcp_destroy_common(msk);
2793 sk_sockets_allocated_dec(sk);
2794 }
2795
2796 static int mptcp_setsockopt_sol_socket(struct mptcp_sock *msk, int optname,
2797 sockptr_t optval, unsigned int optlen)
2798 {
2799 struct sock *sk = (struct sock *)msk;
2800 struct socket *ssock;
2801 int ret;
2802
2803 switch (optname) {
2804 case SO_REUSEPORT:
2805 case SO_REUSEADDR:
2806 lock_sock(sk);
2807 ssock = __mptcp_nmpc_socket(msk);
2808 if (!ssock) {
2809 release_sock(sk);
2810 return -EINVAL;
2811 }
2812
2813 ret = sock_setsockopt(ssock, SOL_SOCKET, optname, optval, optlen);
2814 if (ret == 0) {
2815 if (optname == SO_REUSEPORT)
2816 sk->sk_reuseport = ssock->sk->sk_reuseport;
2817 else if (optname == SO_REUSEADDR)
2818 sk->sk_reuse = ssock->sk->sk_reuse;
2819 }
2820 release_sock(sk);
2821 return ret;
2822 }
2823
2824 return sock_setsockopt(sk->sk_socket, SOL_SOCKET, optname, optval, optlen);
2825 }
2826
2827 static int mptcp_setsockopt_v6(struct mptcp_sock *msk, int optname,
2828 sockptr_t optval, unsigned int optlen)
2829 {
2830 struct sock *sk = (struct sock *)msk;
2831 int ret = -EOPNOTSUPP;
2832 struct socket *ssock;
2833
2834 switch (optname) {
2835 case IPV6_V6ONLY:
2836 lock_sock(sk);
2837 ssock = __mptcp_nmpc_socket(msk);
2838 if (!ssock) {
2839 release_sock(sk);
2840 return -EINVAL;
2841 }
2842
2843 ret = tcp_setsockopt(ssock->sk, SOL_IPV6, optname, optval, optlen);
2844 if (ret == 0)
2845 sk->sk_ipv6only = ssock->sk->sk_ipv6only;
2846
2847 release_sock(sk);
2848 break;
2849 }
2850
2851 return ret;
2852 }
2853
2854 static int mptcp_setsockopt(struct sock *sk, int level, int optname,
2855 sockptr_t optval, unsigned int optlen)
2856 {
2857 struct mptcp_sock *msk = mptcp_sk(sk);
2858 struct sock *ssk;
2859
2860 pr_debug("msk=%p", msk);
2861
2862 if (level == SOL_SOCKET)
2863 return mptcp_setsockopt_sol_socket(msk, optname, optval, optlen);
2864
2865 /* @@ the meaning of setsockopt() when the socket is connected and
2866 * there are multiple subflows is not yet defined. It is up to the
2867 * MPTCP-level socket to configure the subflows until the subflow
2868 * is in TCP fallback, when TCP socket options are passed through
2869 * to the one remaining subflow.
2870 */
2871 lock_sock(sk);
2872 ssk = __mptcp_tcp_fallback(msk);
2873 release_sock(sk);
2874 if (ssk)
2875 return tcp_setsockopt(ssk, level, optname, optval, optlen);
2876
2877 if (level == SOL_IPV6)
2878 return mptcp_setsockopt_v6(msk, optname, optval, optlen);
2879
2880 return -EOPNOTSUPP;
2881 }
2882
2883 static int mptcp_getsockopt(struct sock *sk, int level, int optname,
2884 char __user *optval, int __user *option)
2885 {
2886 struct mptcp_sock *msk = mptcp_sk(sk);
2887 struct sock *ssk;
2888
2889 pr_debug("msk=%p", msk);
2890
2891 /* @@ the meaning of setsockopt() when the socket is connected and
2892 * there are multiple subflows is not yet defined. It is up to the
2893 * MPTCP-level socket to configure the subflows until the subflow
2894 * is in TCP fallback, when socket options are passed through
2895 * to the one remaining subflow.
