projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
mptcp: remove tx_pending_data
[linux-2.6-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
e10a9892
GT		 28#define CREATE_TRACE_POINTS
29#include <trace/events/mptcp.h>
30
b0519de8
FW		 31#if IS_ENABLED(CONFIG_MPTCP_IPV6)
32struct mptcp6_sock {
33 struct mptcp_sock msk;
34 struct ipv6_pinfo np;
35};
36#endif
37
6771bfd9 38struct mptcp_skb_cb {
ab174ad8
PA		 39 u64 map_seq;
40 u64 end_seq;
6771bfd9 41 u32 offset;
b7f653b2 42 u8 has_rxtstamp:1;
6771bfd9
FW		 43};
44
45#define MPTCP_SKB_CB(__skb) ((struct mptcp_skb_cb *)&((__skb)->cb[0]))
46
b7f653b2
FW		 47enum {
48 MPTCP_CMSG_TS = BIT(0),
49};
50
d027236c
PA		 51static struct percpu_counter mptcp_sockets_allocated;
52
e16163b6 53static void __mptcp_destroy_sock(struct sock *sk);
d9ca1de8 54static void __mptcp_check_send_data_fin(struct sock *sk);
e16163b6 55
b19bc294
PA		 56DEFINE_PER_CPU(struct mptcp_delegated_action, mptcp_delegated_actions);
57static struct net_device mptcp_napi_dev;
58
2303f994
PK		 59/* If msk has an initial subflow socket, and the MP_CAPABLE handshake has not
60 * completed yet or has failed, return the subflow socket.
61 * Otherwise return NULL.
62 */
1729cf18 63struct socket *__mptcp_nmpc_socket(const struct mptcp_sock *msk)
2303f994 64{
d22f4988 65 if (!msk->subflow || READ_ONCE(msk->can_ack))
2303f994
PK		 66 return NULL;
67
68 return msk->subflow;
69}
70
6f8a612a
FW		 71/* Returns end sequence number of the receiver's advertised window */
72static u64 mptcp_wnd_end(const struct mptcp_sock *msk)
73{
7439d687 74 return READ_ONCE(msk->wnd_end);
6f8a612a
FW		 75}
76
d2f77c53 77static bool mptcp_is_tcpsk(struct sock *sk)
0b4f33de
FW		 78{
79 struct socket *sock = sk->sk_socket;
80
0b4f33de
FW		 81 if (unlikely(sk->sk_prot == &tcp_prot)) {
82 /* we are being invoked after mptcp_accept() has
83 * accepted a non-mp-capable flow: sk is a tcp_sk,
84 * not an mptcp one.
85 *
86 * Hand the socket over to tcp so all further socket ops
87 * bypass mptcp.
88 */
89 sock->ops = &inet_stream_ops;
d2f77c53 90 return true;
0b4f33de
FW		 91#if IS_ENABLED(CONFIG_MPTCP_IPV6)
92 } else if (unlikely(sk->sk_prot == &tcpv6_prot)) {
93 sock->ops = &inet6_stream_ops;
d2f77c53 94 return true;
0b4f33de
FW		 95#endif
96 }
97
d2f77c53 98 return false;
0b4f33de
FW		 99}
100
fa68018d 101static int __mptcp_socket_create(struct mptcp_sock *msk)
2303f994
PK		 102{
103 struct mptcp_subflow_context *subflow;
104 struct sock *sk = (struct sock *)msk;
105 struct socket *ssock;
106 int err;
107
2303f994
PK		 108 err = mptcp_subflow_create_socket(sk, &ssock);
109 if (err)
fa68018d 110 return err;
2303f994 111
8ab183de 112 msk->first = ssock->sk;
2303f994
PK		 113 msk->subflow = ssock;
114 subflow = mptcp_subflow_ctx(ssock->sk);
cec37a6e 115 list_add(&subflow->node, &msk->conn_list);
e16163b6 116 sock_hold(ssock->sk);
2303f994 117 subflow->request_mptcp = 1;
866f26f2 118 mptcp_sock_graft(msk->first, sk->sk_socket);
e1ff9e82 119
fa68018d 120 return 0;
2303f994
PK		 121}
122
ab174ad8
PA		 123static void mptcp_drop(struct sock *sk, struct sk_buff *skb)
124{
125 sk_drops_add(sk, skb);
126 __kfree_skb(skb);
127}
128
8268ed4c
PA		 129static bool mptcp_try_coalesce(struct sock *sk, struct sk_buff *to,
130 struct sk_buff *from)
131{
132 bool fragstolen;
133 int delta;
134
135 if (MPTCP_SKB_CB(from)->offset ||
136 !skb_try_coalesce(to, from, &fragstolen, &delta))
137 return false;
138
06242e44
PA		 139 pr_debug("colesced seq %llx into %llx new len %d new end seq %llx",
140 MPTCP_SKB_CB(from)->map_seq, MPTCP_SKB_CB(to)->map_seq,
141 to->len, MPTCP_SKB_CB(from)->end_seq);
ab174ad8 142 MPTCP_SKB_CB(to)->end_seq = MPTCP_SKB_CB(from)->end_seq;
8268ed4c
PA		 143 kfree_skb_partial(from, fragstolen);
144 atomic_add(delta, &sk->sk_rmem_alloc);
145 sk_mem_charge(sk, delta);
146 return true;
147}
148
ab174ad8
PA		 149static bool mptcp_ooo_try_coalesce(struct mptcp_sock *msk, struct sk_buff *to,
150 struct sk_buff *from)
151{
152 if (MPTCP_SKB_CB(from)->map_seq != MPTCP_SKB_CB(to)->end_seq)
153 return false;
154
155 return mptcp_try_coalesce((struct sock *)msk, to, from);
156}
157
158/* "inspired" by tcp_data_queue_ofo(), main differences:
159 * - use mptcp seqs
160 * - don't cope with sacks
161 */
162static void mptcp_data_queue_ofo(struct mptcp_sock *msk, struct sk_buff *skb)
163{
164 struct sock *sk = (struct sock *)msk;
165 struct rb_node **p, *parent;
166 u64 seq, end_seq, max_seq;
167 struct sk_buff *skb1;
168
169 seq = MPTCP_SKB_CB(skb)->map_seq;
170 end_seq = MPTCP_SKB_CB(skb)->end_seq;
fa3fe2b1 171 max_seq = READ_ONCE(msk->rcv_wnd_sent);
ab174ad8 172
06242e44
PA		 173 pr_debug("msk=%p seq=%llx limit=%llx empty=%d", msk, seq, max_seq,
174 RB_EMPTY_ROOT(&msk->out_of_order_queue));
fa3fe2b1 175 if (after64(end_seq, max_seq)) {
ab174ad8
PA		 176 /* out of window */
177 mptcp_drop(sk, skb);
fa3fe2b1
FW		 178 pr_debug("oow by %lld, rcv_wnd_sent %llu\n",
179 (unsigned long long)end_seq - (unsigned long)max_seq,
180 (unsigned long long)msk->rcv_wnd_sent);
06242e44 181 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_NODSSWINDOW);
ab174ad8
PA		 182 return;
183 }
184
185 p = &msk->out_of_order_queue.rb_node;
06242e44 186 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOQUEUE);
ab174ad8
PA		 187 if (RB_EMPTY_ROOT(&msk->out_of_order_queue)) {
188 rb_link_node(&skb->rbnode, NULL, p);
189 rb_insert_color(&skb->rbnode, &msk->out_of_order_queue);
190 msk->ooo_last_skb = skb;
191 goto end;
192 }
193
194 /* with 2 subflows, adding at end of ooo queue is quite likely
195 * Use of ooo_last_skb avoids the O(Log(N)) rbtree lookup.
196 */
06242e44
PA		 197 if (mptcp_ooo_try_coalesce(msk, msk->ooo_last_skb, skb)) {
198 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOMERGE);
199 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOQUEUETAIL);
ab174ad8 200 return;
06242e44 201 }
ab174ad8
PA		 202
203 /* Can avoid an rbtree lookup if we are adding skb after ooo_last_skb */
204 if (!before64(seq, MPTCP_SKB_CB(msk->ooo_last_skb)->end_seq)) {
06242e44 205 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOQUEUETAIL);
ab174ad8
PA		 206 parent = &msk->ooo_last_skb->rbnode;
207 p = &parent->rb_right;
208 goto insert;
209 }
210
211 /* Find place to insert this segment. Handle overlaps on the way. */
212 parent = NULL;
213 while (*p) {
214 parent = *p;
215 skb1 = rb_to_skb(parent);
216 if (before64(seq, MPTCP_SKB_CB(skb1)->map_seq)) {
217 p = &parent->rb_left;
218 continue;
219 }
220 if (before64(seq, MPTCP_SKB_CB(skb1)->end_seq)) {
221 if (!after64(end_seq, MPTCP_SKB_CB(skb1)->end_seq)) {
222 /* All the bits are present. Drop. */
223 mptcp_drop(sk, skb);
06242e44 224 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
ab174ad8
PA		 225 return;
226 }
227 if (after64(seq, MPTCP_SKB_CB(skb1)->map_seq)) {
228 /* partial overlap:
229 * | skb |
230 * | skb1 |
231 * continue traversing
232 */
233 } else {
234 /* skb's seq == skb1's seq and skb covers skb1.
235 * Replace skb1 with skb.
236 */
237 rb_replace_node(&skb1->rbnode, &skb->rbnode,
238 &msk->out_of_order_queue);
239 mptcp_drop(sk, skb1);
06242e44 240 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
ab174ad8
PA		 241 goto merge_right;
242 }
243 } else if (mptcp_ooo_try_coalesce(msk, skb1, skb)) {
06242e44 244 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOMERGE);
ab174ad8
PA		 245 return;
246 }
247 p = &parent->rb_right;
248 }
06242e44 249
ab174ad8
PA		 250insert:
251 /* Insert segment into RB tree. */
252 rb_link_node(&skb->rbnode, parent, p);
253 rb_insert_color(&skb->rbnode, &msk->out_of_order_queue);
254
255merge_right:
256 /* Remove other segments covered by skb. */
257 while ((skb1 = skb_rb_next(skb)) != NULL) {
258 if (before64(end_seq, MPTCP_SKB_CB(skb1)->end_seq))
259 break;
260 rb_erase(&skb1->rbnode, &msk->out_of_order_queue);
261 mptcp_drop(sk, skb1);
06242e44 262 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
ab174ad8
PA		 263 }
264 /* If there is no skb after us, we are the last_skb ! */
265 if (!skb1)
266 msk->ooo_last_skb = skb;
267
268end:
269 skb_condense(skb);
270 skb_set_owner_r(skb, sk);
271}
272
273static bool __mptcp_move_skb(struct mptcp_sock *msk, struct sock *ssk,
274 struct sk_buff *skb, unsigned int offset,
275 size_t copy_len)
6771bfd9 276{
ab174ad8 277 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
6771bfd9 278 struct sock *sk = (struct sock *)msk;
4e637c70 279 struct sk_buff *tail;
b7f653b2 280 bool has_rxtstamp;
6771bfd9
FW		 281
282 __skb_unlink(skb, &ssk->sk_receive_queue);
6771bfd9 283
4e637c70
FW		 284 skb_ext_reset(skb);
285 skb_orphan(skb);
ab174ad8 286
9c3f94e1
PA		 287 /* try to fetch required memory from subflow */
288 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
72f96132
PA		 289 int amount = sk_mem_pages(skb->truesize) << SK_MEM_QUANTUM_SHIFT;
290
291 if (ssk->sk_forward_alloc < amount)
9c3f94e1 292 goto drop;
72f96132
PA		 293
294 ssk->sk_forward_alloc -= amount;
295 sk->sk_forward_alloc += amount;
9c3f94e1
PA		 296 }
297
b7f653b2
FW		 298 has_rxtstamp = TCP_SKB_CB(skb)->has_rxtstamp;
299
ab174ad8
PA		 300 /* the skb map_seq accounts for the skb offset:
301 * mptcp_subflow_get_mapped_dsn() is based on the current tp->copied_seq
302 * value
303 */
304 MPTCP_SKB_CB(skb)->map_seq = mptcp_subflow_get_mapped_dsn(subflow);
305 MPTCP_SKB_CB(skb)->end_seq = MPTCP_SKB_CB(skb)->map_seq + copy_len;
8268ed4c 306 MPTCP_SKB_CB(skb)->offset = offset;
b7f653b2 307 MPTCP_SKB_CB(skb)->has_rxtstamp = has_rxtstamp;
4e637c70 308
ab174ad8
PA		 309 if (MPTCP_SKB_CB(skb)->map_seq == msk->ack_seq) {
310 /* in sequence */
8b0308fe 311 WRITE_ONCE(msk->ack_seq, msk->ack_seq + copy_len);
ab174ad8
PA		 312 tail = skb_peek_tail(&sk->sk_receive_queue);
313 if (tail && mptcp_try_coalesce(sk, tail, skb))
314 return true;
4e637c70 315
ab174ad8
PA		 316 skb_set_owner_r(skb, sk);
317 __skb_queue_tail(&sk->sk_receive_queue, skb);
318 return true;
319 } else if (after64(MPTCP_SKB_CB(skb)->map_seq, msk->ack_seq)) {
320 mptcp_data_queue_ofo(msk, skb);
321 return false;
322 }
323
324 /* old data, keep it simple and drop the whole pkt, sender
325 * will retransmit as needed, if needed.
326 */
06242e44 327 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
9c3f94e1 328drop:
ab174ad8
PA		 329 mptcp_drop(sk, skb);
330 return false;
6771bfd9
FW		 331}
332
16a9a9da
MM		 333static void mptcp_stop_timer(struct sock *sk)
334{
335 struct inet_connection_sock *icsk = inet_csk(sk);
336
337 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
338 mptcp_sk(sk)->timer_ival = 0;
339}
340
e16163b6
PA		 341static void mptcp_close_wake_up(struct sock *sk)
342{
343 if (sock_flag(sk, SOCK_DEAD))
344 return;
345
346 sk->sk_state_change(sk);
347 if (sk->sk_shutdown == SHUTDOWN_MASK ||
348 sk->sk_state == TCP_CLOSE)
349 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_HUP);
350 else
351 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
352}
353
6e628cd3 354static bool mptcp_pending_data_fin_ack(struct sock *sk)
16a9a9da
MM		 355{
356 struct mptcp_sock *msk = mptcp_sk(sk);
357
6e628cd3
PA		 358 return !__mptcp_check_fallback(msk) &&
359 ((1 << sk->sk_state) &
360 (TCPF_FIN_WAIT1 | TCPF_CLOSING | TCPF_LAST_ACK)) &&
361 msk->write_seq == READ_ONCE(msk->snd_una);
362}
363
364static void mptcp_check_data_fin_ack(struct sock *sk)
365{
366 struct mptcp_sock *msk = mptcp_sk(sk);
16a9a9da
MM		 367
368 /* Look for an acknowledged DATA_FIN */
6e628cd3 369 if (mptcp_pending_data_fin_ack(sk)) {
16a9a9da
MM		 370 WRITE_ONCE(msk->snd_data_fin_enable, 0);
371
372 switch (sk->sk_state) {
373 case TCP_FIN_WAIT1:
374 inet_sk_state_store(sk, TCP_FIN_WAIT2);
16a9a9da
MM		 375 break;
376 case TCP_CLOSING:
16a9a9da
MM		 377 case TCP_LAST_ACK:
378 inet_sk_state_store(sk, TCP_CLOSE);
16a9a9da
MM		 379 break;
380 }
381
e16163b6 382 mptcp_close_wake_up(sk);
16a9a9da
MM		 383 }
384}
385
3721b9b6
MM		 386static bool mptcp_pending_data_fin(struct sock *sk, u64 *seq)
387{
388 struct mptcp_sock *msk = mptcp_sk(sk);
389
390 if (READ_ONCE(msk->rcv_data_fin) &&
391 ((1 << sk->sk_state) &
392 (TCPF_ESTABLISHED | TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2))) {
393 u64 rcv_data_fin_seq = READ_ONCE(msk->rcv_data_fin_seq);
394
395 if (msk->ack_seq == rcv_data_fin_seq) {
396 if (seq)
397 *seq = rcv_data_fin_seq;
398
399 return true;
400 }
401 }
402
403 return false;
404}
405
6477dd39
MM		 406static void mptcp_set_datafin_timeout(const struct sock *sk)
407{
408 struct inet_connection_sock *icsk = inet_csk(sk);
409
410 mptcp_sk(sk)->timer_ival = min(TCP_RTO_MAX,
411 TCP_RTO_MIN << icsk->icsk_retransmits);
412}
413
33d41c9c 414static void __mptcp_set_timeout(struct sock *sk, long tout)
3721b9b6 415{
3721b9b6
MM		 416 mptcp_sk(sk)->timer_ival = tout > 0 ? tout : TCP_RTO_MIN;
417}
418
33d41c9c
PA		 419static long mptcp_timeout_from_subflow(const struct mptcp_subflow_context *subflow)
420{
421 const struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
422
71b7dec2
PA		 423 return inet_csk(ssk)->icsk_pending && !subflow->stale_count ?
424 inet_csk(ssk)->icsk_timeout - jiffies : 0;
33d41c9c
PA		 425}
426
427static void mptcp_set_timeout(struct sock *sk)
428{
429 struct mptcp_subflow_context *subflow;
430 long tout = 0;
431
432 mptcp_for_each_subflow(mptcp_sk(sk), subflow)
433 tout = max(tout, mptcp_timeout_from_subflow(subflow));
434 __mptcp_set_timeout(sk, tout);
435}
436
fd897679
PA		 437static bool tcp_can_send_ack(const struct sock *ssk)
438{
439 return !((1 << inet_sk_state_load(ssk)) &
20bc80b6 440 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_TIME_WAIT | TCPF_CLOSE | TCPF_LISTEN));
fd897679
PA		 441}
442
340fa666
MM		 443void mptcp_subflow_send_ack(struct sock *ssk)
444{
445 bool slow;
446
447 slow = lock_sock_fast(ssk);
448 if (tcp_can_send_ack(ssk))
449 tcp_send_ack(ssk);
450 unlock_sock_fast(ssk, slow);
451}
452
fd897679 453static void mptcp_send_ack(struct mptcp_sock *msk)
7ed90803
PA		 454{
455 struct mptcp_subflow_context *subflow;
456
340fa666
MM		 457 mptcp_for_each_subflow(msk, subflow)
458 mptcp_subflow_send_ack(mptcp_subflow_tcp_sock(subflow));
fd897679
PA		 459}
460
fde56eea 461static void mptcp_subflow_cleanup_rbuf(struct sock *ssk)
fd897679 462{
75e908c3 463 bool slow;
fd897679 464
75e908c3 465 slow = lock_sock_fast(ssk);
fde56eea 466 if (tcp_can_send_ack(ssk))
fd897679 467 tcp_cleanup_rbuf(ssk, 1);
75e908c3 468 unlock_sock_fast(ssk, slow);
fde56eea
PA		 469}
470
471static bool mptcp_subflow_could_cleanup(const struct sock *ssk, bool rx_empty)
472{
473 const struct inet_connection_sock *icsk = inet_csk(ssk);
c4512c63 474 u8 ack_pending = READ_ONCE(icsk->icsk_ack.pending);
fde56eea
PA		 475 const struct tcp_sock *tp = tcp_sk(ssk);
476
477 return (ack_pending & ICSK_ACK_SCHED) &&
478 ((READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->rcv_wup) >
479 READ_ONCE(icsk->icsk_ack.rcv_mss)) ||
480 (rx_empty && ack_pending &
481 (ICSK_ACK_PUSHED2 | ICSK_ACK_PUSHED)));
fd897679
PA		 482}
483
484static void mptcp_cleanup_rbuf(struct mptcp_sock *msk)
485{
e3859603 486 int old_space = READ_ONCE(msk->old_wspace);
fd897679 487 struct mptcp_subflow_context *subflow;
e3859603 488 struct sock *sk = (struct sock *)msk;
fde56eea
PA		 489 int space = __mptcp_space(sk);
490 bool cleanup, rx_empty;
e3859603 491
fde56eea 492 cleanup = (space > 0) && (space >= (old_space << 1));
ce599c51 493 rx_empty = !__mptcp_rmem(sk);
fd897679 494
fde56eea
PA		 495 mptcp_for_each_subflow(msk, subflow) {
496 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
fd897679 497
fde56eea
PA		 498 if (cleanup || mptcp_subflow_could_cleanup(ssk, rx_empty))
499 mptcp_subflow_cleanup_rbuf(ssk);
7ed90803
PA		 500 }
501}
502
503static bool mptcp_check_data_fin(struct sock *sk)
3721b9b6
MM		 504{
505 struct mptcp_sock *msk = mptcp_sk(sk);
506 u64 rcv_data_fin_seq;
7ed90803 507 bool ret = false;
3721b9b6 508
781bf13d 509 if (__mptcp_check_fallback(msk))
7ed90803 510 return ret;
3721b9b6
MM		 511
512 /* Need to ack a DATA_FIN received from a peer while this side
513 * of the connection is in ESTABLISHED, FIN_WAIT1, or FIN_WAIT2.
