projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
mptcp: drop legacy code around RX EOF
[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>
644807e3 25#include <asm/ioctls.h>
f870fa0b 26#include "protocol.h"
fc518953 27#include "mib.h"
f870fa0b 28
e10a9892
GT		 29#define CREATE_TRACE_POINTS
30#include <trace/events/mptcp.h>
31
b0519de8
FW		 32#if IS_ENABLED(CONFIG_MPTCP_IPV6)
33struct mptcp6_sock {
34 struct mptcp_sock msk;
35 struct ipv6_pinfo np;
36};
37#endif
38
b7f653b2
FW		 39enum {
40 MPTCP_CMSG_TS = BIT(0),
2c9e7765 41 MPTCP_CMSG_INQ = BIT(1),
b7f653b2
FW		 42};
43
91b6d325 44static struct percpu_counter mptcp_sockets_allocated ____cacheline_aligned_in_smp;
d027236c 45
e16163b6 46static void __mptcp_destroy_sock(struct sock *sk);
81c1d029 47static void mptcp_check_send_data_fin(struct sock *sk);
e16163b6 48
b19bc294
PA		 49DEFINE_PER_CPU(struct mptcp_delegated_action, mptcp_delegated_actions);
50static struct net_device mptcp_napi_dev;
51
6f8a612a
FW		 52/* Returns end sequence number of the receiver's advertised window */
53static u64 mptcp_wnd_end(const struct mptcp_sock *msk)
54{
7439d687 55 return READ_ONCE(msk->wnd_end);
6f8a612a
FW		 56}
57
d2f77c53 58static bool mptcp_is_tcpsk(struct sock *sk)
0b4f33de
FW		 59{
60 struct socket *sock = sk->sk_socket;
61
0b4f33de
FW		 62 if (unlikely(sk->sk_prot == &tcp_prot)) {
63 /* we are being invoked after mptcp_accept() has
64 * accepted a non-mp-capable flow: sk is a tcp_sk,
65 * not an mptcp one.
66 *
67 * Hand the socket over to tcp so all further socket ops
68 * bypass mptcp.
69 */
70 sock->ops = &inet_stream_ops;
d2f77c53 71 return true;
0b4f33de
FW		 72#if IS_ENABLED(CONFIG_MPTCP_IPV6)
73 } else if (unlikely(sk->sk_prot == &tcpv6_prot)) {
74 sock->ops = &inet6_stream_ops;
d2f77c53 75 return true;
0b4f33de
FW		 76#endif
77 }
78
d2f77c53 79 return false;
0b4f33de
FW		 80}
81
fa68018d 82static int __mptcp_socket_create(struct mptcp_sock *msk)
2303f994
PK		 83{
84 struct mptcp_subflow_context *subflow;
85 struct sock *sk = (struct sock *)msk;
86 struct socket *ssock;
87 int err;
88
6bc1fe7d 89 err = mptcp_subflow_create_socket(sk, sk->sk_family, &ssock);
2303f994 90 if (err)
fa68018d 91 return err;
2303f994 92
1b1b43ee 93 WRITE_ONCE(msk->first, ssock->sk);
5b825727 94 WRITE_ONCE(msk->subflow, ssock);
2303f994 95 subflow = mptcp_subflow_ctx(ssock->sk);
cec37a6e 96 list_add(&subflow->node, &msk->conn_list);
e16163b6 97 sock_hold(ssock->sk);
2303f994 98 subflow->request_mptcp = 1;
4cf86ae8
PA		 99
100 /* This is the first subflow, always with id 0 */
101 subflow->local_id_valid = 1;
866f26f2 102 mptcp_sock_graft(msk->first, sk->sk_socket);
e1ff9e82 103
fa68018d 104 return 0;
2303f994
PK		 105}
106
ddb1a072
PA		 107/* If the MPC handshake is not started, returns the first subflow,
108 * eventually allocating it.
109 */
110struct socket *__mptcp_nmpc_socket(struct mptcp_sock *msk)
111{
112 struct sock *sk = (struct sock *)msk;
113 int ret;
114
115 if (!((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)))
116 return ERR_PTR(-EINVAL);
117
118 if (!msk->subflow) {
119 if (msk->first)
120 return ERR_PTR(-EINVAL);
121
122 ret = __mptcp_socket_create(msk);
123 if (ret)
124 return ERR_PTR(ret);
125
126 mptcp_sockopt_sync(msk, msk->first);
127 }
128
129 return msk->subflow;
130}
131
ab174ad8
PA		 132static void mptcp_drop(struct sock *sk, struct sk_buff *skb)
133{
134 sk_drops_add(sk, skb);
135 __kfree_skb(skb);
136}
137
6511882c
PA		 138static void mptcp_rmem_charge(struct sock *sk, int size)
139{
140 mptcp_sk(sk)->rmem_fwd_alloc -= size;
141}
142
8268ed4c
PA		 143static bool mptcp_try_coalesce(struct sock *sk, struct sk_buff *to,
144 struct sk_buff *from)
145{
146 bool fragstolen;
147 int delta;
148
149 if (MPTCP_SKB_CB(from)->offset ||
150 !skb_try_coalesce(to, from, &fragstolen, &delta))
151 return false;
152
06242e44
PA		 153 pr_debug("colesced seq %llx into %llx new len %d new end seq %llx",
154 MPTCP_SKB_CB(from)->map_seq, MPTCP_SKB_CB(to)->map_seq,
155 to->len, MPTCP_SKB_CB(from)->end_seq);
ab174ad8 156 MPTCP_SKB_CB(to)->end_seq = MPTCP_SKB_CB(from)->end_seq;
7288ff6e
PA		 157
158 /* note the fwd memory can reach a negative value after accounting
159 * for the delta, but the later skb free will restore a non
160 * negative one
161 */
8268ed4c 162 atomic_add(delta, &sk->sk_rmem_alloc);
6511882c 163 mptcp_rmem_charge(sk, delta);
7288ff6e
PA		 164 kfree_skb_partial(from, fragstolen);
165
8268ed4c
PA		 166 return true;
167}
168
ab174ad8
PA		 169static bool mptcp_ooo_try_coalesce(struct mptcp_sock *msk, struct sk_buff *to,
170 struct sk_buff *from)
171{
172 if (MPTCP_SKB_CB(from)->map_seq != MPTCP_SKB_CB(to)->end_seq)
173 return false;
174
175 return mptcp_try_coalesce((struct sock *)msk, to, from);
176}
177
6511882c
PA		 178static void __mptcp_rmem_reclaim(struct sock *sk, int amount)
179{
100fdd1f
ED		 180 amount >>= PAGE_SHIFT;
181 mptcp_sk(sk)->rmem_fwd_alloc -= amount << PAGE_SHIFT;
6511882c
PA		 182 __sk_mem_reduce_allocated(sk, amount);
183}
184
185static void mptcp_rmem_uncharge(struct sock *sk, int size)
186{
187 struct mptcp_sock *msk = mptcp_sk(sk);
188 int reclaimable;
189
190 msk->rmem_fwd_alloc += size;
191 reclaimable = msk->rmem_fwd_alloc - sk_unused_reserved_mem(sk);
192
193 /* see sk_mem_uncharge() for the rationale behind the following schema */
d24141fe
PA		 194 if (unlikely(reclaimable >= PAGE_SIZE))
195 __mptcp_rmem_reclaim(sk, reclaimable);
6511882c
PA		 196}
197
198static void mptcp_rfree(struct sk_buff *skb)
199{
200 unsigned int len = skb->truesize;
201 struct sock *sk = skb->sk;
202
203 atomic_sub(len, &sk->sk_rmem_alloc);
204 mptcp_rmem_uncharge(sk, len);
205}
206
36b122ba 207void mptcp_set_owner_r(struct sk_buff *skb, struct sock *sk)
6511882c
PA		 208{
209 skb_orphan(skb);
210 skb->sk = sk;
211 skb->destructor = mptcp_rfree;
212 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
213 mptcp_rmem_charge(sk, skb->truesize);
214}
215
ab174ad8
PA		 216/* "inspired" by tcp_data_queue_ofo(), main differences:
217 * - use mptcp seqs
218 * - don't cope with sacks
219 */
220static void mptcp_data_queue_ofo(struct mptcp_sock *msk, struct sk_buff *skb)
221{
222 struct sock *sk = (struct sock *)msk;
223 struct rb_node **p, *parent;
224 u64 seq, end_seq, max_seq;
225 struct sk_buff *skb1;
226
227 seq = MPTCP_SKB_CB(skb)->map_seq;
228 end_seq = MPTCP_SKB_CB(skb)->end_seq;
f3589be0 229 max_seq = atomic64_read(&msk->rcv_wnd_sent);
ab174ad8 230
06242e44
PA		 231 pr_debug("msk=%p seq=%llx limit=%llx empty=%d", msk, seq, max_seq,
232 RB_EMPTY_ROOT(&msk->out_of_order_queue));
fa3fe2b1 233 if (after64(end_seq, max_seq)) {
ab174ad8
PA		 234 /* out of window */
235 mptcp_drop(sk, skb);
fa3fe2b1
FW		 236 pr_debug("oow by %lld, rcv_wnd_sent %llu\n",
237 (unsigned long long)end_seq - (unsigned long)max_seq,
f3589be0 238 (unsigned long long)atomic64_read(&msk->rcv_wnd_sent));
06242e44 239 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_NODSSWINDOW);
ab174ad8
PA		 240 return;
241 }
242
243 p = &msk->out_of_order_queue.rb_node;
06242e44 244 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOQUEUE);
ab174ad8
PA		 245 if (RB_EMPTY_ROOT(&msk->out_of_order_queue)) {
246 rb_link_node(&skb->rbnode, NULL, p);
247 rb_insert_color(&skb->rbnode, &msk->out_of_order_queue);
248 msk->ooo_last_skb = skb;
249 goto end;
250 }
251
252 /* with 2 subflows, adding at end of ooo queue is quite likely
253 * Use of ooo_last_skb avoids the O(Log(N)) rbtree lookup.