2896 */
2897 lock_sock(sk);
2898 ssk = __mptcp_tcp_fallback(msk);
2899 release_sock(sk);
2900 if (ssk)
2901 return tcp_getsockopt(ssk, level, optname, optval, option);
2902
2903 return -EOPNOTSUPP;
2904 }
2905
2906 void __mptcp_data_acked(struct sock *sk)
2907 {
2908 if (!sock_owned_by_user(sk))
2909 __mptcp_clean_una(sk);
2910 else
2911 set_bit(MPTCP_CLEAN_UNA, &mptcp_sk(sk)->flags);
2912
2913 if (mptcp_pending_data_fin_ack(sk))
2914 mptcp_schedule_work(sk);
2915 }
2916
2917 void __mptcp_check_push(struct sock *sk, struct sock *ssk)
2918 {
2919 if (!mptcp_send_head(sk))
2920 return;
2921
2922 if (!sock_owned_by_user(sk))
2923 __mptcp_subflow_push_pending(sk, ssk);
2924 else
2925 set_bit(MPTCP_PUSH_PENDING, &mptcp_sk(sk)->flags);
2926 }
2927
2928 #define MPTCP_DEFERRED_ALL (TCPF_WRITE_TIMER_DEFERRED)
2929
2930 /* processes deferred events and flush wmem */
2931 static void mptcp_release_cb(struct sock *sk)
2932 {
2933 unsigned long flags, nflags;
2934
2935 /* push_pending may touch wmem_reserved, do it before the later
2936 * cleanup
2937 */
2938 if (test_and_clear_bit(MPTCP_CLEAN_UNA, &mptcp_sk(sk)->flags))
2939 __mptcp_clean_una(sk);
2940 if (test_and_clear_bit(MPTCP_PUSH_PENDING, &mptcp_sk(sk)->flags)) {
2941 /* mptcp_push_pending() acquires the subflow socket lock
2942 *
2943 * 1) can't be invoked in atomic scope
2944 * 2) must avoid ABBA deadlock with msk socket spinlock: the RX
2945 * datapath acquires the msk socket spinlock while helding
2946 * the subflow socket lock
2947 */
2948
2949 spin_unlock_bh(&sk->sk_lock.slock);
2950 mptcp_push_pending(sk, 0);
2951 spin_lock_bh(&sk->sk_lock.slock);
2952 }
2953
2954 /* clear any wmem reservation and errors */
2955 __mptcp_update_wmem(sk);
2956 __mptcp_update_rmem(sk);
2957
2958 do {
2959 flags = sk->sk_tsq_flags;
2960 if (!(flags & MPTCP_DEFERRED_ALL))
2961 return;
2962 nflags = flags & ~MPTCP_DEFERRED_ALL;
2963 } while (cmpxchg(&sk->sk_tsq_flags, flags, nflags) != flags);
2964
2965 sock_release_ownership(sk);
2966
2967 if (flags & TCPF_WRITE_TIMER_DEFERRED) {
2968 mptcp_retransmit_handler(sk);
2969 __sock_put(sk);
2970 }
2971 }
2972
2973 static int mptcp_hash(struct sock *sk)
2974 {
2975 /* should never be called,
2976 * we hash the TCP subflows not the master socket
2977 */
2978 WARN_ON_ONCE(1);
2979 return 0;
2980 }
2981
2982 static void mptcp_unhash(struct sock *sk)
2983 {
2984 /* called from sk_common_release(), but nothing to do here */