514 * msk->rcv_data_fin was set when parsing the incoming options
515 * at the subflow level and the msk lock was not held, so this
516 * is the first opportunity to act on the DATA_FIN and change
517 * the msk state.
518 *
519 * If we are caught up to the sequence number of the incoming
520 * DATA_FIN, send the DATA_ACK now and do state transition. If
521 * not caught up, do nothing and let the recv code send DATA_ACK
522 * when catching up.
523 */
524
525 if (mptcp_pending_data_fin(sk, &rcv_data_fin_seq)) {
917944da 526 WRITE_ONCE(msk->ack_seq, msk->ack_seq + 1);
3721b9b6
MM		 527 WRITE_ONCE(msk->rcv_data_fin, 0);
528
529 sk->sk_shutdown |= RCV_SHUTDOWN;
16a9a9da
MM		 530 smp_mb__before_atomic(); /* SHUTDOWN must be visible first */
531 set_bit(MPTCP_DATA_READY, &msk->flags);
3721b9b6
MM		 532
533 switch (sk->sk_state) {
534 case TCP_ESTABLISHED:
535 inet_sk_state_store(sk, TCP_CLOSE_WAIT);
536 break;
537 case TCP_FIN_WAIT1:
538 inet_sk_state_store(sk, TCP_CLOSING);
539 break;
540 case TCP_FIN_WAIT2:
541 inet_sk_state_store(sk, TCP_CLOSE);
3721b9b6
MM		 542 break;
543 default:
544 /* Other states not expected */
545 WARN_ON_ONCE(1);
546 break;
547 }
548
7ed90803 549 ret = true;
fd897679 550 mptcp_send_ack(msk);
e16163b6 551 mptcp_close_wake_up(sk);
3721b9b6 552 }
7ed90803 553 return ret;
3721b9b6
MM		 554}
555
6771bfd9
FW		 556static bool __mptcp_move_skbs_from_subflow(struct mptcp_sock *msk,
557 struct sock *ssk,
558 unsigned int *bytes)
559{
560 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
600911ff 561 struct sock *sk = (struct sock *)msk;
6771bfd9
FW		 562 unsigned int moved = 0;
563 bool more_data_avail;
564 struct tcp_sock *tp;
565 bool done = false;
13c7ba0c
FW		 566 int sk_rbuf;
567
568 sk_rbuf = READ_ONCE(sk->sk_rcvbuf);
569
570 if (!(sk->sk_userlocks & SOCK_RCVBUF_LOCK)) {
571 int ssk_rbuf = READ_ONCE(ssk->sk_rcvbuf);
572
573 if (unlikely(ssk_rbuf > sk_rbuf)) {
574 WRITE_ONCE(sk->sk_rcvbuf, ssk_rbuf);
575 sk_rbuf = ssk_rbuf;
576 }
577 }
600911ff 578
ab174ad8 579 pr_debug("msk=%p ssk=%p", msk, ssk);
6771bfd9
FW		 580 tp = tcp_sk(ssk);
581 do {
582 u32 map_remaining, offset;
583 u32 seq = tp->copied_seq;
584 struct sk_buff *skb;
585 bool fin;
586
587 /* try to move as much data as available */
588 map_remaining = subflow->map_data_len -
589 mptcp_subflow_get_map_offset(subflow);
590
591 skb = skb_peek(&ssk->sk_receive_queue);
d9fb8c50
PA		 592 if (!skb) {
593 /* if no data is found, a racing workqueue/recvmsg
594 * already processed the new data, stop here or we
595 * can enter an infinite loop
596 */
597 if (!moved)
598 done = true;
6771bfd9 599 break;
d9fb8c50 600 }
6771bfd9 601
e1ff9e82
DC		 602 if (__mptcp_check_fallback(msk)) {
603 /* if we are running under the workqueue, TCP could have
604 * collapsed skbs between dummy map creation and now
605 * be sure to adjust the size
606 */
607 map_remaining = skb->len;
608 subflow->map_data_len = skb->len;
609 }
610
6771bfd9
FW		 611 offset = seq - TCP_SKB_CB(skb)->seq;
612 fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
613 if (fin) {
614 done = true;
615 seq++;
616 }
617
618 if (offset < skb->len) {
619 size_t len = skb->len - offset;
620
621 if (tp->urg_data)
622 done = true;
623
ab174ad8
PA		 624 if (__mptcp_move_skb(msk, ssk, skb, offset, len))
625 moved += len;
6771bfd9 626 seq += len;
6771bfd9
FW		 627
628 if (WARN_ON_ONCE(map_remaining < len))
629 break;
630 } else {
631 WARN_ON_ONCE(!fin);
632 sk_eat_skb(ssk, skb);
633 done = true;
634 }
635
636 WRITE_ONCE(tp->copied_seq, seq);
637 more_data_avail = mptcp_subflow_data_available(ssk);
600911ff 638
13c7ba0c 639 if (atomic_read(&sk->sk_rmem_alloc) > sk_rbuf) {
600911ff
FW		 640 done = true;
641 break;
642 }
6771bfd9
FW		 643 } while (more_data_avail);
644
6719331c 645 *bytes += moved;
6771bfd9
FW		 646 return done;
647}
648
87952603 649static bool __mptcp_ofo_queue(struct mptcp_sock *msk)
ab174ad8
PA		 650{
651 struct sock *sk = (struct sock *)msk;
652 struct sk_buff *skb, *tail;
653 bool moved = false;
654 struct rb_node *p;
655 u64 end_seq;
656
657 p = rb_first(&msk->out_of_order_queue);
06242e44 658 pr_debug("msk=%p empty=%d", msk, RB_EMPTY_ROOT(&msk->out_of_order_queue));
ab174ad8
PA		 659 while (p) {
660 skb = rb_to_skb(p);
661 if (after64(MPTCP_SKB_CB(skb)->map_seq, msk->ack_seq))
662 break;
663
664 p = rb_next(p);
665 rb_erase(&skb->rbnode, &msk->out_of_order_queue);
666
667 if (unlikely(!after64(MPTCP_SKB_CB(skb)->end_seq,
668 msk->ack_seq))) {
669 mptcp_drop(sk, skb);
06242e44 670 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
ab174ad8
PA		 671 continue;
672 }
673
674 end_seq = MPTCP_SKB_CB(skb)->end_seq;
675 tail = skb_peek_tail(&sk->sk_receive_queue);
676 if (!tail || !mptcp_ooo_try_coalesce(msk, tail, skb)) {
677 int delta = msk->ack_seq - MPTCP_SKB_CB(skb)->map_seq;
678
679 /* skip overlapping data, if any */
06242e44
PA		 680 pr_debug("uncoalesced seq=%llx ack seq=%llx delta=%d",
681 MPTCP_SKB_CB(skb)->map_seq, msk->ack_seq,
682 delta);
ab174ad8
PA		 683 MPTCP_SKB_CB(skb)->offset += delta;
684 __skb_queue_tail(&sk->sk_receive_queue, skb);
685 }
686 msk->ack_seq = end_seq;
687 moved = true;
688 }
689 return moved;
690}
691
2e52213c
FW		 692/* In most cases we will be able to lock the mptcp socket. If its already
693 * owned, we need to defer to the work queue to avoid ABBA deadlock.
694 */
99d1055c 695static bool move_skbs_to_msk(struct mptcp_sock *msk, struct sock *ssk)
2e52213c
FW		 696{
697 struct sock *sk = (struct sock *)msk;
698 unsigned int moved = 0;
699
87952603
PA		 700 __mptcp_move_skbs_from_subflow(msk, ssk, &moved);
701 __mptcp_ofo_queue(msk);
499ada50
PA		 702 if (unlikely(ssk->sk_err)) {
703 if (!sock_owned_by_user(sk))
704 __mptcp_error_report(sk);
705 else
706 set_bit(MPTCP_ERROR_REPORT, &msk->flags);
707 }
2e52213c 708
87952603
PA		 709 /* If the moves have caught up with the DATA_FIN sequence number
710 * it's time to ack the DATA_FIN and change socket state, but
711 * this is not a good place to change state. Let the workqueue
712 * do it.
713 */
714 if (mptcp_pending_data_fin(sk, NULL))
715 mptcp_schedule_work(sk);
99d1055c 716 return moved > 0;
2e52213c
FW		 717}
718
719void mptcp_data_ready(struct sock *sk, struct sock *ssk)
101f6f85 720{
6719331c 721 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
101f6f85 722 struct mptcp_sock *msk = mptcp_sk(sk);
13c7ba0c 723 int sk_rbuf, ssk_rbuf;
101f6f85 724
d7b1bfd0
PA		 725 /* The peer can send data while we are shutting down this
726 * subflow at msk destruction time, but we must avoid enqueuing
727 * more data to the msk receive queue
728 */
729 if (unlikely(subflow->disposable))
730 return;
731
13c7ba0c
FW		 732 ssk_rbuf = READ_ONCE(ssk->sk_rcvbuf);
733 sk_rbuf = READ_ONCE(sk->sk_rcvbuf);
734 if (unlikely(ssk_rbuf > sk_rbuf))
735 sk_rbuf = ssk_rbuf;
736
99d1055c 737 /* over limit? can't append more skbs to msk, Also, no need to wake-up*/
ce599c51
PA		 738 if (__mptcp_rmem(sk) > sk_rbuf) {
739 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_RCVPRUNED);
99d1055c 740 return;
ce599c51 741 }
600911ff 742
99d1055c
PA		 743 /* Wake-up the reader only for in-sequence data */
744 mptcp_data_lock(sk);
745 if (move_skbs_to_msk(msk, ssk)) {
746 set_bit(MPTCP_DATA_READY, &msk->flags);
6719331c 747 sk->sk_data_ready(sk);
99d1055c
PA		 748 }
749 mptcp_data_unlock(sk);
101f6f85
FW		 750}
751
78962489 752static bool mptcp_do_flush_join_list(struct mptcp_sock *msk)
ec3edaa7 753{
5cf92bba 754 struct mptcp_subflow_context *subflow;
df00b087 755 bool ret = false;
5cf92bba 756
ec3edaa7 757 if (likely(list_empty(&msk->join_list)))
78962489 758 return false;
ec3edaa7
PK		 759
760 spin_lock_bh(&msk->join_list_lock);
df00b087
FW		 761 list_for_each_entry(subflow, &msk->join_list, node) {
762 u32 sseq = READ_ONCE(subflow->setsockopt_seq);
78962489 763
df00b087
FW		 764 mptcp_propagate_sndbuf((struct sock *)msk, mptcp_subflow_tcp_sock(subflow));
765 if (READ_ONCE(msk->setsockopt_seq) != sseq)
766 ret = true;
767 }
ec3edaa7
PK		 768 list_splice_tail_init(&msk->join_list, &msk->conn_list);
769 spin_unlock_bh(&msk->join_list_lock);
78962489 770
df00b087 771 return ret;
78962489
FW		 772}
773
774void __mptcp_flush_join_list(struct mptcp_sock *msk)
775{
776 if (likely(!mptcp_do_flush_join_list(msk)))
777 return;
778
779 if (!test_and_set_bit(MPTCP_WORK_SYNC_SETSOCKOPT, &msk->flags))
780 mptcp_schedule_work((struct sock *)msk);
781}
782
783static void mptcp_flush_join_list(struct mptcp_sock *msk)
784{
785 bool sync_needed = test_and_clear_bit(MPTCP_WORK_SYNC_SETSOCKOPT, &msk->flags);
786
787 might_sleep();
788
789 if (!mptcp_do_flush_join_list(msk) && !sync_needed)
790 return;
791
792 mptcp_sockopt_sync_all(msk);
ec3edaa7
PK		 793}
794
b51f9b80
PA		 795static bool mptcp_timer_pending(struct sock *sk)
796{
797 return timer_pending(&inet_csk(sk)->icsk_retransmit_timer);
798}
799
800static void mptcp_reset_timer(struct sock *sk)
801{
802 struct inet_connection_sock *icsk = inet_csk(sk);
803 unsigned long tout;
804
e16163b6
PA		 805 /* prevent rescheduling on close */
806 if (unlikely(inet_sk_state_load(sk) == TCP_CLOSE))
807 return;
808
33d41c9c 809 tout = mptcp_sk(sk)->timer_ival;
b51f9b80
PA		 810 sk_reset_timer(sk, &icsk->icsk_retransmit_timer, jiffies + tout);
811}
812
ba8f48f7
PA		 813bool mptcp_schedule_work(struct sock *sk)
814{
815 if (inet_sk_state_load(sk) != TCP_CLOSE &&
816 schedule_work(&mptcp_sk(sk)->work)) {
817 /* each subflow already holds a reference to the sk, and the
818 * workqueue is invoked by a subflow, so sk can't go away here.
819 */
820 sock_hold(sk);
821 return true;
822 }
823 return false;
824}
825
59832e24
FW		 826void mptcp_subflow_eof(struct sock *sk)
827{
ba8f48f7
PA		 828 if (!test_and_set_bit(MPTCP_WORK_EOF, &mptcp_sk(sk)->flags))
829 mptcp_schedule_work(sk);
59832e24
FW		 830}
831
5969856a
PA		 832static void mptcp_check_for_eof(struct mptcp_sock *msk)
833{
834 struct mptcp_subflow_context *subflow;
835 struct sock *sk = (struct sock *)msk;
836 int receivers = 0;
837
838 mptcp_for_each_subflow(msk, subflow)
839 receivers += !subflow->rx_eof;
e16163b6
PA		 840 if (receivers)
841 return;
5969856a 842
e16163b6 843 if (!(sk->sk_shutdown & RCV_SHUTDOWN)) {
5969856a
PA		 844 /* hopefully temporary hack: propagate shutdown status
845 * to msk, when all subflows agree on it
846 */
847 sk->sk_shutdown |= RCV_SHUTDOWN;
848
849 smp_mb__before_atomic(); /* SHUTDOWN must be visible first */
850 set_bit(MPTCP_DATA_READY, &msk->flags);
851 sk->sk_data_ready(sk);
852 }
e16163b6
PA		 853
854 switch (sk->sk_state) {
855 case TCP_ESTABLISHED:
856 inet_sk_state_store(sk, TCP_CLOSE_WAIT);
857 break;
858 case TCP_FIN_WAIT1:
26aa2314
PA		 859 inet_sk_state_store(sk, TCP_CLOSING);
860 break;
861 case TCP_FIN_WAIT2:
e16163b6
PA		 862 inet_sk_state_store(sk, TCP_CLOSE);
863 break;
864 default:
865 return;
866 }
867 mptcp_close_wake_up(sk);
5969856a
PA		 868}
869
7a6a6cbc
PA		 870static struct sock *mptcp_subflow_recv_lookup(const struct mptcp_sock *msk)
871{
872 struct mptcp_subflow_context *subflow;
873 struct sock *sk = (struct sock *)msk;
874
875 sock_owned_by_me(sk);
876
877 mptcp_for_each_subflow(msk, subflow) {
99d1055c 878 if (READ_ONCE(subflow->data_avail))
7a6a6cbc
PA		 879 return mptcp_subflow_tcp_sock(subflow);
880 }
881
882 return NULL;
883}
884
3f8e0aae
PA		 885static bool mptcp_skb_can_collapse_to(u64 write_seq,
886 const struct sk_buff *skb,
887 const struct mptcp_ext *mpext)
57040755
PA		 888{
889 if (!tcp_skb_can_collapse_to(skb))
890 return false;
891
5a369ca6
PA		 892 /* can collapse only if MPTCP level sequence is in order and this
893 * mapping has not been xmitted yet
894 */
895 return mpext && mpext->data_seq + mpext->data_len == write_seq &&
896 !mpext->frozen;
57040755
PA		 897}
898
29249eac
PA		 899/* we can append data to the given data frag if:
900 * - there is space available in the backing page_frag
901 * - the data frag tail matches the current page_frag free offset
902 * - the data frag end sequence number matches the current write seq
903 */
18b683bf
PA		 904static bool mptcp_frag_can_collapse_to(const struct mptcp_sock *msk,
905 const struct page_frag *pfrag,
906 const struct mptcp_data_frag *df)
907{
908 return df && pfrag->page == df->page &&
d9ca1de8 909 pfrag->size - pfrag->offset > 0 &&
29249eac 910 pfrag->offset == (df->offset + df->data_len) &&
18b683bf
PA		 911 df->data_seq + df->data_len == msk->write_seq;
912}
913
8ce568ed 914static int mptcp_wmem_with_overhead(int size)
e93da928 915{
8ce568ed 916 return size + ((sizeof(struct mptcp_data_frag) * size) >> PAGE_SHIFT);
e93da928
PA		 917}
918
919static void __mptcp_wmem_reserve(struct sock *sk, int size)
920{
8ce568ed 921 int amount = mptcp_wmem_with_overhead(size);
e93da928
PA		 922 struct mptcp_sock *msk = mptcp_sk(sk);
923
924 WARN_ON_ONCE(msk->wmem_reserved);
e7579d5d
DC		 925 if (WARN_ON_ONCE(amount < 0))
926 amount = 0;
927
e93da928
PA		 928 if (amount <= sk->sk_forward_alloc)
929 goto reserve;
930
931 /* under memory pressure try to reserve at most a single page
932 * otherwise try to reserve the full estimate and fallback
933 * to a single page before entering the error path
934 */
935 if ((tcp_under_memory_pressure(sk) && amount > PAGE_SIZE) ||
936 !sk_wmem_schedule(sk, amount)) {
937 if (amount <= PAGE_SIZE)
938 goto nomem;
939
940 amount = PAGE_SIZE;
941 if (!sk_wmem_schedule(sk, amount))
942 goto nomem;
943 }
944
945reserve:
946 msk->wmem_reserved = amount;
947 sk->sk_forward_alloc -= amount;
948 return;
949
950nomem:
951 /* we will wait for memory on next allocation */
952 msk->wmem_reserved = -1;
953}
954
955static void __mptcp_update_wmem(struct sock *sk)
956{
957 struct mptcp_sock *msk = mptcp_sk(sk);
958
765ff425 959 lockdep_assert_held_once(&sk->sk_lock.slock);
b5941f06 960
e93da928
PA		 961 if (!msk->wmem_reserved)
962 return;
963
964 if (msk->wmem_reserved < 0)
965 msk->wmem_reserved = 0;
966 if (msk->wmem_reserved > 0) {
967 sk->sk_forward_alloc += msk->wmem_reserved;
968 msk->wmem_reserved = 0;
969 }
970}
971
972static bool mptcp_wmem_alloc(struct sock *sk, int size)
973{
974 struct mptcp_sock *msk = mptcp_sk(sk);
975
976 /* check for pre-existing error condition */
977 if (msk->wmem_reserved < 0)
978 return false;
979
980 if (msk->wmem_reserved >= size)
981 goto account;
982
87952603
PA		 983 mptcp_data_lock(sk);
984 if (!sk_wmem_schedule(sk, size)) {
985 mptcp_data_unlock(sk);
e93da928 986 return false;
87952603 987 }
e93da928
PA		 988
989 sk->sk_forward_alloc -= size;
990 msk->wmem_reserved += size;
87952603 991 mptcp_data_unlock(sk);
e93da928
PA		 992
993account:
994 msk->wmem_reserved -= size;
995 return true;
996}
997
87952603
PA		 998static void mptcp_wmem_uncharge(struct sock *sk, int size)
999{
1000 struct mptcp_sock *msk = mptcp_sk(sk);
1001
1002 if (msk->wmem_reserved < 0)
1003 msk->wmem_reserved = 0;
1004 msk->wmem_reserved += size;
1005}
1006
1094c6fe
PA		 1007static void __mptcp_mem_reclaim_partial(struct sock *sk)
1008{
1009 lockdep_assert_held_once(&sk->sk_lock.slock);
1010 __mptcp_update_wmem(sk);
1011 sk_mem_reclaim_partial(sk);
1012}
1013
724cfd2e
PA		 1014static void mptcp_mem_reclaim_partial(struct sock *sk)
1015{
1016 struct mptcp_sock *msk = mptcp_sk(sk);
1017
1018 /* if we are experiencing a transint allocation error,
1019 * the forward allocation memory has been already
1020 * released
1021 */
1022 if (msk->wmem_reserved < 0)
1023 return;
1024
1025 mptcp_data_lock(sk);
1026 sk->sk_forward_alloc += msk->wmem_reserved;
1027 sk_mem_reclaim_partial(sk);
1028 msk->wmem_reserved = sk->sk_forward_alloc;
1029 sk->sk_forward_alloc = 0;
1030 mptcp_data_unlock(sk);
1031}
1032
d027236c
PA		 1033static void dfrag_uncharge(struct sock *sk, int len)
1034{
1035 sk_mem_uncharge(sk, len);
7948f6cc 1036 sk_wmem_queued_add(sk, -len);
d027236c
PA		 1037}
1038
1039static void dfrag_clear(struct sock *sk, struct mptcp_data_frag *dfrag)
18b683bf 1040{
d027236c
PA		 1041 int len = dfrag->data_len + dfrag->overhead;
1042
18b683bf 1043 list_del(&dfrag->list);
d027236c 1044 dfrag_uncharge(sk, len);
18b683bf
PA		 1045 put_page(dfrag->page);
1046}
1047
6e628cd3 1048static void __mptcp_clean_una(struct sock *sk)
18b683bf
PA		 1049{
1050 struct mptcp_sock *msk = mptcp_sk(sk);
1051 struct mptcp_data_frag *dtmp, *dfrag;