254 */
06242e44
PA		 255 if (mptcp_ooo_try_coalesce(msk, msk->ooo_last_skb, skb)) {
256 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOMERGE);
257 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOQUEUETAIL);
ab174ad8 258 return;
06242e44 259 }
ab174ad8
PA		 260
261 /* Can avoid an rbtree lookup if we are adding skb after ooo_last_skb */
262 if (!before64(seq, MPTCP_SKB_CB(msk->ooo_last_skb)->end_seq)) {
06242e44 263 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOQUEUETAIL);
ab174ad8
PA		 264 parent = &msk->ooo_last_skb->rbnode;
265 p = &parent->rb_right;
266 goto insert;
267 }
268
269 /* Find place to insert this segment. Handle overlaps on the way. */
270 parent = NULL;
271 while (*p) {
272 parent = *p;
273 skb1 = rb_to_skb(parent);
274 if (before64(seq, MPTCP_SKB_CB(skb1)->map_seq)) {
275 p = &parent->rb_left;
276 continue;
277 }
278 if (before64(seq, MPTCP_SKB_CB(skb1)->end_seq)) {
279 if (!after64(end_seq, MPTCP_SKB_CB(skb1)->end_seq)) {
280 /* All the bits are present. Drop. */
281 mptcp_drop(sk, skb);
06242e44 282 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
ab174ad8
PA		 283 return;
284 }
285 if (after64(seq, MPTCP_SKB_CB(skb1)->map_seq)) {
286 /* partial overlap:
287 * | skb |
288 * | skb1 |
289 * continue traversing
290 */
291 } else {
292 /* skb's seq == skb1's seq and skb covers skb1.
293 * Replace skb1 with skb.
294 */
295 rb_replace_node(&skb1->rbnode, &skb->rbnode,
296 &msk->out_of_order_queue);
297 mptcp_drop(sk, skb1);
06242e44 298 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
ab174ad8
PA		 299 goto merge_right;
300 }
301 } else if (mptcp_ooo_try_coalesce(msk, skb1, skb)) {
06242e44 302 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOMERGE);
ab174ad8
PA		 303 return;
304 }
305 p = &parent->rb_right;
306 }
06242e44 307
ab174ad8
PA		 308insert:
309 /* Insert segment into RB tree. */
310 rb_link_node(&skb->rbnode, parent, p);
311 rb_insert_color(&skb->rbnode, &msk->out_of_order_queue);
312
313merge_right:
314 /* Remove other segments covered by skb. */
315 while ((skb1 = skb_rb_next(skb)) != NULL) {
316 if (before64(end_seq, MPTCP_SKB_CB(skb1)->end_seq))
317 break;
318 rb_erase(&skb1->rbnode, &msk->out_of_order_queue);
319 mptcp_drop(sk, skb1);
06242e44 320 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
ab174ad8
PA		 321 }
322 /* If there is no skb after us, we are the last_skb ! */
323 if (!skb1)
324 msk->ooo_last_skb = skb;
325
326end:
327 skb_condense(skb);
6511882c
PA		 328 mptcp_set_owner_r(skb, sk);
329}
330
331static bool mptcp_rmem_schedule(struct sock *sk, struct sock *ssk, int size)
332{
333 struct mptcp_sock *msk = mptcp_sk(sk);
334 int amt, amount;
335
69d93dae 336 if (size <= msk->rmem_fwd_alloc)
6511882c
PA		 337 return true;
338
69d93dae 339 size -= msk->rmem_fwd_alloc;
6511882c 340 amt = sk_mem_pages(size);
100fdd1f 341 amount = amt << PAGE_SHIFT;
4aaa1685
PA		 342 if (!__sk_mem_raise_allocated(sk, size, amt, SK_MEM_RECV))
343 return false;
6511882c 344
4aaa1685 345 msk->rmem_fwd_alloc += amount;
6511882c 346 return true;
ab174ad8
PA		 347}
348
349static bool __mptcp_move_skb(struct mptcp_sock *msk, struct sock *ssk,
350 struct sk_buff *skb, unsigned int offset,
351 size_t copy_len)
6771bfd9 352{
ab174ad8 353 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
6771bfd9 354 struct sock *sk = (struct sock *)msk;
4e637c70 355 struct sk_buff *tail;
b7f653b2 356 bool has_rxtstamp;
6771bfd9
FW		 357
358 __skb_unlink(skb, &ssk->sk_receive_queue);
6771bfd9 359
4e637c70
FW		 360 skb_ext_reset(skb);
361 skb_orphan(skb);
ab174ad8 362
9c3f94e1 363 /* try to fetch required memory from subflow */
6511882c
PA		 364 if (!mptcp_rmem_schedule(sk, ssk, skb->truesize))
365 goto drop;
9c3f94e1 366
b7f653b2
FW		 367 has_rxtstamp = TCP_SKB_CB(skb)->has_rxtstamp;
368
ab174ad8
PA		 369 /* the skb map_seq accounts for the skb offset:
370 * mptcp_subflow_get_mapped_dsn() is based on the current tp->copied_seq
371 * value
372 */
373 MPTCP_SKB_CB(skb)->map_seq = mptcp_subflow_get_mapped_dsn(subflow);
374 MPTCP_SKB_CB(skb)->end_seq = MPTCP_SKB_CB(skb)->map_seq + copy_len;
8268ed4c 375 MPTCP_SKB_CB(skb)->offset = offset;
b7f653b2 376 MPTCP_SKB_CB(skb)->has_rxtstamp = has_rxtstamp;
4e637c70 377
ab174ad8
PA		 378 if (MPTCP_SKB_CB(skb)->map_seq == msk->ack_seq) {
379 /* in sequence */
8b0308fe 380 WRITE_ONCE(msk->ack_seq, msk->ack_seq + copy_len);
ab174ad8
PA		 381 tail = skb_peek_tail(&sk->sk_receive_queue);
382 if (tail && mptcp_try_coalesce(sk, tail, skb))
383 return true;
4e637c70 384
6511882c 385 mptcp_set_owner_r(skb, sk);
ab174ad8
PA		 386 __skb_queue_tail(&sk->sk_receive_queue, skb);
387 return true;
388 } else if (after64(MPTCP_SKB_CB(skb)->map_seq, msk->ack_seq)) {
389 mptcp_data_queue_ofo(msk, skb);
390 return false;
391 }
392
393 /* old data, keep it simple and drop the whole pkt, sender
394 * will retransmit as needed, if needed.
395 */
06242e44 396 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
9c3f94e1 397drop:
ab174ad8
PA		 398 mptcp_drop(sk, skb);
399 return false;
6771bfd9
FW		 400}
401
16a9a9da
MM		 402static void mptcp_stop_timer(struct sock *sk)
403{
404 struct inet_connection_sock *icsk = inet_csk(sk);
405
406 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
407 mptcp_sk(sk)->timer_ival = 0;
408}
409
e16163b6
PA		 410static void mptcp_close_wake_up(struct sock *sk)
411{
412 if (sock_flag(sk, SOCK_DEAD))
413 return;
414
415 sk->sk_state_change(sk);
416 if (sk->sk_shutdown == SHUTDOWN_MASK ||
417 sk->sk_state == TCP_CLOSE)
418 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_HUP);
419 else
420 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
421}
422
6e628cd3 423static bool mptcp_pending_data_fin_ack(struct sock *sk)
16a9a9da
MM		 424{
425 struct mptcp_sock *msk = mptcp_sk(sk);
426
81c1d029 427 return ((1 << sk->sk_state) &
6e628cd3
PA		 428 (TCPF_FIN_WAIT1 | TCPF_CLOSING | TCPF_LAST_ACK)) &&
429 msk->write_seq == READ_ONCE(msk->snd_una);
430}
431
432static void mptcp_check_data_fin_ack(struct sock *sk)
433{
434 struct mptcp_sock *msk = mptcp_sk(sk);
16a9a9da
MM		 435
436 /* Look for an acknowledged DATA_FIN */
6e628cd3 437 if (mptcp_pending_data_fin_ack(sk)) {
16a9a9da
MM		 438 WRITE_ONCE(msk->snd_data_fin_enable, 0);
439
440 switch (sk->sk_state) {
441 case TCP_FIN_WAIT1:
442 inet_sk_state_store(sk, TCP_FIN_WAIT2);
16a9a9da
MM		 443 break;
444 case TCP_CLOSING:
16a9a9da
MM		 445 case TCP_LAST_ACK:
446 inet_sk_state_store(sk, TCP_CLOSE);
16a9a9da
MM		 447 break;
448 }
449
e16163b6 450 mptcp_close_wake_up(sk);
16a9a9da
MM		 451 }
452}
453
3721b9b6
MM		 454static bool mptcp_pending_data_fin(struct sock *sk, u64 *seq)
455{
456 struct mptcp_sock *msk = mptcp_sk(sk);
457
458 if (READ_ONCE(msk->rcv_data_fin) &&
459 ((1 << sk->sk_state) &
460 (TCPF_ESTABLISHED | TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2))) {
461 u64 rcv_data_fin_seq = READ_ONCE(msk->rcv_data_fin_seq);
462
463 if (msk->ack_seq == rcv_data_fin_seq) {
464 if (seq)
465 *seq = rcv_data_fin_seq;
466
467 return true;
468 }
469 }
470
471 return false;
472}
473
403a40f2 474static void mptcp_set_datafin_timeout(struct sock *sk)
6477dd39
MM		 475{
476 struct inet_connection_sock *icsk = inet_csk(sk);
877d11f0 477 u32 retransmits;
6477dd39 478
877d11f0
MM		 479 retransmits = min_t(u32, icsk->icsk_retransmits,
480 ilog2(TCP_RTO_MAX / TCP_RTO_MIN));
481
482 mptcp_sk(sk)->timer_ival = TCP_RTO_MIN << retransmits;
6477dd39
MM		 483}
484
33d41c9c 485static void __mptcp_set_timeout(struct sock *sk, long tout)
3721b9b6 486{
3721b9b6
MM		 487 mptcp_sk(sk)->timer_ival = tout > 0 ? tout : TCP_RTO_MIN;
488}
489
33d41c9c
PA		 490static long mptcp_timeout_from_subflow(const struct mptcp_subflow_context *subflow)
491{
492 const struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
493
71b7dec2
PA		 494 return inet_csk(ssk)->icsk_pending && !subflow->stale_count ?
