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