d027236c 1052 bool cleaned = false;
e1ff9e82
DC		 1053 u64 snd_una;
1054
1055 /* on fallback we just need to ignore snd_una, as this is really
1056 * plain TCP
1057 */
1058 if (__mptcp_check_fallback(msk))
7439d687 1059 msk->snd_una = READ_ONCE(msk->snd_nxt);
6f8a612a 1060
7439d687 1061 snd_una = msk->snd_una;
18b683bf
PA		 1062 list_for_each_entry_safe(dfrag, dtmp, &msk->rtx_queue, list) {
1063 if (after64(dfrag->data_seq + dfrag->data_len, snd_una))
1064 break;
1065
1e1d9d6f
PA		 1066 if (unlikely(dfrag == msk->first_pending)) {
1067 /* in recovery mode can see ack after the current snd head */
1068 if (WARN_ON_ONCE(!msk->recovery))
1069 break;
1070
1071 WRITE_ONCE(msk->first_pending, mptcp_send_next(sk));
1072 }
1073
d027236c
PA		 1074 dfrag_clear(sk, dfrag);
1075 cleaned = true;
1076 }
1077
7948f6cc
FW		 1078 dfrag = mptcp_rtx_head(sk);
1079 if (dfrag && after64(snd_una, dfrag->data_seq)) {
53eb4c38
PA		 1080 u64 delta = snd_una - dfrag->data_seq;
1081
1e1d9d6f
PA		 1082 /* prevent wrap around in recovery mode */
1083 if (unlikely(delta > dfrag->already_sent)) {
1084 if (WARN_ON_ONCE(!msk->recovery))
1085 goto out;
1086 if (WARN_ON_ONCE(delta > dfrag->data_len))
1087 goto out;
1088 dfrag->already_sent += delta - dfrag->already_sent;
1089 }
7948f6cc
FW		 1090
1091 dfrag->data_seq += delta;
53eb4c38 1092 dfrag->offset += delta;
7948f6cc 1093 dfrag->data_len -= delta;
d9ca1de8 1094 dfrag->already_sent -= delta;
7948f6cc
FW		 1095
1096 dfrag_uncharge(sk, delta);
1097 cleaned = true;
1098 }
1099
1e1d9d6f
PA		 1100 /* all retransmitted data acked, recovery completed */
1101 if (unlikely(msk->recovery) && after64(msk->snd_una, msk->recovery_snd_nxt))
1102 msk->recovery = false;
1103
53eb4c38 1104out:
1094c6fe
PA		 1105 if (cleaned && tcp_under_memory_pressure(sk))
1106 __mptcp_mem_reclaim_partial(sk);
95ed690e 1107
1e1d9d6f 1108 if (snd_una == READ_ONCE(msk->snd_nxt) && !msk->recovery) {
33d41c9c 1109 if (mptcp_timer_pending(sk) && !mptcp_data_fin_enabled(msk))
6e628cd3
PA		 1110 mptcp_stop_timer(sk);
1111 } else {
1112 mptcp_reset_timer(sk);
18b683bf
PA		 1113 }
1114}
1115
417789df
PA		 1116static void __mptcp_clean_una_wakeup(struct sock *sk)
1117{
765ff425
PA		 1118 lockdep_assert_held_once(&sk->sk_lock.slock);
1119
417789df
PA		 1120 __mptcp_clean_una(sk);
1121 mptcp_write_space(sk);
1122}
1123
b5941f06
PA		 1124static void mptcp_clean_una_wakeup(struct sock *sk)
1125{
1126 mptcp_data_lock(sk);
1127 __mptcp_clean_una_wakeup(sk);
1128 mptcp_data_unlock(sk);
1129}
1130
724cfd2e 1131static void mptcp_enter_memory_pressure(struct sock *sk)
18b683bf 1132{
d9ca1de8
PA		 1133 struct mptcp_subflow_context *subflow;
1134 struct mptcp_sock *msk = mptcp_sk(sk);
1135 bool first = true;
1136
18b683bf 1137 sk_stream_moderate_sndbuf(sk);
d9ca1de8
PA		 1138 mptcp_for_each_subflow(msk, subflow) {
1139 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1140
1141 if (first)
1142 tcp_enter_memory_pressure(ssk);
1143 sk_stream_moderate_sndbuf(ssk);
1144 first = false;
1145 }
724cfd2e
PA		 1146}
1147
1148/* ensure we get enough memory for the frag hdr, beyond some minimal amount of
1149 * data
1150 */
1151static bool mptcp_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1152{
1153 if (likely(skb_page_frag_refill(32U + sizeof(struct mptcp_data_frag),
1154 pfrag, sk->sk_allocation)))
1155 return true;
1156
1157 mptcp_enter_memory_pressure(sk);
18b683bf
PA		 1158 return false;
1159}
1160
1161static struct mptcp_data_frag *
1162mptcp_carve_data_frag(const struct mptcp_sock *msk, struct page_frag *pfrag,
1163 int orig_offset)
1164{
1165 int offset = ALIGN(orig_offset, sizeof(long));
1166 struct mptcp_data_frag *dfrag;
1167
1168 dfrag = (struct mptcp_data_frag *)(page_to_virt(pfrag->page) + offset);
1169 dfrag->data_len = 0;
1170 dfrag->data_seq = msk->write_seq;
1171 dfrag->overhead = offset - orig_offset + sizeof(struct mptcp_data_frag);
1172 dfrag->offset = offset + sizeof(struct mptcp_data_frag);
d9ca1de8 1173 dfrag->already_sent = 0;
18b683bf
PA		 1174 dfrag->page = pfrag->page;
1175
1176 return dfrag;
1177}
1178
caf971df
PA		 1179struct mptcp_sendmsg_info {
1180 int mss_now;
1181 int size_goal;
d9ca1de8
PA		 1182 u16 limit;
1183 u16 sent;
1184 unsigned int flags;
1094c6fe 1185 bool data_lock_held;
caf971df
PA		 1186};
1187
6f8a612a
FW		 1188static int mptcp_check_allowed_size(struct mptcp_sock *msk, u64 data_seq,
1189 int avail_size)
1190{
1191 u64 window_end = mptcp_wnd_end(msk);
1192
1193 if (__mptcp_check_fallback(msk))
1194 return avail_size;
1195
1196 if (!before64(data_seq + avail_size, window_end)) {
1197 u64 allowed_size = window_end - data_seq;
1198
1199 return min_t(unsigned int, allowed_size, avail_size);
1200 }
1201
1202 return avail_size;
1203}
1204
724cfd2e
PA		 1205static bool __mptcp_add_ext(struct sk_buff *skb, gfp_t gfp)
1206{
1207 struct skb_ext *mpext = __skb_ext_alloc(gfp);
1208
1209 if (!mpext)
1210 return false;
1211 __skb_ext_set(skb, SKB_EXT_MPTCP, mpext);
1212 return true;
1213}
1214
6e628cd3 1215static struct sk_buff *__mptcp_do_alloc_tx_skb(struct sock *sk, gfp_t gfp)
724cfd2e
PA		 1216{
1217 struct sk_buff *skb;
1218
6e628cd3 1219 skb = alloc_skb_fclone(MAX_TCP_HEADER, gfp);
724cfd2e 1220 if (likely(skb)) {
6e628cd3 1221 if (likely(__mptcp_add_ext(skb, gfp))) {
724cfd2e
PA		 1222 skb_reserve(skb, MAX_TCP_HEADER);
1223 skb->reserved_tailroom = skb->end - skb->tail;
f70cad10 1224 INIT_LIST_HEAD(&skb->tcp_tsorted_anchor);
724cfd2e
PA		 1225 return skb;
1226 }
1227 __kfree_skb(skb);
1228 } else {
1229 mptcp_enter_memory_pressure(sk);
1230 }
1231 return NULL;
1232}
1233
f70cad10 1234static struct sk_buff *__mptcp_alloc_tx_skb(struct sock *sk, struct sock *ssk, gfp_t gfp)
724cfd2e 1235{
724cfd2e
PA		 1236 struct sk_buff *skb;
1237
6e628cd3 1238 skb = __mptcp_do_alloc_tx_skb(sk, gfp);
724cfd2e 1239 if (!skb)
f70cad10 1240 return NULL;
724cfd2e
PA		 1241
1242 if (likely(sk_wmem_schedule(ssk, skb->truesize))) {
f70cad10
PA		 1243 tcp_skb_entail(ssk, skb);
1244 return skb;
724cfd2e
PA		 1245 }
1246 kfree_skb(skb);
f70cad10 1247 return NULL;
724cfd2e
PA		 1248}
1249
f70cad10 1250static struct sk_buff *mptcp_alloc_tx_skb(struct sock *sk, struct sock *ssk, bool data_lock_held)
724cfd2e 1251{
1094c6fe 1252 gfp_t gfp = data_lock_held ? GFP_ATOMIC : sk->sk_allocation;
724cfd2e 1253
1094c6fe
PA		 1254 if (unlikely(tcp_under_memory_pressure(sk))) {
1255 if (data_lock_held)
1256 __mptcp_mem_reclaim_partial(sk);
1257 else
1258 mptcp_mem_reclaim_partial(sk);
1259 }
1260 return __mptcp_alloc_tx_skb(sk, ssk, gfp);
724cfd2e
PA		 1261}
1262
d0cc2987
GT		 1263/* note: this always recompute the csum on the whole skb, even
1264 * if we just appended a single frag. More status info needed
1265 */
1266static void mptcp_update_data_checksum(struct sk_buff *skb, int added)
1267{
1268 struct mptcp_ext *mpext = mptcp_get_ext(skb);
1269 __wsum csum = ~csum_unfold(mpext->csum);
1270 int offset = skb->len - added;
1271
1272 mpext->csum = csum_fold(csum_block_add(csum, skb_checksum(skb, offset, added, 0), offset));
1273}
1274
6d0060f6 1275static int mptcp_sendmsg_frag(struct sock *sk, struct sock *ssk,
d9ca1de8 1276 struct mptcp_data_frag *dfrag,
caf971df 1277 struct mptcp_sendmsg_info *info)
6d0060f6 1278{
d9ca1de8 1279 u64 data_seq = dfrag->data_seq + info->sent;
f70cad10 1280 int offset = dfrag->offset + info->sent;
6d0060f6 1281 struct mptcp_sock *msk = mptcp_sk(sk);
6f8a612a 1282 bool zero_window_probe = false;
6d0060f6 1283 struct mptcp_ext *mpext = NULL;
f70cad10
PA		 1284 bool can_coalesce = false;
1285 bool reuse_skb = true;
1286 struct sk_buff *skb;
1287 size_t copy;
1288 int i;
6d0060f6 1289
e4b61351 1290 pr_debug("msk=%p ssk=%p sending dfrag at seq=%llu len=%u already sent=%u",
d9ca1de8
PA		 1291 msk, ssk, dfrag->data_seq, dfrag->data_len, info->sent);
1292
f70cad10
PA		 1293 if (WARN_ON_ONCE(info->sent > info->limit ||
1294 info->limit > dfrag->data_len))
1295 return 0;
1296
d9ca1de8
PA		 1297 /* compute send limit */
1298 info->mss_now = tcp_send_mss(ssk, &info->size_goal, info->flags);
f70cad10
PA		 1299 copy = info->size_goal;
1300
57040755 1301 skb = tcp_write_queue_tail(ssk);
f70cad10 1302 if (skb && copy > skb->len) {
57040755
PA		 1303 /* Limit the write to the size available in the
1304 * current skb, if any, so that we create at most a new skb.
1305 * Explicitly tells TCP internals to avoid collapsing on later
1306 * queue management operation, to avoid breaking the ext <->
1307 * SSN association set here
1308 */
d9ca1de8 1309 mpext = skb_ext_find(skb, SKB_EXT_MPTCP);
1094c6fe 1310 if (!mptcp_skb_can_collapse_to(data_seq, skb, mpext)) {
57040755 1311 TCP_SKB_CB(skb)->eor = 1;
1094c6fe
PA		 1312 goto alloc_skb;
1313 }
1314
f70cad10
PA		 1315 i = skb_shinfo(skb)->nr_frags;
1316 can_coalesce = skb_can_coalesce(skb, i, dfrag->page, offset);
1317 if (!can_coalesce && i >= sysctl_max_skb_frags) {
1318 tcp_mark_push(tcp_sk(ssk), skb);
1319 goto alloc_skb;
15e6ca97 1320 }
18b683bf 1321
f70cad10
PA		 1322 copy -= skb->len;
1323 } else {
1094c6fe 1324alloc_skb:
f70cad10
PA		 1325 skb = mptcp_alloc_tx_skb(sk, ssk, info->data_lock_held);
1326 if (!skb)
1327 return -ENOMEM;
1328
1329 i = skb_shinfo(skb)->nr_frags;
1330 reuse_skb = false;
1331 mpext = skb_ext_find(skb, SKB_EXT_MPTCP);
1332 }
1094c6fe 1333
6f8a612a 1334 /* Zero window and all data acked? Probe. */
f70cad10
PA		 1335 copy = mptcp_check_allowed_size(msk, data_seq, copy);
1336 if (copy == 0) {
7439d687
PA		 1337 u64 snd_una = READ_ONCE(msk->snd_una);
1338
f70cad10
PA		 1339 if (snd_una != msk->snd_nxt) {
1340 tcp_remove_empty_skb(ssk, tcp_write_queue_tail(ssk));
6f8a612a 1341 return 0;
f70cad10
PA		 1342 }
1343
6f8a612a 1344 zero_window_probe = true;
7439d687 1345 data_seq = snd_una - 1;
f70cad10 1346 copy = 1;
6f8a612a 1347
f70cad10
PA		 1348 /* all mptcp-level data is acked, no skbs should be present into the
1349 * ssk write queue
1350 */
1351 WARN_ON_ONCE(reuse_skb);
1352 }
d027236c 1353
f70cad10
PA		 1354 copy = min_t(size_t, copy, info->limit - info->sent);
1355 if (!sk_wmem_schedule(ssk, copy)) {
1356 tcp_remove_empty_skb(ssk, tcp_write_queue_tail(ssk));
e2223995 1357 return -ENOMEM;
35759383 1358 }
18b683bf 1359
f70cad10
PA		 1360 if (can_coalesce) {
1361 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1362 } else {
1363 get_page(dfrag->page);
1364 skb_fill_page_desc(skb, i, dfrag->page, offset, copy);
57040755
PA		 1365 }
1366
f70cad10
PA		 1367 skb->len += copy;
1368 skb->data_len += copy;
1369 skb->truesize += copy;
1370 sk_wmem_queued_add(ssk, copy);
1371 sk_mem_charge(ssk, copy);
1372 skb->ip_summed = CHECKSUM_PARTIAL;
1373 WRITE_ONCE(tcp_sk(ssk)->write_seq, tcp_sk(ssk)->write_seq + copy);
1374 TCP_SKB_CB(skb)->end_seq += copy;
1375 tcp_skb_pcount_set(skb, 0);
1376
1377 /* on skb reuse we just need to update the DSS len */
1378 if (reuse_skb) {
1379 TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH;
1380 mpext->data_len += copy;
1381 WARN_ON_ONCE(zero_window_probe);
1382 goto out;
724cfd2e 1383 }
6d0060f6
MM		 1384
1385 memset(mpext, 0, sizeof(*mpext));
d9ca1de8 1386 mpext->data_seq = data_seq;
6d0060f6 1387 mpext->subflow_seq = mptcp_subflow_ctx(ssk)->rel_write_seq;
f70cad10 1388 mpext->data_len = copy;
6d0060f6
MM		 1389 mpext->use_map = 1;
1390 mpext->dsn64 = 1;
1391
1392 pr_debug("data_seq=%llu subflow_seq=%u data_len=%u dsn64=%d",
1393 mpext->data_seq, mpext->subflow_seq, mpext->data_len,
1394 mpext->dsn64);
1395
6f8a612a 1396 if (zero_window_probe) {
f70cad10 1397 mptcp_subflow_ctx(ssk)->rel_write_seq += copy;
6f8a612a 1398 mpext->frozen = 1;
d0cc2987 1399 if (READ_ONCE(msk->csum_enabled))
f70cad10 1400 mptcp_update_data_checksum(skb, copy);
6f8a612a 1401 tcp_push_pending_frames(ssk);
d0cc2987 1402 return 0;
6f8a612a 1403 }
57040755 1404out:
d0cc2987 1405 if (READ_ONCE(msk->csum_enabled))
f70cad10
PA		 1406 mptcp_update_data_checksum(skb, copy);
1407 mptcp_subflow_ctx(ssk)->rel_write_seq += copy;
1408 return copy;
6d0060f6
MM		 1409}
1410
d5f49190
PA		 1411#define MPTCP_SEND_BURST_SIZE ((1 << 16) - \
1412 sizeof(struct tcphdr) - \
1413 MAX_TCP_OPTION_SPACE - \
1414 sizeof(struct ipv6hdr) - \
1415 sizeof(struct frag_hdr))
1416
1417struct subflow_send_info {
1418 struct sock *ssk;
1419 u64 ratio;
1420};
1421
ff5a0b42
PA		 1422void mptcp_subflow_set_active(struct mptcp_subflow_context *subflow)
1423{
1424 if (!subflow->stale)
1425 return;
1426
1427 subflow->stale = 0;
fc1b4e3b 1428 MPTCP_INC_STATS(sock_net(mptcp_subflow_tcp_sock(subflow)), MPTCP_MIB_SUBFLOWRECOVER);
ff5a0b42
PA		 1429}
1430
1431bool mptcp_subflow_active(struct mptcp_subflow_context *subflow)
1432{
1433 if (unlikely(subflow->stale)) {
1434 u32 rcv_tstamp = READ_ONCE(tcp_sk(mptcp_subflow_tcp_sock(subflow))->rcv_tstamp);
1435
1436 if (subflow->stale_rcv_tstamp == rcv_tstamp)
1437 return false;
1438
1439 mptcp_subflow_set_active(subflow);
1440 }
1441 return __mptcp_subflow_active(subflow);
1442}
1443
33d41c9c
PA		 1444/* implement the mptcp packet scheduler;
1445 * returns the subflow that will transmit the next DSS
1446 * additionally updates the rtx timeout
1447 */
5cf92bba 1448static struct sock *mptcp_subflow_get_send(struct mptcp_sock *msk)
f296234c 1449{
d5f49190 1450 struct subflow_send_info send_info[2];
f296234c 1451 struct mptcp_subflow_context *subflow;
33d41c9c 1452 struct sock *sk = (struct sock *)msk;
d5f49190
PA		 1453 int i, nr_active = 0;
1454 struct sock *ssk;
33d41c9c 1455 long tout = 0;
d5f49190
PA		 1456 u64 ratio;
1457 u32 pace;
f296234c 1458
33d41c9c 1459 sock_owned_by_me(sk);
f296234c 1460
d5f49190
PA		 1461 if (__mptcp_check_fallback(msk)) {
1462 if (!msk->first)
f296234c 1463 return NULL;
d5f49190
PA		 1464 return sk_stream_memory_free(msk->first) ? msk->first : NULL;
1465 }
1466
1467 /* re-use last subflow, if the burst allow that */
1468 if (msk->last_snd && msk->snd_burst > 0 &&
1469 sk_stream_memory_free(msk->last_snd) &&
33d41c9c
PA		 1470 mptcp_subflow_active(mptcp_subflow_ctx(msk->last_snd))) {
1471 mptcp_set_timeout(sk);
d5f49190 1472 return msk->last_snd;
33d41c9c 1473 }
f296234c 1474
d5f49190
PA		 1475 /* pick the subflow with the lower wmem/wspace ratio */
1476 for (i = 0; i < 2; ++i) {
1477 send_info[i].ssk = NULL;
1478 send_info[i].ratio = -1;
1479 }
1480 mptcp_for_each_subflow(msk, subflow) {
e10a9892 1481 trace_mptcp_subflow_get_send(subflow);
d5f49190
PA		 1482 ssk = mptcp_subflow_tcp_sock(subflow);
1483 if (!mptcp_subflow_active(subflow))
1484 continue;
1485
33d41c9c 1486 tout = max(tout, mptcp_timeout_from_subflow(subflow));
d5f49190 1487 nr_active += !subflow->backup;
ec369c3a 1488 if (!sk_stream_memory_free(subflow->tcp_sock) || !tcp_sk(ssk)->snd_wnd)
d5f49190 1489 continue;
f296234c 1490
d5f49190
PA		 1491 pace = READ_ONCE(ssk->sk_pacing_rate);
1492 if (!pace)
f296234c 1493 continue;
f296234c 1494
d5f49190
PA		 1495 ratio = div_u64((u64)READ_ONCE(ssk->sk_wmem_queued) << 32,
1496 pace);
1497 if (ratio < send_info[subflow->backup].ratio) {
1498 send_info[subflow->backup].ssk = ssk;
1499 send_info[subflow->backup].ratio = ratio;
1500 }
f296234c 1501 }
33d41c9c 1502 __mptcp_set_timeout(sk, tout);
f296234c 1503
d5f49190
PA		 1504 /* pick the best backup if no other subflow is active */
1505 if (!nr_active)
1506 send_info[0].ssk = send_info[1].ssk;
1507
1508 if (send_info[0].ssk) {
1509 msk->last_snd = send_info[0].ssk;
1510 msk->snd_burst = min_t(int, MPTCP_SEND_BURST_SIZE,
ec369c3a 1511 tcp_sk(msk->last_snd)->snd_wnd);
d5f49190
PA		 1512 return msk->last_snd;
1513 }
5cf92bba 1514
d5f49190 1515 return NULL;
f296234c
PK		 1516}
1517
d9ca1de8
PA		 1518static void mptcp_push_release(struct sock *sk, struct sock *ssk,
1519 struct mptcp_sendmsg_info *info)
1520{
d9ca1de8
PA		 1521 tcp_push(ssk, 0, info->mss_now, tcp_sk(ssk)->nonagle, info->size_goal);
1522 release_sock(ssk);
1523}
1524
0d199e43
FW		 1525static void mptcp_update_post_push(struct mptcp_sock *msk,
1526 struct mptcp_data_frag *dfrag,
1527 u32 sent)
1528{
1529 u64 snd_nxt_new = dfrag->data_seq;
1530
1531 dfrag->already_sent += sent;
1532
1533 msk->snd_burst -= sent;
0d199e43
FW		 1534
1535 snd_nxt_new += dfrag->already_sent;
1536
1537 /* snd_nxt_new can be smaller than snd_nxt in case mptcp
1538 * is recovering after a failover. In that event, this re-sends
1539 * old segments.