495 inet_csk(ssk)->icsk_timeout - jiffies : 0;
33d41c9c
PA		 496}
497
498static void mptcp_set_timeout(struct sock *sk)
499{
500 struct mptcp_subflow_context *subflow;
501 long tout = 0;
502
503 mptcp_for_each_subflow(mptcp_sk(sk), subflow)
504 tout = max(tout, mptcp_timeout_from_subflow(subflow));
505 __mptcp_set_timeout(sk, tout);
506}
507
76a13b31 508static inline bool tcp_can_send_ack(const struct sock *ssk)
fd897679
PA		 509{
510 return !((1 << inet_sk_state_load(ssk)) &
20bc80b6 511 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_TIME_WAIT | TCPF_CLOSE | TCPF_LISTEN));
fd897679
PA		 512}
513
a6574302
MM		 514void __mptcp_subflow_send_ack(struct sock *ssk)
515{
516 if (tcp_can_send_ack(ssk))
517 tcp_send_ack(ssk);
518}
519
f5360e9b 520static void mptcp_subflow_send_ack(struct sock *ssk)
340fa666
MM		 521{
522 bool slow;
523
524 slow = lock_sock_fast(ssk);
a6574302 525 __mptcp_subflow_send_ack(ssk);
340fa666
MM		 526 unlock_sock_fast(ssk, slow);
527}
528
fd897679 529static void mptcp_send_ack(struct mptcp_sock *msk)
7ed90803
PA		 530{
531 struct mptcp_subflow_context *subflow;
532
340fa666
MM		 533 mptcp_for_each_subflow(msk, subflow)
534 mptcp_subflow_send_ack(mptcp_subflow_tcp_sock(subflow));
fd897679
PA		 535}
536
fde56eea 537static void mptcp_subflow_cleanup_rbuf(struct sock *ssk)
fd897679 538{
75e908c3 539 bool slow;
fd897679 540
75e908c3 541 slow = lock_sock_fast(ssk);
fde56eea 542 if (tcp_can_send_ack(ssk))
fd897679 543 tcp_cleanup_rbuf(ssk, 1);
75e908c3 544 unlock_sock_fast(ssk, slow);
fde56eea
PA		 545}
546
547static bool mptcp_subflow_could_cleanup(const struct sock *ssk, bool rx_empty)
548{
549 const struct inet_connection_sock *icsk = inet_csk(ssk);
c4512c63 550 u8 ack_pending = READ_ONCE(icsk->icsk_ack.pending);
fde56eea
PA		 551 const struct tcp_sock *tp = tcp_sk(ssk);
552
553 return (ack_pending & ICSK_ACK_SCHED) &&
554 ((READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->rcv_wup) >
555 READ_ONCE(icsk->icsk_ack.rcv_mss)) ||
556 (rx_empty && ack_pending &
557 (ICSK_ACK_PUSHED2 | ICSK_ACK_PUSHED)));
fd897679
PA		 558}
559
560static void mptcp_cleanup_rbuf(struct mptcp_sock *msk)
561{
e3859603 562 int old_space = READ_ONCE(msk->old_wspace);
fd897679 563 struct mptcp_subflow_context *subflow;
e3859603 564 struct sock *sk = (struct sock *)msk;
fde56eea
PA		 565 int space = __mptcp_space(sk);
566 bool cleanup, rx_empty;
e3859603 567
fde56eea 568 cleanup = (space > 0) && (space >= (old_space << 1));
ce599c51 569 rx_empty = !__mptcp_rmem(sk);
fd897679 570
fde56eea
PA		 571 mptcp_for_each_subflow(msk, subflow) {
572 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
fd897679 573
fde56eea
PA		 574 if (cleanup || mptcp_subflow_could_cleanup(ssk, rx_empty))
575 mptcp_subflow_cleanup_rbuf(ssk);
7ed90803
PA		 576 }
577}
578
579static bool mptcp_check_data_fin(struct sock *sk)
3721b9b6
MM		 580{
581 struct mptcp_sock *msk = mptcp_sk(sk);
582 u64 rcv_data_fin_seq;
7ed90803 583 bool ret = false;
3721b9b6 584
3721b9b6
MM		 585 /* Need to ack a DATA_FIN received from a peer while this side
586 * of the connection is in ESTABLISHED, FIN_WAIT1, or FIN_WAIT2.
587 * msk->rcv_data_fin was set when parsing the incoming options
588 * at the subflow level and the msk lock was not held, so this
589 * is the first opportunity to act on the DATA_FIN and change
590 * the msk state.
591 *
592 * If we are caught up to the sequence number of the incoming
593 * DATA_FIN, send the DATA_ACK now and do state transition. If
594 * not caught up, do nothing and let the recv code send DATA_ACK
595 * when catching up.
596 */
597
598 if (mptcp_pending_data_fin(sk, &rcv_data_fin_seq)) {
917944da 599 WRITE_ONCE(msk->ack_seq, msk->ack_seq + 1);
3721b9b6
MM		 600 WRITE_ONCE(msk->rcv_data_fin, 0);
601
6b9831bf 602 WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | RCV_SHUTDOWN);
16a9a9da 603 smp_mb__before_atomic(); /* SHUTDOWN must be visible first */
3721b9b6
MM		 604
605 switch (sk->sk_state) {
606 case TCP_ESTABLISHED:
607 inet_sk_state_store(sk, TCP_CLOSE_WAIT);
608 break;
609 case TCP_FIN_WAIT1:
610 inet_sk_state_store(sk, TCP_CLOSING);
611 break;
612 case TCP_FIN_WAIT2:
613 inet_sk_state_store(sk, TCP_CLOSE);
3721b9b6
MM		 614 break;
615 default:
616 /* Other states not expected */
617 WARN_ON_ONCE(1);
618 break;
619 }
620
7ed90803 621 ret = true;
81c1d029
PA		 622 if (!__mptcp_check_fallback(msk))
623 mptcp_send_ack(msk);
e16163b6 624 mptcp_close_wake_up(sk);
3721b9b6 625 }
7ed90803 626 return ret;
3721b9b6
MM		 627}
628
6771bfd9
FW		 629static bool __mptcp_move_skbs_from_subflow(struct mptcp_sock *msk,
630 struct sock *ssk,
631 unsigned int *bytes)
632{
633 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
600911ff 634 struct sock *sk = (struct sock *)msk;
6771bfd9
FW		 635 unsigned int moved = 0;
636 bool more_data_avail;
637 struct tcp_sock *tp;
638 bool done = false;
13c7ba0c
FW		 639 int sk_rbuf;
640
641 sk_rbuf = READ_ONCE(sk->sk_rcvbuf);
642
643 if (!(sk->sk_userlocks & SOCK_RCVBUF_LOCK)) {
644 int ssk_rbuf = READ_ONCE(ssk->sk_rcvbuf);
645
646 if (unlikely(ssk_rbuf > sk_rbuf)) {
647 WRITE_ONCE(sk->sk_rcvbuf, ssk_rbuf);
648 sk_rbuf = ssk_rbuf;
649 }
650 }
600911ff 651
ab174ad8 652 pr_debug("msk=%p ssk=%p", msk, ssk);
6771bfd9
FW		 653 tp = tcp_sk(ssk);
654 do {
655 u32 map_remaining, offset;
656 u32 seq = tp->copied_seq;
657 struct sk_buff *skb;
658 bool fin;
659
660 /* try to move as much data as available */
661 map_remaining = subflow->map_data_len -
662 mptcp_subflow_get_map_offset(subflow);
663
664 skb = skb_peek(&ssk->sk_receive_queue);
d9fb8c50 665 if (!skb) {
d89e3ed7
PA		 666 /* With racing move_skbs_to_msk() and __mptcp_move_skbs(),
667 * a different CPU can have already processed the pending
668 * data, stop here or we can enter an infinite loop
d9fb8c50
PA		 669 */
670 if (!moved)
671 done = true;
6771bfd9 672 break;
d9fb8c50 673 }
6771bfd9 674
e1ff9e82 675 if (__mptcp_check_fallback(msk)) {
d89e3ed7
PA		 676 /* Under fallback skbs have no MPTCP extension and TCP could
677 * collapse them between the dummy map creation and the
678 * current dequeue. Be sure to adjust the map size.
e1ff9e82
DC		 679 */
680 map_remaining = skb->len;
681 subflow->map_data_len = skb->len;
682 }
683
6771bfd9
FW		 684 offset = seq - TCP_SKB_CB(skb)->seq;
685 fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
686 if (fin) {
687 done = true;
688 seq++;
689 }
690
691 if (offset < skb->len) {
692 size_t len = skb->len - offset;
693
694 if (tp->urg_data)
695 done = true;
696
ab174ad8
PA		 697 if (__mptcp_move_skb(msk, ssk, skb, offset, len))
698 moved += len;
6771bfd9 699 seq += len;
6771bfd9
FW		 700
701 if (WARN_ON_ONCE(map_remaining < len))
702 break;
703 } else {
704 WARN_ON_ONCE(!fin);
705 sk_eat_skb(ssk, skb);
706 done = true;
707 }
708
709 WRITE_ONCE(tp->copied_seq, seq);
710 more_data_avail = mptcp_subflow_data_available(ssk);
600911ff 711
13c7ba0c 712 if (atomic_read(&sk->sk_rmem_alloc) > sk_rbuf) {
600911ff
FW		 713 done = true;
714 break;
715 }
6771bfd9
FW		 716 } while (more_data_avail);
717
6719331c 718 *bytes += moved;
6771bfd9
FW		 719 return done;
720}
721
87952603 722static bool __mptcp_ofo_queue(struct mptcp_sock *msk)
ab174ad8
PA		 723{
724 struct sock *sk = (struct sock *)msk;
725 struct sk_buff *skb, *tail;
726 bool moved = false;
727 struct rb_node *p;
728 u64 end_seq;
729
730 p = rb_first(&msk->out_of_order_queue);
06242e44 731 pr_debug("msk=%p empty=%d", msk, RB_EMPTY_ROOT(&msk->out_of_order_queue));
ab174ad8
PA		 732 while (p) {
733 skb = rb_to_skb(p);
734 if (after64(MPTCP_SKB_CB(skb)->map_seq, msk->ack_seq))
735 break;
736
737 p = rb_next(p);
738 rb_erase(&skb->rbnode, &msk->out_of_order_queue);
739
740 if (unlikely(!after64(MPTCP_SKB_CB(skb)->end_seq,
741 msk->ack_seq))) {
742 mptcp_drop(sk, skb);
06242e44 743 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
ab174ad8
PA		 744 continue;
745 }
746
747 end_seq = MPTCP_SKB_CB(skb)->end_seq;
748 tail = skb_peek_tail(&sk->sk_receive_queue);
749 if (!tail || !mptcp_ooo_try_coalesce(msk, tail, skb)) {
750 int delta = msk->ack_seq - MPTCP_SKB_CB(skb)->map_seq;
751
752 /* skip overlapping data, if any */
06242e44
PA		 753 pr_debug("uncoalesced seq=%llx ack seq=%llx delta=%d",
754 MPTCP_SKB_CB(skb)->map_seq, msk->ack_seq,
755 delta);
ab174ad8 756 MPTCP_SKB_CB(skb)->offset += delta;
2c9e7765 757 MPTCP_SKB_CB(skb)->map_seq += delta;
ab174ad8
PA		 758 __skb_queue_tail(&sk->sk_receive_queue, skb);
759 }
760 msk->ack_seq = end_seq;
761 moved = true;
762 }
763 return moved;
764}
765
2e52213c
FW		 766/* In most cases we will be able to lock the mptcp socket. If its already
767 * owned, we need to defer to the work queue to avoid ABBA deadlock.
