projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
mptcp: deliver ssk errors to msk
[linux-block.git] / net / mptcp / protocol.c
CommitLineData
f870fa0b
MM		 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>
7a6a6cbc
PA		 12#include <linux/sched/signal.h>
13#include <linux/atomic.h>
f870fa0b
MM		 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>
3721b9b6 19#include <net/tcp_states.h>
cf7da0d6
PK		 20#if IS_ENABLED(CONFIG_MPTCP_IPV6)
21#include <net/transp_v6.h>
22#endif
f870fa0b 23#include <net/mptcp.h>
e16163b6 24#include <net/xfrm.h>
f870fa0b 25#include "protocol.h"
fc518953 26#include "mib.h"
f870fa0b 27
b0519de8
FW		 28#if IS_ENABLED(CONFIG_MPTCP_IPV6)
29struct mptcp6_sock {
30 struct mptcp_sock msk;
31 struct ipv6_pinfo np;
32};
33#endif
34
6771bfd9 35struct mptcp_skb_cb {
ab174ad8
PA		 36 u64 map_seq;
37 u64 end_seq;
6771bfd9
FW		 38 u32 offset;
39};
40
41#define MPTCP_SKB_CB(__skb) ((struct mptcp_skb_cb *)&((__skb)->cb[0]))
42
d027236c
PA		 43static struct percpu_counter mptcp_sockets_allocated;
44
e16163b6 45static void __mptcp_destroy_sock(struct sock *sk);
d9ca1de8 46static void __mptcp_check_send_data_fin(struct sock *sk);
e16163b6 47
2303f994
PK		 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 */
52static struct socket *__mptcp_nmpc_socket(const struct mptcp_sock *msk)
53{
d22f4988 54 if (!msk->subflow || READ_ONCE(msk->can_ack))
2303f994
PK		 55 return NULL;
56
57 return msk->subflow;
58}
59
6f8a612a
FW		 60/* Returns end sequence number of the receiver's advertised window */
61static u64 mptcp_wnd_end(const struct mptcp_sock *msk)
62{
7439d687 63 return READ_ONCE(msk->wnd_end);
6f8a612a
FW		 64}
65
d2f77c53 66static bool mptcp_is_tcpsk(struct sock *sk)
0b4f33de
FW		 67{
68 struct socket *sock = sk->sk_socket;
69
0b4f33de
FW		 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;
d2f77c53 79 return true;
0b4f33de
FW		 80#if IS_ENABLED(CONFIG_MPTCP_IPV6)
81 } else if (unlikely(sk->sk_prot == &tcpv6_prot)) {
82 sock->ops = &inet6_stream_ops;
d2f77c53 83 return true;
0b4f33de
FW		 84#endif
85 }
86
d2f77c53 87 return false;
0b4f33de
FW		 88}
89
76660afb 90static struct sock *__mptcp_tcp_fallback(struct mptcp_sock *msk)
cec37a6e 91{
cec37a6e
PK		 92 sock_owned_by_me((const struct sock *)msk);
93
e1ff9e82 94 if (likely(!__mptcp_check_fallback(msk)))
cec37a6e
PK		 95 return NULL;
96
76660afb 97 return msk->first;
cec37a6e
PK		 98}
99
fa68018d 100static int __mptcp_socket_create(struct mptcp_sock *msk)
2303f994
PK		 101{
102 struct mptcp_subflow_context *subflow;
103 struct sock *sk = (struct sock *)msk;
104 struct socket *ssock;
105 int err;
106
2303f994
PK		 107 err = mptcp_subflow_create_socket(sk, &ssock);
108 if (err)
fa68018d 109 return err;
2303f994 110
8ab183de 111 msk->first = ssock->sk;
2303f994
PK		 112 msk->subflow = ssock;
113 subflow = mptcp_subflow_ctx(ssock->sk);
cec37a6e 114 list_add(&subflow->node, &msk->conn_list);
e16163b6 115 sock_hold(ssock->sk);
2303f994
PK		 116 subflow->request_mptcp = 1;
117
e1ff9e82
DC		 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
fa68018d 123 return 0;
2303f994
PK		 124}
125
ab174ad8
PA		 126static void mptcp_drop(struct sock *sk, struct sk_buff *skb)
127{
128 sk_drops_add(sk, skb);
129 __kfree_skb(skb);
130}
131
8268ed4c
PA		 132static 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
06242e44
PA		 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);
ab174ad8 145 MPTCP_SKB_CB(to)->end_seq = MPTCP_SKB_CB(from)->end_seq;
8268ed4c
PA		 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
ab174ad8
PA		 152static 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 */
165static 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;
fa3fe2b1 174 max_seq = READ_ONCE(msk->rcv_wnd_sent);
ab174ad8 175
06242e44
PA		 176 pr_debug("msk=%p seq=%llx limit=%llx empty=%d", msk, seq, max_seq,
177 RB_EMPTY_ROOT(&msk->out_of_order_queue));
fa3fe2b1 178 if (after64(end_seq, max_seq)) {
ab174ad8
PA		 179 /* out of window */
180 mptcp_drop(sk, skb);
fa3fe2b1
FW		 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);
06242e44 184 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_NODSSWINDOW);
ab174ad8
PA		 185 return;
186 }
187
188 p = &msk->out_of_order_queue.rb_node;
06242e44 189 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOQUEUE);
ab174ad8
PA		 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 */
06242e44
PA		 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);
ab174ad8 203 return;
06242e44 204 }
ab174ad8
PA		 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)) {
06242e44 208 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOQUEUETAIL);
ab174ad8
PA		 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);
06242e44 227 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
ab174ad8
PA		 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);
06242e44 243 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
ab174ad8
PA		 244 goto merge_right;
245 }
246 } else if (mptcp_ooo_try_coalesce(msk, skb1, skb)) {
06242e44 247 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOMERGE);
ab174ad8
PA		 248 return;
249 }
250 p = &parent->rb_right;
251 }
06242e44 252
ab174ad8
PA		 253insert:
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
258merge_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);
06242e44 265 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
ab174ad8
PA		 266 }
267 /* If there is no skb after us, we are the last_skb ! */
268 if (!skb1)
269 msk->ooo_last_skb = skb;
270
271end:
272 skb_condense(skb);
273 skb_set_owner_r(skb, sk);
274}
275
276static bool __mptcp_move_skb(struct mptcp_sock *msk, struct sock *ssk,
277 struct sk_buff *skb, unsigned int offset,
278 size_t copy_len)
6771bfd9 279{
ab174ad8 280 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
6771bfd9 281 struct sock *sk = (struct sock *)msk;
4e637c70 282 struct sk_buff *tail;
6771bfd9
FW		 283
284 __skb_unlink(skb, &ssk->sk_receive_queue);
6771bfd9 285
4e637c70
FW		 286 skb_ext_reset(skb);
287 skb_orphan(skb);
ab174ad8 288
9c3f94e1
PA		 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
ab174ad8
PA		 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;
8268ed4c 304 MPTCP_SKB_CB(skb)->offset = offset;
4e637c70 305
ab174ad8
PA		 306 if (MPTCP_SKB_CB(skb)->map_seq == msk->ack_seq) {
307 /* in sequence */
8b0308fe 308 WRITE_ONCE(msk->ack_seq, msk->ack_seq + copy_len);
ab174ad8
PA		 309 tail = skb_peek_tail(&sk->sk_receive_queue);
310 if (tail && mptcp_try_coalesce(sk, tail, skb))
311 return true;
4e637c70 312
ab174ad8
PA		 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 */
06242e44 324 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
9c3f94e1 325drop:
ab174ad8
PA		 326 mptcp_drop(sk, skb);
327 return false;
6771bfd9
FW		 328}
329
16a9a9da
MM		 330static 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
e16163b6
PA		 338static 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
6e628cd3 351static bool mptcp_pending_data_fin_ack(struct sock *sk)
16a9a9da
MM		 352{
353 struct mptcp_sock *msk = mptcp_sk(sk);
354
6e628cd3
PA		 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
361static void mptcp_check_data_fin_ack(struct sock *sk)
362{
363 struct mptcp_sock *msk = mptcp_sk(sk);
16a9a9da
MM		 364
365 /* Look for an acknowledged DATA_FIN */
6e628cd3 366 if (mptcp_pending_data_fin_ack(sk)) {
16a9a9da
MM		 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);
16a9a9da
MM		 374 break;
375 case TCP_CLOSING:
16a9a9da
MM		 376 case TCP_LAST_ACK:
377 inet_sk_state_store(sk, TCP_CLOSE);
16a9a9da
MM		 378 break;
379 }
380
e16163b6 381 mptcp_close_wake_up(sk);
16a9a9da
MM		 382 }
383}
384
3721b9b6
MM		 385static 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
405static 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
ea4ca586
PA		 415static 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
fd897679
PA		 427static bool tcp_can_send_ack(const struct sock *ssk)
428{
429 return !((1 << inet_sk_state_load(ssk)) &
20bc80b6 430 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_TIME_WAIT | TCPF_CLOSE | TCPF_LISTEN));
fd897679
PA		 431}
432
433static void mptcp_send_ack(struct mptcp_sock *msk)
7ed90803
PA		 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
fd897679
PA		 440 lock_sock(ssk);
441 if (tcp_can_send_ack(ssk))
ea4ca586 442 tcp_send_ack(ssk);
fd897679 443 release_sock(ssk);
ea4ca586 444 }
fd897679
PA		 445}
446
447static 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
459static void mptcp_cleanup_rbuf(struct mptcp_sock *msk)
460{
87952603 461 struct sock *ack_hint = READ_ONCE(msk->ack_hint);
fd897679
PA		 462 struct mptcp_subflow_context *subflow;
463
464 /* if the hinted ssk is still active, try to use it */
87952603 465 if (likely(ack_hint)) {
fd897679
PA		 466 mptcp_for_each_subflow(msk, subflow) {
467 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
468
87952603 469 if (ack_hint == ssk && mptcp_subflow_cleanup_rbuf(ssk))
fd897679
PA		 470 return;
471 }
7ed90803 472 }
fd897679
PA		 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;
7ed90803
PA		 478}
479
480static bool mptcp_check_data_fin(struct sock *sk)
3721b9b6
MM		 481{
482 struct mptcp_sock *msk = mptcp_sk(sk);
483 u64 rcv_data_fin_seq;
7ed90803 484 bool ret = false;
3721b9b6
MM		 485
486 if (__mptcp_check_fallback(msk) || !msk->first)
7ed90803 487 return ret;
3721b9b6
MM		 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)) {
917944da 503 WRITE_ONCE(msk->ack_seq, msk->ack_seq + 1);
3721b9b6
MM		 504 WRITE_ONCE(msk->rcv_data_fin, 0);
505
506 sk->sk_shutdown |= RCV_SHUTDOWN;
16a9a9da
MM		 507 smp_mb__before_atomic(); /* SHUTDOWN must be visible first */
508 set_bit(MPTCP_DATA_READY, &msk->flags);
3721b9b6
MM		 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);
3721b9b6
MM		 519 break;
520 default:
521 /* Other states not expected */
522 WARN_ON_ONCE(1);
523 break;
524 }
525
7ed90803 526 ret = true;
3721b9b6 527 mptcp_set_timeout(sk, NULL);
fd897679 528 mptcp_send_ack(msk);
e16163b6 529 mptcp_close_wake_up(sk);
3721b9b6 530 }
7ed90803 531 return ret;
3721b9b6
MM		 532}
533
6771bfd9
FW		 534static 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);
600911ff 539 struct sock *sk = (struct sock *)msk;
6771bfd9
FW		 540 unsigned int moved = 0;
541 bool more_data_avail;
542 struct tcp_sock *tp;
543 bool done = false;
13c7ba0c
FW		 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 }
600911ff 556
ab174ad8 557 pr_debug("msk=%p ssk=%p", msk, ssk);
6771bfd9
FW		 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);
d9fb8c50
PA		 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;
6771bfd9 577 break;
d9fb8c50 578 }
6771bfd9 579
e1ff9e82
DC		 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
6771bfd9
FW		 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
ab174ad8
PA		 602 if (__mptcp_move_skb(msk, ssk, skb, offset, len))
603 moved += len;
6771bfd9 604 seq += len;
6771bfd9
FW		 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);
600911ff 616
13c7ba0c 617 if (atomic_read(&sk->sk_rmem_alloc) > sk_rbuf) {
600911ff
FW		 618 done = true;
619 break;
620 }
6771bfd9 621 } while (more_data_avail);
87952603 622 WRITE_ONCE(msk->ack_hint, ssk);
6771bfd9 623
6719331c 624 *bytes += moved;
6771bfd9
FW		 625 return done;
626}
627
87952603 628static bool __mptcp_ofo_queue(struct mptcp_sock *msk)
ab174ad8
PA		 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);
06242e44 637 pr_debug("msk=%p empty=%d", msk, RB_EMPTY_ROOT(&msk->out_of_order_queue));
ab174ad8
PA		 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);
06242e44 649 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
ab174ad8
PA		 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 */
06242e44
PA		 659 pr_debug("uncoalesced seq=%llx ack seq=%llx delta=%d",
660 MPTCP_SKB_CB(skb)->map_seq, msk->ack_seq,
661 delta);
ab174ad8
PA		 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
2e52213c
FW		 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 */
87952603 674static void move_skbs_to_msk(struct mptcp_sock *msk, struct sock *ssk)
2e52213c
FW		 675{
676 struct sock *sk = (struct sock *)msk;
677 unsigned int moved = 0;
678
87952603
PA		 679 if (inet_sk_state_load(sk) == TCP_CLOSE)
680 return;
ab174ad8 681
87952603 682 mptcp_data_lock(sk);
2e52213c 683
87952603
PA		 684 __mptcp_move_skbs_from_subflow(msk, ssk, &moved);
685 __mptcp_ofo_queue(msk);
2e52213c 686
87952603
PA		 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);
2e52213c
FW		 695}
696
697void mptcp_data_ready(struct sock *sk, struct sock *ssk)
101f6f85 698{
6719331c 699 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
101f6f85 700 struct mptcp_sock *msk = mptcp_sk(sk);
13c7ba0c 701 int sk_rbuf, ssk_rbuf;
6719331c 702 bool wake;
101f6f85 703
d7b1bfd0
PA		 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
6719331c
PA		 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);
6771bfd9 718
13c7ba0c
FW		 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)
2e52213c
FW		 726 goto wake;
727
ea4ca586 728 move_skbs_to_msk(msk, ssk);
600911ff 729
600911ff 730wake:
6719331c
PA		 731 if (wake)
732 sk->sk_data_ready(sk);
101f6f85
FW		 733}
734
84dfe367 735void __mptcp_flush_join_list(struct mptcp_sock *msk)
ec3edaa7
PK		 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
b51f9b80
PA		 745static bool mptcp_timer_pending(struct sock *sk)
746{
747 return timer_pending(&inet_csk(sk)->icsk_retransmit_timer);
748}
749
750static void mptcp_reset_timer(struct sock *sk)
751{
752 struct inet_connection_sock *icsk = inet_csk(sk);
753 unsigned long tout;
754
e16163b6
PA		 755 /* prevent rescheduling on close */