1540 *
1541 * Thus compute snd_nxt_new candidate based on
1542 * the dfrag->data_seq that was sent and the data
1543 * that has been handed to the subflow for transmission
1544 * and skip update in case it was old dfrag.
1545 */
1546 if (likely(after64(snd_nxt_new, msk->snd_nxt)))
1547 msk->snd_nxt = snd_nxt_new;
1548}
1549
ff5a0b42 1550void __mptcp_push_pending(struct sock *sk, unsigned int flags)
f870fa0b 1551{
d9ca1de8 1552 struct sock *prev_ssk = NULL, *ssk = NULL;
f870fa0b 1553 struct mptcp_sock *msk = mptcp_sk(sk);
caf971df 1554 struct mptcp_sendmsg_info info = {
d9ca1de8 1555 .flags = flags,
caf971df 1556 };
d9ca1de8
PA		 1557 struct mptcp_data_frag *dfrag;
1558 int len, copied = 0;
d9ca1de8
PA		 1559
1560 while ((dfrag = mptcp_send_head(sk))) {
1561 info.sent = dfrag->already_sent;
1562 info.limit = dfrag->data_len;
1563 len = dfrag->data_len - dfrag->already_sent;
1564 while (len > 0) {
1565 int ret = 0;
1566
1567 prev_ssk = ssk;
78962489 1568 mptcp_flush_join_list(msk);
5cf92bba 1569 ssk = mptcp_subflow_get_send(msk);
d9ca1de8 1570
9758f40e
PA		 1571 /* First check. If the ssk has changed since
1572 * the last round, release prev_ssk
d9ca1de8
PA		 1573 */
1574 if (ssk != prev_ssk && prev_ssk)
1575 mptcp_push_release(sk, prev_ssk, &info);
1576 if (!ssk)
1577 goto out;
1578
9758f40e
PA		 1579 /* Need to lock the new subflow only if different
1580 * from the previous one, otherwise we are still
1581 * helding the relevant lock
1582 */
1583 if (ssk != prev_ssk)
d9ca1de8
PA		 1584 lock_sock(ssk);
1585
1586 ret = mptcp_sendmsg_frag(sk, ssk, dfrag, &info);
1587 if (ret <= 0) {
1588 mptcp_push_release(sk, ssk, &info);
1589 goto out;
1590 }
1591
1592 info.sent += ret;
d9ca1de8
PA		 1593 copied += ret;
1594 len -= ret;
0d199e43
FW		 1595
1596 mptcp_update_post_push(msk, dfrag, ret);
d9ca1de8
PA		 1597 }
1598 WRITE_ONCE(msk->first_pending, mptcp_send_next(sk));
1599 }
1600
1601 /* at this point we held the socket lock for the last subflow we used */
1602 if (ssk)
1603 mptcp_push_release(sk, ssk, &info);
1604
1605out:
33d41c9c
PA		 1606 /* ensure the rtx timer is running */
1607 if (!mptcp_timer_pending(sk))
1608 mptcp_reset_timer(sk);
1609 if (copied)
d9ca1de8
PA		 1610 __mptcp_check_send_data_fin(sk);
1611}
1612
6e628cd3
PA		 1613static void __mptcp_subflow_push_pending(struct sock *sk, struct sock *ssk)
1614{
1615 struct mptcp_sock *msk = mptcp_sk(sk);
1094c6fe
PA		 1616 struct mptcp_sendmsg_info info = {
1617 .data_lock_held = true,
1618 };
6e628cd3 1619 struct mptcp_data_frag *dfrag;
b19bc294 1620 struct sock *xmit_ssk;
6e628cd3 1621 int len, copied = 0;
b19bc294 1622 bool first = true;
6e628cd3
PA		 1623
1624 info.flags = 0;
1625 while ((dfrag = mptcp_send_head(sk))) {
1626 info.sent = dfrag->already_sent;
1627 info.limit = dfrag->data_len;
1628 len = dfrag->data_len - dfrag->already_sent;
1629 while (len > 0) {
1630 int ret = 0;
1631
b19bc294
PA		 1632 /* the caller already invoked the packet scheduler,
1633 * check for a different subflow usage only after
1634 * spooling the first chunk of data
1635 */
1636 xmit_ssk = first ? ssk : mptcp_subflow_get_send(mptcp_sk(sk));
1637 if (!xmit_ssk)
1638 goto out;
1639 if (xmit_ssk != ssk) {
1640 mptcp_subflow_delegate(mptcp_subflow_ctx(xmit_ssk));
1641 goto out;
1642 }
1643
6e628cd3
PA		 1644 ret = mptcp_sendmsg_frag(sk, ssk, dfrag, &info);
1645 if (ret <= 0)
1646 goto out;
1647
1648 info.sent += ret;
6e628cd3
PA		 1649 copied += ret;
1650 len -= ret;
b19bc294 1651 first = false;
0d199e43
FW		 1652
1653 mptcp_update_post_push(msk, dfrag, ret);
6e628cd3
PA		 1654 }
1655 WRITE_ONCE(msk->first_pending, mptcp_send_next(sk));
1656 }
1657
1658out:
1659 /* __mptcp_alloc_tx_skb could have released some wmem and we are
1660 * not going to flush it via release_sock()
1661 */
1662 __mptcp_update_wmem(sk);
1663 if (copied) {
6e628cd3
PA		 1664 tcp_push(ssk, 0, info.mss_now, tcp_sk(ssk)->nonagle,
1665 info.size_goal);
d09d818e
PA		 1666 if (!mptcp_timer_pending(sk))
1667 mptcp_reset_timer(sk);
1668
6e628cd3
PA		 1669 if (msk->snd_data_fin_enable &&
1670 msk->snd_nxt + 1 == msk->write_seq)
1671 mptcp_schedule_work(sk);
1672 }
1673}
1674
5cf92bba
PA		 1675static void mptcp_set_nospace(struct sock *sk)
1676{
1677 /* enable autotune */
1678 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1679
1680 /* will be cleared on avail space */
1681 set_bit(MPTCP_NOSPACE, &mptcp_sk(sk)->flags);
1682}
1683
d9ca1de8
PA		 1684static int mptcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1685{
1686 struct mptcp_sock *msk = mptcp_sk(sk);
17091708 1687 struct page_frag *pfrag;
6d0060f6 1688 size_t copied = 0;
caf971df 1689 int ret = 0;
6d0060f6 1690 long timeo;
f870fa0b 1691
987858e5
PA		 1692 /* we don't support FASTOPEN yet */
1693 if (msg->msg_flags & MSG_FASTOPEN)
f870fa0b
MM		 1694 return -EOPNOTSUPP;
1695
987858e5
PA		 1696 /* silently ignore everything else */
1697 msg->msg_flags &= MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL;
1698
e7579d5d 1699 mptcp_lock_sock(sk, __mptcp_wmem_reserve(sk, min_t(size_t, 1 << 20, len)));
1954b860
MM		 1700
1701 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1702
1703 if ((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) {
1704 ret = sk_stream_wait_connect(sk, &timeo);
1705 if (ret)
1706 goto out;
1707 }
1708
17091708 1709 pfrag = sk_page_frag(sk);
18b683bf 1710
d9ca1de8 1711 while (msg_data_left(msg)) {
724cfd2e 1712 int total_ts, frag_truesize = 0;
d9ca1de8 1713 struct mptcp_data_frag *dfrag;
d9ca1de8
PA		 1714 bool dfrag_collapsed;
1715 size_t psize, offset;
18b683bf 1716
d9ca1de8
PA		 1717 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN)) {
1718 ret = -EPIPE;
f296234c
PK		 1719 goto out;
1720 }
da51aef5 1721
d9ca1de8
PA		 1722 /* reuse tail pfrag, if possible, or carve a new one from the
1723 * page allocator
1724 */
1725 dfrag = mptcp_pending_tail(sk);
1726 dfrag_collapsed = mptcp_frag_can_collapse_to(msk, pfrag, dfrag);
1727 if (!dfrag_collapsed) {
6e628cd3
PA		 1728 if (!sk_stream_memory_free(sk))
1729 goto wait_for_memory;
1730
d9ca1de8
PA		 1731 if (!mptcp_page_frag_refill(sk, pfrag))
1732 goto wait_for_memory;
1733
1734 dfrag = mptcp_carve_data_frag(msk, pfrag, pfrag->offset);
1735 frag_truesize = dfrag->overhead;
72511aab 1736 }
6d0060f6 1737
d9ca1de8
PA		 1738 /* we do not bound vs wspace, to allow a single packet.
1739 * memory accounting will prevent execessive memory usage
1740 * anyway
d5f49190 1741 */
d9ca1de8
PA		 1742 offset = dfrag->offset + dfrag->data_len;
1743 psize = pfrag->size - offset;
1744 psize = min_t(size_t, psize, msg_data_left(msg));
724cfd2e 1745 total_ts = psize + frag_truesize;
d9ca1de8 1746
8ce568ed 1747 if (!mptcp_wmem_alloc(sk, total_ts))
724cfd2e 1748 goto wait_for_memory;
724cfd2e 1749
d9ca1de8
PA		 1750 if (copy_page_from_iter(dfrag->page, offset, psize,
1751 &msg->msg_iter) != psize) {
87952603 1752 mptcp_wmem_uncharge(sk, psize + frag_truesize);
d9ca1de8
PA		 1753 ret = -EFAULT;
1754 goto out;
72511aab
FW		 1755 }
1756
d9ca1de8
PA		 1757 /* data successfully copied into the write queue */
1758 copied += psize;
1759 dfrag->data_len += psize;
1760 frag_truesize += psize;
1761 pfrag->offset += frag_truesize;
1762 WRITE_ONCE(msk->write_seq, msk->write_seq + psize);
1763
1764 /* charge data on mptcp pending queue to the msk socket
1765 * Note: we charge such data both to sk and ssk
fb529e62 1766 */
d9ca1de8 1767 sk_wmem_queued_add(sk, frag_truesize);
d9ca1de8
PA		 1768 if (!dfrag_collapsed) {
1769 get_page(dfrag->page);
1770 list_add_tail(&dfrag->list, &msk->rtx_queue);
1771 if (!msk->first_pending)
1772 WRITE_ONCE(msk->first_pending, dfrag);
fb529e62 1773 }
e4b61351 1774 pr_debug("msk=%p dfrag at seq=%llu len=%u sent=%u new=%d", msk,
d9ca1de8
PA		 1775 dfrag->data_seq, dfrag->data_len, dfrag->already_sent,
1776 !dfrag_collapsed);
6d0060f6 1777
d9ca1de8 1778 continue;
b51f9b80 1779
d9ca1de8 1780wait_for_memory:
5cf92bba 1781 mptcp_set_nospace(sk);
c2e6048f 1782 __mptcp_push_pending(sk, msg->msg_flags);
d9ca1de8
PA		 1783 ret = sk_stream_wait_memory(sk, &timeo);
1784 if (ret)
1785 goto out;
57040755 1786 }
6d0060f6 1787
13e16037 1788 if (copied)
c2e6048f 1789 __mptcp_push_pending(sk, msg->msg_flags);
d9ca1de8 1790
1954b860 1791out:
cec37a6e 1792 release_sock(sk);
8555c6bf 1793 return copied ? : ret;
f870fa0b
MM		 1794}
1795
7a6a6cbc
PA		 1796static void mptcp_wait_data(struct sock *sk, long *timeo)
1797{
1798 DEFINE_WAIT_FUNC(wait, woken_wake_function);
1799 struct mptcp_sock *msk = mptcp_sk(sk);
1800
1801 add_wait_queue(sk_sleep(sk), &wait);
1802 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1803
1804 sk_wait_event(sk, timeo,
3c90e377 1805 test_bit(MPTCP_DATA_READY, &msk->flags), &wait);
7a6a6cbc
PA		 1806
1807 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1808 remove_wait_queue(sk_sleep(sk), &wait);
1809}
1810
6771bfd9
FW		 1811static int __mptcp_recvmsg_mskq(struct mptcp_sock *msk,
1812 struct msghdr *msg,
b7f653b2
FW		 1813 size_t len, int flags,
1814 struct scm_timestamping_internal *tss,
1815 int *cmsg_flags)
6771bfd9 1816{
ca4fb892 1817 struct sk_buff *skb, *tmp;
6771bfd9
FW		 1818 int copied = 0;
1819
ca4fb892 1820 skb_queue_walk_safe(&msk->receive_queue, skb, tmp) {
6771bfd9
FW		 1821 u32 offset = MPTCP_SKB_CB(skb)->offset;
1822 u32 data_len = skb->len - offset;
1823 u32 count = min_t(size_t, len - copied, data_len);
1824 int err;
1825
d976092c
PA		 1826 if (!(flags & MSG_TRUNC)) {
1827 err = skb_copy_datagram_msg(skb, offset, msg, count);
1828 if (unlikely(err < 0)) {
1829 if (!copied)
1830 return err;
1831 break;
1832 }
6771bfd9
FW		 1833 }
1834
b7f653b2
FW		 1835 if (MPTCP_SKB_CB(skb)->has_rxtstamp) {
1836 tcp_update_recv_tstamps(skb, tss);
1837 *cmsg_flags |= MPTCP_CMSG_TS;
1838 }
1839
6771bfd9
FW		 1840 copied += count;
1841
1842 if (count < data_len) {
ca4fb892
YL		 1843 if (!(flags & MSG_PEEK))
1844 MPTCP_SKB_CB(skb)->offset += count;
6771bfd9
FW		 1845 break;
1846 }
1847
ca4fb892
YL		 1848 if (!(flags & MSG_PEEK)) {
1849 /* we will bulk release the skb memory later */
1850 skb->destructor = NULL;
ce599c51 1851 WRITE_ONCE(msk->rmem_released, msk->rmem_released + skb->truesize);
ca4fb892
YL		 1852 __skb_unlink(skb, &msk->receive_queue);
1853 __kfree_skb(skb);
1854 }
6771bfd9
FW		 1855
1856 if (copied >= len)
1857 break;
1858 }
1859
1860 return copied;
1861}
1862
a6b118fe
FW		 1863/* receive buffer autotuning. See tcp_rcv_space_adjust for more information.
1864 *
1865 * Only difference: Use highest rtt estimate of the subflows in use.