768 */
99d1055c 769static bool move_skbs_to_msk(struct mptcp_sock *msk, struct sock *ssk)
2e52213c
FW		 770{
771 struct sock *sk = (struct sock *)msk;
772 unsigned int moved = 0;
773
87952603
PA		 774 __mptcp_move_skbs_from_subflow(msk, ssk, &moved);
775 __mptcp_ofo_queue(msk);
499ada50
PA		 776 if (unlikely(ssk->sk_err)) {
777 if (!sock_owned_by_user(sk))
778 __mptcp_error_report(sk);
779 else
e9d09bac 780 __set_bit(MPTCP_ERROR_REPORT, &msk->cb_flags);
499ada50 781 }
2e52213c 782
87952603
PA		 783 /* If the moves have caught up with the DATA_FIN sequence number
784 * it's time to ack the DATA_FIN and change socket state, but
785 * this is not a good place to change state. Let the workqueue
786 * do it.
787 */
788 if (mptcp_pending_data_fin(sk, NULL))
789 mptcp_schedule_work(sk);
99d1055c 790 return moved > 0;
2e52213c
FW		 791}
792
793void mptcp_data_ready(struct sock *sk, struct sock *ssk)
101f6f85 794{
6719331c 795 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
101f6f85 796 struct mptcp_sock *msk = mptcp_sk(sk);
13c7ba0c 797 int sk_rbuf, ssk_rbuf;
101f6f85 798
d7b1bfd0
PA		 799 /* The peer can send data while we are shutting down this
800 * subflow at msk destruction time, but we must avoid enqueuing
801 * more data to the msk receive queue
802 */
803 if (unlikely(subflow->disposable))
804 return;
805
13c7ba0c
FW		 806 ssk_rbuf = READ_ONCE(ssk->sk_rcvbuf);
807 sk_rbuf = READ_ONCE(sk->sk_rcvbuf);
808 if (unlikely(ssk_rbuf > sk_rbuf))
809 sk_rbuf = ssk_rbuf;
810
99d1055c 811 /* over limit? can't append more skbs to msk, Also, no need to wake-up*/
ce599c51
PA		 812 if (__mptcp_rmem(sk) > sk_rbuf) {
813 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_RCVPRUNED);
99d1055c 814 return;
ce599c51 815 }
600911ff 816
99d1055c
PA		 817 /* Wake-up the reader only for in-sequence data */
818 mptcp_data_lock(sk);
612f71d7 819 if (move_skbs_to_msk(msk, ssk))
6719331c 820 sk->sk_data_ready(sk);
612f71d7 821
99d1055c 822 mptcp_data_unlock(sk);
101f6f85
FW		 823}
824
55b47ca7
PA		 825static void mptcp_subflow_joined(struct mptcp_sock *msk, struct sock *ssk)
826{
827 mptcp_subflow_ctx(ssk)->map_seq = READ_ONCE(msk->ack_seq);
828 WRITE_ONCE(msk->allow_infinite_fallback, false);
829 mptcp_event(MPTCP_EVENT_SUB_ESTABLISHED, msk, ssk, GFP_ATOMIC);
830}
831
3e501490 832static bool __mptcp_finish_join(struct mptcp_sock *msk, struct sock *ssk)
ec3edaa7 833{
3e501490 834 struct sock *sk = (struct sock *)msk;
5cf92bba 835
3e501490 836 if (sk->sk_state != TCP_ESTABLISHED)
78962489 837 return false;
ec3edaa7 838
3e501490
PA		 839 /* attach to msk socket only after we are sure we will deal with it
840 * at close time
841 */
842 if (sk->sk_socket && !ssk->sk_socket)
843 mptcp_sock_graft(ssk, sk->sk_socket);
78962489 844
3e501490 845 mptcp_sockopt_sync_locked(msk, ssk);
55b47ca7 846 mptcp_subflow_joined(msk, ssk);
3e501490 847 return true;
78962489
FW		 848}
849
56a666c4 850static void __mptcp_flush_join_list(struct sock *sk, struct list_head *join_list)
78962489 851{
3e501490
PA		 852 struct mptcp_subflow_context *tmp, *subflow;
853 struct mptcp_sock *msk = mptcp_sk(sk);
78962489 854
56a666c4 855 list_for_each_entry_safe(subflow, tmp, join_list, node) {
3e501490
PA		 856 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
857 bool slow = lock_sock_fast(ssk);
78962489 858
3e501490
PA		 859 list_move_tail(&subflow->node, &msk->conn_list);
860 if (!__mptcp_finish_join(msk, ssk))
861 mptcp_subflow_reset(ssk);
862 unlock_sock_fast(ssk, slow);
863 }
ec3edaa7
PK		 864}
865
b51f9b80
PA		 866static bool mptcp_timer_pending(struct sock *sk)
867{
868 return timer_pending(&inet_csk(sk)->icsk_retransmit_timer);
869}
870
871static void mptcp_reset_timer(struct sock *sk)
872{
873 struct inet_connection_sock *icsk = inet_csk(sk);
874 unsigned long tout;
875
e16163b6
PA		 876 /* prevent rescheduling on close */
877 if (unlikely(inet_sk_state_load(sk) == TCP_CLOSE))
878 return;
879
33d41c9c 880 tout = mptcp_sk(sk)->timer_ival;
b51f9b80
PA		 881 sk_reset_timer(sk, &icsk->icsk_retransmit_timer, jiffies + tout);
882}
883
ba8f48f7
PA		 884bool mptcp_schedule_work(struct sock *sk)
885{
886 if (inet_sk_state_load(sk) != TCP_CLOSE &&
887 schedule_work(&mptcp_sk(sk)->work)) {
888 /* each subflow already holds a reference to the sk, and the
889 * workqueue is invoked by a subflow, so sk can't go away here.
890 */
891 sock_hold(sk);
892 return true;
893 }
894 return false;
895}
896
7a6a6cbc
PA		 897static struct sock *mptcp_subflow_recv_lookup(const struct mptcp_sock *msk)
898{
899 struct mptcp_subflow_context *subflow;
7a6a6cbc 900
109cdeb8 901 msk_owned_by_me(msk);
7a6a6cbc
PA		 902
903 mptcp_for_each_subflow(msk, subflow) {
99d1055c 904 if (READ_ONCE(subflow->data_avail))
7a6a6cbc
PA		 905 return mptcp_subflow_tcp_sock(subflow);
906 }
907
908 return NULL;
909}
910
3f8e0aae
PA		 911static bool mptcp_skb_can_collapse_to(u64 write_seq,
912 const struct sk_buff *skb,
913 const struct mptcp_ext *mpext)
57040755
PA		 914{
915 if (!tcp_skb_can_collapse_to(skb))
916 return false;
917
5a369ca6
PA		 918 /* can collapse only if MPTCP level sequence is in order and this
919 * mapping has not been xmitted yet
920 */
921 return mpext && mpext->data_seq + mpext->data_len == write_seq &&
922 !mpext->frozen;
57040755
PA		 923}
924
29249eac
PA		 925/* we can append data to the given data frag if:
926 * - there is space available in the backing page_frag
927 * - the data frag tail matches the current page_frag free offset
928 * - the data frag end sequence number matches the current write seq
929 */
18b683bf
PA		 930static bool mptcp_frag_can_collapse_to(const struct mptcp_sock *msk,
931 const struct page_frag *pfrag,
932 const struct mptcp_data_frag *df)
933{
934 return df && pfrag->page == df->page &&
d9ca1de8 935 pfrag->size - pfrag->offset > 0 &&
29249eac 936 pfrag->offset == (df->offset + df->data_len) &&
18b683bf
PA		 937 df->data_seq + df->data_len == msk->write_seq;
938}
939
d027236c
PA		 940static void dfrag_uncharge(struct sock *sk, int len)
941{
942 sk_mem_uncharge(sk, len);
7948f6cc 943 sk_wmem_queued_add(sk, -len);
d027236c
PA		 944}
945
946static void dfrag_clear(struct sock *sk, struct mptcp_data_frag *dfrag)
18b683bf 947{
d027236c
PA		 948 int len = dfrag->data_len + dfrag->overhead;
949
18b683bf 950 list_del(&dfrag->list);
d027236c 951 dfrag_uncharge(sk, len);
18b683bf
PA		 952 put_page(dfrag->page);
953}
954
6e628cd3 955static void __mptcp_clean_una(struct sock *sk)
18b683bf
PA		 956{
957 struct mptcp_sock *msk = mptcp_sk(sk);
958 struct mptcp_data_frag *dtmp, *dfrag;
e1ff9e82
DC		 959 u64 snd_una;
960
961 /* on fallback we just need to ignore snd_una, as this is really
962 * plain TCP
963 */
964 if (__mptcp_check_fallback(msk))
7439d687 965 msk->snd_una = READ_ONCE(msk->snd_nxt);
6f8a612a 966
7439d687 967 snd_una = msk->snd_una;
18b683bf
PA		 968 list_for_each_entry_safe(dfrag, dtmp, &msk->rtx_queue, list) {
969 if (after64(dfrag->data_seq + dfrag->data_len, snd_una))
970 break;
971
1e1d9d6f
PA		 972 if (unlikely(dfrag == msk->first_pending)) {
973 /* in recovery mode can see ack after the current snd head */
974 if (WARN_ON_ONCE(!msk->recovery))
975 break;
976
977 WRITE_ONCE(msk->first_pending, mptcp_send_next(sk));
978 }
979
d027236c 980 dfrag_clear(sk, dfrag);
d027236c
PA		 981 }
982
7948f6cc
FW		 983 dfrag = mptcp_rtx_head(sk);
984 if (dfrag && after64(snd_una, dfrag->data_seq)) {
53eb4c38
PA		 985 u64 delta = snd_una - dfrag->data_seq;
986
1e1d9d6f
PA		 987 /* prevent wrap around in recovery mode */
988 if (unlikely(delta > dfrag->already_sent)) {
989 if (WARN_ON_ONCE(!msk->recovery))
990 goto out;
991 if (WARN_ON_ONCE(delta > dfrag->data_len))
992 goto out;