756 if (unlikely(inet_sk_state_load(sk) == TCP_CLOSE))
757 return;
758
b51f9b80
PA		 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
ba8f48f7
PA		 766bool 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
59832e24
FW		 779void mptcp_subflow_eof(struct sock *sk)
780{
ba8f48f7
PA		 781 if (!test_and_set_bit(MPTCP_WORK_EOF, &mptcp_sk(sk)->flags))
782 mptcp_schedule_work(sk);
59832e24
FW		 783}
784
5969856a
PA		 785static 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;
e16163b6
PA		 793 if (receivers)
794 return;
5969856a 795
e16163b6 796 if (!(sk->sk_shutdown & RCV_SHUTDOWN)) {
5969856a
PA		 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 }
e16163b6
PA		 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:
26aa2314
PA		 812 inet_sk_state_store(sk, TCP_CLOSING);
813 break;
814 case TCP_FIN_WAIT2:
e16163b6
PA		 815 inet_sk_state_store(sk, TCP_CLOSE);
816 break;
817 default:
818 return;
819 }
820 mptcp_close_wake_up(sk);
5969856a
PA		 821}
822
7a6a6cbc
PA		 823static 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
3f8e0aae
PA		 838static bool mptcp_skb_can_collapse_to(u64 write_seq,
839 const struct sk_buff *skb,
840 const struct mptcp_ext *mpext)
57040755
PA		 841{
842 if (!tcp_skb_can_collapse_to(skb))
843 return false;
844
5a369ca6
PA		 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;
57040755
PA		 850}
851
18b683bf
PA		 852static 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 &&
d9ca1de8 857 pfrag->size - pfrag->offset > 0 &&
18b683bf
PA		 858 df->data_seq + df->data_len == msk->write_seq;
859}
860
724cfd2e 861static int mptcp_wmem_with_overhead(struct sock *sk, int size)
e93da928 862{
724cfd2e
PA		 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);
e93da928
PA		 872}
873
874static void __mptcp_wmem_reserve(struct sock *sk, int size)
875{
724cfd2e 876 int amount = mptcp_wmem_with_overhead(sk, size);
e93da928
PA		 877 struct mptcp_sock *msk = mptcp_sk(sk);
878
879 WARN_ON_ONCE(msk->wmem_reserved);
e7579d5d
DC		 880 if (WARN_ON_ONCE(amount < 0))
881 amount = 0;
882
e93da928
PA		 883 if (amount <= sk->sk_forward_alloc)
884 goto reserve;
885
886 /* under memory pressure try to reserve at most a single page
887 * otherwise try to reserve the full estimate and fallback
888 * to a single page before entering the error path
889 */
890 if ((tcp_under_memory_pressure(sk) && amount > PAGE_SIZE) ||
891 !sk_wmem_schedule(sk, amount)) {
892 if (amount <= PAGE_SIZE)
893 goto nomem;
894
895 amount = PAGE_SIZE;
896 if (!sk_wmem_schedule(sk, amount))
897 goto nomem;
898 }
899
900reserve:
901 msk->wmem_reserved = amount;
902 sk->sk_forward_alloc -= amount;
903 return;
904
905nomem:
906 /* we will wait for memory on next allocation */
907 msk->wmem_reserved = -1;
908}
909
910static void __mptcp_update_wmem(struct sock *sk)
911{
912 struct mptcp_sock *msk = mptcp_sk(sk);
913
914 if (!msk->wmem_reserved)
915 return;
916
917 if (msk->wmem_reserved < 0)
918 msk->wmem_reserved = 0;
919 if (msk->wmem_reserved > 0) {
920 sk->sk_forward_alloc += msk->wmem_reserved;
921 msk->wmem_reserved = 0;
922 }
923}
924
925static bool mptcp_wmem_alloc(struct sock *sk, int size)
926{
927 struct mptcp_sock *msk = mptcp_sk(sk);
928
929 /* check for pre-existing error condition */
930 if (msk->wmem_reserved < 0)
931 return false;
932
933 if (msk->wmem_reserved >= size)
934 goto account;
935
87952603
PA		 936 mptcp_data_lock(sk);
937 if (!sk_wmem_schedule(sk, size)) {
938 mptcp_data_unlock(sk);
e93da928 939 return false;
87952603 940 }
e93da928
PA		 941
942 sk->sk_forward_alloc -= size;
943 msk->wmem_reserved += size;
87952603 944 mptcp_data_unlock(sk);
e93da928
PA		 945
946account:
947 msk->wmem_reserved -= size;
948 return true;
949}
950
87952603
PA		 951static void mptcp_wmem_uncharge(struct sock *sk, int size)
952{
953 struct mptcp_sock *msk = mptcp_sk(sk);
954
955 if (msk->wmem_reserved < 0)
956 msk->wmem_reserved = 0;
957 msk->wmem_reserved += size;
958}
959
724cfd2e
PA		 960static void mptcp_mem_reclaim_partial(struct sock *sk)
961{
962 struct mptcp_sock *msk = mptcp_sk(sk);
963
964 /* if we are experiencing a transint allocation error,
965 * the forward allocation memory has been already
966 * released
967 */
968 if (msk->wmem_reserved < 0)
969 return;
970
971 mptcp_data_lock(sk);
972 sk->sk_forward_alloc += msk->wmem_reserved;
973 sk_mem_reclaim_partial(sk);
974 msk->wmem_reserved = sk->sk_forward_alloc;
975 sk->sk_forward_alloc = 0;
976 mptcp_data_unlock(sk);
977}
978
d027236c
PA		 979static void dfrag_uncharge(struct sock *sk, int len)
980{
981 sk_mem_uncharge(sk, len);
7948f6cc 982 sk_wmem_queued_add(sk, -len);
d027236c
PA		 983}
984
985static void dfrag_clear(struct sock *sk, struct mptcp_data_frag *dfrag)
18b683bf 986{
d027236c
PA		 987 int len = dfrag->data_len + dfrag->overhead;
988
18b683bf 989 list_del(&dfrag->list);
d027236c 990 dfrag_uncharge(sk, len);
18b683bf
PA		 991 put_page(dfrag->page);
992}
993
6e628cd3 994static void __mptcp_clean_una(struct sock *sk)
18b683bf
PA		 995{
996 struct mptcp_sock *msk = mptcp_sk(sk);
997 struct mptcp_data_frag *dtmp, *dfrag;
d027236c 998 bool cleaned = false;
e1ff9e82
DC		 999 u64 snd_una;
1000
1001 /* on fallback we just need to ignore snd_una, as this is really
1002 * plain TCP
1003 */
1004 if (__mptcp_check_fallback(msk))
7439d687 1005 msk->snd_una = READ_ONCE(msk->snd_nxt);
6f8a612a 1006
7439d687 1007 snd_una = msk->snd_una;
18b683bf
PA		 1008 list_for_each_entry_safe(dfrag, dtmp, &msk->rtx_queue, list) {
1009 if (after64(dfrag->data_seq + dfrag->data_len, snd_una))
1010 break;
1011
d9ca1de8
PA		 1012 if (WARN_ON_ONCE(dfrag == msk->first_pending))
1013 break;
d027236c
PA		 1014 dfrag_clear(sk, dfrag);
1015 cleaned = true;
1016 }
1017
7948f6cc
FW		 1018 dfrag = mptcp_rtx_head(sk);
1019 if (dfrag && after64(snd_una, dfrag->data_seq)) {
53eb4c38
PA		 1020 u64 delta = snd_una - dfrag->data_seq;
1021
d9ca1de8 1022 if (WARN_ON_ONCE(delta > dfrag->already_sent))
53eb4c38 1023 goto out;
7948f6cc
FW		 1024
1025 dfrag->data_seq += delta;
53eb4c38 1026 dfrag->offset += delta;
7948f6cc 1027 dfrag->data_len -= delta;
d9ca1de8 1028 dfrag->already_sent -= delta;
7948f6cc
FW		 1029
1030 dfrag_uncharge(sk, delta);
1031 cleaned = true;
1032 }
1033
53eb4c38 1034out:
6e628cd3
PA		 1035 if (cleaned) {
1036 if (tcp_under_memory_pressure(sk)) {
1037 __mptcp_update_wmem(sk);
1038 sk_mem_reclaim_partial(sk);
1039 }
63561a40 1040
6e628cd3
PA		 1041 if (sk_stream_is_writeable(sk)) {
1042 /* pairs with memory barrier in mptcp_poll */
1043 smp_mb();
1044 if (test_and_clear_bit(MPTCP_NOSPACE, &msk->flags))
1045 sk_stream_write_space(sk);
1046 }
1047 }
95ed690e 1048
6e628cd3
PA		 1049 if (snd_una == READ_ONCE(msk->snd_nxt)) {
1050 if (msk->timer_ival)
1051 mptcp_stop_timer(sk);
1052 } else {
1053 mptcp_reset_timer(sk);
18b683bf
PA		 1054 }
1055}
1056
724cfd2e 1057static void mptcp_enter_memory_pressure(struct sock *sk)
18b683bf 1058{
d9ca1de8
PA		 1059 struct mptcp_subflow_context *subflow;
1060 struct mptcp_sock *msk = mptcp_sk(sk);
1061 bool first = true;
1062
18b683bf 1063 sk_stream_moderate_sndbuf(sk);
d9ca1de8
PA		 1064 mptcp_for_each_subflow(msk, subflow) {
1065 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1066
1067 if (first)
1068 tcp_enter_memory_pressure(ssk);
1069 sk_stream_moderate_sndbuf(ssk);
1070 first = false;
1071 }
724cfd2e
PA		 1072}
1073
1074/* ensure we get enough memory for the frag hdr, beyond some minimal amount of
1075 * data
1076 */
1077static bool mptcp_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1078{
1079 if (likely(skb_page_frag_refill(32U + sizeof(struct mptcp_data_frag),
1080 pfrag, sk->sk_allocation)))
1081 return true;
1082
1083 mptcp_enter_memory_pressure(sk);
18b683bf
PA		 1084 return false;
1085}
1086
1087static struct mptcp_data_frag *
1088mptcp_carve_data_frag(const struct mptcp_sock *msk, struct page_frag *pfrag,
1089 int orig_offset)
1090{
1091 int offset = ALIGN(orig_offset, sizeof(long));
1092 struct mptcp_data_frag *dfrag;
1093
1094 dfrag = (struct mptcp_data_frag *)(page_to_virt(pfrag->page) + offset);
1095 dfrag->data_len = 0;
1096 dfrag->data_seq = msk->write_seq;
1097 dfrag->overhead = offset - orig_offset + sizeof(struct mptcp_data_frag);
1098 dfrag->offset = offset + sizeof(struct mptcp_data_frag);
d9ca1de8 1099 dfrag->already_sent = 0;
18b683bf
PA		 1100 dfrag->page = pfrag->page;
1101
1102 return dfrag;
1103}
1104
caf971df
PA		 1105struct mptcp_sendmsg_info {
1106 int mss_now;
1107 int size_goal;
d9ca1de8
PA		 1108 u16 limit;
1109 u16 sent;
1110 unsigned int flags;
caf971df
PA		 1111};
1112
6f8a612a
FW		 1113static int mptcp_check_allowed_size(struct mptcp_sock *msk, u64 data_seq,
1114 int avail_size)
1115{
1116 u64 window_end = mptcp_wnd_end(msk);
1117
1118 if (__mptcp_check_fallback(msk))
1119 return avail_size;
1120
1121 if (!before64(data_seq + avail_size, window_end)) {
1122 u64 allowed_size = window_end - data_seq;
1123
1124 return min_t(unsigned int, allowed_size, avail_size);
1125 }
1126
1127 return avail_size;
1128}
1129
724cfd2e
PA		 1130static bool __mptcp_add_ext(struct sk_buff *skb, gfp_t gfp)
1131{
1132 struct skb_ext *mpext = __skb_ext_alloc(gfp);
1133
1134 if (!mpext)
1135 return false;
1136 __skb_ext_set(skb, SKB_EXT_MPTCP, mpext);
1137 return true;
1138}
1139
6e628cd3 1140static struct sk_buff *__mptcp_do_alloc_tx_skb(struct sock *sk, gfp_t gfp)
724cfd2e
PA		 1141{
1142 struct sk_buff *skb;
1143
6e628cd3 1144 skb = alloc_skb_fclone(MAX_TCP_HEADER, gfp);
724cfd2e 1145 if (likely(skb)) {
6e628cd3 1146 if (likely(__mptcp_add_ext(skb, gfp))) {
724cfd2e
PA		 1147 skb_reserve(skb, MAX_TCP_HEADER);
1148 skb->reserved_tailroom = skb->end - skb->tail;
1149 return skb;
1150 }
1151 __kfree_skb(skb);
1152 } else {
1153 mptcp_enter_memory_pressure(sk);
1154 }
1155 return NULL;
1156}
1157
1158static bool mptcp_tx_cache_refill(struct sock *sk, int size,
1159 struct sk_buff_head *skbs, int *total_ts)
1160{
1161 struct mptcp_sock *msk = mptcp_sk(sk);
1162 struct sk_buff *skb;
1163 int space_needed;
1164
1165 if (unlikely(tcp_under_memory_pressure(sk))) {
1166 mptcp_mem_reclaim_partial(sk);
1167
1168 /* under pressure pre-allocate at most a single skb */
1169 if (msk->skb_tx_cache.qlen)
1170 return true;
1171 space_needed = msk->size_goal_cache;
1172 } else {
1173 space_needed = msk->tx_pending_data + size -
1174 msk->skb_tx_cache.qlen * msk->size_goal_cache;
1175 }
1176
1177 while (space_needed > 0) {
6e628cd3 1178 skb = __mptcp_do_alloc_tx_skb(sk, sk->sk_allocation);
724cfd2e
PA		 1179 if (unlikely(!skb)) {
1180 /* under memory pressure, try to pass the caller a
1181 * single skb to allow forward progress
1182 */
1183 while (skbs->qlen > 1) {
1184 skb = __skb_dequeue_tail(skbs);
1185 __kfree_skb(skb);
1186 }
1187 return skbs->qlen > 0;
1188 }
1189
1190 *total_ts += skb->truesize;
1191 __skb_queue_tail(skbs, skb);
1192 space_needed -= msk->size_goal_cache;
1193 }
1194 return true;
1195}
1196
6e628cd3 1197static bool __mptcp_alloc_tx_skb(struct sock *sk, struct sock *ssk, gfp_t gfp)
724cfd2e
PA		 1198{
1199 struct mptcp_sock *msk = mptcp_sk(sk);
1200 struct sk_buff *skb;
1201
1202 if (ssk->sk_tx_skb_cache) {
1203 skb = ssk->sk_tx_skb_cache;
1204 if (unlikely(!skb_ext_find(skb, SKB_EXT_MPTCP) &&
6e628cd3 1205 !__mptcp_add_ext(skb, gfp)))
724cfd2e
PA		 1206 return false;
1207 return true;
1208 }
1209
1210 skb = skb_peek(&msk->skb_tx_cache);
1211 if (skb) {
1212 if (likely(sk_wmem_schedule(ssk, skb->truesize))) {
1213 skb = __skb_dequeue(&msk->skb_tx_cache);
1214 if (WARN_ON_ONCE(!skb))
1215 return false;
1216
1217 mptcp_wmem_uncharge(sk, skb->truesize);
1218 ssk->sk_tx_skb_cache = skb;
1219 return true;
1220 }
1221
1222 /* over memory limit, no point to try to allocate a new skb */
1223 return false;
1224 }
1225
6e628cd3 1226 skb = __mptcp_do_alloc_tx_skb(sk, gfp);
724cfd2e
PA		 1227 if (!skb)
1228 return false;
1229
1230 if (likely(sk_wmem_schedule(ssk, skb->truesize))) {
1231 ssk->sk_tx_skb_cache = skb;
1232 return true;
1233 }
1234 kfree_skb(skb);
1235 return false;
1236}
1237
1238static bool mptcp_must_reclaim_memory(struct sock *sk, struct sock *ssk)
1239{
1240 return !ssk->sk_tx_skb_cache &&
1241 !skb_peek(&mptcp_sk(sk)->skb_tx_cache) &&
1242 tcp_under_memory_pressure(sk);
1243}
1244
1245static bool mptcp_alloc_tx_skb(struct sock *sk, struct sock *ssk)
1246{
1247 if (unlikely(mptcp_must_reclaim_memory(sk, ssk)))
1248 mptcp_mem_reclaim_partial(sk);
6e628cd3 1249 return __mptcp_alloc_tx_skb(sk, ssk, sk->sk_allocation);
724cfd2e
PA		 1250}
1251
6d0060f6 1252static int mptcp_sendmsg_frag(struct sock *sk, struct sock *ssk,
d9ca1de8 1253 struct mptcp_data_frag *dfrag,
caf971df 1254 struct mptcp_sendmsg_info *info)
6d0060f6 1255{
d9ca1de8 1256 u64 data_seq = dfrag->data_seq + info->sent;
6d0060f6 1257 struct mptcp_sock *msk = mptcp_sk(sk);
6f8a612a 1258 bool zero_window_probe = false;
6d0060f6 1259 struct mptcp_ext *mpext = NULL;
57040755 1260 struct sk_buff *skb, *tail;
d9ca1de8 1261 bool can_collapse = false;
15e6ca97 1262 int size_bias = 0;
d9ca1de8 1263 int avail_size;
724cfd2e 1264 size_t ret = 0;
6d0060f6 1265
d9ca1de8
PA		 1266 pr_debug("msk=%p ssk=%p sending dfrag at seq=%lld len=%d already sent=%d",
1267 msk, ssk, dfrag->data_seq, dfrag->data_len, info->sent);
1268
1269 /* compute send limit */
1270 info->mss_now = tcp_send_mss(ssk, &info->size_goal, info->flags);
caf971df 1271 avail_size = info->size_goal;
724cfd2e 1272 msk->size_goal_cache = info->size_goal;
57040755
PA		 1273 skb = tcp_write_queue_tail(ssk);
1274 if (skb) {
57040755
PA		 1275 /* Limit the write to the size available in the
1276 * current skb, if any, so that we create at most a new skb.