1866 */
1867static void mptcp_rcv_space_adjust(struct mptcp_sock *msk, int copied)
1868{
1869 struct mptcp_subflow_context *subflow;
1870 struct sock *sk = (struct sock *)msk;
1871 u32 time, advmss = 1;
1872 u64 rtt_us, mstamp;
1873
1874 sock_owned_by_me(sk);
1875
1876 if (copied <= 0)
1877 return;
1878
1879 msk->rcvq_space.copied += copied;
1880
1881 mstamp = div_u64(tcp_clock_ns(), NSEC_PER_USEC);
1882 time = tcp_stamp_us_delta(mstamp, msk->rcvq_space.time);
1883
1884 rtt_us = msk->rcvq_space.rtt_us;
1885 if (rtt_us && time < (rtt_us >> 3))
1886 return;
1887
1888 rtt_us = 0;
1889 mptcp_for_each_subflow(msk, subflow) {
1890 const struct tcp_sock *tp;
1891 u64 sf_rtt_us;
1892 u32 sf_advmss;
1893
1894 tp = tcp_sk(mptcp_subflow_tcp_sock(subflow));
1895
1896 sf_rtt_us = READ_ONCE(tp->rcv_rtt_est.rtt_us);
1897 sf_advmss = READ_ONCE(tp->advmss);
1898
1899 rtt_us = max(sf_rtt_us, rtt_us);
1900 advmss = max(sf_advmss, advmss);
1901 }
1902
1903 msk->rcvq_space.rtt_us = rtt_us;
1904 if (time < (rtt_us >> 3) || rtt_us == 0)
1905 return;
1906
1907 if (msk->rcvq_space.copied <= msk->rcvq_space.space)
1908 goto new_measure;
1909
1910 if (sock_net(sk)->ipv4.sysctl_tcp_moderate_rcvbuf &&
1911 !(sk->sk_userlocks & SOCK_RCVBUF_LOCK)) {
1912 int rcvmem, rcvbuf;
1913 u64 rcvwin, grow;
1914
1915 rcvwin = ((u64)msk->rcvq_space.copied << 1) + 16 * advmss;
1916
1917 grow = rcvwin * (msk->rcvq_space.copied - msk->rcvq_space.space);
1918
1919 do_div(grow, msk->rcvq_space.space);
1920 rcvwin += (grow << 1);
1921
1922 rcvmem = SKB_TRUESIZE(advmss + MAX_TCP_HEADER);
1923 while (tcp_win_from_space(sk, rcvmem) < advmss)
1924 rcvmem += 128;
1925
1926 do_div(rcvwin, advmss);
1927 rcvbuf = min_t(u64, rcvwin * rcvmem,
1928 sock_net(sk)->ipv4.sysctl_tcp_rmem[2]);
1929
1930 if (rcvbuf > sk->sk_rcvbuf) {
1931 u32 window_clamp;
1932
1933 window_clamp = tcp_win_from_space(sk, rcvbuf);
1934 WRITE_ONCE(sk->sk_rcvbuf, rcvbuf);
1935
1936 /* Make subflows follow along. If we do not do this, we
1937 * get drops at subflow level if skbs can't be moved to
1938 * the mptcp rx queue fast enough (announced rcv_win can
1939 * exceed ssk->sk_rcvbuf).
1940 */
1941 mptcp_for_each_subflow(msk, subflow) {
1942 struct sock *ssk;
c76c6956 1943 bool slow;
a6b118fe
FW		 1944
1945 ssk = mptcp_subflow_tcp_sock(subflow);
c76c6956 1946 slow = lock_sock_fast(ssk);
a6b118fe
FW		 1947 WRITE_ONCE(ssk->sk_rcvbuf, rcvbuf);
1948 tcp_sk(ssk)->window_clamp = window_clamp;
c76c6956
PA		 1949 tcp_cleanup_rbuf(ssk, 1);
1950 unlock_sock_fast(ssk, slow);
a6b118fe
FW		 1951 }
1952 }
1953 }
1954
1955 msk->rcvq_space.space = msk->rcvq_space.copied;
1956new_measure:
1957 msk->rcvq_space.copied = 0;
1958 msk->rcvq_space.time = mstamp;
1959}
1960
87952603
PA		 1961static void __mptcp_update_rmem(struct sock *sk)
1962{
1963 struct mptcp_sock *msk = mptcp_sk(sk);
1964
1965 if (!msk->rmem_released)
1966 return;
1967
1968 atomic_sub(msk->rmem_released, &sk->sk_rmem_alloc);
1969 sk_mem_uncharge(sk, msk->rmem_released);
ce599c51 1970 WRITE_ONCE(msk->rmem_released, 0);
87952603
PA		 1971}
1972
1973static void __mptcp_splice_receive_queue(struct sock *sk)
1974{
1975 struct mptcp_sock *msk = mptcp_sk(sk);
1976
1977 skb_queue_splice_tail_init(&sk->sk_receive_queue, &msk->receive_queue);
1978}
1979
e3859603 1980static bool __mptcp_move_skbs(struct mptcp_sock *msk)
6771bfd9 1981{
87952603 1982 struct sock *sk = (struct sock *)msk;
6771bfd9 1983 unsigned int moved = 0;
87952603 1984 bool ret, done;
d5f49190 1985
78962489 1986 mptcp_flush_join_list(msk);
6771bfd9
FW		 1987 do {
1988 struct sock *ssk = mptcp_subflow_recv_lookup(msk);
65f49fe7 1989 bool slowpath;
6771bfd9 1990
87952603
PA		 1991 /* we can have data pending in the subflows only if the msk
1992 * receive buffer was full at subflow_data_ready() time,
1993 * that is an unlikely slow path.
1994 */
1995 if (likely(!ssk))
6771bfd9
FW		 1996 break;
1997
65f49fe7 1998 slowpath = lock_sock_fast(ssk);
87952603 1999 mptcp_data_lock(sk);
e3859603 2000 __mptcp_update_rmem(sk);
6771bfd9 2001 done = __mptcp_move_skbs_from_subflow(msk, ssk, &moved);
87952603 2002 mptcp_data_unlock(sk);
499ada50
PA		 2003
2004 if (unlikely(ssk->sk_err))
2005 __mptcp_error_report(sk);
65f49fe7 2006 unlock_sock_fast(ssk, slowpath);
6771bfd9
FW		 2007 } while (!done);
2008
87952603
PA		 2009 /* acquire the data lock only if some input data is pending */
2010 ret = moved > 0;
2011 if (!RB_EMPTY_ROOT(&msk->out_of_order_queue) ||
2012 !skb_queue_empty_lockless(&sk->sk_receive_queue)) {
2013 mptcp_data_lock(sk);
2014 __mptcp_update_rmem(sk);
2015 ret |= __mptcp_ofo_queue(msk);
2016 __mptcp_splice_receive_queue(sk);
2017 mptcp_data_unlock(sk);
ab174ad8 2018 }
87952603
PA		 2019 if (ret)
2020 mptcp_check_data_fin((struct sock *)msk);
2021 return !skb_queue_empty(&msk->receive_queue);
6771bfd9
FW		 2022}
2023
f870fa0b
MM		 2024static int mptcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
2025 int nonblock, int flags, int *addr_len)
2026{
2027 struct mptcp_sock *msk = mptcp_sk(sk);
b7f653b2
FW		 2028 struct scm_timestamping_internal tss;
2029 int copied = 0, cmsg_flags = 0;
7a6a6cbc
PA		 2030 int target;
2031 long timeo;
f870fa0b 2032
cb9d80f4
PA		 2033 /* MSG_ERRQUEUE is really a no-op till we support IP_RECVERR */
2034 if (unlikely(flags & MSG_ERRQUEUE))
2035 return inet_recv_error(sk, msg, len, addr_len);
2036
87952603 2037 mptcp_lock_sock(sk, __mptcp_splice_receive_queue(sk));
fd897679
PA		 2038 if (unlikely(sk->sk_state == TCP_LISTEN)) {
2039 copied = -ENOTCONN;
2040 goto out_err;
2041 }
2042
7a6a6cbc
PA		 2043 timeo = sock_rcvtimeo(sk, nonblock);
2044
2045 len = min_t(size_t, len, INT_MAX);
2046 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
2047
05e3ecea 2048 while (copied < len) {
e3859603 2049 int bytes_read;
7a6a6cbc 2050
b7f653b2 2051 bytes_read = __mptcp_recvmsg_mskq(msk, msg, len - copied, flags, &tss, &cmsg_flags);
6771bfd9
FW		 2052 if (unlikely(bytes_read < 0)) {
2053 if (!copied)
2054 copied = bytes_read;
2055 goto out_err;
2056 }
7a6a6cbc 2057
6771bfd9 2058 copied += bytes_read;
7a6a6cbc 2059
ea4ca586 2060 /* be sure to advertise window change */
e3859603
PA		 2061 mptcp_cleanup_rbuf(msk);
2062
2063 if (skb_queue_empty(&msk->receive_queue) && __mptcp_move_skbs(msk))
2064 continue;
ea4ca586 2065
7a6a6cbc
PA		 2066 /* only the master socket status is relevant here. The exit
2067 * conditions mirror closely tcp_recvmsg()
2068 */
2069 if (copied >= target)
2070 break;
2071
2072 if (copied) {
2073 if (sk->sk_err ||
2074 sk->sk_state == TCP_CLOSE ||
2075 (sk->sk_shutdown & RCV_SHUTDOWN) ||
2076 !timeo ||
2077 signal_pending(current))
2078 break;
2079 } else {
2080 if (sk->sk_err) {
2081 copied = sock_error(sk);
2082 break;
2083 }
2084
5969856a
PA		 2085 if (test_and_clear_bit(MPTCP_WORK_EOF, &msk->flags))
2086 mptcp_check_for_eof(msk);
2087
87952603
PA		 2088 if (sk->sk_shutdown & RCV_SHUTDOWN) {
2089 /* race breaker: the shutdown could be after the
2090 * previous receive queue check
2091 */
e3859603 2092 if (__mptcp_move_skbs(msk))
87952603 2093 continue;
7a6a6cbc 2094 break;
87952603 2095 }
7a6a6cbc
PA		 2096
2097 if (sk->sk_state == TCP_CLOSE) {
2098 copied = -ENOTCONN;
2099 break;
2100 }
2101
2102 if (!timeo) {
2103 copied = -EAGAIN;
2104 break;
2105 }
2106
2107 if (signal_pending(current)) {
2108 copied = sock_intr_errno(timeo);
2109 break;
2110 }
2111 }
2112
2113 pr_debug("block timeout %ld", timeo);
7a6a6cbc 2114 mptcp_wait_data(sk, &timeo);
cec37a6e
PK		 2115 }
2116
87952603
PA		 2117 if (skb_queue_empty_lockless(&sk->sk_receive_queue) &&
2118 skb_queue_empty(&msk->receive_queue)) {
6771bfd9 2119 /* entire backlog drained, clear DATA_READY. */
7a6a6cbc 2120 clear_bit(MPTCP_DATA_READY, &msk->flags);
cec37a6e 2121
6771bfd9
FW		 2122 /* .. race-breaker: ssk might have gotten new data
2123 * after last __mptcp_move_skbs() returned false.
7a6a6cbc 2124 */
e3859603 2125 if (unlikely(__mptcp_move_skbs(msk)))
7a6a6cbc
PA		 2126 set_bit(MPTCP_DATA_READY, &msk->flags);
2127 }
3c90e377 2128
6771bfd9 2129out_err:
b7f653b2
FW		 2130 if (cmsg_flags && copied >= 0) {
2131 if (cmsg_flags & MPTCP_CMSG_TS)
2132 tcp_recv_timestamp(msg, sk, &tss);
2133 }
2134
6719331c
PA		 2135 pr_debug("msk=%p data_ready=%d rx queue empty=%d copied=%d",
2136 msk, test_bit(MPTCP_DATA_READY, &msk->flags),
87952603 2137 skb_queue_empty_lockless(&sk->sk_receive_queue), copied);
ca4fb892
YL		 2138 if (!(flags & MSG_PEEK))
2139 mptcp_rcv_space_adjust(msk, copied);
a6b118fe 2140
7a6a6cbc 2141 release_sock(sk);
cec37a6e
PK		 2142 return copied;
2143}
2144
b51f9b80
PA		 2145static void mptcp_retransmit_timer(struct timer_list *t)
2146{
2147 struct inet_connection_sock *icsk = from_timer(icsk, t,
2148 icsk_retransmit_timer);
2149 struct sock *sk = &icsk->icsk_inet.sk;
2d6f5a2b 2150 struct mptcp_sock *msk = mptcp_sk(sk);
b51f9b80
PA		 2151
2152 bh_lock_sock(sk);
2153 if (!sock_owned_by_user(sk)) {
2d6f5a2b
PA		 2154 /* we need a process context to retransmit */
2155 if (!test_and_set_bit(MPTCP_WORK_RTX, &msk->flags))
2156 mptcp_schedule_work(sk);
b51f9b80
PA		 2157 } else {
2158 /* delegate our work to tcp_release_cb() */
2d6f5a2b 2159 set_bit(MPTCP_RETRANSMIT, &msk->flags);
b51f9b80
PA		 2160 }
2161 bh_unlock_sock(sk);
2162 sock_put(sk);
2163}
2164
e16163b6
PA		 2165static void mptcp_timeout_timer(struct timer_list *t)
2166{
2167 struct sock *sk = from_timer(sk, t, sk_timer);
2168
2169 mptcp_schedule_work(sk);
b6d69fc8 2170 sock_put(sk);
e16163b6
PA		 2171}
2172
3b1d6210
PA		 2173/* Find an idle subflow. Return NULL if there is unacked data at tcp
2174 * level.
2175 *
2176 * A backup subflow is returned only if that is the only kind available.
2177 */
ff5a0b42 2178static struct sock *mptcp_subflow_get_retrans(struct mptcp_sock *msk)
3b1d6210 2179{
71b7dec2 2180 struct sock *backup = NULL, *pick = NULL;
3b1d6210 2181 struct mptcp_subflow_context *subflow;
71b7dec2 2182 int min_stale_count = INT_MAX;
3b1d6210
PA		 2183
2184 sock_owned_by_me((const struct sock *)msk);
2185
d5f49190 2186 if (__mptcp_check_fallback(msk))
d9ca1de8 2187 return NULL;
d5f49190 2188
3b1d6210
PA		 2189 mptcp_for_each_subflow(msk, subflow) {
2190 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2191
ff5a0b42 2192 if (!__mptcp_subflow_active(subflow))
d5f49190
PA		 2193 continue;
2194
71b7dec2
PA		 2195 /* still data outstanding at TCP level? skip this */
2196 if (!tcp_rtx_and_write_queues_empty(ssk)) {
2197 mptcp_pm_subflow_chk_stale(msk, ssk);
2198 min_stale_count = min_t(int, min_stale_count, subflow->stale_count);
2199 continue;
860975c6 2200 }
3b1d6210
PA		 2201
2202 if (subflow->backup) {
2203 if (!backup)
2204 backup = ssk;
2205 continue;
2206 }
2207
71b7dec2
PA		 2208 if (!pick)
2209 pick = ssk;
3b1d6210
PA		 2210 }
2211
71b7dec2
PA		 2212 if (pick)
2213 return pick;
2214
2215 /* use backup only if there are no progresses anywhere */
2216 return min_stale_count > 1 ? backup : NULL;
3b1d6210
PA		 2217}
2218
17aee05d
FW		 2219static void mptcp_dispose_initial_subflow(struct mptcp_sock *msk)
2220{
2221 if (msk->subflow) {
2222 iput(SOCK_INODE(msk->subflow));
2223 msk->subflow = NULL;
2224 }
2225}
2226
1e1d9d6f
PA		 2227bool __mptcp_retransmit_pending_data(struct sock *sk)
2228{
2229 struct mptcp_data_frag *cur, *rtx_head;
2230 struct mptcp_sock *msk = mptcp_sk(sk);
2231
2232 if (__mptcp_check_fallback(mptcp_sk(sk)))
2233 return false;
2234
2235 if (tcp_rtx_and_write_queues_empty(sk))
2236 return false;
2237
2238 /* the closing socket has some data untransmitted and/or unacked:
2239 * some data in the mptcp rtx queue has not really xmitted yet.
2240 * keep it simple and re-inject the whole mptcp level rtx queue
2241 */
2242 mptcp_data_lock(sk);
2243 __mptcp_clean_una_wakeup(sk);
2244 rtx_head = mptcp_rtx_head(sk);
2245 if (!rtx_head) {
2246 mptcp_data_unlock(sk);
2247 return false;
2248 }
2249
0d199e43 2250 msk->recovery_snd_nxt = msk->snd_nxt;
1e1d9d6f
PA		 2251 msk->recovery = true;
2252 mptcp_data_unlock(sk);
2253
2254 msk->first_pending = rtx_head;
1e1d9d6f
PA		 2255 msk->snd_burst = 0;
2256
2257 /* be sure to clear the "sent status" on all re-injected fragments */
2258 list_for_each_entry(cur, &msk->rtx_queue, list) {
2259 if (!cur->already_sent)
2260 break;
2261 cur->already_sent = 0;
2262 }
2263
2264 return true;
2265}
2266
cec37a6e
PK		 2267/* subflow sockets can be either outgoing (connect) or incoming
2268 * (accept).
2269 *
2270 * Outgoing subflows use in-kernel sockets.
2271 * Incoming subflows do not have their own 'struct socket' allocated,
2272 * so we need to use tcp_close() after detaching them from the mptcp
2273 * parent socket.