2985 }
2986
2987 static int mptcp_get_port(struct sock *sk, unsigned short snum)
2988 {
2989 struct mptcp_sock *msk = mptcp_sk(sk);
2990 struct socket *ssock;
2991
2992 ssock = __mptcp_nmpc_socket(msk);
2993 pr_debug("msk=%p, subflow=%p", msk, ssock);
2994 if (WARN_ON_ONCE(!ssock))
2995 return -EINVAL;
2996
2997 return inet_csk_get_port(ssock->sk, snum);
2998 }
2999
3000 void mptcp_finish_connect(struct sock *ssk)
3001 {
3002 struct mptcp_subflow_context *subflow;
3003 struct mptcp_sock *msk;
3004 struct sock *sk;
3005 u64 ack_seq;
3006
3007 subflow = mptcp_subflow_ctx(ssk);
3008 sk = subflow->conn;
3009 msk = mptcp_sk(sk);
3010
3011 pr_debug("msk=%p, token=%u", sk, subflow->token);
3012
3013 mptcp_crypto_key_sha(subflow->remote_key, NULL, &ack_seq);
3014 ack_seq++;
3015 subflow->map_seq = ack_seq;
3016 subflow->map_subflow_seq = 1;
3017
3018 /* the socket is not connected yet, no msk/subflow ops can access/race
3019 * accessing the field below
3020 */
3021 WRITE_ONCE(msk->remote_key, subflow->remote_key);
3022 WRITE_ONCE(msk->local_key, subflow->local_key);
3023 WRITE_ONCE(msk->write_seq, subflow->idsn + 1);
3024 WRITE_ONCE(msk->snd_nxt, msk->write_seq);
3025 WRITE_ONCE(msk->ack_seq, ack_seq);
3026 WRITE_ONCE(msk->rcv_wnd_sent, ack_seq);
3027 WRITE_ONCE(msk->can_ack, 1);
3028 WRITE_ONCE(msk->snd_una, msk->write_seq);
3029
3030 mptcp_pm_new_connection(msk, 0);
3031
3032 mptcp_rcv_space_init(msk, ssk);
3033 }
3034
3035 static void mptcp_sock_graft(struct sock *sk, struct socket *parent)
3036 {
3037 write_lock_bh(&sk->sk_callback_lock);
3038 rcu_assign_pointer(sk->sk_wq, &parent->wq);
3039 sk_set_socket(sk, parent);
3040 sk->sk_uid = SOCK_INODE(parent)->i_uid;
3041 write_unlock_bh(&sk->sk_callback_lock);
3042 }
3043
3044 bool mptcp_finish_join(struct sock *ssk)
3045 {
3046 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
3047 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
3048 struct sock *parent = (void *)msk;
3049 struct socket *parent_sock;
3050 bool ret;
3051
3052 pr_debug("msk=%p, subflow=%p", msk, subflow);
3053
3054 /* mptcp socket already closing? */
3055 if (!mptcp_is_fully_established(parent))
3056 return false;
3057
3058 if (!msk->pm.server_side)
3059 return true;
3060
3061 if (!mptcp_pm_allow_new_subflow(msk))
3062 return false;
3063
3064 /* active connections are already on conn_list, and we can't acquire
3065 * msk lock here.