993 dfrag->already_sent += delta - dfrag->already_sent;
994 }
7948f6cc
FW		 995
996 dfrag->data_seq += delta;
53eb4c38 997 dfrag->offset += delta;
7948f6cc 998 dfrag->data_len -= delta;
d9ca1de8 999 dfrag->already_sent -= delta;
7948f6cc
FW		 1000
1001 dfrag_uncharge(sk, delta);
7948f6cc
FW		 1002 }
1003
1e1d9d6f
PA		 1004 /* all retransmitted data acked, recovery completed */
1005 if (unlikely(msk->recovery) && after64(msk->snd_una, msk->recovery_snd_nxt))
1006 msk->recovery = false;
1007
53eb4c38 1008out:
3241a9c0
FW		 1009 if (snd_una == READ_ONCE(msk->snd_nxt) &&
1010 snd_una == READ_ONCE(msk->write_seq)) {
33d41c9c 1011 if (mptcp_timer_pending(sk) && !mptcp_data_fin_enabled(msk))
6e628cd3
PA		 1012 mptcp_stop_timer(sk);
1013 } else {
1014 mptcp_reset_timer(sk);
18b683bf
PA		 1015 }
1016}
1017
417789df
PA		 1018static void __mptcp_clean_una_wakeup(struct sock *sk)
1019{
765ff425
PA		 1020 lockdep_assert_held_once(&sk->sk_lock.slock);
1021
417789df
PA		 1022 __mptcp_clean_una(sk);
1023 mptcp_write_space(sk);
1024}
1025
b5941f06
PA		 1026static void mptcp_clean_una_wakeup(struct sock *sk)
1027{
1028 mptcp_data_lock(sk);
1029 __mptcp_clean_una_wakeup(sk);
1030 mptcp_data_unlock(sk);
1031}
1032
724cfd2e 1033static void mptcp_enter_memory_pressure(struct sock *sk)
18b683bf 1034{
d9ca1de8
PA		 1035 struct mptcp_subflow_context *subflow;
1036 struct mptcp_sock *msk = mptcp_sk(sk);
1037 bool first = true;
1038
18b683bf 1039 sk_stream_moderate_sndbuf(sk);
d9ca1de8
PA		 1040 mptcp_for_each_subflow(msk, subflow) {
1041 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1042
1043 if (first)
1044 tcp_enter_memory_pressure(ssk);
1045 sk_stream_moderate_sndbuf(ssk);
1046 first = false;
1047 }
724cfd2e
PA		 1048}
1049
1050/* ensure we get enough memory for the frag hdr, beyond some minimal amount of
1051 * data
1052 */
1053static bool mptcp_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1054{
1055 if (likely(skb_page_frag_refill(32U + sizeof(struct mptcp_data_frag),
1056 pfrag, sk->sk_allocation)))
1057 return true;
1058
1059 mptcp_enter_memory_pressure(sk);
18b683bf
PA		 1060 return false;
1061}
1062
1063static struct mptcp_data_frag *
1064mptcp_carve_data_frag(const struct mptcp_sock *msk, struct page_frag *pfrag,
1065 int orig_offset)
1066{
1067 int offset = ALIGN(orig_offset, sizeof(long));
1068 struct mptcp_data_frag *dfrag;
1069
1070 dfrag = (struct mptcp_data_frag *)(page_to_virt(pfrag->page) + offset);
1071 dfrag->data_len = 0;
1072 dfrag->data_seq = msk->write_seq;
1073 dfrag->overhead = offset - orig_offset + sizeof(struct mptcp_data_frag);
1074 dfrag->offset = offset + sizeof(struct mptcp_data_frag);
d9ca1de8 1075 dfrag->already_sent = 0;
18b683bf
PA		 1076 dfrag->page = pfrag->page;
1077
1078 return dfrag;
1079}
1080
caf971df
PA		 1081struct mptcp_sendmsg_info {
1082 int mss_now;
1083 int size_goal;
d9ca1de8
PA		 1084 u16 limit;
1085 u16 sent;
1086 unsigned int flags;
1094c6fe 1087 bool data_lock_held;
caf971df
PA		 1088};
1089
b713d006
PA		 1090static int mptcp_check_allowed_size(const struct mptcp_sock *msk, struct sock *ssk,
1091 u64 data_seq, int avail_size)
6f8a612a
FW		 1092{
1093 u64 window_end = mptcp_wnd_end(msk);
b713d006 1094 u64 mptcp_snd_wnd;
6f8a612a
FW		 1095
1096 if (__mptcp_check_fallback(msk))
1097 return avail_size;
1098
b713d006
PA		 1099 mptcp_snd_wnd = window_end - data_seq;
1100 avail_size = min_t(unsigned int, mptcp_snd_wnd, avail_size);
6f8a612a 1101
92be2f52 1102 if (unlikely(tcp_sk(ssk)->snd_wnd < mptcp_snd_wnd)) {
b713d006 1103 tcp_sk(ssk)->snd_wnd = min_t(u64, U32_MAX, mptcp_snd_wnd);
92be2f52
PA		 1104 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_SNDWNDSHARED);
1105 }
6f8a612a
FW		 1106
1107 return avail_size;
1108}
1109
724cfd2e
PA		 1110static bool __mptcp_add_ext(struct sk_buff *skb, gfp_t gfp)
1111{
1112 struct skb_ext *mpext = __skb_ext_alloc(gfp);
1113
1114 if (!mpext)
1115 return false;
1116 __skb_ext_set(skb, SKB_EXT_MPTCP, mpext);
1117 return true;
1118}
1119
6e628cd3 1120static struct sk_buff *__mptcp_do_alloc_tx_skb(struct sock *sk, gfp_t gfp)
724cfd2e
PA		 1121{
1122 struct sk_buff *skb;
1123
6e628cd3 1124 skb = alloc_skb_fclone(MAX_TCP_HEADER, gfp);
724cfd2e 1125 if (likely(skb)) {
6e628cd3 1126 if (likely(__mptcp_add_ext(skb, gfp))) {
724cfd2e 1127 skb_reserve(skb, MAX_TCP_HEADER);
a52fe46e 1128 skb->ip_summed = CHECKSUM_PARTIAL;
f70cad10 1129 INIT_LIST_HEAD(&skb->tcp_tsorted_anchor);
724cfd2e
PA		 1130 return skb;
1131 }
1132 __kfree_skb(skb);
1133 } else {
1134 mptcp_enter_memory_pressure(sk);
1135 }
1136 return NULL;
1137}
1138
f70cad10 1139static struct sk_buff *__mptcp_alloc_tx_skb(struct sock *sk, struct sock *ssk, gfp_t gfp)
724cfd2e 1140{
724cfd2e
PA		 1141 struct sk_buff *skb;
1142
6e628cd3 1143 skb = __mptcp_do_alloc_tx_skb(sk, gfp);
724cfd2e 1144 if (!skb)
f70cad10 1145 return NULL;
724cfd2e
PA		 1146
1147 if (likely(sk_wmem_schedule(ssk, skb->truesize))) {
f70cad10
PA		 1148 tcp_skb_entail(ssk, skb);
1149 return skb;
724cfd2e 1150 }
3ef3905a 1151 tcp_skb_tsorted_anchor_cleanup(skb);
724cfd2e 1152 kfree_skb(skb);
f70cad10 1153 return NULL;
724cfd2e
PA		 1154}
1155
f70cad10 1156static struct sk_buff *mptcp_alloc_tx_skb(struct sock *sk, struct sock *ssk, bool data_lock_held)
724cfd2e 1157{
1094c6fe 1158 gfp_t gfp = data_lock_held ? GFP_ATOMIC : sk->sk_allocation;
724cfd2e 1159
1094c6fe 1160 return __mptcp_alloc_tx_skb(sk, ssk, gfp);
724cfd2e
PA		 1161}
1162
d0cc2987
GT		 1163/* note: this always recompute the csum on the whole skb, even
1164 * if we just appended a single frag. More status info needed
1165 */
1166static void mptcp_update_data_checksum(struct sk_buff *skb, int added)
1167{
1168 struct mptcp_ext *mpext = mptcp_get_ext(skb);
1169 __wsum csum = ~csum_unfold(mpext->csum);
1170 int offset = skb->len - added;
1171
1172 mpext->csum = csum_fold(csum_block_add(csum, skb_checksum(skb, offset, added, 0), offset));
1173}
1174
1e39e5a3
GT		 1175static void mptcp_update_infinite_map(struct mptcp_sock *msk,
1176 struct sock *ssk,
1177 struct mptcp_ext *mpext)
1178{
1179 if (!mpext)
1180 return;
1181
1182 mpext->infinite_map = 1;
1183 mpext->data_len = 0;
1184
104125b8 1185 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPTX);
1e39e5a3
GT		 1186 mptcp_subflow_ctx(ssk)->send_infinite_map = 0;
1187 pr_fallback(msk);
d51991e2 1188 mptcp_do_fallback(ssk);
1e39e5a3
GT		 1189}
1190
6d0060f6 1191static int mptcp_sendmsg_frag(struct sock *sk, struct sock *ssk,
d9ca1de8 1192 struct mptcp_data_frag *dfrag,
caf971df 1193 struct mptcp_sendmsg_info *info)
6d0060f6 1194{
d9ca1de8 1195 u64 data_seq = dfrag->data_seq + info->sent;
f70cad10 1196 int offset = dfrag->offset + info->sent;
6d0060f6 1197 struct mptcp_sock *msk = mptcp_sk(sk);
6f8a612a 1198 bool zero_window_probe = false;
6d0060f6 1199 struct mptcp_ext *mpext = NULL;
f70cad10
PA		 1200 bool can_coalesce = false;
1201 bool reuse_skb = true;
1202 struct sk_buff *skb;
1203 size_t copy;
1204 int i;
6d0060f6 1205
e4b61351 1206 pr_debug("msk=%p ssk=%p sending dfrag at seq=%llu len=%u already sent=%u",
d9ca1de8
PA		 1207 msk, ssk, dfrag->data_seq, dfrag->data_len, info->sent);
1208
f70cad10
PA		 1209 if (WARN_ON_ONCE(info->sent > info->limit ||
1210 info->limit > dfrag->data_len))
1211 return 0;
1212
c886d702
PA		 1213 if (unlikely(!__tcp_can_send(ssk)))
1214 return -EAGAIN;
1215
d9ca1de8
PA		 1216 /* compute send limit */
1217 info->mss_now = tcp_send_mss(ssk, &info->size_goal, info->flags);
f70cad10
PA		 1218 copy = info->size_goal;
1219
57040755 1220 skb = tcp_write_queue_tail(ssk);
f70cad10 1221 if (skb && copy > skb->len) {
57040755
PA		 1222 /* Limit the write to the size available in the
1223 * current skb, if any, so that we create at most a new skb.