2274 */
a141e02e
FW		 2275static void __mptcp_close_ssk(struct sock *sk, struct sock *ssk,
2276 struct mptcp_subflow_context *subflow)
cec37a6e 2277{
e0be4931 2278 struct mptcp_sock *msk = mptcp_sk(sk);
1e1d9d6f 2279 bool need_push;
e0be4931 2280
cec37a6e
PK		 2281 list_del(&subflow->node);
2282
3f8b2667 2283 lock_sock_nested(ssk, SINGLE_DEPTH_NESTING);
e16163b6
PA		 2284
2285 /* if we are invoked by the msk cleanup code, the subflow is
2286 * already orphaned
2287 */
866f26f2 2288 if (ssk->sk_socket)
e16163b6 2289 sock_orphan(ssk);
e16163b6 2290
1e1d9d6f 2291 need_push = __mptcp_retransmit_pending_data(sk);
d7b1bfd0
PA		 2292 subflow->disposable = 1;
2293
e16163b6
PA		 2294 /* if ssk hit tcp_done(), tcp_cleanup_ulp() cleared the related ops
2295 * the ssk has been already destroyed, we just need to release the
2296 * reference owned by msk;
2297 */
2298 if (!inet_csk(ssk)->icsk_ulp_ops) {
2299 kfree_rcu(subflow, rcu);
cec37a6e 2300 } else {
d7b1bfd0 2301 /* otherwise tcp will dispose of the ssk and subflow ctx */
e16163b6
PA		 2302 __tcp_close(ssk, 0);
2303
2304 /* close acquired an extra ref */
2305 __sock_put(ssk);
cec37a6e 2306 }
e16163b6 2307 release_sock(ssk);
e16163b6
PA		 2308
2309 sock_put(ssk);
e0be4931
FW		 2310
2311 if (ssk == msk->last_snd)
2312 msk->last_snd = NULL;
17aee05d 2313
c8fe62f0
FW		 2314 if (ssk == msk->first)
2315 msk->first = NULL;
2316
17aee05d
FW		 2317 if (msk->subflow && ssk == msk->subflow->sk)
2318 mptcp_dispose_initial_subflow(msk);
1e1d9d6f
PA		 2319
2320 if (need_push)
2321 __mptcp_push_pending(sk, 0);
f870fa0b
MM		 2322}
2323
a141e02e
FW		 2324void mptcp_close_ssk(struct sock *sk, struct sock *ssk,
2325 struct mptcp_subflow_context *subflow)
2326{
b911c97c
FW		 2327 if (sk->sk_state == TCP_ESTABLISHED)
2328 mptcp_event(MPTCP_EVENT_SUB_CLOSED, mptcp_sk(sk), ssk, GFP_KERNEL);
a141e02e
FW		 2329 __mptcp_close_ssk(sk, ssk, subflow);
2330}
2331
dc24f8b4
PA		 2332static unsigned int mptcp_sync_mss(struct sock *sk, u32 pmtu)
2333{
2334 return 0;
2335}
2336
0e4f35d7
PA		 2337static void __mptcp_close_subflow(struct mptcp_sock *msk)
2338{
2339 struct mptcp_subflow_context *subflow, *tmp;
2340
3abc05d9
FW		 2341 might_sleep();
2342
0e4f35d7
PA		 2343 list_for_each_entry_safe(subflow, tmp, &msk->conn_list, node) {
2344 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2345
2346 if (inet_sk_state_load(ssk) != TCP_CLOSE)
2347 continue;
2348
40947e13
FW		 2349 /* 'subflow_data_ready' will re-sched once rx queue is empty */
2350 if (!skb_queue_empty_lockless(&ssk->sk_receive_queue))
2351 continue;
2352
a141e02e 2353 mptcp_close_ssk((struct sock *)msk, ssk, subflow);
0e4f35d7
PA		 2354 }
2355}
2356
e16163b6
PA		 2357static bool mptcp_check_close_timeout(const struct sock *sk)
2358{
2359 s32 delta = tcp_jiffies32 - inet_csk(sk)->icsk_mtup.probe_timestamp;
2360 struct mptcp_subflow_context *subflow;
2361
2362 if (delta >= TCP_TIMEWAIT_LEN)
2363 return true;
2364
2365 /* if all subflows are in closed status don't bother with additional
2366 * timeout
2367 */
2368 mptcp_for_each_subflow(mptcp_sk(sk), subflow) {
2369 if (inet_sk_state_load(mptcp_subflow_tcp_sock(subflow)) !=
2370 TCP_CLOSE)
2371 return false;
2372 }
2373 return true;
2374}
2375
50c504a2
FW		 2376static void mptcp_check_fastclose(struct mptcp_sock *msk)
2377{
2378 struct mptcp_subflow_context *subflow, *tmp;
2379 struct sock *sk = &msk->sk.icsk_inet.sk;
2380
2381 if (likely(!READ_ONCE(msk->rcv_fastclose)))
2382 return;
2383
2384 mptcp_token_destroy(msk);
2385
2386 list_for_each_entry_safe(subflow, tmp, &msk->conn_list, node) {
2387 struct sock *tcp_sk = mptcp_subflow_tcp_sock(subflow);
75e908c3 2388 bool slow;
50c504a2 2389
75e908c3 2390 slow = lock_sock_fast(tcp_sk);
50c504a2
FW		 2391 if (tcp_sk->sk_state != TCP_CLOSE) {
2392 tcp_send_active_reset(tcp_sk, GFP_ATOMIC);
2393 tcp_set_state(tcp_sk, TCP_CLOSE);
2394 }
75e908c3 2395 unlock_sock_fast(tcp_sk, slow);
50c504a2
FW		 2396 }
2397
2398 inet_sk_state_store(sk, TCP_CLOSE);
2399 sk->sk_shutdown = SHUTDOWN_MASK;
2400 smp_mb__before_atomic(); /* SHUTDOWN must be visible first */
2401 set_bit(MPTCP_DATA_READY, &msk->flags);
2402 set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags);
2403
2404 mptcp_close_wake_up(sk);
2405}
2406
2948d0a1 2407static void __mptcp_retrans(struct sock *sk)
80992017 2408{
2948d0a1 2409 struct mptcp_sock *msk = mptcp_sk(sk);
caf971df 2410 struct mptcp_sendmsg_info info = {};
3b1d6210 2411 struct mptcp_data_frag *dfrag;
3b1d6210 2412 size_t copied = 0;
2948d0a1
PA		 2413 struct sock *ssk;
2414 int ret;
3b1d6210 2415
b5941f06 2416 mptcp_clean_una_wakeup(sk);
3b1d6210 2417 dfrag = mptcp_rtx_head(sk);
6477dd39
MM		 2418 if (!dfrag) {
2419 if (mptcp_data_fin_enabled(msk)) {
2420 struct inet_connection_sock *icsk = inet_csk(sk);
2421
2422 icsk->icsk_retransmits++;
2423 mptcp_set_datafin_timeout(sk);
2424 mptcp_send_ack(msk);
2425
2426 goto reset_timer;
2427 }
2428
2948d0a1 2429 return;
6477dd39 2430 }
3b1d6210
PA		 2431
2432 ssk = mptcp_subflow_get_retrans(msk);
2433 if (!ssk)
2948d0a1 2434 goto reset_timer;
3b1d6210
PA		 2435
2436 lock_sock(ssk);
2437
d9ca1de8
PA		 2438 /* limit retransmission to the bytes already sent on some subflows */
2439 info.sent = 0;
4e14867d
PA		 2440 info.limit = READ_ONCE(msk->csum_enabled) ? dfrag->data_len : dfrag->already_sent;
2441 while (info.sent < info.limit) {
d9ca1de8 2442 ret = mptcp_sendmsg_frag(sk, ssk, dfrag, &info);
6f8a612a 2443 if (ret <= 0)
3b1d6210
PA		 2444 break;
2445
fc518953 2446 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_RETRANSSEGS);
3b1d6210 2447 copied += ret;
d9ca1de8 2448 info.sent += ret;
3b1d6210 2449 }
4e14867d
PA		 2450 if (copied) {
2451 dfrag->already_sent = max(dfrag->already_sent, info.sent);
caf971df
PA		 2452 tcp_push(ssk, 0, info.mss_now, tcp_sk(ssk)->nonagle,
2453 info.size_goal);
4e14867d 2454 }
3b1d6210 2455
3b1d6210
PA		 2456 release_sock(ssk);
2457
2948d0a1 2458reset_timer:
3b1d6210
PA		 2459 if (!mptcp_timer_pending(sk))
2460 mptcp_reset_timer(sk);
2948d0a1
PA		 2461}
2462
2463static void mptcp_worker(struct work_struct *work)
2464{
2465 struct mptcp_sock *msk = container_of(work, struct mptcp_sock, work);
2466 struct sock *sk = &msk->sk.icsk_inet.sk;
2467 int state;
2468
2469 lock_sock(sk);
2470 state = sk->sk_state;
2471 if (unlikely(state == TCP_CLOSE))
2472 goto unlock;
2473
2474 mptcp_check_data_fin_ack(sk);
78962489 2475 mptcp_flush_join_list(msk);
2948d0a1
PA		 2476
2477 mptcp_check_fastclose(msk);
2478
2479 if (msk->pm.status)
2480 mptcp_pm_nl_work(msk);
2481
2482 if (test_and_clear_bit(MPTCP_WORK_EOF, &msk->flags))
2483 mptcp_check_for_eof(msk);
2484
2485 __mptcp_check_send_data_fin(sk);
2486 mptcp_check_data_fin(sk);
2487
2488 /* There is no point in keeping around an orphaned sk timedout or
2489 * closed, but we need the msk around to reply to incoming DATA_FIN,
2490 * even if it is orphaned and in FIN_WAIT2 state
2491 */
2492 if (sock_flag(sk, SOCK_DEAD) &&
2493 (mptcp_check_close_timeout(sk) || sk->sk_state == TCP_CLOSE)) {
2494 inet_sk_state_store(sk, TCP_CLOSE);
2495 __mptcp_destroy_sock(sk);
2496 goto unlock;
2497 }
2498
2499 if (test_and_clear_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags))
2500 __mptcp_close_subflow(msk);
2501
2502 if (test_and_clear_bit(MPTCP_WORK_RTX, &msk->flags))
2503 __mptcp_retrans(sk);
3b1d6210
PA		 2504
2505unlock:
80992017
PA		 2506 release_sock(sk);
2507 sock_put(sk);
2508}
2509
784325e9 2510static int __mptcp_init_sock(struct sock *sk)
f870fa0b 2511{
cec37a6e
PK		 2512 struct mptcp_sock *msk = mptcp_sk(sk);
2513
ec3edaa7
PK		 2514 spin_lock_init(&msk->join_list_lock);
2515
cec37a6e 2516 INIT_LIST_HEAD(&msk->conn_list);
ec3edaa7 2517 INIT_LIST_HEAD(&msk->join_list);
18b683bf 2518 INIT_LIST_HEAD(&msk->rtx_queue);
80992017 2519 INIT_WORK(&msk->work, mptcp_worker);
87952603 2520 __skb_queue_head_init(&msk->receive_queue);
ab174ad8 2521 msk->out_of_order_queue = RB_ROOT;
f0e6a4cf 2522 msk->first_pending = NULL;
e93da928 2523 msk->wmem_reserved = 0;
ce599c51 2524 WRITE_ONCE(msk->rmem_released, 0);
33d41c9c 2525 msk->timer_ival = TCP_RTO_MIN;
cec37a6e 2526
8ab183de 2527 msk->first = NULL;
dc24f8b4 2528 inet_csk(sk)->icsk_sync_mss = mptcp_sync_mss;
752e9067 2529 WRITE_ONCE(msk->csum_enabled, mptcp_is_checksum_enabled(sock_net(sk)));
1e1d9d6f 2530 msk->recovery = false;
8ab183de 2531
1b1c7a0e
PK		 2532 mptcp_pm_data_init(msk);
2533
b51f9b80
PA		 2534 /* re-use the csk retrans timer for MPTCP-level retrans */
2535 timer_setup(&msk->sk.icsk_retransmit_timer, mptcp_retransmit_timer, 0);
e16163b6 2536 timer_setup(&sk->sk_timer, mptcp_timeout_timer, 0);
aa1fbd94 2537
f870fa0b
MM		 2538 return 0;
2539}
2540
784325e9
MB		 2541static int mptcp_init_sock(struct sock *sk)
2542{
20b5759f 2543 struct inet_connection_sock *icsk = inet_csk(sk);
fc518953
FW		 2544 struct net *net = sock_net(sk);
2545 int ret;
18b683bf 2546
b6c08380
GT		 2547 ret = __mptcp_init_sock(sk);
2548 if (ret)
2549 return ret;
2550
fc518953
FW		 2551 if (!mptcp_is_enabled(net))
2552 return -ENOPROTOOPT;
2553
2554 if (unlikely(!net->mib.mptcp_statistics) && !mptcp_mib_alloc(net))
2555 return -ENOMEM;
2556
fa68018d
PA		 2557 ret = __mptcp_socket_create(mptcp_sk(sk));
2558 if (ret)
2559 return ret;
2560
20b5759f
PA		 2561 /* fetch the ca name; do it outside __mptcp_init_sock(), so that clone will
2562 * propagate the correct value
2563 */
2564 tcp_assign_congestion_control(sk);
2565 strcpy(mptcp_sk(sk)->ca_name, icsk->icsk_ca_ops->name);
2566
2567 /* no need to keep a reference to the ops, the name will suffice */
2568 tcp_cleanup_congestion_control(sk);
2569 icsk->icsk_ca_ops = NULL;
2570
d027236c 2571 sk_sockets_allocated_inc(sk);
a6b118fe 2572 sk->sk_rcvbuf = sock_net(sk)->ipv4.sysctl_tcp_rmem[1];
da51aef5 2573 sk->sk_sndbuf = sock_net(sk)->ipv4.sysctl_tcp_wmem[1];
d027236c 2574
18b683bf
PA		 2575 return 0;
2576}
2577
2578static void __mptcp_clear_xmit(struct sock *sk)
2579{
2580 struct mptcp_sock *msk = mptcp_sk(sk);
2581 struct mptcp_data_frag *dtmp, *dfrag;
2582
d9ca1de8 2583 WRITE_ONCE(msk->first_pending, NULL);
18b683bf 2584 list_for_each_entry_safe(dfrag, dtmp, &msk->rtx_queue, list)
d027236c 2585 dfrag_clear(sk, dfrag);
784325e9
MB		 2586}
2587
80992017
PA		 2588static void mptcp_cancel_work(struct sock *sk)
2589{
2590 struct mptcp_sock *msk = mptcp_sk(sk);
2591
b2771d24 2592 if (cancel_work_sync(&msk->work))
e16163b6 2593 __sock_put(sk);
80992017
PA		 2594}
2595
d0876b22 2596void mptcp_subflow_shutdown(struct sock *sk, struct sock *ssk, int how)
21498490
PK		 2597{
2598 lock_sock(ssk);
2599
2600 switch (ssk->sk_state) {
2601 case TCP_LISTEN:
2602 if (!(how & RCV_SHUTDOWN))
2603 break;
df561f66 2604 fallthrough;
21498490
PK		 2605 case TCP_SYN_SENT:
2606 tcp_disconnect(ssk, O_NONBLOCK);
2607 break;
2608 default:
43b54c6e
MM		 2609 if (__mptcp_check_fallback(mptcp_sk(sk))) {
2610 pr_debug("Fallback");
2611 ssk->sk_shutdown |= how;
2612 tcp_shutdown(ssk, how);
2613 } else {
2614 pr_debug("Sending DATA_FIN on subflow %p", ssk);
43b54c6e 2615 tcp_send_ack(ssk);
6477dd39
MM		 2616 if (!mptcp_timer_pending(sk))
2617 mptcp_reset_timer(sk);
43b54c6e 2618 }
21498490
PK		 2619 break;
2620 }
2621
21498490
PK		 2622 release_sock(ssk);
2623}
2624
6920b851
MM		 2625static const unsigned char new_state[16] = {
2626 /* current state: new state: action: */
2627 [0 /* (Invalid) */] = TCP_CLOSE,
2628 [TCP_ESTABLISHED] = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2629 [TCP_SYN_SENT] = TCP_CLOSE,
2630 [TCP_SYN_RECV] = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2631 [TCP_FIN_WAIT1] = TCP_FIN_WAIT1,
2632 [TCP_FIN_WAIT2] = TCP_FIN_WAIT2,
2633 [TCP_TIME_WAIT] = TCP_CLOSE, /* should not happen ! */
2634 [TCP_CLOSE] = TCP_CLOSE,
2635 [TCP_CLOSE_WAIT] = TCP_LAST_ACK | TCP_ACTION_FIN,
2636 [TCP_LAST_ACK] = TCP_LAST_ACK,
2637 [TCP_LISTEN] = TCP_CLOSE,
2638 [TCP_CLOSING] = TCP_CLOSING,
2639 [TCP_NEW_SYN_RECV] = TCP_CLOSE, /* should not happen ! */
2640};
2641
2642static int mptcp_close_state(struct sock *sk)
2643{
2644 int next = (int)new_state[sk->sk_state];
2645 int ns = next & TCP_STATE_MASK;
2646
2647 inet_sk_state_store(sk, ns);
2648
2649 return next & TCP_ACTION_FIN;
2650}
2651
e16163b6 2652static void __mptcp_check_send_data_fin(struct sock *sk)
f870fa0b 2653{
e16163b6 2654 struct mptcp_subflow_context *subflow;
f870fa0b
MM		 2655 struct mptcp_sock *msk = mptcp_sk(sk);
2656
e16163b6
PA		 2657 pr_debug("msk=%p snd_data_fin_enable=%d pending=%d snd_nxt=%llu write_seq=%llu",
2658 msk, msk->snd_data_fin_enable, !!mptcp_send_head(sk),
2659 msk->snd_nxt, msk->write_seq);
43b54c6e 2660
e16163b6
PA		 2661 /* we still need to enqueue subflows or not really shutting down,
2662 * skip this
2663 */
2664 if (!msk->snd_data_fin_enable || msk->snd_nxt + 1 != msk->write_seq ||
2665 mptcp_send_head(sk))
2666 return;
2667
2668 WRITE_ONCE(msk->snd_nxt, msk->write_seq);
2669
26aa2314
PA		 2670 /* fallback socket will not get data_fin/ack, can move to the next
2671 * state now
2672 */
2673 if (__mptcp_check_fallback(msk)) {
2674 if ((1 << sk->sk_state) & (TCPF_CLOSING | TCPF_LAST_ACK)) {
2675 inet_sk_state_store(sk, TCP_CLOSE);
2676 mptcp_close_wake_up(sk);
2677 } else if (sk->sk_state == TCP_FIN_WAIT1) {
2678 inet_sk_state_store(sk, TCP_FIN_WAIT2);
2679 }
43b54c6e
MM		 2680 }
2681
78962489 2682 mptcp_flush_join_list(msk);
e16163b6
PA		 2683 mptcp_for_each_subflow(msk, subflow) {
2684 struct sock *tcp_sk = mptcp_subflow_tcp_sock(subflow);
43b54c6e 2685
e16163b6 2686 mptcp_subflow_shutdown(sk, tcp_sk, SEND_SHUTDOWN);
43b54c6e 2687 }
e16163b6 2688}
2c22c06c 2689
e16163b6
PA		 2690static void __mptcp_wr_shutdown(struct sock *sk)
2691{
2692 struct mptcp_sock *msk = mptcp_sk(sk);
43b54c6e 2693
e16163b6
PA		 2694 pr_debug("msk=%p snd_data_fin_enable=%d shutdown=%x state=%d pending=%d",
2695 msk, msk->snd_data_fin_enable, sk->sk_shutdown, sk->sk_state,
2696 !!mptcp_send_head(sk));
2697
2698 /* will be ignored by fallback sockets */
2699 WRITE_ONCE(msk->write_seq, msk->write_seq + 1);
2700 WRITE_ONCE(msk->snd_data_fin_enable, 1);
2701
2702 __mptcp_check_send_data_fin(sk);
2703}
2704
2705static void __mptcp_destroy_sock(struct sock *sk)
2706{
2707 struct mptcp_subflow_context *subflow, *tmp;
2708 struct mptcp_sock *msk = mptcp_sk(sk);
2709 LIST_HEAD(conn_list);
2710
2711 pr_debug("msk=%p", msk);
f870fa0b 2712
3abc05d9
FW		 2713 might_sleep();
2714
10f6d46c
PA		 2715 /* be sure to always acquire the join list lock, to sync vs
2716 * mptcp_finish_join().