1277 * Explicitly tells TCP internals to avoid collapsing on later
1278 * queue management operation, to avoid breaking the ext <->
1279 * SSN association set here
1280 */
d9ca1de8 1281 mpext = skb_ext_find(skb, SKB_EXT_MPTCP);
caf971df 1282 can_collapse = (info->size_goal - skb->len > 0) &&
d9ca1de8 1283 mptcp_skb_can_collapse_to(data_seq, skb, mpext);
15e6ca97 1284 if (!can_collapse) {
57040755 1285 TCP_SKB_CB(skb)->eor = 1;
15e6ca97
PA		 1286 } else {
1287 size_bias = skb->len;
caf971df 1288 avail_size = info->size_goal - skb->len;
15e6ca97 1289 }
57040755 1290 }
18b683bf 1291
6f8a612a
FW		 1292 /* Zero window and all data acked? Probe. */
1293 avail_size = mptcp_check_allowed_size(msk, data_seq, avail_size);
1294 if (avail_size == 0) {
7439d687
PA		 1295 u64 snd_una = READ_ONCE(msk->snd_una);
1296
1297 if (skb || snd_una != msk->snd_nxt)
6f8a612a
FW		 1298 return 0;
1299 zero_window_probe = true;
7439d687 1300 data_seq = snd_una - 1;
6f8a612a
FW		 1301 avail_size = 1;
1302 }
1303
d9ca1de8
PA		 1304 if (WARN_ON_ONCE(info->sent > info->limit ||
1305 info->limit > dfrag->data_len))
1306 return 0;
d027236c 1307
d9ca1de8 1308 ret = info->limit - info->sent;
15e6ca97
PA		 1309 tail = tcp_build_frag(ssk, avail_size + size_bias, info->flags,
1310 dfrag->page, dfrag->offset + info->sent, &ret);
e2223995
PA		 1311 if (!tail) {
1312 tcp_remove_empty_skb(sk, tcp_write_queue_tail(ssk));
1313 return -ENOMEM;
35759383 1314 }
18b683bf 1315
e2223995 1316 /* if the tail skb is still the cached one, collapsing really happened.
57040755 1317 */
e2223995 1318 if (skb == tail) {
15e6ca97 1319 TCP_SKB_CB(tail)->tcp_flags &= ~TCPHDR_PSH;
57040755 1320 mpext->data_len += ret;
15e6ca97 1321 WARN_ON_ONCE(!can_collapse);
6f8a612a 1322 WARN_ON_ONCE(zero_window_probe);
57040755
PA		 1323 goto out;
1324 }
1325
724cfd2e
PA		 1326 mpext = skb_ext_find(tail, SKB_EXT_MPTCP);
1327 if (WARN_ON_ONCE(!mpext)) {
1328 /* should never reach here, stream corrupted */
1329 return -EINVAL;
1330 }
6d0060f6
MM		 1331
1332 memset(mpext, 0, sizeof(*mpext));
d9ca1de8 1333 mpext->data_seq = data_seq;
6d0060f6
MM		 1334 mpext->subflow_seq = mptcp_subflow_ctx(ssk)->rel_write_seq;
1335 mpext->data_len = ret;
1336 mpext->use_map = 1;
1337 mpext->dsn64 = 1;
1338
1339 pr_debug("data_seq=%llu subflow_seq=%u data_len=%u dsn64=%d",
1340 mpext->data_seq, mpext->subflow_seq, mpext->data_len,
1341 mpext->dsn64);
1342
6f8a612a
FW		 1343 if (zero_window_probe) {
1344 mptcp_subflow_ctx(ssk)->rel_write_seq += ret;
1345 mpext->frozen = 1;
1346 ret = 0;
1347 tcp_push_pending_frames(ssk);
1348 }
57040755 1349out:
6d0060f6 1350 mptcp_subflow_ctx(ssk)->rel_write_seq += ret;
6d0060f6
MM		 1351 return ret;
1352}
1353
d5f49190
PA		 1354#define MPTCP_SEND_BURST_SIZE ((1 << 16) - \
1355 sizeof(struct tcphdr) - \
1356 MAX_TCP_OPTION_SPACE - \
1357 sizeof(struct ipv6hdr) - \
1358 sizeof(struct frag_hdr))
1359
1360struct subflow_send_info {
1361 struct sock *ssk;
1362 u64 ratio;
1363};
1364
da51aef5
PA		 1365static struct sock *mptcp_subflow_get_send(struct mptcp_sock *msk,
1366 u32 *sndbuf)
f296234c 1367{
d5f49190 1368 struct subflow_send_info send_info[2];
f296234c 1369 struct mptcp_subflow_context *subflow;
d5f49190
PA		 1370 int i, nr_active = 0;
1371 struct sock *ssk;
1372 u64 ratio;
1373 u32 pace;
f296234c 1374
d5f49190 1375 sock_owned_by_me((struct sock *)msk);
f296234c 1376
da51aef5 1377 *sndbuf = 0;
d5f49190
PA		 1378 if (__mptcp_check_fallback(msk)) {
1379 if (!msk->first)
f296234c 1380 return NULL;
d5f49190
PA		 1381 *sndbuf = msk->first->sk_sndbuf;
1382 return sk_stream_memory_free(msk->first) ? msk->first : NULL;
1383 }
1384
1385 /* re-use last subflow, if the burst allow that */
1386 if (msk->last_snd && msk->snd_burst > 0 &&
1387 sk_stream_memory_free(msk->last_snd) &&
1388 mptcp_subflow_active(mptcp_subflow_ctx(msk->last_snd))) {
1389 mptcp_for_each_subflow(msk, subflow) {
1390 ssk = mptcp_subflow_tcp_sock(subflow);
1391 *sndbuf = max(tcp_sk(ssk)->snd_wnd, *sndbuf);
f296234c 1392 }
d5f49190
PA		 1393 return msk->last_snd;
1394 }
f296234c 1395
d5f49190
PA		 1396 /* pick the subflow with the lower wmem/wspace ratio */
1397 for (i = 0; i < 2; ++i) {
1398 send_info[i].ssk = NULL;
1399 send_info[i].ratio = -1;
1400 }
1401 mptcp_for_each_subflow(msk, subflow) {
1402 ssk = mptcp_subflow_tcp_sock(subflow);
1403 if (!mptcp_subflow_active(subflow))
1404 continue;
1405
1406 nr_active += !subflow->backup;
da51aef5 1407 *sndbuf = max(tcp_sk(ssk)->snd_wnd, *sndbuf);
d5f49190
PA		 1408 if (!sk_stream_memory_free(subflow->tcp_sock))
1409 continue;
f296234c 1410
d5f49190
PA		 1411 pace = READ_ONCE(ssk->sk_pacing_rate);
1412 if (!pace)
f296234c 1413 continue;
f296234c 1414
d5f49190
PA		 1415 ratio = div_u64((u64)READ_ONCE(ssk->sk_wmem_queued) << 32,
1416 pace);
1417 if (ratio < send_info[subflow->backup].ratio) {
1418 send_info[subflow->backup].ssk = ssk;
1419 send_info[subflow->backup].ratio = ratio;
1420 }
f296234c
PK		 1421 }
1422
d5f49190
PA		 1423 pr_debug("msk=%p nr_active=%d ssk=%p:%lld backup=%p:%lld",
1424 msk, nr_active, send_info[0].ssk, send_info[0].ratio,
1425 send_info[1].ssk, send_info[1].ratio);
1426
1427 /* pick the best backup if no other subflow is active */
1428 if (!nr_active)
1429 send_info[0].ssk = send_info[1].ssk;
1430
1431 if (send_info[0].ssk) {
1432 msk->last_snd = send_info[0].ssk;
1433 msk->snd_burst = min_t(int, MPTCP_SEND_BURST_SIZE,
1434 sk_stream_wspace(msk->last_snd));
1435 return msk->last_snd;
1436 }
1437 return NULL;
f296234c
PK		 1438}
1439
d9ca1de8
PA		 1440static void mptcp_push_release(struct sock *sk, struct sock *ssk,
1441 struct mptcp_sendmsg_info *info)
1442{
1443 mptcp_set_timeout(sk, ssk);
1444 tcp_push(ssk, 0, info->mss_now, tcp_sk(ssk)->nonagle, info->size_goal);
1445 release_sock(ssk);
1446}
1447
1448static void mptcp_push_pending(struct sock *sk, unsigned int flags)
f870fa0b 1449{
d9ca1de8 1450 struct sock *prev_ssk = NULL, *ssk = NULL;
f870fa0b 1451 struct mptcp_sock *msk = mptcp_sk(sk);
caf971df 1452 struct mptcp_sendmsg_info info = {
d9ca1de8 1453 .flags = flags,
caf971df 1454 };
d9ca1de8
PA		 1455 struct mptcp_data_frag *dfrag;
1456 int len, copied = 0;
1457 u32 sndbuf;
1458
1459 while ((dfrag = mptcp_send_head(sk))) {
1460 info.sent = dfrag->already_sent;
1461 info.limit = dfrag->data_len;
1462 len = dfrag->data_len - dfrag->already_sent;
1463 while (len > 0) {
1464 int ret = 0;
1465
1466 prev_ssk = ssk;
1467 __mptcp_flush_join_list(msk);
1468 ssk = mptcp_subflow_get_send(msk, &sndbuf);
1469
1470 /* do auto tuning */
1471 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK) &&
1472 sndbuf > READ_ONCE(sk->sk_sndbuf))
1473 WRITE_ONCE(sk->sk_sndbuf, sndbuf);
1474
1475 /* try to keep the subflow socket lock across
1476 * consecutive xmit on the same socket
1477 */
1478 if (ssk != prev_ssk && prev_ssk)
1479 mptcp_push_release(sk, prev_ssk, &info);
1480 if (!ssk)
1481 goto out;
1482
1483 if (ssk != prev_ssk || !prev_ssk)
1484 lock_sock(ssk);
1485
724cfd2e
PA		 1486 /* keep it simple and always provide a new skb for the
1487 * subflow, even if we will not use it when collapsing
1488 * on the pending one
1489 */
1490 if (!mptcp_alloc_tx_skb(sk, ssk)) {
1491 mptcp_push_release(sk, ssk, &info);
1492 goto out;
1493 }
1494
d9ca1de8
PA		 1495 ret = mptcp_sendmsg_frag(sk, ssk, dfrag, &info);
1496 if (ret <= 0) {
1497 mptcp_push_release(sk, ssk, &info);
1498 goto out;
1499 }
1500
1501 info.sent += ret;
1502 dfrag->already_sent += ret;
1503 msk->snd_nxt += ret;
1504 msk->snd_burst -= ret;
724cfd2e 1505 msk->tx_pending_data -= ret;
d9ca1de8
PA		 1506 copied += ret;
1507 len -= ret;
1508 }
1509 WRITE_ONCE(msk->first_pending, mptcp_send_next(sk));
1510 }
1511
1512 /* at this point we held the socket lock for the last subflow we used */
1513 if (ssk)
1514 mptcp_push_release(sk, ssk, &info);
1515
1516out:
b680a214
PA		 1517 if (copied) {
1518 /* start the timer, if it's not pending */
1519 if (!mptcp_timer_pending(sk))
1520 mptcp_reset_timer(sk);
d9ca1de8 1521 __mptcp_check_send_data_fin(sk);
b680a214 1522 }
d9ca1de8
PA		 1523}
1524
6e628cd3
PA		 1525static void __mptcp_subflow_push_pending(struct sock *sk, struct sock *ssk)
1526{
1527 struct mptcp_sock *msk = mptcp_sk(sk);
1528 struct mptcp_sendmsg_info info;
1529 struct mptcp_data_frag *dfrag;
1530 int len, copied = 0;
1531
1532 info.flags = 0;
1533 while ((dfrag = mptcp_send_head(sk))) {
1534 info.sent = dfrag->already_sent;
1535 info.limit = dfrag->data_len;
1536 len = dfrag->data_len - dfrag->already_sent;
1537 while (len > 0) {
1538 int ret = 0;
1539
1540 /* do auto tuning */
1541 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK) &&
1542 ssk->sk_sndbuf > READ_ONCE(sk->sk_sndbuf))
1543 WRITE_ONCE(sk->sk_sndbuf, ssk->sk_sndbuf);
1544
1545 if (unlikely(mptcp_must_reclaim_memory(sk, ssk))) {
1546 __mptcp_update_wmem(sk);
1547 sk_mem_reclaim_partial(sk);
1548 }
1549 if (!__mptcp_alloc_tx_skb(sk, ssk, GFP_ATOMIC))
1550 goto out;
1551
1552 ret = mptcp_sendmsg_frag(sk, ssk, dfrag, &info);
1553 if (ret <= 0)
1554 goto out;
1555
1556 info.sent += ret;
1557 dfrag->already_sent += ret;
1558 msk->snd_nxt += ret;
1559 msk->snd_burst -= ret;
1560 msk->tx_pending_data -= ret;
1561 copied += ret;
1562 len -= ret;
1563 }
1564 WRITE_ONCE(msk->first_pending, mptcp_send_next(sk));
1565 }
1566
1567out:
1568 /* __mptcp_alloc_tx_skb could have released some wmem and we are
1569 * not going to flush it via release_sock()
1570 */
1571 __mptcp_update_wmem(sk);
1572 if (copied) {
1573 mptcp_set_timeout(sk, ssk);
1574 tcp_push(ssk, 0, info.mss_now, tcp_sk(ssk)->nonagle,
1575 info.size_goal);
1576 if (msk->snd_data_fin_enable &&
1577 msk->snd_nxt + 1 == msk->write_seq)
1578 mptcp_schedule_work(sk);
1579 }
1580}
1581
d9ca1de8
PA		 1582static int mptcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1583{
1584 struct mptcp_sock *msk = mptcp_sk(sk);
17091708 1585 struct page_frag *pfrag;
6d0060f6 1586 size_t copied = 0;
caf971df 1587 int ret = 0;
6d0060f6 1588 long timeo;
f870fa0b
MM		 1589
1590 if (msg->msg_flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL))
1591 return -EOPNOTSUPP;
1592
e7579d5d 1593 mptcp_lock_sock(sk, __mptcp_wmem_reserve(sk, min_t(size_t, 1 << 20, len)));
1954b860
MM		 1594
1595 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1596
1597 if ((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) {
1598 ret = sk_stream_wait_connect(sk, &timeo);
1599 if (ret)
1600 goto out;
1601 }
1602
17091708 1603 pfrag = sk_page_frag(sk);
18b683bf 1604
d9ca1de8 1605 while (msg_data_left(msg)) {
724cfd2e 1606 int total_ts, frag_truesize = 0;
d9ca1de8 1607 struct mptcp_data_frag *dfrag;
724cfd2e 1608 struct sk_buff_head skbs;
d9ca1de8
PA		 1609 bool dfrag_collapsed;
1610 size_t psize, offset;
18b683bf 1611
d9ca1de8
PA		 1612 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN)) {
1613 ret = -EPIPE;
f296234c
PK		 1614 goto out;
1615 }
da51aef5 1616
d9ca1de8
PA		 1617 /* reuse tail pfrag, if possible, or carve a new one from the
1618 * page allocator
1619 */
1620 dfrag = mptcp_pending_tail(sk);
1621 dfrag_collapsed = mptcp_frag_can_collapse_to(msk, pfrag, dfrag);
1622 if (!dfrag_collapsed) {
6e628cd3
PA		 1623 if (!sk_stream_memory_free(sk))
1624 goto wait_for_memory;
1625
d9ca1de8
PA		 1626 if (!mptcp_page_frag_refill(sk, pfrag))
1627 goto wait_for_memory;
1628
1629 dfrag = mptcp_carve_data_frag(msk, pfrag, pfrag->offset);
1630 frag_truesize = dfrag->overhead;
72511aab 1631 }
6d0060f6 1632
d9ca1de8
PA		 1633 /* we do not bound vs wspace, to allow a single packet.