3066 * use the join list lock as synchronization point and double-check
3067 * msk status to avoid racing with __mptcp_destroy_sock()
3068 */
3069 spin_lock_bh(&msk->join_list_lock);
3070 ret = inet_sk_state_load(parent) == TCP_ESTABLISHED;
3071 if (ret && !WARN_ON_ONCE(!list_empty(&subflow->node))) {
3072 list_add_tail(&subflow->node, &msk->join_list);
3073 sock_hold(ssk);
3074 }
3075 spin_unlock_bh(&msk->join_list_lock);
3076 if (!ret)
3077 return false;
3078
3079 /* attach to msk socket only after we are sure he will deal with us
3080 * at close time
3081 */
3082 parent_sock = READ_ONCE(parent->sk_socket);
3083 if (parent_sock && !ssk->sk_socket)
3084 mptcp_sock_graft(ssk, parent_sock);
3085 subflow->map_seq = READ_ONCE(msk->ack_seq);
3086 return true;
3087 }
3088
3089 static struct proto mptcp_prot = {
3090 .name = "MPTCP",
3091 .owner = THIS_MODULE,
3092 .init = mptcp_init_sock,
3093 .disconnect = mptcp_disconnect,
3094 .close = mptcp_close,
3095 .accept = mptcp_accept,
3096 .setsockopt = mptcp_setsockopt,
3097 .getsockopt = mptcp_getsockopt,
3098 .shutdown = tcp_shutdown,
3099 .destroy = mptcp_destroy,
3100 .sendmsg = mptcp_sendmsg,
3101 .recvmsg = mptcp_recvmsg,
3102 .release_cb = mptcp_release_cb,
3103 .hash = mptcp_hash,
3104 .unhash = mptcp_unhash,
3105 .get_port = mptcp_get_port,
3106 .sockets_allocated = &mptcp_sockets_allocated,
3107 .memory_allocated = &tcp_memory_allocated,
3108 .memory_pressure = &tcp_memory_pressure,
3109 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_tcp_wmem),
3110 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_tcp_rmem),
3111 .sysctl_mem = sysctl_tcp_mem,
3112 .obj_size = sizeof(struct mptcp_sock),
3113 .slab_flags = SLAB_TYPESAFE_BY_RCU,
3114 .no_autobind = true,
3115 };
3116
3117 static int mptcp_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3118 {
3119 struct mptcp_sock *msk = mptcp_sk(sock->sk);
3120 struct socket *ssock;
3121 int err;
3122
3123 lock_sock(sock->sk);
3124 ssock = __mptcp_nmpc_socket(msk);
3125 if (!ssock) {
3126 err = -EINVAL;
3127 goto unlock;
3128 }
3129
3130 err = ssock->ops->bind(ssock, uaddr, addr_len);
3131 if (!err)
3132 mptcp_copy_inaddrs(sock->sk, ssock->sk);
3133
3134 unlock:
3135 release_sock(sock->sk);
3136 return err;
3137 }
3138
3139 static void mptcp_subflow_early_fallback(struct mptcp_sock *msk,
3140 struct mptcp_subflow_context *subflow)
3141 {
3142 subflow->request_mptcp = 0;
3143 __mptcp_do_fallback(msk);
3144 }
3145
3146 static int mptcp_stream_connect(struct socket *sock, struct sockaddr *uaddr,
3147 int addr_len, int flags)
3148 {
3149 struct mptcp_sock *msk = mptcp_sk(sock->sk);
3150 struct mptcp_subflow_context *subflow;
3151 struct socket *ssock;
3152 int err;
3153
3154 lock_sock(sock->sk);
3155 if (sock->state != SS_UNCONNECTED && msk->subflow) {
3156 /* pending connection or invalid state, let existing subflow
3157 * cope with that
3158 */
3159 ssock = msk->subflow;
3160 goto do_connect;
3161 }
3162
3163 ssock = __mptcp_nmpc_socket(msk);
3164 if (!ssock) {
3165 err = -EINVAL;
3166 goto unlock;
3167 }
3168
3169 mptcp_token_destroy(msk);
3170 inet_sk_state_store(sock->sk, TCP_SYN_SENT);
3171 subflow = mptcp_subflow_ctx(ssock->sk);
3172 #ifdef CONFIG_TCP_MD5SIG
3173 /* no MPTCP if MD5SIG is enabled on this socket or we may run out of
3174 * TCP option space.