2717 */
2718 spin_lock_bh(&msk->join_list_lock);
2719 list_splice_tail_init(&msk->join_list, &msk->conn_list);
2720 spin_unlock_bh(&msk->join_list_lock);
b2c5b614
FW		 2721 list_splice_init(&msk->conn_list, &conn_list);
2722
6e628cd3 2723 sk_stop_timer(sk, &msk->sk.icsk_retransmit_timer);
e16163b6
PA		 2724 sk_stop_timer(sk, &sk->sk_timer);
2725 msk->pm.status = 0;
b2c5b614
FW		 2726
2727 list_for_each_entry_safe(subflow, tmp, &conn_list, node) {
cec37a6e 2728 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
e16163b6 2729 __mptcp_close_ssk(sk, ssk, subflow);
f870fa0b
MM		 2730 }
2731
e16163b6 2732 sk->sk_prot->destroy(sk);
80992017 2733
e93da928 2734 WARN_ON_ONCE(msk->wmem_reserved);
87952603 2735 WARN_ON_ONCE(msk->rmem_released);
e16163b6
PA		 2736 sk_stream_kill_queues(sk);
2737 xfrm_sk_free_policy(sk);
aa1fbd94 2738
e16163b6 2739 sk_refcnt_debug_release(sk);
17aee05d 2740 mptcp_dispose_initial_subflow(msk);
e16163b6
PA		 2741 sock_put(sk);
2742}
2743
2744static void mptcp_close(struct sock *sk, long timeout)
2745{
2746 struct mptcp_subflow_context *subflow;
2747 bool do_cancel_work = false;
2748
2749 lock_sock(sk);
2750 sk->sk_shutdown = SHUTDOWN_MASK;
2751
2752 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE)) {
2753 inet_sk_state_store(sk, TCP_CLOSE);
2754 goto cleanup;
2755 }
6771bfd9 2756
e16163b6
PA		 2757 if (mptcp_close_state(sk))
2758 __mptcp_wr_shutdown(sk);
2759
2760 sk_stream_wait_close(sk, timeout);
2761
2762cleanup:
2763 /* orphan all the subflows */
2764 inet_csk(sk)->icsk_mtup.probe_timestamp = tcp_jiffies32;
44227915 2765 mptcp_for_each_subflow(mptcp_sk(sk), subflow) {
e16163b6 2766 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
866f26f2 2767 bool slow = lock_sock_fast(ssk);
e16163b6 2768
e16163b6
PA		 2769 sock_orphan(ssk);
2770 unlock_sock_fast(ssk, slow);
e16163b6
PA		 2771 }
2772 sock_orphan(sk);
2773
2774 sock_hold(sk);
2775 pr_debug("msk=%p state=%d", sk, sk->sk_state);
2776 if (sk->sk_state == TCP_CLOSE) {
2777 __mptcp_destroy_sock(sk);
2778 do_cancel_work = true;
2779 } else {
2780 sk_reset_timer(sk, &sk->sk_timer, jiffies + TCP_TIMEWAIT_LEN);
2781 }
2782 release_sock(sk);
2783 if (do_cancel_work)
2784 mptcp_cancel_work(sk);
b911c97c
FW		 2785
2786 if (mptcp_sk(sk)->token)
2787 mptcp_event(MPTCP_EVENT_CLOSED, mptcp_sk(sk), NULL, GFP_KERNEL);
2788
e16163b6 2789 sock_put(sk);
f870fa0b
MM		 2790}
2791
cf7da0d6
PK		 2792static void mptcp_copy_inaddrs(struct sock *msk, const struct sock *ssk)
2793{
2794#if IS_ENABLED(CONFIG_MPTCP_IPV6)
2795 const struct ipv6_pinfo *ssk6 = inet6_sk(ssk);
2796 struct ipv6_pinfo *msk6 = inet6_sk(msk);
2797
2798 msk->sk_v6_daddr = ssk->sk_v6_daddr;
2799 msk->sk_v6_rcv_saddr = ssk->sk_v6_rcv_saddr;
2800
2801 if (msk6 && ssk6) {
2802 msk6->saddr = ssk6->saddr;
2803 msk6->flow_label = ssk6->flow_label;
2804 }
2805#endif
2806
2807 inet_sk(msk)->inet_num = inet_sk(ssk)->inet_num;
2808 inet_sk(msk)->inet_dport = inet_sk(ssk)->inet_dport;
2809 inet_sk(msk)->inet_sport = inet_sk(ssk)->inet_sport;
2810 inet_sk(msk)->inet_daddr = inet_sk(ssk)->inet_daddr;
2811 inet_sk(msk)->inet_saddr = inet_sk(ssk)->inet_saddr;
2812 inet_sk(msk)->inet_rcv_saddr = inet_sk(ssk)->inet_rcv_saddr;
2813}
2814
18b683bf
PA		 2815static int mptcp_disconnect(struct sock *sk, int flags)
2816{
76e2a55d
PA		 2817 struct mptcp_subflow_context *subflow;
2818 struct mptcp_sock *msk = mptcp_sk(sk);
2819
78962489
FW		 2820 mptcp_do_flush_join_list(msk);
2821
13a9499e
PA		 2822 mptcp_for_each_subflow(msk, subflow) {
2823 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2824
2825 lock_sock(ssk);
2826 tcp_disconnect(ssk, flags);
2827 release_sock(ssk);
2828 }
42c556fe 2829 return 0;
18b683bf
PA		 2830}
2831
b0519de8
FW		 2832#if IS_ENABLED(CONFIG_MPTCP_IPV6)
2833static struct ipv6_pinfo *mptcp_inet6_sk(const struct sock *sk)
2834{
2835 unsigned int offset = sizeof(struct mptcp6_sock) - sizeof(struct ipv6_pinfo);
2836
2837 return (struct ipv6_pinfo *)(((u8 *)sk) + offset);
2838}
2839#endif
2840
fca5c82c 2841struct sock *mptcp_sk_clone(const struct sock *sk,
cfde141e 2842 const struct mptcp_options_received *mp_opt,
fca5c82c 2843 struct request_sock *req)
b0519de8 2844{
58b09919 2845 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
b0519de8 2846 struct sock *nsk = sk_clone_lock(sk, GFP_ATOMIC);
58b09919
PA		 2847 struct mptcp_sock *msk;
2848 u64 ack_seq;
b0519de8
FW		 2849
2850 if (!nsk)
2851 return NULL;
2852
2853#if IS_ENABLED(CONFIG_MPTCP_IPV6)
2854 if (nsk->sk_family == AF_INET6)
2855 inet_sk(nsk)->pinet6 = mptcp_inet6_sk(nsk);
2856#endif
2857
58b09919
PA		 2858 __mptcp_init_sock(nsk);
2859
2860 msk = mptcp_sk(nsk);
2861 msk->local_key = subflow_req->local_key;
2862 msk->token = subflow_req->token;
2863 msk->subflow = NULL;
b93df08c 2864 WRITE_ONCE(msk->fully_established, false);
74c7dfbe 2865 if (mp_opt->suboptions & OPTION_MPTCP_CSUMREQD)
06251181 2866 WRITE_ONCE(msk->csum_enabled, true);
58b09919 2867
58b09919 2868 msk->write_seq = subflow_req->idsn + 1;
eaa2ffab 2869 msk->snd_nxt = msk->write_seq;
7439d687
PA		 2870 msk->snd_una = msk->write_seq;
2871 msk->wnd_end = msk->snd_nxt + req->rsk_rcv_wnd;
df00b087 2872 msk->setsockopt_seq = mptcp_sk(sk)->setsockopt_seq;
6f8a612a 2873
74c7dfbe 2874 if (mp_opt->suboptions & OPTIONS_MPTCP_MPC) {
58b09919 2875 msk->can_ack = true;
cfde141e 2876 msk->remote_key = mp_opt->sndr_key;
58b09919
PA		 2877 mptcp_crypto_key_sha(msk->remote_key, NULL, &ack_seq);
2878 ack_seq++;
917944da 2879 WRITE_ONCE(msk->ack_seq, ack_seq);
fa3fe2b1 2880 WRITE_ONCE(msk->rcv_wnd_sent, ack_seq);
58b09919 2881 }
7f20d5fc 2882
5e20087d 2883 sock_reset_flag(nsk, SOCK_RCU_FREE);
7f20d5fc
PA		 2884 /* will be fully established after successful MPC subflow creation */
2885 inet_sk_state_store(nsk, TCP_SYN_RECV);
0c148460
PA		 2886
2887 security_inet_csk_clone(nsk, req);
58b09919
PA		 2888 bh_unlock_sock(nsk);
2889
2890 /* keep a single reference */
2891 __sock_put(nsk);
b0519de8
FW		 2892 return nsk;
2893}
2894
a6b118fe
FW		 2895void mptcp_rcv_space_init(struct mptcp_sock *msk, const struct sock *ssk)
2896{
2897 const struct tcp_sock *tp = tcp_sk(ssk);
2898
2899 msk->rcvq_space.copied = 0;
2900 msk->rcvq_space.rtt_us = 0;
2901
2902 msk->rcvq_space.time = tp->tcp_mstamp;
2903
2904 /* initial rcv_space offering made to peer */
2905 msk->rcvq_space.space = min_t(u32, tp->rcv_wnd,
2906 TCP_INIT_CWND * tp->advmss);
2907 if (msk->rcvq_space.space == 0)
2908 msk->rcvq_space.space = TCP_INIT_CWND * TCP_MSS_DEFAULT;
6f8a612a 2909
7439d687 2910 WRITE_ONCE(msk->wnd_end, msk->snd_nxt + tcp_sk(ssk)->snd_wnd);
a6b118fe
FW		 2911}
2912
cf7da0d6
PK		 2913static struct sock *mptcp_accept(struct sock *sk, int flags, int *err,
2914 bool kern)
2915{
2916 struct mptcp_sock *msk = mptcp_sk(sk);
2917 struct socket *listener;
2918 struct sock *newsk;
2919
2920 listener = __mptcp_nmpc_socket(msk);
2921 if (WARN_ON_ONCE(!listener)) {
2922 *err = -EINVAL;
2923 return NULL;
2924 }
2925
2926 pr_debug("msk=%p, listener=%p", msk, mptcp_subflow_ctx(listener->sk));
2927 newsk = inet_csk_accept(listener->sk, flags, err, kern);
2928 if (!newsk)
2929 return NULL;
2930
2931 pr_debug("msk=%p, subflow is mptcp=%d", msk, sk_is_mptcp(newsk));
cf7da0d6
PK		 2932 if (sk_is_mptcp(newsk)) {
2933 struct mptcp_subflow_context *subflow;
2934 struct sock *new_mptcp_sock;
cf7da0d6
PK		 2935
2936 subflow = mptcp_subflow_ctx(newsk);
58b09919 2937 new_mptcp_sock = subflow->conn;
cf7da0d6 2938
58b09919
PA		 2939 /* is_mptcp should be false if subflow->conn is missing, see
2940 * subflow_syn_recv_sock()
2941 */
2942 if (WARN_ON_ONCE(!new_mptcp_sock)) {
2943 tcp_sk(newsk)->is_mptcp = 0;
2944 return newsk;
cf7da0d6
PK		 2945 }
2946
58b09919
PA		 2947 /* acquire the 2nd reference for the owning socket */
2948 sock_hold(new_mptcp_sock);
cf7da0d6 2949 newsk = new_mptcp_sock;
0397c6d8 2950 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPCAPABLEPASSIVEACK);
fc518953
FW		 2951 } else {
2952 MPTCP_INC_STATS(sock_net(sk),
2953 MPTCP_MIB_MPCAPABLEPASSIVEFALLBACK);
cf7da0d6
PK		 2954 }
2955
2956 return newsk;
2957}
2958
5c8c1640
GT		 2959void mptcp_destroy_common(struct mptcp_sock *msk)
2960{
87952603
PA		 2961 struct sock *sk = (struct sock *)msk;
2962
6e628cd3
PA		 2963 __mptcp_clear_xmit(sk);
2964
87952603
PA		 2965 /* move to sk_receive_queue, sk_stream_kill_queues will purge it */
2966 skb_queue_splice_tail_init(&msk->receive_queue, &sk->sk_receive_queue);
2967
5c8c1640
GT		 2968 skb_rbtree_purge(&msk->out_of_order_queue);
2969 mptcp_token_destroy(msk);
2970 mptcp_pm_free_anno_list(msk);
2971}
2972
79c0949e
PK		 2973static void mptcp_destroy(struct sock *sk)
2974{
c9fd9c5f
FW		 2975 struct mptcp_sock *msk = mptcp_sk(sk);
2976
5c8c1640 2977 mptcp_destroy_common(msk);
d027236c 2978 sk_sockets_allocated_dec(sk);
79c0949e
PK		 2979}
2980
6e628cd3
PA		 2981void __mptcp_data_acked(struct sock *sk)
2982{
2983 if (!sock_owned_by_user(sk))
2984 __mptcp_clean_una(sk);
2985 else
2986 set_bit(MPTCP_CLEAN_UNA, &mptcp_sk(sk)->flags);
2987
2988 if (mptcp_pending_data_fin_ack(sk))
2989 mptcp_schedule_work(sk);
2990}
2991
219d0499 2992void __mptcp_check_push(struct sock *sk, struct sock *ssk)
6e628cd3
PA		 2993{
2994 if (!mptcp_send_head(sk))
2995 return;
2996
40dc9416 2997 if (!sock_owned_by_user(sk)) {
b19bc294
PA		 2998 struct sock *xmit_ssk = mptcp_subflow_get_send(mptcp_sk(sk));
2999
3000 if (xmit_ssk == ssk)
40dc9416 3001 __mptcp_subflow_push_pending(sk, ssk);
b19bc294
PA		 3002 else if (xmit_ssk)
3003 mptcp_subflow_delegate(mptcp_subflow_ctx(xmit_ssk));
40dc9416 3004 } else {
6e628cd3 3005 set_bit(MPTCP_PUSH_PENDING, &mptcp_sk(sk)->flags);
40dc9416 3006 }
6e628cd3
PA		 3007}
3008
e93da928 3009/* processes deferred events and flush wmem */
14c441b5
PA		 3010static void mptcp_release_cb(struct sock *sk)
3011{
c2e6048f 3012 for (;;) {
2d6f5a2b
PA		 3013 unsigned long flags = 0;
3014
c2e6048f 3015 if (test_and_clear_bit(MPTCP_PUSH_PENDING, &mptcp_sk(sk)->flags))
2e5de7e0 3016 flags |= BIT(MPTCP_PUSH_PENDING);
2d6f5a2b
PA		 3017 if (test_and_clear_bit(MPTCP_RETRANSMIT, &mptcp_sk(sk)->flags))
3018 flags |= BIT(MPTCP_RETRANSMIT);
c2e6048f
PA		 3019 if (!flags)
3020 break;
3021
3022 /* the following actions acquire the subflow socket lock
6e628cd3
PA		 3023 *
3024 * 1) can't be invoked in atomic scope
3025 * 2) must avoid ABBA deadlock with msk socket spinlock: the RX
3026 * datapath acquires the msk socket spinlock while helding
3027 * the subflow socket lock
3028 */
3029
3030 spin_unlock_bh(&sk->sk_lock.slock);
2e5de7e0 3031 if (flags & BIT(MPTCP_PUSH_PENDING))
c2e6048f 3032 __mptcp_push_pending(sk, 0);
2d6f5a2b
PA		 3033 if (flags & BIT(MPTCP_RETRANSMIT))
3034 __mptcp_retrans(sk);
c2e6048f
PA		 3035
3036 cond_resched();
6e628cd3
PA		 3037 spin_lock_bh(&sk->sk_lock.slock);
3038 }
c2e6048f 3039
490274b4
PA		 3040 /* be sure to set the current sk state before tacking actions
3041 * depending on sk_state
3042 */
3043 if (test_and_clear_bit(MPTCP_CONNECTED, &mptcp_sk(sk)->flags))
3044 __mptcp_set_connected(sk);
c2e6048f 3045 if (test_and_clear_bit(MPTCP_CLEAN_UNA, &mptcp_sk(sk)->flags))
417789df 3046 __mptcp_clean_una_wakeup(sk);
15cc1045
PA		 3047 if (test_and_clear_bit(MPTCP_ERROR_REPORT, &mptcp_sk(sk)->flags))
3048 __mptcp_error_report(sk);
6e628cd3 3049
c2e6048f
PA		 3050 /* push_pending may touch wmem_reserved, ensure we do the cleanup
3051 * later
3052 */
e93da928 3053 __mptcp_update_wmem(sk);
87952603 3054 __mptcp_update_rmem(sk);
14c441b5
PA		 3055}
3056
b19bc294
PA		 3057void mptcp_subflow_process_delegated(struct sock *ssk)
3058{
3059 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
3060 struct sock *sk = subflow->conn;
3061
3062 mptcp_data_lock(sk);
3063 if (!sock_owned_by_user(sk))
3064 __mptcp_subflow_push_pending(sk, ssk);
3065 else
3066 set_bit(MPTCP_PUSH_PENDING, &mptcp_sk(sk)->flags);
3067 mptcp_data_unlock(sk);
3068 mptcp_subflow_delegated_done(subflow);
3069}
3070
2c5ebd00
PA		 3071static int mptcp_hash(struct sock *sk)
3072{
3073 /* should never be called,
3074 * we hash the TCP subflows not the master socket
3075 */
3076 WARN_ON_ONCE(1);
3077 return 0;
3078}
3079
3080static void mptcp_unhash(struct sock *sk)
3081{
3082 /* called from sk_common_release(), but nothing to do here */
3083}
3084
cec37a6e 3085static int mptcp_get_port(struct sock *sk, unsigned short snum)
f870fa0b
MM		 3086{
3087 struct mptcp_sock *msk = mptcp_sk(sk);
cec37a6e 3088 struct socket *ssock;
f870fa0b 3089
cec37a6e
PK		 3090 ssock = __mptcp_nmpc_socket(msk);
3091 pr_debug("msk=%p, subflow=%p", msk, ssock);
3092 if (WARN_ON_ONCE(!ssock))
3093 return -EINVAL;
f870fa0b 3094
cec37a6e
PK		 3095 return inet_csk_get_port(ssock->sk, snum);
3096}
f870fa0b 3097
cec37a6e
PK		 3098void mptcp_finish_connect(struct sock *ssk)
3099{
3100 struct mptcp_subflow_context *subflow;
3101 struct mptcp_sock *msk;
3102 struct sock *sk;
6d0060f6 3103 u64 ack_seq;
f870fa0b 3104
cec37a6e 3105 subflow = mptcp_subflow_ctx(ssk);
cec37a6e
PK		 3106 sk = subflow->conn;
3107 msk = mptcp_sk(sk);
3108
648ef4b8
MM		 3109 pr_debug("msk=%p, token=%u", sk, subflow->token);
3110
6d0060f6
MM		 3111 mptcp_crypto_key_sha(subflow->remote_key, NULL, &ack_seq);
3112 ack_seq++;
648ef4b8
MM		 3113 subflow->map_seq = ack_seq;
3114 subflow->map_subflow_seq = 1;
6d0060f6 3115
cec37a6e
PK		 3116 /* the socket is not connected yet, no msk/subflow ops can access/race
3117 * accessing the field below
3118 */
3119 WRITE_ONCE(msk->remote_key, subflow->remote_key);
3120 WRITE_ONCE(msk->local_key, subflow->local_key);
6d0060f6 3121 WRITE_ONCE(msk->write_seq, subflow->idsn + 1);
eaa2ffab 3122 WRITE_ONCE(msk->snd_nxt, msk->write_seq);
6d0060f6 3123 WRITE_ONCE(msk->ack_seq, ack_seq);
fa3fe2b1 3124 WRITE_ONCE(msk->rcv_wnd_sent, ack_seq);
d22f4988 3125 WRITE_ONCE(msk->can_ack, 1);
7439d687 3126 WRITE_ONCE(msk->snd_una, msk->write_seq);
1b1c7a0e 3127
6c714f1b 3128 mptcp_pm_new_connection(msk, ssk, 0);
a6b118fe
FW		 3129
3130 mptcp_rcv_space_init(msk, ssk);
f870fa0b
MM		 3131}
3132
866f26f2 3133void mptcp_sock_graft(struct sock *sk, struct socket *parent)
cf7da0d6
PK		 3134{
3135 write_lock_bh(&sk->sk_callback_lock);
3136 rcu_assign_pointer(sk->sk_wq, &parent->wq);
3137 sk_set_socket(sk, parent);
3138 sk->sk_uid = SOCK_INODE(parent)->i_uid;
3139 write_unlock_bh(&sk->sk_callback_lock);
3140}
3141
e16163b6 3142bool mptcp_finish_join(struct sock *ssk)
f296234c 3143{
e16163b6 3144 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
f296234c
PK		 3145 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
3146 struct sock *parent = (void *)msk;
3147 struct socket *parent_sock;
ec3edaa7 3148 bool ret;
f296234c
PK		 3149
3150 pr_debug("msk=%p, subflow=%p", msk, subflow);
3151
3152 /* mptcp socket already closing? */
dc87efdb
FW		 3153 if (!mptcp_is_fully_established(parent)) {
3154 subflow->reset_reason = MPTCP_RST_EMPTCP;
f296234c 3155 return false;
dc87efdb 3156 }
f296234c
PK		 3157
3158 if (!msk->pm.server_side)
b911c97c 3159 goto out;
f296234c 3160
dc87efdb
FW		 3161 if (!mptcp_pm_allow_new_subflow(msk)) {
3162 subflow->reset_reason = MPTCP_RST_EPROHIBIT;
10f6d46c 3163 return false;
dc87efdb 3164 }
10f6d46c
PA		 3165
3166 /* active connections are already on conn_list, and we can't acquire
3167 * msk lock here.