1634 * memory accounting will prevent execessive memory usage
1635 * anyway
d5f49190 1636 */
d9ca1de8
PA		 1637 offset = dfrag->offset + dfrag->data_len;
1638 psize = pfrag->size - offset;
1639 psize = min_t(size_t, psize, msg_data_left(msg));
724cfd2e
PA		 1640 total_ts = psize + frag_truesize;
1641 __skb_queue_head_init(&skbs);
1642 if (!mptcp_tx_cache_refill(sk, psize, &skbs, &total_ts))
d9ca1de8
PA		 1643 goto wait_for_memory;
1644
724cfd2e
PA		 1645 if (!mptcp_wmem_alloc(sk, total_ts)) {
1646 __skb_queue_purge(&skbs);
1647 goto wait_for_memory;
1648 }
1649
1650 skb_queue_splice_tail(&skbs, &msk->skb_tx_cache);
d9ca1de8
PA		 1651 if (copy_page_from_iter(dfrag->page, offset, psize,
1652 &msg->msg_iter) != psize) {
87952603 1653 mptcp_wmem_uncharge(sk, psize + frag_truesize);
d9ca1de8
PA		 1654 ret = -EFAULT;
1655 goto out;
72511aab
FW		 1656 }
1657
d9ca1de8
PA		 1658 /* data successfully copied into the write queue */
1659 copied += psize;
1660 dfrag->data_len += psize;
1661 frag_truesize += psize;
1662 pfrag->offset += frag_truesize;
1663 WRITE_ONCE(msk->write_seq, msk->write_seq + psize);
13e16037 1664 msk->tx_pending_data += psize;
d9ca1de8
PA		 1665
1666 /* charge data on mptcp pending queue to the msk socket
1667 * Note: we charge such data both to sk and ssk
fb529e62 1668 */
d9ca1de8 1669 sk_wmem_queued_add(sk, frag_truesize);
d9ca1de8
PA		 1670 if (!dfrag_collapsed) {
1671 get_page(dfrag->page);
1672 list_add_tail(&dfrag->list, &msk->rtx_queue);
1673 if (!msk->first_pending)
1674 WRITE_ONCE(msk->first_pending, dfrag);
fb529e62 1675 }
d9ca1de8
PA		 1676 pr_debug("msk=%p dfrag at seq=%lld len=%d sent=%d new=%d", msk,
1677 dfrag->data_seq, dfrag->data_len, dfrag->already_sent,
1678 !dfrag_collapsed);
6d0060f6 1679
d9ca1de8 1680 continue;
b51f9b80 1681
d9ca1de8 1682wait_for_memory:
6e628cd3
PA		 1683 set_bit(MPTCP_NOSPACE, &msk->flags);
1684 mptcp_push_pending(sk, msg->msg_flags);
d9ca1de8
PA		 1685 ret = sk_stream_wait_memory(sk, &timeo);
1686 if (ret)
1687 goto out;
57040755 1688 }
6d0060f6 1689
13e16037 1690 if (copied)
d9ca1de8
PA		 1691 mptcp_push_pending(sk, msg->msg_flags);
1692
1954b860 1693out:
cec37a6e 1694 release_sock(sk);
8555c6bf 1695 return copied ? : ret;
f870fa0b
MM		 1696}
1697
7a6a6cbc
PA		 1698static void mptcp_wait_data(struct sock *sk, long *timeo)
1699{
1700 DEFINE_WAIT_FUNC(wait, woken_wake_function);
1701 struct mptcp_sock *msk = mptcp_sk(sk);
1702
1703 add_wait_queue(sk_sleep(sk), &wait);
1704 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1705
1706 sk_wait_event(sk, timeo,
1707 test_and_clear_bit(MPTCP_DATA_READY, &msk->flags), &wait);
1708
1709 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1710 remove_wait_queue(sk_sleep(sk), &wait);
1711}
1712
6771bfd9
FW		 1713static int __mptcp_recvmsg_mskq(struct mptcp_sock *msk,
1714 struct msghdr *msg,
1715 size_t len)
1716{
6771bfd9
FW		 1717 struct sk_buff *skb;
1718 int copied = 0;
1719
87952603 1720 while ((skb = skb_peek(&msk->receive_queue)) != NULL) {
6771bfd9
FW		 1721 u32 offset = MPTCP_SKB_CB(skb)->offset;
1722 u32 data_len = skb->len - offset;
1723 u32 count = min_t(size_t, len - copied, data_len);
1724 int err;
1725
1726 err = skb_copy_datagram_msg(skb, offset, msg, count);
1727 if (unlikely(err < 0)) {
1728 if (!copied)
1729 return err;
1730 break;
1731 }
1732
1733 copied += count;
1734
1735 if (count < data_len) {
1736 MPTCP_SKB_CB(skb)->offset += count;
1737 break;
1738 }
1739
87952603
PA		 1740 /* we will bulk release the skb memory later */
1741 skb->destructor = NULL;
1742 msk->rmem_released += skb->truesize;
1743 __skb_unlink(skb, &msk->receive_queue);
6771bfd9
FW		 1744 __kfree_skb(skb);
1745
1746 if (copied >= len)
1747 break;
1748 }
1749
1750 return copied;
1751}
1752
a6b118fe
FW		 1753/* receive buffer autotuning. See tcp_rcv_space_adjust for more information.
1754 *
1755 * Only difference: Use highest rtt estimate of the subflows in use.
1756 */
1757static void mptcp_rcv_space_adjust(struct mptcp_sock *msk, int copied)
1758{
1759 struct mptcp_subflow_context *subflow;
1760 struct sock *sk = (struct sock *)msk;
1761 u32 time, advmss = 1;
1762 u64 rtt_us, mstamp;
1763
1764 sock_owned_by_me(sk);
1765
1766 if (copied <= 0)
1767 return;
1768
1769 msk->rcvq_space.copied += copied;
1770
1771 mstamp = div_u64(tcp_clock_ns(), NSEC_PER_USEC);
1772 time = tcp_stamp_us_delta(mstamp, msk->rcvq_space.time);
1773
1774 rtt_us = msk->rcvq_space.rtt_us;
1775 if (rtt_us && time < (rtt_us >> 3))
1776 return;
1777
1778 rtt_us = 0;
1779 mptcp_for_each_subflow(msk, subflow) {
1780 const struct tcp_sock *tp;
1781 u64 sf_rtt_us;
1782 u32 sf_advmss;
1783
1784 tp = tcp_sk(mptcp_subflow_tcp_sock(subflow));
1785
1786 sf_rtt_us = READ_ONCE(tp->rcv_rtt_est.rtt_us);
1787 sf_advmss = READ_ONCE(tp->advmss);
1788
1789 rtt_us = max(sf_rtt_us, rtt_us);
1790 advmss = max(sf_advmss, advmss);
1791 }
1792
1793 msk->rcvq_space.rtt_us = rtt_us;
1794 if (time < (rtt_us >> 3) || rtt_us == 0)
1795 return;
1796
1797 if (msk->rcvq_space.copied <= msk->rcvq_space.space)
1798 goto new_measure;
1799
1800 if (sock_net(sk)->ipv4.sysctl_tcp_moderate_rcvbuf &&
1801 !(sk->sk_userlocks & SOCK_RCVBUF_LOCK)) {
1802 int rcvmem, rcvbuf;
1803 u64 rcvwin, grow;
1804
1805 rcvwin = ((u64)msk->rcvq_space.copied << 1) + 16 * advmss;
1806
1807 grow = rcvwin * (msk->rcvq_space.copied - msk->rcvq_space.space);
1808
1809 do_div(grow, msk->rcvq_space.space);
1810 rcvwin += (grow << 1);
1811
1812 rcvmem = SKB_TRUESIZE(advmss + MAX_TCP_HEADER);
1813 while (tcp_win_from_space(sk, rcvmem) < advmss)
1814 rcvmem += 128;
1815
1816 do_div(rcvwin, advmss);
1817 rcvbuf = min_t(u64, rcvwin * rcvmem,
1818 sock_net(sk)->ipv4.sysctl_tcp_rmem[2]);
1819
1820 if (rcvbuf > sk->sk_rcvbuf) {
1821 u32 window_clamp;
1822
1823 window_clamp = tcp_win_from_space(sk, rcvbuf);
1824 WRITE_ONCE(sk->sk_rcvbuf, rcvbuf);
1825
1826 /* Make subflows follow along. If we do not do this, we
1827 * get drops at subflow level if skbs can't be moved to
1828 * the mptcp rx queue fast enough (announced rcv_win can
1829 * exceed ssk->sk_rcvbuf).
1830 */
1831 mptcp_for_each_subflow(msk, subflow) {
1832 struct sock *ssk;
c76c6956 1833 bool slow;
a6b118fe
FW		 1834
1835 ssk = mptcp_subflow_tcp_sock(subflow);
c76c6956 1836 slow = lock_sock_fast(ssk);
a6b118fe
FW		 1837 WRITE_ONCE(ssk->sk_rcvbuf, rcvbuf);
1838 tcp_sk(ssk)->window_clamp = window_clamp;
c76c6956
PA		 1839 tcp_cleanup_rbuf(ssk, 1);
1840 unlock_sock_fast(ssk, slow);
a6b118fe
FW		 1841 }
1842 }
1843 }
1844
1845 msk->rcvq_space.space = msk->rcvq_space.copied;
1846new_measure:
1847 msk->rcvq_space.copied = 0;
1848 msk->rcvq_space.time = mstamp;
1849}
1850
87952603
PA		 1851static void __mptcp_update_rmem(struct sock *sk)
1852{
1853 struct mptcp_sock *msk = mptcp_sk(sk);
1854
1855 if (!msk->rmem_released)
1856 return;
1857
1858 atomic_sub(msk->rmem_released, &sk->sk_rmem_alloc);
1859 sk_mem_uncharge(sk, msk->rmem_released);
1860 msk->rmem_released = 0;
1861}
1862
1863static void __mptcp_splice_receive_queue(struct sock *sk)
1864{
1865 struct mptcp_sock *msk = mptcp_sk(sk);
1866
1867 skb_queue_splice_tail_init(&sk->sk_receive_queue, &msk->receive_queue);
1868}
1869
ea4ca586 1870static bool __mptcp_move_skbs(struct mptcp_sock *msk, unsigned int rcv)
6771bfd9 1871{
87952603 1872 struct sock *sk = (struct sock *)msk;
6771bfd9 1873 unsigned int moved = 0;
87952603 1874 bool ret, done;
d5f49190
PA		 1875
1876 __mptcp_flush_join_list(msk);
6771bfd9
FW		 1877 do {
1878 struct sock *ssk = mptcp_subflow_recv_lookup(msk);
65f49fe7 1879 bool slowpath;
6771bfd9 1880
87952603
PA		 1881 /* we can have data pending in the subflows only if the msk
1882 * receive buffer was full at subflow_data_ready() time,
1883 * that is an unlikely slow path.
1884 */
1885 if (likely(!ssk))
6771bfd9
FW		 1886 break;
1887
65f49fe7 1888 slowpath = lock_sock_fast(ssk);
87952603 1889 mptcp_data_lock(sk);
6771bfd9 1890 done = __mptcp_move_skbs_from_subflow(msk, ssk, &moved);
87952603 1891 mptcp_data_unlock(sk);
ea4ca586
PA		 1892 if (moved && rcv) {
1893 WRITE_ONCE(msk->rmem_pending, min(rcv, moved));
1894 tcp_cleanup_rbuf(ssk, 1);
1895 WRITE_ONCE(msk->rmem_pending, 0);
1896 }
65f49fe7 1897 unlock_sock_fast(ssk, slowpath);
6771bfd9
FW		 1898 } while (!done);
1899
87952603
PA		 1900 /* acquire the data lock only if some input data is pending */
1901 ret = moved > 0;
1902 if (!RB_EMPTY_ROOT(&msk->out_of_order_queue) ||
1903 !skb_queue_empty_lockless(&sk->sk_receive_queue)) {
1904 mptcp_data_lock(sk);
1905 __mptcp_update_rmem(sk);
1906 ret |= __mptcp_ofo_queue(msk);
1907 __mptcp_splice_receive_queue(sk);
1908 mptcp_data_unlock(sk);
ab174ad8 1909 }
87952603
PA		 1910 if (ret)
1911 mptcp_check_data_fin((struct sock *)msk);
1912 return !skb_queue_empty(&msk->receive_queue);
6771bfd9
FW		 1913}
1914
f870fa0b
MM		 1915static int mptcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
1916 int nonblock, int flags, int *addr_len)
1917{
1918 struct mptcp_sock *msk = mptcp_sk(sk);
cec37a6e 1919 int copied = 0;
7a6a6cbc
PA		 1920 int target;
1921 long timeo;
f870fa0b
MM		 1922
1923 if (msg->msg_flags & ~(MSG_WAITALL | MSG_DONTWAIT))
1924 return -EOPNOTSUPP;
1925
87952603 1926 mptcp_lock_sock(sk, __mptcp_splice_receive_queue(sk));
fd897679
PA		 1927 if (unlikely(sk->sk_state == TCP_LISTEN)) {
1928 copied = -ENOTCONN;
1929 goto out_err;
1930 }
1931
7a6a6cbc
PA		 1932 timeo = sock_rcvtimeo(sk, nonblock);
1933
1934 len = min_t(size_t, len, INT_MAX);
1935 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1936
05e3ecea 1937 while (copied < len) {
ea4ca586 1938 int bytes_read, old_space;
7a6a6cbc 1939
6771bfd9
FW		 1940 bytes_read = __mptcp_recvmsg_mskq(msk, msg, len - copied);
1941 if (unlikely(bytes_read < 0)) {
1942 if (!copied)
1943 copied = bytes_read;
1944 goto out_err;
1945 }
7a6a6cbc 1946
6771bfd9 1947 copied += bytes_read;
7a6a6cbc 1948
87952603 1949 if (skb_queue_empty(&msk->receive_queue) &&
ea4ca586 1950 __mptcp_move_skbs(msk, len - copied))
6771bfd9 1951 continue;
7a6a6cbc 1952
ea4ca586
PA		 1953 /* be sure to advertise window change */
1954 old_space = READ_ONCE(msk->old_wspace);
1955 if ((tcp_space(sk) - old_space) >= old_space)
fd897679 1956 mptcp_cleanup_rbuf(msk);
ea4ca586 1957
7a6a6cbc
PA		 1958 /* only the master socket status is relevant here. The exit
1959 * conditions mirror closely tcp_recvmsg()
1960 */
1961 if (copied >= target)
1962 break;
1963
1964 if (copied) {
1965 if (sk->sk_err ||
1966 sk->sk_state == TCP_CLOSE ||
1967 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1968 !timeo ||
1969 signal_pending(current))
1970 break;
1971 } else {
1972 if (sk->sk_err) {
1973 copied = sock_error(sk);
1974 break;
1975 }
1976
5969856a
PA		 1977 if (test_and_clear_bit(MPTCP_WORK_EOF, &msk->flags))
1978 mptcp_check_for_eof(msk);
1979
87952603
PA		 1980 if (sk->sk_shutdown & RCV_SHUTDOWN) {
1981 /* race breaker: the shutdown could be after the
1982 * previous receive queue check
1983 */
1984 if (__mptcp_move_skbs(msk, len - copied))
1985 continue;
7a6a6cbc 1986 break;
87952603 1987 }
7a6a6cbc
PA		 1988
1989 if (sk->sk_state == TCP_CLOSE) {
1990 copied = -ENOTCONN;
1991 break;
1992 }
1993
1994 if (!timeo) {
1995 copied = -EAGAIN;
1996 break;
1997 }
1998
1999 if (signal_pending(current)) {
2000 copied = sock_intr_errno(timeo);
2001 break;
2002 }
2003 }
2004
2005 pr_debug("block timeout %ld", timeo);
7a6a6cbc 2006 mptcp_wait_data(sk, &timeo);
cec37a6e
PK		 2007 }
2008
87952603
PA		 2009 if (skb_queue_empty_lockless(&sk->sk_receive_queue) &&
2010 skb_queue_empty(&msk->receive_queue)) {
6771bfd9 2011 /* entire backlog drained, clear DATA_READY. */
7a6a6cbc 2012 clear_bit(MPTCP_DATA_READY, &msk->flags);
cec37a6e 2013
6771bfd9
FW		 2014 /* .. race-breaker: ssk might have gotten new data
2015 * after last __mptcp_move_skbs() returned false.