3175 */
3176 if (rcu_access_pointer(tcp_sk(ssock->sk)->md5sig_info))
3177 mptcp_subflow_early_fallback(msk, subflow);
3178 #endif
3179 if (subflow->request_mptcp && mptcp_token_new_connect(ssock->sk))
3180 mptcp_subflow_early_fallback(msk, subflow);
3181
3182 do_connect:
3183 err = ssock->ops->connect(ssock, uaddr, addr_len, flags);
3184 sock->state = ssock->state;
3185
3186 /* on successful connect, the msk state will be moved to established by
3187 * subflow_finish_connect()
3188 */
3189 if (!err || err == -EINPROGRESS)
3190 mptcp_copy_inaddrs(sock->sk, ssock->sk);
3191 else
3192 inet_sk_state_store(sock->sk, inet_sk_state_load(ssock->sk));
3193
3194 unlock:
3195 release_sock(sock->sk);
3196 return err;
3197 }
3198
3199 static int mptcp_listen(struct socket *sock, int backlog)
3200 {
3201 struct mptcp_sock *msk = mptcp_sk(sock->sk);
3202 struct socket *ssock;
3203 int err;
3204
3205 pr_debug("msk=%p", msk);
3206
3207 lock_sock(sock->sk);
3208 ssock = __mptcp_nmpc_socket(msk);
3209 if (!ssock) {
3210 err = -EINVAL;
3211 goto unlock;
3212 }
3213
3214 mptcp_token_destroy(msk);
3215 inet_sk_state_store(sock->sk, TCP_LISTEN);
3216 sock_set_flag(sock->sk, SOCK_RCU_FREE);
3217
3218 err = ssock->ops->listen(ssock, backlog);
3219 inet_sk_state_store(sock->sk, inet_sk_state_load(ssock->sk));
3220 if (!err)
3221 mptcp_copy_inaddrs(sock->sk, ssock->sk);
3222
3223 unlock:
3224 release_sock(sock->sk);
3225 return err;
3226 }
3227
3228 static int mptcp_stream_accept(struct socket *sock, struct socket *newsock,
3229 int flags, bool kern)
3230 {
3231 struct mptcp_sock *msk = mptcp_sk(sock->sk);
3232 struct socket *ssock;
3233 int err;
3234
3235 pr_debug("msk=%p", msk);
3236
3237 lock_sock(sock->sk);
3238 if (sock->sk->sk_state != TCP_LISTEN)
3239 goto unlock_fail;
3240
3241 ssock = __mptcp_nmpc_socket(msk);
3242 if (!ssock)
3243 goto unlock_fail;
3244
3245 clear_bit(MPTCP_DATA_READY, &msk->flags);
3246 sock_hold(ssock->sk);
3247 release_sock(sock->sk);
3248
3249 err = ssock->ops->accept(sock, newsock, flags, kern);
3250 if (err == 0 && !mptcp_is_tcpsk(newsock->sk)) {
3251 struct mptcp_sock *msk = mptcp_sk(newsock->sk);
3252 struct mptcp_subflow_context *subflow;
3253 struct sock *newsk = newsock->sk;
3254 bool slowpath;
3255
3256 slowpath = lock_sock_fast(newsk);
3257
3258 /* PM/worker can now acquire the first subflow socket
3259 * lock without racing with listener queue cleanup,
3260 * we can notify it, if needed.
3261 */
3262 subflow = mptcp_subflow_ctx(msk->first);
3263 list_add(&subflow->node, &msk->conn_list);
3264 sock_hold(msk->first);
3265 if (mptcp_is_fully_established(newsk))
3266 mptcp_pm_fully_established(msk);
3267
3268 mptcp_copy_inaddrs(newsk, msk->first);
3269 mptcp_rcv_space_init(msk, msk->first);
3270
3271 /* set ssk->sk_socket of accept()ed flows to mptcp socket.
3272 * This is needed so NOSPACE flag can be set from tcp stack.