3168 * use the join list lock as synchronization point and double-check
e16163b6 3169 * msk status to avoid racing with __mptcp_destroy_sock()
10f6d46c
PA		 3170 */
3171 spin_lock_bh(&msk->join_list_lock);
3172 ret = inet_sk_state_load(parent) == TCP_ESTABLISHED;
e16163b6 3173 if (ret && !WARN_ON_ONCE(!list_empty(&subflow->node))) {
10f6d46c 3174 list_add_tail(&subflow->node, &msk->join_list);
e16163b6
PA		 3175 sock_hold(ssk);
3176 }
10f6d46c 3177 spin_unlock_bh(&msk->join_list_lock);
dc87efdb
FW		 3178 if (!ret) {
3179 subflow->reset_reason = MPTCP_RST_EPROHIBIT;
10f6d46c 3180 return false;
dc87efdb 3181 }
10f6d46c
PA		 3182
3183 /* attach to msk socket only after we are sure he will deal with us
3184 * at close time
3185 */
f296234c 3186 parent_sock = READ_ONCE(parent->sk_socket);
e16163b6
PA		 3187 if (parent_sock && !ssk->sk_socket)
3188 mptcp_sock_graft(ssk, parent_sock);
917944da 3189 subflow->map_seq = READ_ONCE(msk->ack_seq);
b911c97c
FW		 3190out:
3191 mptcp_event(MPTCP_EVENT_SUB_ESTABLISHED, msk, ssk, GFP_ATOMIC);
10f6d46c 3192 return true;
f296234c
PK		 3193}
3194
76e2a55d
PA		 3195static void mptcp_shutdown(struct sock *sk, int how)
3196{
3197 pr_debug("sk=%p, how=%d", sk, how);
3198
3199 if ((how & SEND_SHUTDOWN) && mptcp_close_state(sk))
3200 __mptcp_wr_shutdown(sk);
3201}
3202
f870fa0b
MM		 3203static struct proto mptcp_prot = {
3204 .name = "MPTCP",
3205 .owner = THIS_MODULE,
3206 .init = mptcp_init_sock,
18b683bf 3207 .disconnect = mptcp_disconnect,
f870fa0b 3208 .close = mptcp_close,
cf7da0d6 3209 .accept = mptcp_accept,
717e79c8
PK		 3210 .setsockopt = mptcp_setsockopt,
3211 .getsockopt = mptcp_getsockopt,
76e2a55d 3212 .shutdown = mptcp_shutdown,
79c0949e 3213 .destroy = mptcp_destroy,
f870fa0b
MM		 3214 .sendmsg = mptcp_sendmsg,
3215 .recvmsg = mptcp_recvmsg,
14c441b5 3216 .release_cb = mptcp_release_cb,
2c5ebd00
PA		 3217 .hash = mptcp_hash,
3218 .unhash = mptcp_unhash,
cec37a6e 3219 .get_port = mptcp_get_port,
d027236c
PA		 3220 .sockets_allocated = &mptcp_sockets_allocated,
3221 .memory_allocated = &tcp_memory_allocated,
3222 .memory_pressure = &tcp_memory_pressure,
d027236c 3223 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_tcp_wmem),
989ef49b 3224 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_tcp_rmem),
d027236c 3225 .sysctl_mem = sysctl_tcp_mem,
f870fa0b 3226 .obj_size = sizeof(struct mptcp_sock),
2c5ebd00 3227 .slab_flags = SLAB_TYPESAFE_BY_RCU,
f870fa0b
MM		 3228 .no_autobind = true,
3229};
3230
2303f994
PK		 3231static int mptcp_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3232{
3233 struct mptcp_sock *msk = mptcp_sk(sock->sk);
3234 struct socket *ssock;
cf7da0d6 3235 int err;
2303f994
PK		 3236
3237 lock_sock(sock->sk);
fa68018d
PA		 3238 ssock = __mptcp_nmpc_socket(msk);
3239 if (!ssock) {
3240 err = -EINVAL;
2303f994
PK		 3241 goto unlock;
3242 }
3243
3244 err = ssock->ops->bind(ssock, uaddr, addr_len);
cf7da0d6
PK		 3245 if (!err)
3246 mptcp_copy_inaddrs(sock->sk, ssock->sk);
2303f994
PK		 3247
3248unlock:
3249 release_sock(sock->sk);
3250 return err;
3251}
3252
0235d075
PA		 3253static void mptcp_subflow_early_fallback(struct mptcp_sock *msk,
3254 struct mptcp_subflow_context *subflow)
3255{
3256 subflow->request_mptcp = 0;
3257 __mptcp_do_fallback(msk);
3258}
3259
2303f994
PK		 3260static int mptcp_stream_connect(struct socket *sock, struct sockaddr *uaddr,
3261 int addr_len, int flags)
3262{
3263 struct mptcp_sock *msk = mptcp_sk(sock->sk);
2c5ebd00 3264 struct mptcp_subflow_context *subflow;
2303f994
PK		 3265 struct socket *ssock;
3266 int err;
3267
3268 lock_sock(sock->sk);
41be81a8
PA		 3269 if (sock->state != SS_UNCONNECTED && msk->subflow) {
3270 /* pending connection or invalid state, let existing subflow
3271 * cope with that
3272 */
3273 ssock = msk->subflow;
3274 goto do_connect;
3275 }
3276
fa68018d
PA		 3277 ssock = __mptcp_nmpc_socket(msk);
3278 if (!ssock) {
3279 err = -EINVAL;
2303f994
PK		 3280 goto unlock;
3281 }
3282
fa68018d
PA		 3283 mptcp_token_destroy(msk);
3284 inet_sk_state_store(sock->sk, TCP_SYN_SENT);
2c5ebd00 3285 subflow = mptcp_subflow_ctx(ssock->sk);
cf7da0d6
PK		 3286#ifdef CONFIG_TCP_MD5SIG
3287 /* no MPTCP if MD5SIG is enabled on this socket or we may run out of
3288 * TCP option space.
3289 */
3290 if (rcu_access_pointer(tcp_sk(ssock->sk)->md5sig_info))
0235d075 3291 mptcp_subflow_early_fallback(msk, subflow);
cf7da0d6 3292#endif
a16195e3
PA		 3293 if (subflow->request_mptcp && mptcp_token_new_connect(ssock->sk)) {
3294 MPTCP_INC_STATS(sock_net(ssock->sk), MPTCP_MIB_TOKENFALLBACKINIT);
0235d075 3295 mptcp_subflow_early_fallback(msk, subflow);
a16195e3 3296 }
5695eb88
PA		 3297 if (likely(!__mptcp_check_fallback(msk)))
3298 MPTCP_INC_STATS(sock_net(sock->sk), MPTCP_MIB_MPCAPABLEACTIVE);
cf7da0d6 3299
41be81a8 3300do_connect:
2303f994 3301 err = ssock->ops->connect(ssock, uaddr, addr_len, flags);
41be81a8
PA		 3302 sock->state = ssock->state;
3303
3304 /* on successful connect, the msk state will be moved to established by
3305 * subflow_finish_connect()
3306 */
367fe04e 3307 if (!err || err == -EINPROGRESS)
41be81a8
PA		 3308 mptcp_copy_inaddrs(sock->sk, ssock->sk);
3309 else
3310 inet_sk_state_store(sock->sk, inet_sk_state_load(ssock->sk));
2303f994
PK		 3311
3312unlock:
3313 release_sock(sock->sk);
3314 return err;
3315}
3316
cf7da0d6
PK		 3317static int mptcp_listen(struct socket *sock, int backlog)
3318{
3319 struct mptcp_sock *msk = mptcp_sk(sock->sk);
3320 struct socket *ssock;
3321 int err;
3322
3323 pr_debug("msk=%p", msk);
3324
3325 lock_sock(sock->sk);
fa68018d
PA		 3326 ssock = __mptcp_nmpc_socket(msk);
3327 if (!ssock) {
3328 err = -EINVAL;
cf7da0d6
PK		 3329 goto unlock;
3330 }
3331
fa68018d
PA		 3332 mptcp_token_destroy(msk);
3333 inet_sk_state_store(sock->sk, TCP_LISTEN);
5e20087d
FW		 3334 sock_set_flag(sock->sk, SOCK_RCU_FREE);
3335
cf7da0d6
PK		 3336 err = ssock->ops->listen(ssock, backlog);
3337 inet_sk_state_store(sock->sk, inet_sk_state_load(ssock->sk));
3338 if (!err)
3339 mptcp_copy_inaddrs(sock->sk, ssock->sk);
3340
3341unlock:
3342 release_sock(sock->sk);
3343 return err;
3344}
3345
cf7da0d6
PK		 3346static int mptcp_stream_accept(struct socket *sock, struct socket *newsock,
3347 int flags, bool kern)
3348{
3349 struct mptcp_sock *msk = mptcp_sk(sock->sk);
3350 struct socket *ssock;
3351 int err;
3352
3353 pr_debug("msk=%p", msk);
3354
3355 lock_sock(sock->sk);
3356 if (sock->sk->sk_state != TCP_LISTEN)
3357 goto unlock_fail;
3358
3359 ssock = __mptcp_nmpc_socket(msk);
3360 if (!ssock)
3361 goto unlock_fail;
3362
8a05661b 3363 clear_bit(MPTCP_DATA_READY, &msk->flags);
cf7da0d6
PK		 3364 sock_hold(ssock->sk);
3365 release_sock(sock->sk);
3366
3367 err = ssock->ops->accept(sock, newsock, flags, kern);
d2f77c53 3368 if (err == 0 && !mptcp_is_tcpsk(newsock->sk)) {
cf7da0d6
PK		 3369 struct mptcp_sock *msk = mptcp_sk(newsock->sk);
3370 struct mptcp_subflow_context *subflow;
0397c6d8 3371 struct sock *newsk = newsock->sk;
0397c6d8 3372
4d54cc32 3373 lock_sock(newsk);
5b950ff4
PA		 3374
3375 /* PM/worker can now acquire the first subflow socket
3376 * lock without racing with listener queue cleanup,
3377 * we can notify it, if needed.
c8fe62f0
FW		 3378 *
3379 * Even if remote has reset the initial subflow by now
3380 * the refcnt is still at least one.
5b950ff4
PA		 3381 */
3382 subflow = mptcp_subflow_ctx(msk->first);
3383 list_add(&subflow->node, &msk->conn_list);
3384 sock_hold(msk->first);
3385 if (mptcp_is_fully_established(newsk))
6c714f1b 3386 mptcp_pm_fully_established(msk, msk->first, GFP_KERNEL);
5b950ff4 3387
0397c6d8
PA		 3388 mptcp_copy_inaddrs(newsk, msk->first);
3389 mptcp_rcv_space_init(msk, msk->first);
5cf92bba 3390 mptcp_propagate_sndbuf(newsk, msk->first);
cf7da0d6
PK		 3391
3392 /* set ssk->sk_socket of accept()ed flows to mptcp socket.
3393 * This is needed so NOSPACE flag can be set from tcp stack.
3394 */
78962489 3395 mptcp_flush_join_list(msk);
190f8b06 3396 mptcp_for_each_subflow(msk, subflow) {
cf7da0d6
PK		 3397 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
3398
3399 if (!ssk->sk_socket)
3400 mptcp_sock_graft(ssk, newsock);
3401 }
4d54cc32 3402 release_sock(newsk);
cf7da0d6
PK		 3403 }
3404
8a05661b
PA		 3405 if (inet_csk_listen_poll(ssock->sk))
3406 set_bit(MPTCP_DATA_READY, &msk->flags);
cf7da0d6
PK		 3407 sock_put(ssock->sk);
3408 return err;
3409
3410unlock_fail:
3411 release_sock(sock->sk);
3412 return -EINVAL;
3413}
3414
8a05661b
PA		 3415static __poll_t mptcp_check_readable(struct mptcp_sock *msk)
3416{
3417 return test_bit(MPTCP_DATA_READY, &msk->flags) ? EPOLLIN | EPOLLRDNORM :
3418 0;
3419}
3420
8edf0864
FW		 3421static __poll_t mptcp_check_writeable(struct mptcp_sock *msk)
3422{
3423 struct sock *sk = (struct sock *)msk;
8edf0864
FW		 3424
3425 if (unlikely(sk->sk_shutdown & SEND_SHUTDOWN))
dd913410 3426 return EPOLLOUT | EPOLLWRNORM;
8edf0864
FW		 3427
3428 if (sk_stream_is_writeable(sk))
3429 return EPOLLOUT | EPOLLWRNORM;
3430
5cf92bba 3431 mptcp_set_nospace(sk);
6e628cd3
PA		 3432 smp_mb__after_atomic(); /* msk->flags is changed by write_space cb */
3433 if (sk_stream_is_writeable(sk))
3434 return EPOLLOUT | EPOLLWRNORM;
8edf0864 3435
6e628cd3 3436 return 0;
8edf0864
FW		 3437}
3438
2303f994
PK		 3439static __poll_t mptcp_poll(struct file *file, struct socket *sock,
3440 struct poll_table_struct *wait)
3441{
1891c4a0 3442 struct sock *sk = sock->sk;
8ab183de 3443 struct mptcp_sock *msk;
2303f994 3444 __poll_t mask = 0;
8a05661b 3445 int state;
2303f994 3446
1891c4a0 3447 msk = mptcp_sk(sk);
1891c4a0 3448 sock_poll_wait(file, sock, wait);
1891c4a0 3449
8a05661b 3450 state = inet_sk_state_load(sk);
6719331c 3451 pr_debug("msk=%p state=%d flags=%lx", msk, state, msk->flags);
8a05661b
PA		 3452 if (state == TCP_LISTEN)
3453 return mptcp_check_readable(msk);
3454
3455 if (state != TCP_SYN_SENT && state != TCP_SYN_RECV) {
3456 mask |= mptcp_check_readable(msk);
8edf0864 3457 mask |= mptcp_check_writeable(msk);
8a05661b 3458 }
dd913410
PA		 3459 if (sk->sk_shutdown == SHUTDOWN_MASK || state == TCP_CLOSE)
3460 mask |= EPOLLHUP;
1891c4a0
FW		 3461 if (sk->sk_shutdown & RCV_SHUTDOWN)
3462 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
3463
15cc1045
PA		 3464 /* This barrier is coupled with smp_wmb() in tcp_reset() */
3465 smp_rmb();
3466 if (sk->sk_err)
3467 mask |= EPOLLERR;
3468
2303f994
PK		 3469 return mask;
3470}
3471
e42f1ac6
FW		 3472static const struct proto_ops mptcp_stream_ops = {
3473 .family = PF_INET,
3474 .owner = THIS_MODULE,
0a3cc579 3475 .release = inet_release,
e42f1ac6
FW		 3476 .bind = mptcp_bind,
3477 .connect = mptcp_stream_connect,
3478 .socketpair = sock_no_socketpair,
3479 .accept = mptcp_stream_accept,
d2f77c53 3480 .getname = inet_getname,
e42f1ac6
FW		 3481 .poll = mptcp_poll,
3482 .ioctl = inet_ioctl,
3483 .gettstamp = sock_gettstamp,
3484 .listen = mptcp_listen,
76e2a55d 3485 .shutdown = inet_shutdown,
e42f1ac6
FW		 3486 .setsockopt = sock_common_setsockopt,
3487 .getsockopt = sock_common_getsockopt,
3488 .sendmsg = inet_sendmsg,
3489 .recvmsg = inet_recvmsg,
3490 .mmap = sock_no_mmap,
3491 .sendpage = inet_sendpage,
e42f1ac6 3492};
2303f994 3493
f870fa0b
MM		 3494static struct inet_protosw mptcp_protosw = {
3495 .type = SOCK_STREAM,
3496 .protocol = IPPROTO_MPTCP,
3497 .prot = &mptcp_prot,
2303f994
PK		 3498 .ops = &mptcp_stream_ops,
3499 .flags = INET_PROTOSW_ICSK,
f870fa0b
MM		 3500};
3501
b19bc294
PA		 3502static int mptcp_napi_poll(struct napi_struct *napi, int budget)
3503{
3504 struct mptcp_delegated_action *delegated;
3505 struct mptcp_subflow_context *subflow;
3506 int work_done = 0;
3507
3508 delegated = container_of(napi, struct mptcp_delegated_action, napi);
3509 while ((subflow = mptcp_subflow_delegated_next(delegated)) != NULL) {
3510 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
3511
3512 bh_lock_sock_nested(ssk);
3513 if (!sock_owned_by_user(ssk) &&
3514 mptcp_subflow_has_delegated_action(subflow))
3515 mptcp_subflow_process_delegated(ssk);
3516 /* ... elsewhere tcp_release_cb_override already processed
3517 * the action or will do at next release_sock().
3518 * In both case must dequeue the subflow here - on the same
3519 * CPU that scheduled it.
3520 */
3521 bh_unlock_sock(ssk);
3522 sock_put(ssk);
3523
3524 if (++work_done == budget)
3525 return budget;
3526 }
3527
3528 /* always provide a 0 'work_done' argument, so that napi_complete_done
3529 * will not try accessing the NULL napi->dev ptr
3530 */
3531 napi_complete_done(napi, 0);
3532 return work_done;
3533}
3534
d39dceca 3535void __init mptcp_proto_init(void)
f870fa0b 3536{
b19bc294
PA		 3537 struct mptcp_delegated_action *delegated;
3538 int cpu;
3539
2303f994 3540 mptcp_prot.h.hashinfo = tcp_prot.h.hashinfo;
2303f994 3541
d027236c
PA		 3542 if (percpu_counter_init(&mptcp_sockets_allocated, 0, GFP_KERNEL))
3543 panic("Failed to allocate MPTCP pcpu counter\n");
3544
b19bc294
PA		 3545 init_dummy_netdev(&mptcp_napi_dev);
3546 for_each_possible_cpu(cpu) {
3547 delegated = per_cpu_ptr(&mptcp_delegated_actions, cpu);
3548 INIT_LIST_HEAD(&delegated->head);
3549 netif_tx_napi_add(&mptcp_napi_dev, &delegated->napi, mptcp_napi_poll,
3550 NAPI_POLL_WEIGHT);
3551 napi_enable(&delegated->napi);
3552 }
3553
2303f994 3554 mptcp_subflow_init();
1b1c7a0e 3555 mptcp_pm_init();
2c5ebd00 3556 mptcp_token_init();
2303f994 3557
f870fa0b
MM		 3558 if (proto_register(&mptcp_prot, 1) != 0)
3559 panic("Failed to register MPTCP proto.\n");
3560
3561 inet_register_protosw(&mptcp_protosw);
6771bfd9
FW		 3562
3563 BUILD_BUG_ON(sizeof(struct mptcp_skb_cb) > sizeof_field(struct sk_buff, cb));
f870fa0b
MM		 3564}
3565
3566#if IS_ENABLED(CONFIG_MPTCP_IPV6)
e42f1ac6
FW		 3567static const struct proto_ops mptcp_v6_stream_ops = {
3568 .family = PF_INET6,
3569 .owner = THIS_MODULE,
0a3cc579 3570 .release = inet6_release,
e42f1ac6
FW		 3571 .bind = mptcp_bind,
3572 .connect = mptcp_stream_connect,
3573 .socketpair = sock_no_socketpair,
3574 .accept = mptcp_stream_accept,
d2f77c53 3575 .getname = inet6_getname,
e42f1ac6
FW		 3576 .poll = mptcp_poll,
3577 .ioctl = inet6_ioctl,
3578 .gettstamp = sock_gettstamp,
3579 .listen = mptcp_listen,
76e2a55d 3580 .shutdown = inet_shutdown,
e42f1ac6
FW		 3581 .setsockopt = sock_common_setsockopt,
3582 .getsockopt = sock_common_getsockopt,
3583 .sendmsg = inet6_sendmsg,
3584 .recvmsg = inet6_recvmsg,
3585 .mmap = sock_no_mmap,
3586 .sendpage = inet_sendpage,
3587#ifdef CONFIG_COMPAT
3986912f 3588 .compat_ioctl = inet6_compat_ioctl,
e42f1ac6
FW		 3589#endif
3590};
3591
f870fa0b
MM		 3592static struct proto mptcp_v6_prot;
3593
79c0949e
PK		 3594static void mptcp_v6_destroy(struct sock *sk)
3595{
3596 mptcp_destroy(sk);
3597 inet6_destroy_sock(sk);
3598}
3599
f870fa0b
MM		 3600static struct inet_protosw mptcp_v6_protosw = {
3601 .type = SOCK_STREAM,
3602 .protocol = IPPROTO_MPTCP,
3603 .prot = &mptcp_v6_prot,
2303f994 3604 .ops = &mptcp_v6_stream_ops,
f870fa0b
MM		 3605 .flags = INET_PROTOSW_ICSK,
3606};
3607
d39dceca 3608int __init mptcp_proto_v6_init(void)
f870fa0b
MM		 3609{
3610 int err;
3611
3612 mptcp_v6_prot = mptcp_prot;
3613 strcpy(mptcp_v6_prot.name, "MPTCPv6");
3614 mptcp_v6_prot.slab = NULL;
79c0949e 3615 mptcp_v6_prot.destroy = mptcp_v6_destroy;
b0519de8 3616 mptcp_v6_prot.obj_size = sizeof(struct mptcp6_sock);
f870fa0b
MM		 3617
3618 err = proto_register(&mptcp_v6_prot, 1);
3619 if (err)
3620 return err;
3621
3622 err = inet6_register_protosw(&mptcp_v6_protosw);
3623 if (err)
3624 proto_unregister(&mptcp_v6_prot);
3625
3626 return err;
3627}
3628#endif
Linux 6.x block layer and io_uring tree(s)