7a6a6cbc 2016 */
ea4ca586 2017 if (unlikely(__mptcp_move_skbs(msk, 0)))
7a6a6cbc 2018 set_bit(MPTCP_DATA_READY, &msk->flags);
6771bfd9
FW		 2019 } else if (unlikely(!test_bit(MPTCP_DATA_READY, &msk->flags))) {
2020 /* data to read but mptcp_wait_data() cleared DATA_READY */
2021 set_bit(MPTCP_DATA_READY, &msk->flags);
7a6a6cbc 2022 }
6771bfd9 2023out_err:
6719331c
PA		 2024 pr_debug("msk=%p data_ready=%d rx queue empty=%d copied=%d",
2025 msk, test_bit(MPTCP_DATA_READY, &msk->flags),
87952603 2026 skb_queue_empty_lockless(&sk->sk_receive_queue), copied);
a6b118fe
FW		 2027 mptcp_rcv_space_adjust(msk, copied);
2028
7a6a6cbc 2029 release_sock(sk);
cec37a6e
PK		 2030 return copied;
2031}
2032
b51f9b80
PA		 2033static void mptcp_retransmit_handler(struct sock *sk)
2034{
2035 struct mptcp_sock *msk = mptcp_sk(sk);
2036
7439d687
PA		 2037 set_bit(MPTCP_WORK_RTX, &msk->flags);
2038 mptcp_schedule_work(sk);
b51f9b80
PA		 2039}
2040
2041static void mptcp_retransmit_timer(struct timer_list *t)
2042{
2043 struct inet_connection_sock *icsk = from_timer(icsk, t,
2044 icsk_retransmit_timer);
2045 struct sock *sk = &icsk->icsk_inet.sk;
2046
2047 bh_lock_sock(sk);
2048 if (!sock_owned_by_user(sk)) {
2049 mptcp_retransmit_handler(sk);
2050 } else {
2051 /* delegate our work to tcp_release_cb() */
2052 if (!test_and_set_bit(TCP_WRITE_TIMER_DEFERRED,
2053 &sk->sk_tsq_flags))
2054 sock_hold(sk);
2055 }
2056 bh_unlock_sock(sk);
2057 sock_put(sk);
2058}
2059
e16163b6
PA		 2060static void mptcp_timeout_timer(struct timer_list *t)
2061{
2062 struct sock *sk = from_timer(sk, t, sk_timer);
2063
2064 mptcp_schedule_work(sk);
b6d69fc8 2065 sock_put(sk);
e16163b6
PA		 2066}
2067
3b1d6210
PA		 2068/* Find an idle subflow. Return NULL if there is unacked data at tcp
2069 * level.
2070 *
2071 * A backup subflow is returned only if that is the only kind available.
2072 */
2073static struct sock *mptcp_subflow_get_retrans(const struct mptcp_sock *msk)
2074{
2075 struct mptcp_subflow_context *subflow;
2076 struct sock *backup = NULL;
2077
2078 sock_owned_by_me((const struct sock *)msk);
2079
d5f49190 2080 if (__mptcp_check_fallback(msk))
d9ca1de8 2081 return NULL;
d5f49190 2082
3b1d6210
PA		 2083 mptcp_for_each_subflow(msk, subflow) {
2084 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2085
d5f49190
PA		 2086 if (!mptcp_subflow_active(subflow))
2087 continue;
2088
3b1d6210 2089 /* still data outstanding at TCP level? Don't retransmit. */
860975c6
FW		 2090 if (!tcp_write_queue_empty(ssk)) {
2091 if (inet_csk(ssk)->icsk_ca_state >= TCP_CA_Loss)
2092 continue;
3b1d6210 2093 return NULL;
860975c6 2094 }
3b1d6210
PA		 2095
2096 if (subflow->backup) {
2097 if (!backup)
2098 backup = ssk;
2099 continue;
2100 }
2101
2102 return ssk;
2103 }
2104
2105 return backup;
2106}
2107
cec37a6e
PK		 2108/* subflow sockets can be either outgoing (connect) or incoming
2109 * (accept).
2110 *
2111 * Outgoing subflows use in-kernel sockets.
2112 * Incoming subflows do not have their own 'struct socket' allocated,
2113 * so we need to use tcp_close() after detaching them from the mptcp
2114 * parent socket.
2115 */
d0876b22 2116void __mptcp_close_ssk(struct sock *sk, struct sock *ssk,
e16163b6 2117 struct mptcp_subflow_context *subflow)
cec37a6e 2118{
e16163b6
PA		 2119 bool dispose_socket = false;
2120 struct socket *sock;
cec37a6e
PK		 2121
2122 list_del(&subflow->node);
2123
3f8b2667 2124 lock_sock_nested(ssk, SINGLE_DEPTH_NESTING);
e16163b6
PA		 2125
2126 /* if we are invoked by the msk cleanup code, the subflow is
2127 * already orphaned
2128 */
2129 sock = ssk->sk_socket;
2130 if (sock) {
2131 dispose_socket = sock != sk->sk_socket;
2132 sock_orphan(ssk);
2133 }
2134
d7b1bfd0
PA		 2135 subflow->disposable = 1;
2136
e16163b6
PA		 2137 /* if ssk hit tcp_done(), tcp_cleanup_ulp() cleared the related ops
2138 * the ssk has been already destroyed, we just need to release the
2139 * reference owned by msk;
2140 */
2141 if (!inet_csk(ssk)->icsk_ulp_ops) {
2142 kfree_rcu(subflow, rcu);
cec37a6e 2143 } else {
d7b1bfd0 2144 /* otherwise tcp will dispose of the ssk and subflow ctx */
e16163b6
PA		 2145 __tcp_close(ssk, 0);
2146
2147 /* close acquired an extra ref */
2148 __sock_put(ssk);
cec37a6e 2149 }
e16163b6
PA		 2150 release_sock(ssk);
2151 if (dispose_socket)
2152 iput(SOCK_INODE(sock));
2153
2154 sock_put(ssk);
f870fa0b
MM		 2155}
2156
dc24f8b4
PA		 2157static unsigned int mptcp_sync_mss(struct sock *sk, u32 pmtu)
2158{
2159 return 0;
2160}
2161
b416268b
FW		 2162static void pm_work(struct mptcp_sock *msk)
2163{
2164 struct mptcp_pm_data *pm = &msk->pm;
2165
2166 spin_lock_bh(&msk->pm.lock);
2167
2168 pr_debug("msk=%p status=%x", msk, pm->status);
2169 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) {
2170 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED);
2171 mptcp_pm_nl_add_addr_received(msk);
2172 }
84dfe367
GT		 2173 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) {
2174 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK);
2175 mptcp_pm_nl_add_addr_send_ack(msk);
2176 }
d0876b22
GT		 2177 if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) {
2178 pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED);
2179 mptcp_pm_nl_rm_addr_received(msk);
2180 }
b416268b
FW		 2181 if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) {
2182 pm->status &= ~BIT(MPTCP_PM_ESTABLISHED);
2183 mptcp_pm_nl_fully_established(msk);
2184 }
2185 if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) {
2186 pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED);
2187 mptcp_pm_nl_subflow_established(msk);
2188 }
2189
2190 spin_unlock_bh(&msk->pm.lock);
2191}
2192
0e4f35d7
PA		 2193static void __mptcp_close_subflow(struct mptcp_sock *msk)
2194{
2195 struct mptcp_subflow_context *subflow, *tmp;
2196
2197 list_for_each_entry_safe(subflow, tmp, &msk->conn_list, node) {
2198 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2199
2200 if (inet_sk_state_load(ssk) != TCP_CLOSE)
2201 continue;
2202
e16163b6 2203 __mptcp_close_ssk((struct sock *)msk, ssk, subflow);
0e4f35d7
PA		 2204 }
2205}
2206
e16163b6
PA		 2207static bool mptcp_check_close_timeout(const struct sock *sk)
2208{
2209 s32 delta = tcp_jiffies32 - inet_csk(sk)->icsk_mtup.probe_timestamp;
2210 struct mptcp_subflow_context *subflow;
2211
2212 if (delta >= TCP_TIMEWAIT_LEN)
2213 return true;
2214
2215 /* if all subflows are in closed status don't bother with additional
2216 * timeout
2217 */
2218 mptcp_for_each_subflow(mptcp_sk(sk), subflow) {
2219 if (inet_sk_state_load(mptcp_subflow_tcp_sock(subflow)) !=
2220 TCP_CLOSE)
2221 return false;
2222 }
2223 return true;
2224}
2225
50c504a2
FW		 2226static void mptcp_check_fastclose(struct mptcp_sock *msk)
2227{
2228 struct mptcp_subflow_context *subflow, *tmp;
2229 struct sock *sk = &msk->sk.icsk_inet.sk;
2230
2231 if (likely(!READ_ONCE(msk->rcv_fastclose)))
2232 return;
2233
2234 mptcp_token_destroy(msk);
2235
2236 list_for_each_entry_safe(subflow, tmp, &msk->conn_list, node) {
2237 struct sock *tcp_sk = mptcp_subflow_tcp_sock(subflow);
2238
2239 lock_sock(tcp_sk);
2240 if (tcp_sk->sk_state != TCP_CLOSE) {
2241 tcp_send_active_reset(tcp_sk, GFP_ATOMIC);
2242 tcp_set_state(tcp_sk, TCP_CLOSE);
2243 }
2244 release_sock(tcp_sk);
2245 }
2246
2247 inet_sk_state_store(sk, TCP_CLOSE);
2248 sk->sk_shutdown = SHUTDOWN_MASK;
2249 smp_mb__before_atomic(); /* SHUTDOWN must be visible first */
2250 set_bit(MPTCP_DATA_READY, &msk->flags);
2251 set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags);
2252
2253 mptcp_close_wake_up(sk);
2254}
2255
80992017
PA		 2256static void mptcp_worker(struct work_struct *work)
2257{
2258 struct mptcp_sock *msk = container_of(work, struct mptcp_sock, work);
3b1d6210 2259 struct sock *ssk, *sk = &msk->sk.icsk_inet.sk;
caf971df 2260 struct mptcp_sendmsg_info info = {};
3b1d6210 2261 struct mptcp_data_frag *dfrag;
3b1d6210 2262 size_t copied = 0;
e16163b6 2263 int state, ret;
80992017
PA		 2264
2265 lock_sock(sk);
e16163b6
PA		 2266 state = sk->sk_state;
2267 if (unlikely(state == TCP_CLOSE))
2268 goto unlock;
2269
43b54c6e 2270 mptcp_check_data_fin_ack(sk);
ec3edaa7 2271 __mptcp_flush_join_list(msk);
50c504a2
FW		 2272
2273 mptcp_check_fastclose(msk);
2274
0e4f35d7
PA		 2275 if (test_and_clear_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags))
2276 __mptcp_close_subflow(msk);
2277
b416268b
FW		 2278 if (msk->pm.status)
2279 pm_work(msk);
2280
59832e24
FW		 2281 if (test_and_clear_bit(MPTCP_WORK_EOF, &msk->flags))
2282 mptcp_check_for_eof(msk);
2283
6e628cd3 2284 __mptcp_check_send_data_fin(sk);
43b54c6e
MM		 2285 mptcp_check_data_fin(sk);
2286
e16163b6
PA		 2287 /* if the msk data is completely acked, or the socket timedout,
2288 * there is no point in keeping around an orphaned sk
2289 */
2290 if (sock_flag(sk, SOCK_DEAD) &&
2291 (mptcp_check_close_timeout(sk) ||
2292 (state != sk->sk_state &&
2293 ((1 << inet_sk_state_load(sk)) & (TCPF_CLOSE | TCPF_FIN_WAIT2))))) {
2294 inet_sk_state_store(sk, TCP_CLOSE);
2295 __mptcp_destroy_sock(sk);
2296 goto unlock;
2297 }
2298
3b1d6210
PA		 2299 if (!test_and_clear_bit(MPTCP_WORK_RTX, &msk->flags))
2300 goto unlock;
2301
2302 dfrag = mptcp_rtx_head(sk);
2303 if (!dfrag)
2304 goto unlock;
2305
2306 ssk = mptcp_subflow_get_retrans(msk);
2307 if (!ssk)
2308 goto reset_unlock;
2309
2310 lock_sock(ssk);
2311
d9ca1de8
PA		 2312 /* limit retransmission to the bytes already sent on some subflows */
2313 info.sent = 0;
2314 info.limit = dfrag->already_sent;
2315 while (info.sent < dfrag->already_sent) {
724cfd2e
PA		 2316 if (!mptcp_alloc_tx_skb(sk, ssk))
2317 break;
2318
d9ca1de8 2319 ret = mptcp_sendmsg_frag(sk, ssk, dfrag, &info);
6f8a612a 2320 if (ret <= 0)
3b1d6210
PA		 2321 break;
2322
fc518953 2323 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_RETRANSSEGS);
3b1d6210 2324 copied += ret;
d9ca1de8 2325 info.sent += ret;
3b1d6210
PA		 2326 }
2327 if (copied)
caf971df
PA		 2328 tcp_push(ssk, 0, info.mss_now, tcp_sk(ssk)->nonagle,
2329 info.size_goal);
3b1d6210 2330
3b1d6210
PA		 2331 mptcp_set_timeout(sk, ssk);
2332 release_sock(ssk);
2333
2334reset_unlock:
2335 if (!mptcp_timer_pending(sk))
2336 mptcp_reset_timer(sk);
2337
2338unlock:
80992017
PA		 2339 release_sock(sk);
2340 sock_put(sk);
2341}
2342
784325e9 2343static int __mptcp_init_sock(struct sock *sk)
f870fa0b 2344{
cec37a6e
PK		 2345 struct mptcp_sock *msk = mptcp_sk(sk);
2346
ec3edaa7
PK		 2347 spin_lock_init(&msk->join_list_lock);
2348
cec37a6e 2349 INIT_LIST_HEAD(&msk->conn_list);
ec3edaa7 2350 INIT_LIST_HEAD(&msk->join_list);
18b683bf 2351 INIT_LIST_HEAD(&msk->rtx_queue);
80992017 2352 INIT_WORK(&msk->work, mptcp_worker);
87952603 2353 __skb_queue_head_init(&msk->receive_queue);
724cfd2e 2354 __skb_queue_head_init(&msk->skb_tx_cache);
ab174ad8 2355 msk->out_of_order_queue = RB_ROOT;
f0e6a4cf 2356 msk->first_pending = NULL;
e93da928 2357 msk->wmem_reserved = 0;
87952603 2358 msk->rmem_released = 0;
724cfd2e
PA		 2359 msk->tx_pending_data = 0;
2360 msk->size_goal_cache = TCP_BASE_MSS;
cec37a6e 2361
ea4ca586 2362 msk->ack_hint = NULL;
8ab183de 2363 msk->first = NULL;
dc24f8b4 2364 inet_csk(sk)->icsk_sync_mss = mptcp_sync_mss;
8ab183de 2365
1b1c7a0e
PK		 2366 mptcp_pm_data_init(msk);
2367
b51f9b80
PA		 2368 /* re-use the csk retrans timer for MPTCP-level retrans */
2369 timer_setup(&msk->sk.icsk_retransmit_timer, mptcp_retransmit_timer, 0);
e16163b6 2370 timer_setup(&sk->sk_timer, mptcp_timeout_timer, 0);
f870fa0b
MM		 2371 return 0;
2372}
2373
784325e9