3273 */
3274 __mptcp_flush_join_list(msk);
3275 mptcp_for_each_subflow(msk, subflow) {
3276 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
3277
3278 if (!ssk->sk_socket)
3279 mptcp_sock_graft(ssk, newsock);
3280 }
3281 unlock_sock_fast(newsk, slowpath);
3282 }
3283
3284 if (inet_csk_listen_poll(ssock->sk))
3285 set_bit(MPTCP_DATA_READY, &msk->flags);
3286 sock_put(ssock->sk);
3287 return err;
3288
3289 unlock_fail:
3290 release_sock(sock->sk);
3291 return -EINVAL;
3292 }
3293
3294 static __poll_t mptcp_check_readable(struct mptcp_sock *msk)
3295 {
3296 return test_bit(MPTCP_DATA_READY, &msk->flags) ? EPOLLIN | EPOLLRDNORM :
3297 0;
3298 }
3299
3300 static __poll_t mptcp_check_writeable(struct mptcp_sock *msk)
3301 {
3302 struct sock *sk = (struct sock *)msk;
3303
3304 if (unlikely(sk->sk_shutdown & SEND_SHUTDOWN))
3305 return 0;
3306
3307 if (sk_stream_is_writeable(sk))
3308 return EPOLLOUT | EPOLLWRNORM;
3309
3310 set_bit(MPTCP_NOSPACE, &msk->flags);
3311 smp_mb__after_atomic(); /* msk->flags is changed by write_space cb */
3312 if (sk_stream_is_writeable(sk))
3313 return EPOLLOUT | EPOLLWRNORM;
3314
3315 return 0;
3316 }
3317
3318 static __poll_t mptcp_poll(struct file *file, struct socket *sock,
3319 struct poll_table_struct *wait)
3320 {
3321 struct sock *sk = sock->sk;
3322 struct mptcp_sock *msk;
3323 __poll_t mask = 0;
3324 int state;
3325
3326 msk = mptcp_sk(sk);
3327 sock_poll_wait(file, sock, wait);
3328
3329 state = inet_sk_state_load(sk);
3330 pr_debug("msk=%p state=%d flags=%lx", msk, state, msk->flags);
3331 if (state == TCP_LISTEN)
3332 return mptcp_check_readable(msk);
3333
3334 if (state != TCP_SYN_SENT && state != TCP_SYN_RECV) {
3335 mask |= mptcp_check_readable(msk);
3336 mask |= mptcp_check_writeable(msk);
3337 }
3338 if (sk->sk_shutdown & RCV_SHUTDOWN)
3339 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
3340
3341 return mask;
3342 }
3343
3344 static int mptcp_shutdown(struct socket *sock, int how)
3345 {
3346 struct mptcp_sock *msk = mptcp_sk(sock->sk);
3347 struct sock *sk = sock->sk;
3348 int ret = 0;
3349
3350 pr_debug("sk=%p, how=%d", msk, how);
3351
3352 lock_sock(sk);
3353
3354 how++;
3355 if ((how & ~SHUTDOWN_MASK) || !how) {
3356 ret = -EINVAL;
3357 goto out_unlock;
3358 }
3359
3360 if (sock->state == SS_CONNECTING) {
3361 if ((1 << sk->sk_state) &
3362 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
3363 sock->state = SS_DISCONNECTING;
3364 else
3365 sock->state = SS_CONNECTED;
3366 }
3367
3368 sk->sk_shutdown |= how;
3369 if ((how & SEND_SHUTDOWN) && mptcp_close_state(sk))
3370 __mptcp_wr_shutdown(sk);
3371
3372 /* Wake up anyone sleeping in poll. */
3373 sk->sk_state_change(sk);
3374
3375 out_unlock:
3376 release_sock(sk);
3377
3378 return ret;
3379 }
3380
3381 static const struct proto_ops mptcp_stream_ops = {
3382 .family = PF_INET,
3383 .owner = THIS_MODULE,
3384 .release = inet_release,
3385 .bind = mptcp_bind,
3386 .connect = mptcp_stream_connect,