1224 * Explicitly tells TCP internals to avoid collapsing on later
1225 * queue management operation, to avoid breaking the ext <->
1226 * SSN association set here
1227 */
d9ca1de8 1228 mpext = skb_ext_find(skb, SKB_EXT_MPTCP);
1094c6fe 1229 if (!mptcp_skb_can_collapse_to(data_seq, skb, mpext)) {
57040755 1230 TCP_SKB_CB(skb)->eor = 1;
1094c6fe
PA		 1231 goto alloc_skb;
1232 }
1233
f70cad10
PA		 1234 i = skb_shinfo(skb)->nr_frags;
1235 can_coalesce = skb_can_coalesce(skb, i, dfrag->page, offset);
657b991a 1236 if (!can_coalesce && i >= READ_ONCE(sysctl_max_skb_frags)) {
f70cad10
PA		 1237 tcp_mark_push(tcp_sk(ssk), skb);
1238 goto alloc_skb;
15e6ca97 1239 }
18b683bf 1240
f70cad10
PA		 1241 copy -= skb->len;
1242 } else {
1094c6fe 1243alloc_skb:
f70cad10
PA		 1244 skb = mptcp_alloc_tx_skb(sk, ssk, info->data_lock_held);
1245 if (!skb)
1246 return -ENOMEM;
1247
1248 i = skb_shinfo(skb)->nr_frags;
1249 reuse_skb = false;
1250 mpext = skb_ext_find(skb, SKB_EXT_MPTCP);
1251 }
1094c6fe 1252
6f8a612a 1253 /* Zero window and all data acked? Probe. */
b713d006 1254 copy = mptcp_check_allowed_size(msk, ssk, data_seq, copy);
f70cad10 1255 if (copy == 0) {
7439d687
PA		 1256 u64 snd_una = READ_ONCE(msk->snd_una);
1257
f70cad10 1258 if (snd_una != msk->snd_nxt) {
27728ba8 1259 tcp_remove_empty_skb(ssk);
6f8a612a 1260 return 0;
f70cad10
PA		 1261 }
1262
6f8a612a 1263 zero_window_probe = true;
7439d687 1264 data_seq = snd_una - 1;
f70cad10 1265 copy = 1;
6f8a612a 1266
f70cad10
PA		 1267 /* all mptcp-level data is acked, no skbs should be present into the
1268 * ssk write queue
1269 */
1270 WARN_ON_ONCE(reuse_skb);
1271 }
d027236c 1272
f70cad10
PA		 1273 copy = min_t(size_t, copy, info->limit - info->sent);
1274 if (!sk_wmem_schedule(ssk, copy)) {
27728ba8 1275 tcp_remove_empty_skb(ssk);
e2223995 1276 return -ENOMEM;
35759383 1277 }
18b683bf 1278
f70cad10
PA		 1279 if (can_coalesce) {
1280 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1281 } else {
1282 get_page(dfrag->page);
1283 skb_fill_page_desc(skb, i, dfrag->page, offset, copy);
57040755
PA		 1284 }
1285
f70cad10
PA		 1286 skb->len += copy;
1287 skb->data_len += copy;
1288 skb->truesize += copy;
1289 sk_wmem_queued_add(ssk, copy);
1290 sk_mem_charge(ssk, copy);
f70cad10
PA		 1291 WRITE_ONCE(tcp_sk(ssk)->write_seq, tcp_sk(ssk)->write_seq + copy);
1292 TCP_SKB_CB(skb)->end_seq += copy;
1293 tcp_skb_pcount_set(skb, 0);
1294
1295 /* on skb reuse we just need to update the DSS len */
1296 if (reuse_skb) {
1297 TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH;
1298 mpext->data_len += copy;
1299 WARN_ON_ONCE(zero_window_probe);
1300 goto out;
724cfd2e 1301 }
6d0060f6
MM		 1302
1303 memset(mpext, 0, sizeof(*mpext));
d9ca1de8 1304 mpext->data_seq = data_seq;
6d0060f6 1305 mpext->subflow_seq = mptcp_subflow_ctx(ssk)->rel_write_seq;
f70cad10 1306 mpext->data_len = copy;
6d0060f6
MM		 1307 mpext->use_map = 1;
1308 mpext->dsn64 = 1;
1309
1310 pr_debug("data_seq=%llu subflow_seq=%u data_len=%u dsn64=%d",
1311 mpext->data_seq, mpext->subflow_seq, mpext->data_len,
1312 mpext->dsn64);
1313
6f8a612a 1314 if (zero_window_probe) {
f70cad10 1315 mptcp_subflow_ctx(ssk)->rel_write_seq += copy;
6f8a612a 1316 mpext->frozen = 1;
d0cc2987 1317 if (READ_ONCE(msk->csum_enabled))
f70cad10 1318 mptcp_update_data_checksum(skb, copy);
6f8a612a 1319 tcp_push_pending_frames(ssk);
d0cc2987 1320 return 0;
6f8a612a 1321 }
57040755 1322out:
d0cc2987 1323 if (READ_ONCE(msk->csum_enabled))
f70cad10 1324 mptcp_update_data_checksum(skb, copy);
1e39e5a3
GT		 1325 if (mptcp_subflow_ctx(ssk)->send_infinite_map)
1326 mptcp_update_infinite_map(msk, ssk, mpext);
0eb4e7ee 1327 trace_mptcp_sendmsg_frag(mpext);
f70cad10
PA		 1328 mptcp_subflow_ctx(ssk)->rel_write_seq += copy;
1329 return copy;
6d0060f6
MM		 1330}
1331
d5f49190
PA		 1332#define MPTCP_SEND_BURST_SIZE ((1 << 16) - \
1333 sizeof(struct tcphdr) - \
1334 MAX_TCP_OPTION_SPACE - \
1335 sizeof(struct ipv6hdr) - \
1336 sizeof(struct frag_hdr))
1337
1338struct subflow_send_info {
1339 struct sock *ssk;
3ce0852c 1340 u64 linger_time;
d5f49190
PA		 1341};
1342
ff5a0b42
PA		 1343void mptcp_subflow_set_active(struct mptcp_subflow_context *subflow)
1344{
1345 if (!subflow->stale)
1346 return;
1347
1348 subflow->stale = 0;
fc1b4e3b 1349 MPTCP_INC_STATS(sock_net(mptcp_subflow_tcp_sock(subflow)), MPTCP_MIB_SUBFLOWRECOVER);
ff5a0b42
PA		 1350}
1351
1352bool mptcp_subflow_active(struct mptcp_subflow_context *subflow)
1353{
1354 if (unlikely(subflow->stale)) {
1355 u32 rcv_tstamp = READ_ONCE(tcp_sk(mptcp_subflow_tcp_sock(subflow))->rcv_tstamp);
1356
1357 if (subflow->stale_rcv_tstamp == rcv_tstamp)
1358 return false;
1359
1360 mptcp_subflow_set_active(subflow);
1361 }
1362 return __mptcp_subflow_active(subflow);
1363}
1364
3ce0852c
PA		 1365#define SSK_MODE_ACTIVE 0
1366#define SSK_MODE_BACKUP 1
1367#define SSK_MODE_MAX 2
1368
33d41c9c
PA		 1369/* implement the mptcp packet scheduler;
1370 * returns the subflow that will transmit the next DSS
1371 * additionally updates the rtx timeout
1372 */
5cf92bba 1373static struct sock *mptcp_subflow_get_send(struct mptcp_sock *msk)
f296234c 1374{
3ce0852c 1375 struct subflow_send_info send_info[SSK_MODE_MAX];
f296234c 1376 struct mptcp_subflow_context *subflow;
33d41c9c 1377 struct sock *sk = (struct sock *)msk;
3ce0852c 1378 u32 pace, burst, wmem;
d5f49190
PA		 1379 int i, nr_active = 0;
1380 struct sock *ssk;
3ce0852c 1381 u64 linger_time;
33d41c9c 1382 long tout = 0;
f296234c 1383
109cdeb8 1384 msk_owned_by_me(msk);
f296234c 1385
d5f49190
PA		 1386 if (__mptcp_check_fallback(msk)) {
1387 if (!msk->first)
f296234c 1388 return NULL;
c886d702
PA		 1389 return __tcp_can_send(msk->first) &&
1390 sk_stream_memory_free(msk->first) ? msk->first : NULL;
d5f49190
PA		 1391 }
1392
1393 /* re-use last subflow, if the burst allow that */
1394 if (msk->last_snd && msk->snd_burst > 0 &&
1395 sk_stream_memory_free(msk->last_snd) &&
33d41c9c
PA		 1396 mptcp_subflow_active(mptcp_subflow_ctx(msk->last_snd))) {
1397 mptcp_set_timeout(sk);
d5f49190 1398 return msk->last_snd;
33d41c9c 1399 }
f296234c 1400
d5f49190 1401 /* pick the subflow with the lower wmem/wspace ratio */
3ce0852c 1402 for (i = 0; i < SSK_MODE_MAX; ++i) {
d5f49190 1403 send_info[i].ssk = NULL;
3ce0852c 1404 send_info[i].linger_time = -1;
d5f49190 1405 }
3ce0852c 1406
d5f49190 1407 mptcp_for_each_subflow(msk, subflow) {
e10a9892 1408 trace_mptcp_subflow_get_send(subflow);
d5f49190
PA		 1409 ssk = mptcp_subflow_tcp_sock(subflow);
1410 if (!mptcp_subflow_active(subflow))
1411 continue;
1412
33d41c9c 1413 tout = max(tout, mptcp_timeout_from_subflow(subflow));
d5f49190 1414 nr_active += !subflow->backup;
3ce0852c
PA		 1415 pace = subflow->avg_pacing_rate;
1416 if (unlikely(!pace)) {
1417 /* init pacing rate from socket */
1418 subflow->avg_pacing_rate = READ_ONCE(ssk->sk_pacing_rate);
1419 pace = subflow->avg_pacing_rate;
1420 if (!pace)
1421 continue;
1422 }
f296234c 1423
3ce0852c
PA		 1424 linger_time = div_u64((u64)READ_ONCE(ssk->sk_wmem_queued) << 32, pace);
1425 if (linger_time < send_info[subflow->backup].linger_time) {
d5f49190 1426 send_info[subflow->backup].ssk = ssk;
3ce0852c 1427 send_info[subflow->backup].linger_time = linger_time;
d5f49190 1428 }
f296234c 1429 }
33d41c9c 1430 __mptcp_set_timeout(sk, tout);
f296234c 1431
d5f49190
PA		 1432 /* pick the best backup if no other subflow is active */
1433 if (!nr_active)
3ce0852c
PA		 1434 send_info[SSK_MODE_ACTIVE].ssk = send_info[SSK_MODE_BACKUP].ssk;
1435
1436 /* According to the blest algorithm, to avoid HoL blocking for the
1437 * faster flow, we need to:
1438 * - estimate the faster flow linger time
1439 * - use the above to estimate the amount of byte transferred
1440 * by the faster flow
1441 * - check that the amount of queued data is greter than the above,
1442 * otherwise do not use the picked, slower, subflow
1443 * We select the subflow with the shorter estimated time to flush
1444 * the queued mem, which basically ensure the above. We just need
1445 * to check that subflow has a non empty cwin.