MB		 2374static int mptcp_init_sock(struct sock *sk)
2375{
fc518953
FW		 2376 struct net *net = sock_net(sk);
2377 int ret;
18b683bf 2378
b6c08380
GT		 2379 ret = __mptcp_init_sock(sk);
2380 if (ret)
2381 return ret;
2382
fc518953
FW		 2383 if (!mptcp_is_enabled(net))
2384 return -ENOPROTOOPT;
2385
2386 if (unlikely(!net->mib.mptcp_statistics) && !mptcp_mib_alloc(net))
2387 return -ENOMEM;
2388
fa68018d
PA		 2389 ret = __mptcp_socket_create(mptcp_sk(sk));
2390 if (ret)
2391 return ret;
2392
d027236c 2393 sk_sockets_allocated_inc(sk);
a6b118fe 2394 sk->sk_rcvbuf = sock_net(sk)->ipv4.sysctl_tcp_rmem[1];
da51aef5 2395 sk->sk_sndbuf = sock_net(sk)->ipv4.sysctl_tcp_wmem[1];
d027236c 2396
18b683bf
PA		 2397 return 0;
2398}
2399
2400static void __mptcp_clear_xmit(struct sock *sk)
2401{
2402 struct mptcp_sock *msk = mptcp_sk(sk);
2403 struct mptcp_data_frag *dtmp, *dfrag;
724cfd2e 2404 struct sk_buff *skb;
18b683bf 2405
d9ca1de8 2406 WRITE_ONCE(msk->first_pending, NULL);
18b683bf 2407 list_for_each_entry_safe(dfrag, dtmp, &msk->rtx_queue, list)
d027236c 2408 dfrag_clear(sk, dfrag);
724cfd2e
PA		 2409 while ((skb = __skb_dequeue(&msk->skb_tx_cache)) != NULL) {
2410 sk->sk_forward_alloc += skb->truesize;
2411 kfree_skb(skb);
2412 }
784325e9
MB		 2413}
2414
80992017
PA		 2415static void mptcp_cancel_work(struct sock *sk)
2416{
2417 struct mptcp_sock *msk = mptcp_sk(sk);
2418
b2771d24 2419 if (cancel_work_sync(&msk->work))
e16163b6 2420 __sock_put(sk);
80992017
PA		 2421}
2422
d0876b22 2423void mptcp_subflow_shutdown(struct sock *sk, struct sock *ssk, int how)
21498490
PK		 2424{
2425 lock_sock(ssk);
2426
2427 switch (ssk->sk_state) {
2428 case TCP_LISTEN:
2429 if (!(how & RCV_SHUTDOWN))
2430 break;
df561f66 2431 fallthrough;
21498490
PK		 2432 case TCP_SYN_SENT:
2433 tcp_disconnect(ssk, O_NONBLOCK);
2434 break;
2435 default:
43b54c6e
MM		 2436 if (__mptcp_check_fallback(mptcp_sk(sk))) {
2437 pr_debug("Fallback");
2438 ssk->sk_shutdown |= how;
2439 tcp_shutdown(ssk, how);
2440 } else {
2441 pr_debug("Sending DATA_FIN on subflow %p", ssk);
2442 mptcp_set_timeout(sk, ssk);
2443 tcp_send_ack(ssk);
2444 }
21498490
PK		 2445 break;
2446 }
2447
21498490
PK		 2448 release_sock(ssk);
2449}
2450
6920b851
MM		 2451static const unsigned char new_state[16] = {
2452 /* current state: new state: action: */
2453 [0 /* (Invalid) */] = TCP_CLOSE,
2454 [TCP_ESTABLISHED] = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2455 [TCP_SYN_SENT] = TCP_CLOSE,
2456 [TCP_SYN_RECV] = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2457 [TCP_FIN_WAIT1] = TCP_FIN_WAIT1,
2458 [TCP_FIN_WAIT2] = TCP_FIN_WAIT2,
2459 [TCP_TIME_WAIT] = TCP_CLOSE, /* should not happen ! */
2460 [TCP_CLOSE] = TCP_CLOSE,
2461 [TCP_CLOSE_WAIT] = TCP_LAST_ACK | TCP_ACTION_FIN,
2462 [TCP_LAST_ACK] = TCP_LAST_ACK,
2463 [TCP_LISTEN] = TCP_CLOSE,
2464 [TCP_CLOSING] = TCP_CLOSING,
2465 [TCP_NEW_SYN_RECV] = TCP_CLOSE, /* should not happen ! */
2466};
2467
2468static int mptcp_close_state(struct sock *sk)
2469{
2470 int next = (int)new_state[sk->sk_state];
2471 int ns = next & TCP_STATE_MASK;
2472
2473 inet_sk_state_store(sk, ns);
2474
2475 return next & TCP_ACTION_FIN;
2476}
2477
e16163b6 2478static void __mptcp_check_send_data_fin(struct sock *sk)
f870fa0b 2479{
e16163b6 2480 struct mptcp_subflow_context *subflow;
f870fa0b
MM		 2481 struct mptcp_sock *msk = mptcp_sk(sk);
2482
e16163b6
PA		 2483 pr_debug("msk=%p snd_data_fin_enable=%d pending=%d snd_nxt=%llu write_seq=%llu",
2484 msk, msk->snd_data_fin_enable, !!mptcp_send_head(sk),
2485 msk->snd_nxt, msk->write_seq);
43b54c6e 2486
e16163b6
PA		 2487 /* we still need to enqueue subflows or not really shutting down,
2488 * skip this
2489 */
2490 if (!msk->snd_data_fin_enable || msk->snd_nxt + 1 != msk->write_seq ||
2491 mptcp_send_head(sk))
2492 return;
2493
2494 WRITE_ONCE(msk->snd_nxt, msk->write_seq);
2495
26aa2314
PA		 2496 /* fallback socket will not get data_fin/ack, can move to the next
2497 * state now
2498 */
2499 if (__mptcp_check_fallback(msk)) {
2500 if ((1 << sk->sk_state) & (TCPF_CLOSING | TCPF_LAST_ACK)) {
2501 inet_sk_state_store(sk, TCP_CLOSE);
2502 mptcp_close_wake_up(sk);
2503 } else if (sk->sk_state == TCP_FIN_WAIT1) {
2504 inet_sk_state_store(sk, TCP_FIN_WAIT2);
2505 }
43b54c6e
MM		 2506 }
2507
e16163b6
PA		 2508 __mptcp_flush_join_list(msk);
2509 mptcp_for_each_subflow(msk, subflow) {
2510 struct sock *tcp_sk = mptcp_subflow_tcp_sock(subflow);
43b54c6e 2511
e16163b6 2512 mptcp_subflow_shutdown(sk, tcp_sk, SEND_SHUTDOWN);
43b54c6e 2513 }
e16163b6 2514}
2c22c06c 2515
e16163b6
PA		 2516static void __mptcp_wr_shutdown(struct sock *sk)
2517{
2518 struct mptcp_sock *msk = mptcp_sk(sk);
43b54c6e 2519
e16163b6
PA		 2520 pr_debug("msk=%p snd_data_fin_enable=%d shutdown=%x state=%d pending=%d",
2521 msk, msk->snd_data_fin_enable, sk->sk_shutdown, sk->sk_state,
2522 !!mptcp_send_head(sk));
2523
2524 /* will be ignored by fallback sockets */
2525 WRITE_ONCE(msk->write_seq, msk->write_seq + 1);
2526 WRITE_ONCE(msk->snd_data_fin_enable, 1);
2527
2528 __mptcp_check_send_data_fin(sk);
2529}
2530
2531static void __mptcp_destroy_sock(struct sock *sk)
2532{
2533 struct mptcp_subflow_context *subflow, *tmp;
2534 struct mptcp_sock *msk = mptcp_sk(sk);
2535 LIST_HEAD(conn_list);
2536
2537 pr_debug("msk=%p", msk);
f870fa0b 2538
10f6d46c
PA		 2539 /* be sure to always acquire the join list lock, to sync vs
2540 * mptcp_finish_join().
2541 */
2542 spin_lock_bh(&msk->join_list_lock);
2543 list_splice_tail_init(&msk->join_list, &msk->conn_list);
2544 spin_unlock_bh(&msk->join_list_lock);
b2c5b614
FW		 2545 list_splice_init(&msk->conn_list, &conn_list);
2546
6e628cd3 2547 sk_stop_timer(sk, &msk->sk.icsk_retransmit_timer);
e16163b6
PA		 2548 sk_stop_timer(sk, &sk->sk_timer);
2549 msk->pm.status = 0;
b2c5b614
FW		 2550
2551 list_for_each_entry_safe(subflow, tmp, &conn_list, node) {
cec37a6e 2552 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
e16163b6 2553 __mptcp_close_ssk(sk, ssk, subflow);
f870fa0b
MM		 2554 }
2555
e16163b6 2556 sk->sk_prot->destroy(sk);
80992017 2557
e93da928 2558 WARN_ON_ONCE(msk->wmem_reserved);
87952603 2559 WARN_ON_ONCE(msk->rmem_released);
e16163b6
PA		 2560 sk_stream_kill_queues(sk);
2561 xfrm_sk_free_policy(sk);
2562 sk_refcnt_debug_release(sk);
2563 sock_put(sk);
2564}
2565
2566static void mptcp_close(struct sock *sk, long timeout)
2567{
2568 struct mptcp_subflow_context *subflow;
2569 bool do_cancel_work = false;
2570
2571 lock_sock(sk);
2572 sk->sk_shutdown = SHUTDOWN_MASK;
2573
2574 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE)) {
2575 inet_sk_state_store(sk, TCP_CLOSE);
2576 goto cleanup;
2577 }
6771bfd9 2578
e16163b6
PA		 2579 if (mptcp_close_state(sk))
2580 __mptcp_wr_shutdown(sk);
2581
2582 sk_stream_wait_close(sk, timeout);
2583
2584cleanup:
2585 /* orphan all the subflows */
2586 inet_csk(sk)->icsk_mtup.probe_timestamp = tcp_jiffies32;
2587 list_for_each_entry(subflow, &mptcp_sk(sk)->conn_list, node) {
2588 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2589 bool slow, dispose_socket;
2590 struct socket *sock;
2591
2592 slow = lock_sock_fast(ssk);
2593 sock = ssk->sk_socket;
2594 dispose_socket = sock && sock != sk->sk_socket;
2595 sock_orphan(ssk);
2596 unlock_sock_fast(ssk, slow);
2597
2598 /* for the outgoing subflows we additionally need to free
2599 * the associated socket
2600 */
2601 if (dispose_socket)
2602 iput(SOCK_INODE(sock));
2603 }
2604 sock_orphan(sk);
2605
2606 sock_hold(sk);
2607 pr_debug("msk=%p state=%d", sk, sk->sk_state);
2608 if (sk->sk_state == TCP_CLOSE) {
2609 __mptcp_destroy_sock(sk);
2610 do_cancel_work = true;
2611 } else {
2612 sk_reset_timer(sk, &sk->sk_timer, jiffies + TCP_TIMEWAIT_LEN);
2613 }
2614 release_sock(sk);
2615 if (do_cancel_work)
2616 mptcp_cancel_work(sk);
2617 sock_put(sk);
f870fa0b
MM		 2618}
2619
cf7da0d6
PK		 2620static void mptcp_copy_inaddrs(struct sock *msk, const struct sock *ssk)
2621{
2622#if IS_ENABLED(CONFIG_MPTCP_IPV6)
2623 const struct ipv6_pinfo *ssk6 = inet6_sk(ssk);
2624 struct ipv6_pinfo *msk6 = inet6_sk(msk);
2625
2626 msk->sk_v6_daddr = ssk->sk_v6_daddr;
2627 msk->sk_v6_rcv_saddr = ssk->sk_v6_rcv_saddr;
2628
2629 if (msk6 && ssk6) {
2630 msk6->saddr = ssk6->saddr;
2631 msk6->flow_label = ssk6->flow_label;
2632 }
2633#endif
2634
2635 inet_sk(msk)->inet_num = inet_sk(ssk)->inet_num;
2636 inet_sk(msk)->inet_dport = inet_sk(ssk)->inet_dport;
2637 inet_sk(msk)->inet_sport = inet_sk(ssk)->inet_sport;
2638 inet_sk(msk)->inet_daddr = inet_sk(ssk)->inet_daddr;
2639 inet_sk(msk)->inet_saddr = inet_sk(ssk)->inet_saddr;
2640 inet_sk(msk)->inet_rcv_saddr = inet_sk(ssk)->inet_rcv_saddr;
2641}
2642
18b683bf
PA		 2643static int mptcp_disconnect(struct sock *sk, int flags)
2644{
76e2a55d
PA		 2645 struct mptcp_subflow_context *subflow;
2646 struct mptcp_sock *msk = mptcp_sk(sk);
2647
2648 __mptcp_flush_join_list(msk);
13a9499e
PA		 2649 mptcp_for_each_subflow(msk, subflow) {
2650 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2651
2652 lock_sock(ssk);
2653 tcp_disconnect(ssk, flags);
2654 release_sock(ssk);
2655 }
42c556fe 2656 return 0;
18b683bf
PA		 2657}
2658
b0519de8
FW		 2659#if IS_ENABLED(CONFIG_MPTCP_IPV6)
2660static struct ipv6_pinfo *mptcp_inet6_sk(const struct sock *sk)
2661{
2662 unsigned int offset = sizeof(struct mptcp6_sock) - sizeof(struct ipv6_pinfo);
2663
2664 return (struct ipv6_pinfo *)(((u8 *)sk) + offset);
2665}
2666#endif
2667
fca5c82c 2668struct sock *mptcp_sk_clone(const struct sock *sk,
cfde141e 2669 const struct mptcp_options_received *mp_opt,
fca5c82c 2670 struct request_sock *req)
b0519de8 2671{
58b09919 2672 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
b0519de8 2673 struct sock *nsk = sk_clone_lock(sk, GFP_ATOMIC);
58b09919
PA		 2674 struct mptcp_sock *msk;
2675 u64 ack_seq;
b0519de8
FW		 2676
2677 if (!nsk)
2678 return NULL;
2679
2680#if IS_ENABLED(CONFIG_MPTCP_IPV6)
2681 if (nsk->sk_family == AF_INET6)
2682 inet_sk(nsk)->pinet6 = mptcp_inet6_sk(nsk);
2683#endif
2684
58b09919
PA		 2685 __mptcp_init_sock(nsk);
2686
2687 msk = mptcp_sk(nsk);
2688 msk->local_key = subflow_req->local_key;
2689 msk->token = subflow_req->token;
2690 msk->subflow = NULL;
b93df08c 2691 WRITE_ONCE(msk->fully_established, false);
58b09919 2692
58b09919 2693 msk->write_seq = subflow_req->idsn + 1;
eaa2ffab 2694 msk->snd_nxt = msk->write_seq;
7439d687
PA		 2695 msk->snd_una = msk->write_seq;
2696 msk->wnd_end = msk->snd_nxt + req->rsk_rcv_wnd;
6f8a612a 2697
cfde141e 2698 if (mp_opt->mp_capable) {
58b09919 2699 msk->can_ack = true;
cfde141e 2700 msk->remote_key = mp_opt->sndr_key;
58b09919
PA		 2701 mptcp_crypto_key_sha(msk->remote_key, NULL, &ack_seq);
2702 ack_seq++;
917944da 2703 WRITE_ONCE(msk->ack_seq, ack_seq);
fa3fe2b1 2704 WRITE_ONCE(msk->rcv_wnd_sent, ack_seq);
58b09919 2705 }
7f20d5fc 2706
5e20087d 2707 sock_reset_flag(nsk, SOCK_RCU_FREE);
7f20d5fc
PA		 2708 /* will be fully established after successful MPC subflow creation */
2709 inet_sk_state_store(nsk, TCP_SYN_RECV);
0c148460
PA		 2710
2711 security_inet_csk_clone(nsk, req);
58b09919
PA		 2712 bh_unlock_sock(nsk);
2713
2714 /* keep a single reference */
2715 __sock_put(nsk);
b0519de8
FW		 2716 return nsk;
2717}
2718
a6b118fe
FW		 2719void mptcp_rcv_space_init(struct mptcp_sock *msk, const struct sock *ssk)
2720{
2721 const struct tcp_sock *tp = tcp_sk(ssk);
2722
2723 msk->rcvq_space.copied = 0;
2724 msk->rcvq_space.rtt_us = 0;
2725
2726 msk->rcvq_space.time = tp->tcp_mstamp;
2727