3387 .socketpair = sock_no_socketpair,
3388 .accept = mptcp_stream_accept,
3389 .getname = inet_getname,
3390 .poll = mptcp_poll,
3391 .ioctl = inet_ioctl,
3392 .gettstamp = sock_gettstamp,
3393 .listen = mptcp_listen,
3394 .shutdown = mptcp_shutdown,
3395 .setsockopt = sock_common_setsockopt,
3396 .getsockopt = sock_common_getsockopt,
3397 .sendmsg = inet_sendmsg,
3398 .recvmsg = inet_recvmsg,
3399 .mmap = sock_no_mmap,
3400 .sendpage = inet_sendpage,
3401 };
3402
3403 static struct inet_protosw mptcp_protosw = {
3404 .type = SOCK_STREAM,
3405 .protocol = IPPROTO_MPTCP,
3406 .prot = &mptcp_prot,
3407 .ops = &mptcp_stream_ops,
3408 .flags = INET_PROTOSW_ICSK,
3409 };
3410
3411 void __init mptcp_proto_init(void)
3412 {
3413 mptcp_prot.h.hashinfo = tcp_prot.h.hashinfo;
3414
3415 if (percpu_counter_init(&mptcp_sockets_allocated, 0, GFP_KERNEL))
3416 panic("Failed to allocate MPTCP pcpu counter\n");
3417
3418 mptcp_subflow_init();
3419 mptcp_pm_init();
3420 mptcp_token_init();
3421
3422 if (proto_register(&mptcp_prot, 1) != 0)
3423 panic("Failed to register MPTCP proto.\n");
3424
3425 inet_register_protosw(&mptcp_protosw);
3426
3427 BUILD_BUG_ON(sizeof(struct mptcp_skb_cb) > sizeof_field(struct sk_buff, cb));
3428 }
3429
3430 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
3431 static const struct proto_ops mptcp_v6_stream_ops = {
3432 .family = PF_INET6,
3433 .owner = THIS_MODULE,
3434 .release = inet6_release,
3435 .bind = mptcp_bind,
3436 .connect = mptcp_stream_connect,
3437 .socketpair = sock_no_socketpair,
3438 .accept = mptcp_stream_accept,
3439 .getname = inet6_getname,
3440 .poll = mptcp_poll,
3441 .ioctl = inet6_ioctl,
3442 .gettstamp = sock_gettstamp,
3443 .listen = mptcp_listen,
3444 .shutdown = mptcp_shutdown,
3445 .setsockopt = sock_common_setsockopt,
3446 .getsockopt = sock_common_getsockopt,
3447 .sendmsg = inet6_sendmsg,
3448 .recvmsg = inet6_recvmsg,
3449 .mmap = sock_no_mmap,
3450 .sendpage = inet_sendpage,
3451 #ifdef CONFIG_COMPAT
3452 .compat_ioctl = inet6_compat_ioctl,
3453 #endif
3454 };
3455
3456 static struct proto mptcp_v6_prot;
3457
3458 static void mptcp_v6_destroy(struct sock *sk)
3459 {
3460 mptcp_destroy(sk);
3461 inet6_destroy_sock(sk);
3462 }
3463
3464 static struct inet_protosw mptcp_v6_protosw = {
3465 .type = SOCK_STREAM,
3466 .protocol = IPPROTO_MPTCP,
3467 .prot = &mptcp_v6_prot,
3468 .ops = &mptcp_v6_stream_ops,
3469 .flags = INET_PROTOSW_ICSK,
3470 };
3471
3472 int __init mptcp_proto_v6_init(void)
3473 {
3474 int err;
3475
3476 mptcp_v6_prot = mptcp_prot;
3477 strcpy(mptcp_v6_prot.name, "MPTCPv6");
3478 mptcp_v6_prot.slab = NULL;
3479 mptcp_v6_prot.destroy = mptcp_v6_destroy;
3480 mptcp_v6_prot.obj_size = sizeof(struct mptcp6_sock);
3481
3482 err = proto_register(&mptcp_v6_prot, 1);
3483 if (err)
3484 return err;
3485
3486 err = inet6_register_protosw(&mptcp_v6_protosw);
3487 if (err)
3488 proto_unregister(&mptcp_v6_prot);
3489
3490 return err;
3491 }
3492 #endif
Linux with added FPC-III support