1446 */
1447 ssk = send_info[SSK_MODE_ACTIVE].ssk;
b713d006 1448 if (!ssk || !sk_stream_memory_free(ssk))
3ce0852c 1449 return NULL;
5cf92bba 1450
b713d006 1451 burst = min_t(int, MPTCP_SEND_BURST_SIZE, mptcp_wnd_end(msk) - msk->snd_nxt);
3ce0852c 1452 wmem = READ_ONCE(ssk->sk_wmem_queued);
b713d006
PA		 1453 if (!burst) {
1454 msk->last_snd = NULL;
1455 return ssk;
1456 }
1457
3ce0852c
PA		 1458 subflow = mptcp_subflow_ctx(ssk);
1459 subflow->avg_pacing_rate = div_u64((u64)subflow->avg_pacing_rate * wmem +
1460 READ_ONCE(ssk->sk_pacing_rate) * burst,
1461 burst + wmem);
1462 msk->last_snd = ssk;
1463 msk->snd_burst = burst;
1464 return ssk;
f296234c
PK		 1465}
1466
b8e0def3 1467static void mptcp_push_release(struct sock *ssk, struct mptcp_sendmsg_info *info)
d9ca1de8 1468{
d9ca1de8
PA		 1469 tcp_push(ssk, 0, info->mss_now, tcp_sk(ssk)->nonagle, info->size_goal);
1470 release_sock(ssk);
1471}
1472
0d199e43
FW		 1473static void mptcp_update_post_push(struct mptcp_sock *msk,
1474 struct mptcp_data_frag *dfrag,
1475 u32 sent)
1476{
1477 u64 snd_nxt_new = dfrag->data_seq;
1478
1479 dfrag->already_sent += sent;
1480
1481 msk->snd_burst -= sent;
0d199e43
FW		 1482
1483 snd_nxt_new += dfrag->already_sent;
1484
1485 /* snd_nxt_new can be smaller than snd_nxt in case mptcp
1486 * is recovering after a failover. In that event, this re-sends
1487 * old segments.
1488 *
1489 * Thus compute snd_nxt_new candidate based on
1490 * the dfrag->data_seq that was sent and the data
1491 * that has been handed to the subflow for transmission
1492 * and skip update in case it was old dfrag.
1493 */
1494 if (likely(after64(snd_nxt_new, msk->snd_nxt)))
1495 msk->snd_nxt = snd_nxt_new;
1496}
1497
8b38217a 1498void mptcp_check_and_set_pending(struct sock *sk)
3241a9c0 1499{
e9d09bac
PA		 1500 if (mptcp_send_head(sk))
1501 mptcp_sk(sk)->push_pending |= BIT(MPTCP_PUSH_PENDING);
3241a9c0
FW		 1502}
1503
ff5a0b42 1504void __mptcp_push_pending(struct sock *sk, unsigned int flags)
f870fa0b 1505{
d9ca1de8 1506 struct sock *prev_ssk = NULL, *ssk = NULL;
f870fa0b 1507 struct mptcp_sock *msk = mptcp_sk(sk);
caf971df 1508 struct mptcp_sendmsg_info info = {
d9ca1de8 1509 .flags = flags,
caf971df 1510 };
0522b424 1511 bool do_check_data_fin = false;
d9ca1de8 1512 struct mptcp_data_frag *dfrag;
0522b424 1513 int len;
d9ca1de8
PA		 1514
1515 while ((dfrag = mptcp_send_head(sk))) {
1516 info.sent = dfrag->already_sent;
1517 info.limit = dfrag->data_len;
1518 len = dfrag->data_len - dfrag->already_sent;
1519 while (len > 0) {
1520 int ret = 0;
1521
1522 prev_ssk = ssk;
5cf92bba 1523 ssk = mptcp_subflow_get_send(msk);
d9ca1de8 1524
9758f40e
PA		 1525 /* First check. If the ssk has changed since
1526 * the last round, release prev_ssk
d9ca1de8
PA		 1527 */
1528 if (ssk != prev_ssk && prev_ssk)
b8e0def3 1529 mptcp_push_release(prev_ssk, &info);
d9ca1de8
PA		 1530 if (!ssk)
1531 goto out;
1532
9758f40e
PA		 1533 /* Need to lock the new subflow only if different
1534 * from the previous one, otherwise we are still
1535 * helding the relevant lock
1536 */
1537 if (ssk != prev_ssk)
d9ca1de8
PA		 1538 lock_sock(ssk);
1539
1540 ret = mptcp_sendmsg_frag(sk, ssk, dfrag, &info);
1541 if (ret <= 0) {
c886d702
PA		 1542 if (ret == -EAGAIN)
1543 continue;
b8e0def3 1544 mptcp_push_release(ssk, &info);
d9ca1de8
PA		 1545 goto out;
1546 }
1547
0522b424 1548 do_check_data_fin = true;
d9ca1de8 1549 info.sent += ret;
d9ca1de8 1550 len -= ret;
0d199e43
FW		 1551
1552 mptcp_update_post_push(msk, dfrag, ret);
d9ca1de8
PA		 1553 }
1554 WRITE_ONCE(msk->first_pending, mptcp_send_next(sk));
1555 }
1556
1557 /* at this point we held the socket lock for the last subflow we used */
1558 if (ssk)
b8e0def3 1559 mptcp_push_release(ssk, &info);
d9ca1de8
PA		 1560
1561out:
33d41c9c
PA		 1562 /* ensure the rtx timer is running */
1563 if (!mptcp_timer_pending(sk))
1564 mptcp_reset_timer(sk);
0522b424 1565 if (do_check_data_fin)
81c1d029 1566 mptcp_check_send_data_fin(sk);
d9ca1de8
PA		 1567}
1568
73a0052a 1569static void __mptcp_subflow_push_pending(struct sock *sk, struct sock *ssk, bool first)
6e628cd3
PA		 1570{
1571 struct mptcp_sock *msk = mptcp_sk(sk);
1094c6fe
PA		 1572 struct mptcp_sendmsg_info info = {
1573 .data_lock_held = true,
1574 };
6e628cd3 1575 struct mptcp_data_frag *dfrag;
b19bc294 1576 struct sock *xmit_ssk;
6e628cd3
PA		 1577 int len, copied = 0;
1578
1579 info.flags = 0;
1580 while ((dfrag = mptcp_send_head(sk))) {
1581 info.sent = dfrag->already_sent;
1582 info.limit = dfrag->data_len;
1583 len = dfrag->data_len - dfrag->already_sent;
1584 while (len > 0) {
1585 int ret = 0;
1586
73a0052a 1587 /* check for a different subflow usage only after
b19bc294
PA		 1588 * spooling the first chunk of data
1589 */
00df24f1 1590 xmit_ssk = first ? ssk : mptcp_subflow_get_send(msk);
b19bc294
PA		 1591 if (!xmit_ssk)
1592 goto out;
1593 if (xmit_ssk != ssk) {
bcd97734
PA		 1594 mptcp_subflow_delegate(mptcp_subflow_ctx(xmit_ssk),
1595 MPTCP_DELEGATE_SEND);
b19bc294
PA		 1596 goto out;
1597 }
1598
6e628cd3
PA		 1599 ret = mptcp_sendmsg_frag(sk, ssk, dfrag, &info);
1600 if (ret <= 0)
1601 goto out;
1602
1603 info.sent += ret;
6e628cd3
PA		 1604 copied += ret;
1605 len -= ret;
b19bc294 1606 first = false;
0d199e43
FW		 1607
1608 mptcp_update_post_push(msk, dfrag, ret);
6e628cd3
PA		 1609 }
1610 WRITE_ONCE(msk->first_pending, mptcp_send_next(sk));
1611 }
1612
1613out:
1614 /* __mptcp_alloc_tx_skb could have released some wmem and we are
1615 * not going to flush it via release_sock()
1616 */
6e628cd3 1617 if (copied) {
6e628cd3
PA		 1618 tcp_push(ssk, 0, info.mss_now, tcp_sk(ssk)->nonagle,
1619 info.size_goal);
d09d818e
PA		 1620 if (!mptcp_timer_pending(sk))
1621 mptcp_reset_timer(sk);
1622
6e628cd3
PA		 1623 if (msk->snd_data_fin_enable &&
1624 msk->snd_nxt + 1 == msk->write_seq)
1625 mptcp_schedule_work(sk);
1626 }
1627}
1628
5cf92bba
PA		 1629static void mptcp_set_nospace(struct sock *sk)
1630{
1631 /* enable autotune */
1632 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1633
1634 /* will be cleared on avail space */
1635 set_bit(MPTCP_NOSPACE, &mptcp_sk(sk)->flags);
1636}
1637
7d803344
PA		 1638static int mptcp_disconnect(struct sock *sk, int flags);
1639
a2702a07 1640static int mptcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg,
fa9e5746
PA		 1641 size_t len, int *copied_syn)
1642{
1643 unsigned int saved_flags = msg->msg_flags;
1644 struct mptcp_sock *msk = mptcp_sk(sk);
ddb1a072 1645 struct socket *ssock;
a2702a07 1646 struct sock *ssk;
fa9e5746
PA		 1647 int ret;
1648
a2702a07
PA		 1649 /* on flags based fastopen the mptcp is supposed to create the
1650 * first subflow right now. Otherwise we are in the defer_connect
1651 * path, and the first subflow must be already present.
1652 * Since the defer_connect flag is cleared after the first succsful
1653 * fastopen attempt, no need to check for additional subflow status.
1654 */
ddb1a072
PA		 1655 if (msg->msg_flags & MSG_FASTOPEN) {
1656 ssock = __mptcp_nmpc_socket(msk);
1657 if (IS_ERR(ssock))
1658 return PTR_ERR(ssock);
1659 }
a2702a07
PA		 1660 if (!msk->first)
1661 return -EINVAL;
1662
1663 ssk = msk->first;
1664
fa9e5746
PA		 1665 lock_sock(ssk);
1666 msg->msg_flags |= MSG_DONTWAIT;
7d803344 1667 msk->fastopening = 1;
fa9e5746 1668 ret = tcp_sendmsg_fastopen(ssk, msg, copied_syn, len, NULL);
7d803344 1669 msk->fastopening = 0;
fa9e5746
PA		 1670 msg->msg_flags = saved_flags;
1671 release_sock(ssk);
1672
1673 /* do the blocking bits of inet_stream_connect outside the ssk socket lock */
1674 if (ret == -EINPROGRESS && !(msg->msg_flags & MSG_DONTWAIT)) {
1675 ret = __inet_stream_connect(sk->sk_socket, msg->msg_name,
1676 msg->msg_namelen, msg->msg_flags, 1);
1677
1678 /* Keep the same behaviour of plain TCP: zero the copied bytes in
1679 * case of any error, except timeout or signal
1680 */
1681 if (ret && ret != -EINPROGRESS && ret != -ERESTARTSYS && ret != -EINTR)
1682 *copied_syn = 0;
7d803344 1683 } else if (ret && ret != -EINPROGRESS) {
c2b2ae39
PA		 1684 /* The disconnect() op called by tcp_sendmsg_fastopen()/
1685 * __inet_stream_connect() can fail, due to looking check,
1686 * see mptcp_disconnect().