2728 /* initial rcv_space offering made to peer */
2729 msk->rcvq_space.space = min_t(u32, tp->rcv_wnd,
2730 TCP_INIT_CWND * tp->advmss);
2731 if (msk->rcvq_space.space == 0)
2732 msk->rcvq_space.space = TCP_INIT_CWND * TCP_MSS_DEFAULT;
6f8a612a 2733
7439d687 2734 WRITE_ONCE(msk->wnd_end, msk->snd_nxt + tcp_sk(ssk)->snd_wnd);
a6b118fe
FW		 2735}
2736
cf7da0d6
PK		 2737static struct sock *mptcp_accept(struct sock *sk, int flags, int *err,
2738 bool kern)
2739{
2740 struct mptcp_sock *msk = mptcp_sk(sk);
2741 struct socket *listener;
2742 struct sock *newsk;
2743
2744 listener = __mptcp_nmpc_socket(msk);
2745 if (WARN_ON_ONCE(!listener)) {
2746 *err = -EINVAL;
2747 return NULL;
2748 }
2749
2750 pr_debug("msk=%p, listener=%p", msk, mptcp_subflow_ctx(listener->sk));
2751 newsk = inet_csk_accept(listener->sk, flags, err, kern);
2752 if (!newsk)
2753 return NULL;
2754
2755 pr_debug("msk=%p, subflow is mptcp=%d", msk, sk_is_mptcp(newsk));
cf7da0d6
PK		 2756 if (sk_is_mptcp(newsk)) {
2757 struct mptcp_subflow_context *subflow;
2758 struct sock *new_mptcp_sock;
cf7da0d6
PK		 2759
2760 subflow = mptcp_subflow_ctx(newsk);
58b09919 2761 new_mptcp_sock = subflow->conn;
cf7da0d6 2762
58b09919
PA		 2763 /* is_mptcp should be false if subflow->conn is missing, see
2764 * subflow_syn_recv_sock()
2765 */
2766 if (WARN_ON_ONCE(!new_mptcp_sock)) {
2767 tcp_sk(newsk)->is_mptcp = 0;
2768 return newsk;
cf7da0d6
PK		 2769 }
2770
58b09919
PA		 2771 /* acquire the 2nd reference for the owning socket */
2772 sock_hold(new_mptcp_sock);
cf7da0d6 2773 newsk = new_mptcp_sock;
0397c6d8 2774 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPCAPABLEPASSIVEACK);
fc518953
FW		 2775 } else {
2776 MPTCP_INC_STATS(sock_net(sk),
2777 MPTCP_MIB_MPCAPABLEPASSIVEFALLBACK);
cf7da0d6
PK		 2778 }
2779
2780 return newsk;
2781}
2782
5c8c1640
GT		 2783void mptcp_destroy_common(struct mptcp_sock *msk)
2784{
87952603
PA		 2785 struct sock *sk = (struct sock *)msk;
2786
6e628cd3
PA		 2787 __mptcp_clear_xmit(sk);
2788
87952603
PA		 2789 /* move to sk_receive_queue, sk_stream_kill_queues will purge it */
2790 skb_queue_splice_tail_init(&msk->receive_queue, &sk->sk_receive_queue);
2791
5c8c1640
GT		 2792 skb_rbtree_purge(&msk->out_of_order_queue);
2793 mptcp_token_destroy(msk);
2794 mptcp_pm_free_anno_list(msk);
2795}
2796
79c0949e
PK		 2797static void mptcp_destroy(struct sock *sk)
2798{
c9fd9c5f
FW		 2799 struct mptcp_sock *msk = mptcp_sk(sk);
2800
5c8c1640 2801 mptcp_destroy_common(msk);
d027236c 2802 sk_sockets_allocated_dec(sk);
79c0949e
PK		 2803}
2804
fd1452d8 2805static int mptcp_setsockopt_sol_socket(struct mptcp_sock *msk, int optname,
a7b75c5a 2806 sockptr_t optval, unsigned int optlen)
fd1452d8
FW		 2807{
2808 struct sock *sk = (struct sock *)msk;
2809 struct socket *ssock;
2810 int ret;
2811
2812 switch (optname) {
2813 case SO_REUSEPORT:
2814 case SO_REUSEADDR:
2815 lock_sock(sk);
2816 ssock = __mptcp_nmpc_socket(msk);
2817 if (!ssock) {
2818 release_sock(sk);
2819 return -EINVAL;
2820 }
2821
a7b75c5a 2822 ret = sock_setsockopt(ssock, SOL_SOCKET, optname, optval, optlen);
fd1452d8
FW		 2823 if (ret == 0) {
2824 if (optname == SO_REUSEPORT)
2825 sk->sk_reuseport = ssock->sk->sk_reuseport;
2826 else if (optname == SO_REUSEADDR)
2827 sk->sk_reuse = ssock->sk->sk_reuse;
2828 }
2829 release_sock(sk);
2830 return ret;
2831 }
2832
a7b75c5a 2833 return sock_setsockopt(sk->sk_socket, SOL_SOCKET, optname, optval, optlen);
fd1452d8
FW		 2834}
2835
c9b95a13 2836static int mptcp_setsockopt_v6(struct mptcp_sock *msk, int optname,
a7b75c5a 2837 sockptr_t optval, unsigned int optlen)
c9b95a13
FW		 2838{
2839 struct sock *sk = (struct sock *)msk;
2840 int ret = -EOPNOTSUPP;
2841 struct socket *ssock;
2842
2843 switch (optname) {
2844 case IPV6_V6ONLY:
2845 lock_sock(sk);
2846 ssock = __mptcp_nmpc_socket(msk);
2847 if (!ssock) {
2848 release_sock(sk);
2849 return -EINVAL;
2850 }
2851
2852 ret = tcp_setsockopt(ssock->sk, SOL_IPV6, optname, optval, optlen);
2853 if (ret == 0)
2854 sk->sk_ipv6only = ssock->sk->sk_ipv6only;
2855
2856 release_sock(sk);
2857 break;
2858 }
2859
2860 return ret;
2861}
2862
717e79c8 2863static int mptcp_setsockopt(struct sock *sk, int level, int optname,
a7b75c5a 2864 sockptr_t optval, unsigned int optlen)
717e79c8
PK		 2865{
2866 struct mptcp_sock *msk = mptcp_sk(sk);
76660afb 2867 struct sock *ssk;
717e79c8
PK		 2868
2869 pr_debug("msk=%p", msk);
2870
83f0c10b 2871 if (level == SOL_SOCKET)
fd1452d8 2872 return mptcp_setsockopt_sol_socket(msk, optname, optval, optlen);
83f0c10b 2873
717e79c8 2874 /* @@ the meaning of setsockopt() when the socket is connected and
b6e4a1ae
MM		 2875 * there are multiple subflows is not yet defined. It is up to the
2876 * MPTCP-level socket to configure the subflows until the subflow
2877 * is in TCP fallback, when TCP socket options are passed through
2878 * to the one remaining subflow.
717e79c8
PK		 2879 */
2880 lock_sock(sk);
76660afb 2881 ssk = __mptcp_tcp_fallback(msk);
e154659b 2882 release_sock(sk);
76660afb
PA		 2883 if (ssk)
2884 return tcp_setsockopt(ssk, level, optname, optval, optlen);
50e741bb 2885
c9b95a13
FW		 2886 if (level == SOL_IPV6)
2887 return mptcp_setsockopt_v6(msk, optname, optval, optlen);
2888
b6e4a1ae 2889 return -EOPNOTSUPP;
717e79c8
PK		 2890}
2891
2892static int mptcp_getsockopt(struct sock *sk, int level, int optname,
50e741bb 2893 char __user *optval, int __user *option)
717e79c8
PK		 2894{
2895 struct mptcp_sock *msk = mptcp_sk(sk);
76660afb 2896 struct sock *ssk;
717e79c8
PK		 2897
2898 pr_debug("msk=%p", msk);
2899
b6e4a1ae
MM		 2900 /* @@ the meaning of setsockopt() when the socket is connected and
2901 * there are multiple subflows is not yet defined. It is up to the
2902 * MPTCP-level socket to configure the subflows until the subflow
2903 * is in TCP fallback, when socket options are passed through
2904 * to the one remaining subflow.
717e79c8
PK		 2905 */
2906 lock_sock(sk);
76660afb 2907 ssk = __mptcp_tcp_fallback(msk);
e154659b 2908 release_sock(sk);
76660afb
PA		 2909 if (ssk)
2910 return tcp_getsockopt(ssk, level, optname, optval, option);
50e741bb 2911
b6e4a1ae 2912 return -EOPNOTSUPP;
717e79c8
PK		 2913}
2914
6e628cd3
PA		 2915void __mptcp_data_acked(struct sock *sk)
2916{
2917 if (!sock_owned_by_user(sk))
2918 __mptcp_clean_una(sk);
2919 else
2920 set_bit(MPTCP_CLEAN_UNA, &mptcp_sk(sk)->flags);
2921
2922 if (mptcp_pending_data_fin_ack(sk))
2923 mptcp_schedule_work(sk);
2924}
2925
219d0499 2926void __mptcp_check_push(struct sock *sk, struct sock *ssk)
6e628cd3
PA		 2927{
2928 if (!mptcp_send_head(sk))
2929 return;
2930
2931 if (!sock_owned_by_user(sk))
2932 __mptcp_subflow_push_pending(sk, ssk);
2933 else
2934 set_bit(MPTCP_PUSH_PENDING, &mptcp_sk(sk)->flags);
2935}
2936
ea4ca586 2937#define MPTCP_DEFERRED_ALL (TCPF_WRITE_TIMER_DEFERRED)
14c441b5 2938
e93da928 2939/* processes deferred events and flush wmem */
14c441b5
PA		 2940static void mptcp_release_cb(struct sock *sk)
2941{
2942 unsigned long flags, nflags;
2943
6e628cd3
PA		 2944 /* push_pending may touch wmem_reserved, do it before the later
2945 * cleanup
2946 */
2947 if (test_and_clear_bit(MPTCP_CLEAN_UNA, &mptcp_sk(sk)->flags))
2948 __mptcp_clean_una(sk);
2949 if (test_and_clear_bit(MPTCP_PUSH_PENDING, &mptcp_sk(sk)->flags)) {
2950 /* mptcp_push_pending() acquires the subflow socket lock
2951 *
2952 * 1) can't be invoked in atomic scope
2953 * 2) must avoid ABBA deadlock with msk socket spinlock: the RX
2954 * datapath acquires the msk socket spinlock while helding
2955 * the subflow socket lock
2956 */
2957
2958 spin_unlock_bh(&sk->sk_lock.slock);
2959 mptcp_push_pending(sk, 0);
2960 spin_lock_bh(&sk->sk_lock.slock);
2961 }
15cc1045
PA		 2962 if (test_and_clear_bit(MPTCP_ERROR_REPORT, &mptcp_sk(sk)->flags))
2963 __mptcp_error_report(sk);
6e628cd3 2964
e93da928
PA		 2965 /* clear any wmem reservation and errors */
2966 __mptcp_update_wmem(sk);
87952603 2967 __mptcp_update_rmem(sk);
e93da928 2968
14c441b5
PA		 2969 do {
2970 flags = sk->sk_tsq_flags;
2971 if (!(flags & MPTCP_DEFERRED_ALL))
2972 return;
2973 nflags = flags & ~MPTCP_DEFERRED_ALL;
2974 } while (cmpxchg(&sk->sk_tsq_flags, flags, nflags) != flags);
2975
b51f9b80
PA		 2976 sock_release_ownership(sk);
2977
b51f9b80
PA		 2978 if (flags & TCPF_WRITE_TIMER_DEFERRED) {
2979 mptcp_retransmit_handler(sk);
2980 __sock_put(sk);
2981 }
14c441b5
PA		 2982}
2983
2c5ebd00
PA		 2984static int mptcp_hash(struct sock *sk)
2985{
2986 /* should never be called,
2987 * we hash the TCP subflows not the master socket
2988 */
2989 WARN_ON_ONCE(1);
2990 return 0;
2991}
2992
2993static void mptcp_unhash(struct sock *sk)
2994{
2995 /* called from sk_common_release(), but nothing to do here */
2996}
2997
cec37a6e 2998static int mptcp_get_port(struct sock *sk, unsigned short snum)
f870fa0b
MM		 2999{
3000 struct mptcp_sock *msk = mptcp_sk(sk);
cec37a6e 3001 struct socket *ssock;
f870fa0b 3002
cec37a6e
PK		 3003 ssock = __mptcp_nmpc_socket(msk);
3004 pr_debug("msk=%p, subflow=%p", msk, ssock);
3005 if (WARN_ON_ONCE(!ssock))
3006 return -EINVAL;
f870fa0b 3007
cec37a6e
PK		 3008 return inet_csk_get_port(ssock->sk, snum);
3009}
f870fa0b 3010
cec37a6e
PK		 3011void mptcp_finish_connect(struct sock *ssk)
3012{
3013 struct mptcp_subflow_context *subflow;
3014 struct mptcp_sock *msk;
3015 struct sock *sk;
6d0060f6 3016 u64 ack_seq;
f870fa0b 3017
cec37a6e 3018 subflow = mptcp_subflow_ctx(ssk);
cec37a6e
PK		 3019 sk = subflow->conn;
3020 msk = mptcp_sk(sk);
3021
648ef4b8
MM		 3022 pr_debug("msk=%p, token=%u", sk, subflow->token);
3023
6d0060f6
MM		 3024 mptcp_crypto_key_sha(subflow->remote_key, NULL, &ack_seq);
3025 ack_seq++;
648ef4b8
MM		 3026 subflow->map_seq = ack_seq;
3027 subflow->map_subflow_seq = 1;
6d0060f6 3028
cec37a6e
PK		 3029 /* the socket is not connected yet, no msk/subflow ops can access/race
3030 * accessing the field below
3031 */
3032 WRITE_ONCE(msk->remote_key, subflow->remote_key);
3033 WRITE_ONCE(msk->local_key, subflow->local_key);
6d0060f6 3034 WRITE_ONCE(msk->write_seq, subflow->idsn + 1);
eaa2ffab 3035 WRITE_ONCE(msk->snd_nxt, msk->write_seq);
6d0060f6 3036 WRITE_ONCE(msk->ack_seq, ack_seq);
fa3fe2b1 3037 WRITE_ONCE(msk->rcv_wnd_sent, ack_seq);
d22f4988 3038 WRITE_ONCE(msk->can_ack, 1);
7439d687 3039 WRITE_ONCE(msk->snd_una, msk->write_seq);
1b1c7a0e
PK		 3040
3041 mptcp_pm_new_connection(msk, 0);
a6b118fe
FW		 3042
3043 mptcp_rcv_space_init(msk, ssk);
f870fa0b
MM		 3044}
3045
cf7da0d6
PK		 3046static void mptcp_sock_graft(struct sock *sk, struct socket *parent)
3047{
3048 write_lock_bh(&sk->sk_callback_lock);
3049 rcu_assign_pointer(sk->sk_wq, &parent->wq);
3050 sk_set_socket(sk, parent);
3051 sk->sk_uid = SOCK_INODE(parent)->i_uid;
3052 write_unlock_bh(&sk->sk_callback_lock);
3053}
3054
e16163b6 3055bool mptcp_finish_join(struct sock *ssk)
f296234c 3056{
e16163b6 3057 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
f296234c
PK		 3058 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
3059 struct sock *parent = (void *)msk;
3060 struct socket *parent_sock;
ec3edaa7 3061 bool ret;
f296234c
PK		 3062
3063 pr_debug("msk=%p, subflow=%p", msk, subflow);
3064
3065 /* mptcp socket already closing? */
b93df08c 3066 if (!mptcp_is_fully_established(parent))
f296234c
PK		 3067 return false;
3068
3069 if (!msk->pm.server_side)
3070 return true;
3071
10f6d46c
PA		 3072 if (!mptcp_pm_allow_new_subflow(msk))
3073 return false;
3074
3075 /* active connections are already on conn_list, and we can't acquire
3076 * msk lock here.
3077 * use the join list lock as synchronization point and double-check
e16163b6 3078 * msk status to avoid racing with __mptcp_destroy_sock()
10f6d46c
PA		 3079 */
3080 spin_lock_bh(&msk->join_list_lock);
3081 ret = inet_sk_state_load(parent) == TCP_ESTABLISHED;
e16163b6 3082 if (ret && !WARN_ON_ONCE(!list_empty(&subflow->node))) {
10f6d46c 3083 list_add_tail(&subflow->node, &msk->join_list);
e16163b6
PA		 3084 sock_hold(ssk);
3085 }
10f6d46c
PA		 3086 spin_unlock_bh(&msk->join_list_lock);
3087 if (!ret)
3088 return false;
3089
3090 /* attach to msk socket only after we are sure he will deal with us
3091 * at close time
3092 */
f296234c 3093 parent_sock = READ_ONCE(parent->sk_socket);
e16163b6
PA		 3094 if (parent_sock && !ssk->sk_socket)
3095 mptcp_sock_graft(ssk, parent_sock);
917944da 3096 subflow->map_seq = READ_ONCE(msk->ack_seq);
10f6d46c 3097 return true;
f296234c
PK		 3098}
3099
76e2a55d
PA		 3100static void mptcp_shutdown(struct sock *sk, int how)
3101{
3102 pr_debug("sk=%p, how=%d", sk, how);
3103
3104 if ((how & SEND_SHUTDOWN) && mptcp_close_state(sk))
3105 __mptcp_wr_shutdown(sk);
3106}
3107
f870fa0b
MM		 3108static struct proto mptcp_prot = {
3109 .name = "MPTCP",
3110 .owner = THIS_MODULE,
3111 .init = mptcp_init_sock,
18b683bf 3112 .disconnect = mptcp_disconnect,
f870fa0b 3113 .close = mptcp_close,
cf7da0d6 3114 .accept = mptcp_accept,
717e79c8
PK		 3115 .setsockopt = mptcp_setsockopt,
3116 .getsockopt = mptcp_getsockopt,
76e2a55d 3117 .shutdown = mptcp_shutdown,
79c0949e 3118 .destroy = mptcp_destroy,
f870fa0b
MM		 3119 .sendmsg = mptcp_sendmsg,
3120 .recvmsg = mptcp_recvmsg,
14c441b5 3121 .release_cb = mptcp_release_cb,
2c5ebd00
PA		 3122 .hash = mptcp_hash,
3123 .unhash = mptcp_unhash,
cec37a6e 3124 .get_port = mptcp_get_port,
d027236c
PA		 3125 .sockets_allocated = &mptcp_sockets_allocated,
3126 .memory_allocated = &tcp_memory_allocated,
3127 .memory_pressure = &tcp_memory_pressure,
d027236c 3128 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_tcp_wmem),
989ef49b 3129 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_tcp_rmem),
d027236c 3130 .sysctl_mem = sysctl_tcp_mem,
f870fa0b 3131 .obj_size = sizeof(struct mptcp_sock),
2c5ebd00 3132 .slab_flags = SLAB_TYPESAFE_BY_RCU,
f870fa0b
MM		 3133 .no_autobind = true,
3134};
3135
2303f994
PK		 3136static int mptcp_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3137{
3138 struct mptcp_sock *msk = mptcp_sk(sock->sk);
3139 struct socket *ssock;
cf7da0d6 3140 int err;
2303f994
PK		 3141
3142 lock_sock(sock->sk);
fa68018d
PA		 3143 ssock = __mptcp_nmpc_socket(msk);
3144 if (!ssock) {
3145 err = -EINVAL;
2303f994
PK		 3146 goto unlock;
3147 }
3148
3149 err = ssock->ops->bind(ssock, uaddr, addr_len);
cf7da0d6
PK		 3150 if (!err)
3151 mptcp_copy_inaddrs(sock->sk, ssock->sk);
2303f994
PK		 3152
3153unlock:
3154 release_sock(sock->sk);
3155 return err;
3156}
3157
0235d075
PA		 3158static void mptcp_subflow_early_fallback(struct mptcp_sock *msk,
3159 struct mptcp_subflow_context *subflow)
3160{
3161 subflow->request_mptcp = 0;
3162 __mptcp_do_fallback(msk);
3163}
3164
2303f994
PK		 3165static int mptcp_stream_connect(struct socket *sock, struct sockaddr *uaddr,
3166 int addr_len, int flags)
3167{
3168 struct mptcp_sock *msk = mptcp_sk(sock->sk);
2c5ebd00 3169 struct mptcp_subflow_context *subflow;
2303f994
PK		 3170 struct socket *ssock;
3171 int err;
3172
3173 lock_sock(sock->sk);
41be81a8
PA		 3174 if (sock->state != SS_UNCONNECTED && msk->subflow) {
3175 /* pending connection or invalid state, let existing subflow
3176 * cope with that
3177 */
3178 ssock = msk->subflow;
3179 goto do_connect;
3180 }
3181
fa68018d
PA		 3182 ssock = __mptcp_nmpc_socket(msk);
3183 if (!ssock) {
3184 err = -EINVAL;
2303f994
PK		 3185 goto unlock;
3186 }
3187
fa68018d
PA		 3188 mptcp_token_destroy(msk);
3189 inet_sk_state_store(sock->sk, TCP_SYN_SENT);
2c5ebd00 3190 subflow = mptcp_subflow_ctx(ssock->sk);
cf7da0d6
PK		 3191#ifdef CONFIG_TCP_MD5SIG
3192 /* no MPTCP if MD5SIG is enabled on this socket or we may run out of
3193 * TCP option space.
3194 */
3195 if (rcu_access_pointer(tcp_sk(ssock->sk)->md5sig_info))
0235d075 3196 mptcp_subflow_early_fallback(msk, subflow);
cf7da0d6 3197#endif
2c5ebd00 3198 if (subflow->request_mptcp && mptcp_token_new_connect(ssock->sk))
0235d075 3199 mptcp_subflow_early_fallback(msk, subflow);
cf7da0d6 3200
41be81a8 3201do_connect:
2303f994 3202 err = ssock->ops->connect(ssock, uaddr, addr_len, flags);
41be81a8
PA		 3203 sock->state = ssock->state;
3204
3205 /* on successful connect, the msk state will be moved to established by
3206 * subflow_finish_connect()
3207 */
367fe04e 3208 if (!err || err == -EINPROGRESS)
41be81a8
PA		 3209 mptcp_copy_inaddrs(sock->sk, ssock->sk);
3210 else
3211 inet_sk_state_store(sock->sk, inet_sk_state_load(ssock->sk));
2303f994
PK		 3212
3213unlock:
3214 release_sock(sock->sk);
3215 return err;
3216}
3217
cf7da0d6
PK		 3218static int mptcp_listen(struct socket *sock, int backlog)
3219{
3220 struct mptcp_sock *msk = mptcp_sk(sock->sk);
3221 struct socket *ssock;
3222 int err;
3223
3224 pr_debug("msk=%p", msk);
3225
3226 lock_sock(sock->sk);
fa68018d
PA		 3227 ssock = __mptcp_nmpc_socket(msk);
3228 if (!ssock) {
3229 err = -EINVAL;
cf7da0d6
PK		 3230 goto unlock;
3231 }
3232
fa68018d
PA		 3233 mptcp_token_destroy(msk);
3234 inet_sk_state_store(sock->sk, TCP_LISTEN);
5e20087d
FW		 3235 sock_set_flag(sock->sk, SOCK_RCU_FREE);
3236
cf7da0d6
PK		 3237 err = ssock->ops->listen(ssock, backlog);
3238 inet_sk_state_store(sock->sk, inet_sk_state_load(ssock->sk));
3239 if (!err)
3240 mptcp_copy_inaddrs(sock->sk, ssock->sk);
3241
3242unlock:
3243 release_sock(sock->sk);
3244 return err;
3245}
3246
cf7da0d6
PK		 3247static int mptcp_stream_accept(struct socket *sock, struct socket *newsock,
3248 int flags, bool kern)
3249{
3250 struct mptcp_sock *msk = mptcp_sk(sock->sk);
3251 struct socket *ssock;
3252 int err;
3253
3254 pr_debug("msk=%p", msk);
3255
3256 lock_sock(sock->sk);
3257 if (sock->sk->sk_state != TCP_LISTEN)
3258 goto unlock_fail;
3259
3260 ssock = __mptcp_nmpc_socket(msk);
3261 if (!ssock)
3262 goto unlock_fail;
3263
8a05661b 3264 clear_bit(MPTCP_DATA_READY, &msk->flags);
cf7da0d6
PK		 3265 sock_hold(ssock->sk);
3266 release_sock(sock->sk);
3267
3268 err = ssock->ops->accept(sock, newsock, flags, kern);
d2f77c53 3269 if (err == 0 && !mptcp_is_tcpsk(newsock->sk)) {
cf7da0d6
PK		 3270 struct mptcp_sock *msk = mptcp_sk(newsock->sk);
3271 struct mptcp_subflow_context *subflow;
0397c6d8
PA		 3272 struct sock *newsk = newsock->sk;
3273 bool slowpath;
3274
3275 slowpath = lock_sock_fast(newsk);
5b950ff4
PA		 3276
3277 /* PM/worker can now acquire the first subflow socket
3278 * lock without racing with listener queue cleanup,
3279 * we can notify it, if needed.
3280 */
3281 subflow = mptcp_subflow_ctx(msk->first);
3282 list_add(&subflow->node, &msk->conn_list);
3283 sock_hold(msk->first);
3284 if (mptcp_is_fully_established(newsk))
3285 mptcp_pm_fully_established(msk);
3286
0397c6d8
PA		 3287 mptcp_copy_inaddrs(newsk, msk->first);
3288 mptcp_rcv_space_init(msk, msk->first);
cf7da0d6
PK		 3289
3290 /* set ssk->sk_socket of accept()ed flows to mptcp socket.
3291 * This is needed so NOSPACE flag can be set from tcp stack.
3292 */
ec3edaa7 3293 __mptcp_flush_join_list(msk);
190f8b06 3294 mptcp_for_each_subflow(msk, subflow) {
cf7da0d6
PK		 3295 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
3296
3297 if (!ssk->sk_socket)
3298 mptcp_sock_graft(ssk, newsock);
3299 }
0397c6d8 3300 unlock_sock_fast(newsk, slowpath);
cf7da0d6
PK		 3301 }
3302
8a05661b
PA		 3303 if (inet_csk_listen_poll(ssock->sk))
3304 set_bit(MPTCP_DATA_READY, &msk->flags);
cf7da0d6
PK		 3305 sock_put(ssock->sk);
3306 return err;
3307
3308unlock_fail:
3309 release_sock(sock->sk);
3310 return -EINVAL;
3311}
3312
8a05661b
PA		 3313static __poll_t mptcp_check_readable(struct mptcp_sock *msk)
3314{
3315 return test_bit(MPTCP_DATA_READY, &msk->flags) ? EPOLLIN | EPOLLRDNORM :
3316 0;
3317}
3318
8edf0864
FW		 3319static __poll_t mptcp_check_writeable(struct mptcp_sock *msk)
3320{
3321 struct sock *sk = (struct sock *)msk;
8edf0864
FW		 3322
3323 if (unlikely(sk->sk_shutdown & SEND_SHUTDOWN))
3324 return 0;
3325
3326 if (sk_stream_is_writeable(sk))
3327 return EPOLLOUT | EPOLLWRNORM;
3328
6e628cd3
PA		 3329 set_bit(MPTCP_NOSPACE, &msk->flags);
3330 smp_mb__after_atomic(); /* msk->flags is changed by write_space cb */
3331 if (sk_stream_is_writeable(sk))
3332 return EPOLLOUT | EPOLLWRNORM;
8edf0864 3333
6e628cd3 3334 return 0;
8edf0864
FW		 3335}
3336
2303f994
PK		 3337static __poll_t mptcp_poll(struct file *file, struct socket *sock,
3338 struct poll_table_struct *wait)
3339{
1891c4a0 3340 struct sock *sk = sock->sk;
8ab183de 3341 struct mptcp_sock *msk;
2303f994 3342 __poll_t mask = 0;
8a05661b 3343 int state;
2303f994 3344
1891c4a0 3345 msk = mptcp_sk(sk);
1891c4a0 3346 sock_poll_wait(file, sock, wait);
1891c4a0 3347
8a05661b 3348 state = inet_sk_state_load(sk);
6719331c 3349 pr_debug("msk=%p state=%d flags=%lx", msk, state, msk->flags);
8a05661b
PA		 3350 if (state == TCP_LISTEN)
3351 return mptcp_check_readable(msk);
3352
3353 if (state != TCP_SYN_SENT && state != TCP_SYN_RECV) {
3354 mask |= mptcp_check_readable(msk);
8edf0864 3355 mask |= mptcp_check_writeable(msk);
8a05661b 3356 }
1891c4a0
FW		 3357 if (sk->sk_shutdown & RCV_SHUTDOWN)
3358 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
3359
15cc1045
PA		 3360 /* This barrier is coupled with smp_wmb() in tcp_reset() */
3361 smp_rmb();
3362 if (sk->sk_err)
3363 mask |= EPOLLERR;
3364
2303f994
PK		 3365 return mask;
3366}
3367
e42f1ac6
FW		 3368static const struct proto_ops mptcp_stream_ops = {
3369 .family = PF_INET,
3370 .owner = THIS_MODULE,
3371 .release = inet_release,
3372 .bind = mptcp_bind,
3373 .connect = mptcp_stream_connect,
3374 .socketpair = sock_no_socketpair,
3375 .accept = mptcp_stream_accept,
d2f77c53 3376 .getname = inet_getname,
e42f1ac6
FW		 3377 .poll = mptcp_poll,
3378 .ioctl = inet_ioctl,
3379 .gettstamp = sock_gettstamp,
3380 .listen = mptcp_listen,
76e2a55d 3381 .shutdown = inet_shutdown,
e42f1ac6
FW		 3382 .setsockopt = sock_common_setsockopt,
3383 .getsockopt = sock_common_getsockopt,
3384 .sendmsg = inet_sendmsg,
3385 .recvmsg = inet_recvmsg,
3386 .mmap = sock_no_mmap,
3387 .sendpage = inet_sendpage,
e42f1ac6 3388};
2303f994 3389
f870fa0b
MM		 3390static struct inet_protosw mptcp_protosw = {
3391 .type = SOCK_STREAM,
3392 .protocol = IPPROTO_MPTCP,
3393 .prot = &mptcp_prot,
2303f994
PK		 3394 .ops = &mptcp_stream_ops,
3395 .flags = INET_PROTOSW_ICSK,
f870fa0b
MM		 3396};
3397
d39dceca 3398void __init mptcp_proto_init(void)
f870fa0b 3399{
2303f994 3400 mptcp_prot.h.hashinfo = tcp_prot.h.hashinfo;
2303f994 3401
d027236c
PA		 3402 if (percpu_counter_init(&mptcp_sockets_allocated, 0, GFP_KERNEL))
3403 panic("Failed to allocate MPTCP pcpu counter\n");
3404
2303f994 3405 mptcp_subflow_init();
1b1c7a0e 3406 mptcp_pm_init();
2c5ebd00 3407 mptcp_token_init();
2303f994 3408
f870fa0b
MM		 3409 if (proto_register(&mptcp_prot, 1) != 0)
3410 panic("Failed to register MPTCP proto.\n");
3411
3412 inet_register_protosw(&mptcp_protosw);
6771bfd9
FW		 3413
3414 BUILD_BUG_ON(sizeof(struct mptcp_skb_cb) > sizeof_field(struct sk_buff, cb));
f870fa0b
MM		 3415}
3416
3417#if IS_ENABLED(CONFIG_MPTCP_IPV6)
e42f1ac6
FW		 3418static const struct proto_ops mptcp_v6_stream_ops = {
3419 .family = PF_INET6,
3420 .owner = THIS_MODULE,
3421 .release = inet6_release,
3422 .bind = mptcp_bind,
3423 .connect = mptcp_stream_connect,
3424 .socketpair = sock_no_socketpair,
3425 .accept = mptcp_stream_accept,
d2f77c53 3426 .getname = inet6_getname,
e42f1ac6
FW		 3427 .poll = mptcp_poll,
3428 .ioctl = inet6_ioctl,
3429 .gettstamp = sock_gettstamp,
3430 .listen = mptcp_listen,
76e2a55d 3431 .shutdown = inet_shutdown,
e42f1ac6
FW		 3432 .setsockopt = sock_common_setsockopt,
3433 .getsockopt = sock_common_getsockopt,
3434 .sendmsg = inet6_sendmsg,
3435 .recvmsg = inet6_recvmsg,
3436 .mmap = sock_no_mmap,
3437 .sendpage = inet_sendpage,
3438#ifdef CONFIG_COMPAT
3986912f 3439 .compat_ioctl = inet6_compat_ioctl,
e42f1ac6
FW		 3440#endif
3441};
3442
f870fa0b
MM		 3443static struct proto mptcp_v6_prot;
3444
79c0949e
PK		 3445static void mptcp_v6_destroy(struct sock *sk)
3446{
3447 mptcp_destroy(sk);
3448 inet6_destroy_sock(sk);
3449}
3450
f870fa0b
MM		 3451static struct inet_protosw mptcp_v6_protosw = {
3452 .type = SOCK_STREAM,
3453 .protocol = IPPROTO_MPTCP,
3454 .prot = &mptcp_v6_prot,
2303f994 3455 .ops = &mptcp_v6_stream_ops,
f870fa0b
MM		 3456 .flags = INET_PROTOSW_ICSK,
3457};
3458
d39dceca 3459int __init mptcp_proto_v6_init(void)
f870fa0b
MM		 3460{
3461 int err;
3462
3463 mptcp_v6_prot = mptcp_prot;
3464 strcpy(mptcp_v6_prot.name, "MPTCPv6");
3465 mptcp_v6_prot.slab = NULL;
79c0949e 3466 mptcp_v6_prot.destroy = mptcp_v6_destroy;
b0519de8 3467 mptcp_v6_prot.obj_size = sizeof(struct mptcp6_sock);
f870fa0b
MM		 3468
3469 err = proto_register(&mptcp_v6_prot, 1);
3470 if (err)
3471 return err;
3472
3473 err = inet6_register_protosw(&mptcp_v6_protosw);
3474 if (err)
3475 proto_unregister(&mptcp_v6_prot);
3476
3477 return err;
3478}
3479#endif
Linux 6.x block layer and io_uring tree(s)