1687 * Attempt it again outside the problematic scope.
1688 */
1689 if (!mptcp_disconnect(sk, 0))
1690 sk->sk_socket->state = SS_UNCONNECTED;
fa9e5746 1691 }
a2702a07 1692 inet_sk(sk)->defer_connect = 0;
fa9e5746
PA		 1693
1694 return ret;
1695}
1696
d9ca1de8
PA		 1697static int mptcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1698{
1699 struct mptcp_sock *msk = mptcp_sk(sk);
17091708 1700 struct page_frag *pfrag;
6d0060f6 1701 size_t copied = 0;
caf971df 1702 int ret = 0;
6d0060f6 1703 long timeo;
f870fa0b 1704
987858e5 1705 /* silently ignore everything else */
1e777f39 1706 msg->msg_flags &= MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL | MSG_FASTOPEN;
987858e5 1707
6511882c 1708 lock_sock(sk);
1954b860 1709
a2702a07 1710 if (unlikely(inet_sk(sk)->defer_connect || msg->msg_flags & MSG_FASTOPEN)) {
d98a82a6
DS		 1711 int copied_syn = 0;
1712
a2702a07 1713 ret = mptcp_sendmsg_fastopen(sk, msg, len, &copied_syn);
d98a82a6 1714 copied += copied_syn;
fa9e5746 1715 if (ret == -EINPROGRESS && copied_syn > 0)
d98a82a6 1716 goto out;
fa9e5746 1717 else if (ret)
69800e51 1718 goto do_error;
d98a82a6
DS		 1719 }
1720
1954b860
MM		 1721 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1722
1723 if ((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) {
1724 ret = sk_stream_wait_connect(sk, &timeo);
1725 if (ret)
69800e51 1726 goto do_error;
1954b860
MM		 1727 }
1728
69800e51
PA		 1729 ret = -EPIPE;
1730 if (unlikely(sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN)))
1731 goto do_error;
1732
17091708 1733 pfrag = sk_page_frag(sk);
18b683bf 1734
d9ca1de8 1735 while (msg_data_left(msg)) {
724cfd2e 1736 int total_ts, frag_truesize = 0;
d9ca1de8 1737 struct mptcp_data_frag *dfrag;
d9ca1de8
PA		 1738 bool dfrag_collapsed;
1739 size_t psize, offset;
18b683bf 1740
d9ca1de8
PA		 1741 /* reuse tail pfrag, if possible, or carve a new one from the
1742 * page allocator
1743 */
1744 dfrag = mptcp_pending_tail(sk);
1745 dfrag_collapsed = mptcp_frag_can_collapse_to(msk, pfrag, dfrag);
1746 if (!dfrag_collapsed) {
6e628cd3
PA		 1747 if (!sk_stream_memory_free(sk))
1748 goto wait_for_memory;
1749
d9ca1de8
PA		 1750 if (!mptcp_page_frag_refill(sk, pfrag))
1751 goto wait_for_memory;
1752
1753 dfrag = mptcp_carve_data_frag(msk, pfrag, pfrag->offset);
1754 frag_truesize = dfrag->overhead;
72511aab 1755 }
6d0060f6 1756
d9ca1de8
PA		 1757 /* we do not bound vs wspace, to allow a single packet.
1758 * memory accounting will prevent execessive memory usage
1759 * anyway
d5f49190 1760 */
d9ca1de8
PA		 1761 offset = dfrag->offset + dfrag->data_len;
1762 psize = pfrag->size - offset;
1763 psize = min_t(size_t, psize, msg_data_left(msg));
724cfd2e 1764 total_ts = psize + frag_truesize;
d9ca1de8 1765
6511882c 1766 if (!sk_wmem_schedule(sk, total_ts))
724cfd2e 1767 goto wait_for_memory;
724cfd2e 1768
d9ca1de8
PA		 1769 if (copy_page_from_iter(dfrag->page, offset, psize,
1770 &msg->msg_iter) != psize) {
1771 ret = -EFAULT;
69800e51 1772 goto do_error;
72511aab
FW		 1773 }
1774
d9ca1de8 1775 /* data successfully copied into the write queue */
6511882c 1776 sk->sk_forward_alloc -= total_ts;
d9ca1de8
PA		 1777 copied += psize;
1778 dfrag->data_len += psize;
1779 frag_truesize += psize;
1780 pfrag->offset += frag_truesize;
1781 WRITE_ONCE(msk->write_seq, msk->write_seq + psize);
1782
1783 /* charge data on mptcp pending queue to the msk socket
1784 * Note: we charge such data both to sk and ssk
fb529e62 1785 */
d9ca1de8 1786 sk_wmem_queued_add(sk, frag_truesize);
d9ca1de8
PA		 1787 if (!dfrag_collapsed) {
1788 get_page(dfrag->page);
1789 list_add_tail(&dfrag->list, &msk->rtx_queue);
1790 if (!msk->first_pending)
1791 WRITE_ONCE(msk->first_pending, dfrag);
fb529e62 1792 }
e4b61351 1793 pr_debug("msk=%p dfrag at seq=%llu len=%u sent=%u new=%d", msk,
d9ca1de8
PA		 1794 dfrag->data_seq, dfrag->data_len, dfrag->already_sent,
1795 !dfrag_collapsed);
6d0060f6 1796
d9ca1de8 1797 continue;
b51f9b80 1798
d9ca1de8 1799wait_for_memory:
5cf92bba 1800 mptcp_set_nospace(sk);
c2e6048f 1801 __mptcp_push_pending(sk, msg->msg_flags);
d9ca1de8
PA		 1802 ret = sk_stream_wait_memory(sk, &timeo);
1803 if (ret)
69800e51 1804 goto do_error;
57040755 1805 }
6d0060f6 1806
13e16037 1807 if (copied)
c2e6048f 1808 __mptcp_push_pending(sk, msg->msg_flags);
d9ca1de8 1809
1954b860 1810out:
cec37a6e 1811 release_sock(sk);
69800e51
PA		 1812 return copied;
1813
1814do_error:
1815 if (copied)
1816 goto out;
1817
1818 copied = sk_stream_error(sk, msg->msg_flags, ret);
1819 goto out;
f870fa0b
MM		 1820}
1821
6771bfd9
FW		 1822static int __mptcp_recvmsg_mskq(struct mptcp_sock *msk,
1823 struct msghdr *msg,
b7f653b2
FW		 1824 size_t len, int flags,
1825 struct scm_timestamping_internal *tss,
1826 int *cmsg_flags)
6771bfd9 1827{
ca4fb892 1828 struct sk_buff *skb, *tmp;
6771bfd9
FW		 1829 int copied = 0;
1830
ca4fb892 1831 skb_queue_walk_safe(&msk->receive_queue, skb, tmp) {
6771bfd9
FW		 1832 u32 offset = MPTCP_SKB_CB(skb)->offset;
1833 u32 data_len = skb->len - offset;
1834 u32 count = min_t(size_t, len - copied, data_len);
1835 int err;
1836
d976092c
PA		 1837 if (!(flags & MSG_TRUNC)) {
1838 err = skb_copy_datagram_msg(skb, offset, msg, count);
1839 if (unlikely(err < 0)) {
1840 if (!copied)
1841 return err;
1842 break;
1843 }
6771bfd9
FW		 1844 }
1845
b7f653b2
FW		 1846 if (MPTCP_SKB_CB(skb)->has_rxtstamp) {
1847 tcp_update_recv_tstamps(skb, tss);
1848 *cmsg_flags |= MPTCP_CMSG_TS;
1849 }
1850
6771bfd9
FW		 1851 copied += count;
1852
1853 if (count < data_len) {
2c9e7765 1854 if (!(flags & MSG_PEEK)) {
ca4fb892 1855 MPTCP_SKB_CB(skb)->offset += count;
2c9e7765
FW		 1856 MPTCP_SKB_CB(skb)->map_seq += count;
1857 }
6771bfd9
FW		 1858 break;
1859 }
1860
ca4fb892
YL		 1861 if (!(flags & MSG_PEEK)) {
1862 /* we will bulk release the skb memory later */
1863 skb->destructor = NULL;
ce599c51 1864 WRITE_ONCE(msk->rmem_released, msk->rmem_released + skb->truesize);
ca4fb892
YL		 1865 __skb_unlink(skb, &msk->receive_queue);
1866 __kfree_skb(skb);
1867 }
6771bfd9
FW		 1868
1869 if (copied >= len)
1870 break;
1871 }
1872
1873 return copied;
1874}
1875
a6b118fe
FW		 1876/* receive buffer autotuning. See tcp_rcv_space_adjust for more information.
1877 *
1878 * Only difference: Use highest rtt estimate of the subflows in use.
1879 */
1880static void mptcp_rcv_space_adjust(struct mptcp_sock *msk, int copied)
1881{
1882 struct mptcp_subflow_context *subflow;
1883 struct sock *sk = (struct sock *)msk;
1884 u32 time, advmss = 1;
1885 u64 rtt_us, mstamp;
1886
109cdeb8 1887 msk_owned_by_me(msk);
a6b118fe
FW		 1888
1889 if (copied <= 0)
1890 return;
1891
1892 msk->rcvq_space.copied += copied;
1893
1894 mstamp = div_u64(tcp_clock_ns(), NSEC_PER_USEC);
1895 time = tcp_stamp_us_delta(mstamp, msk->rcvq_space.time);
1896
1897 rtt_us = msk->rcvq_space.rtt_us;
1898 if (rtt_us && time < (rtt_us >> 3))
1899 return;
1900
1901 rtt_us = 0;
1902 mptcp_for_each_subflow(msk, subflow) {
1903 const struct tcp_sock *tp;
1904 u64 sf_rtt_us;
1905 u32 sf_advmss;
1906
1907 tp = tcp_sk(mptcp_subflow_tcp_sock(subflow));
1908
1909 sf_rtt_us = READ_ONCE(tp->rcv_rtt_est.rtt_us);
1910 sf_advmss = READ_ONCE(tp->advmss);
1911
1912 rtt_us = max(sf_rtt_us, rtt_us);
1913 advmss = max(sf_advmss, advmss);
1914 }
1915
1916 msk->rcvq_space.rtt_us = rtt_us;
1917 if (time < (rtt_us >> 3) || rtt_us == 0)
1918 return;
1919
1920 if (msk->rcvq_space.copied <= msk->rcvq_space.space)
1921 goto new_measure;
1922
78047648 1923 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_moderate_rcvbuf) &&
a6b118fe
FW		 1924 !(sk->sk_userlocks & SOCK_RCVBUF_LOCK)) {
1925 int rcvmem, rcvbuf;
1926 u64 rcvwin, grow;
1927
1928 rcvwin = ((u64)msk->rcvq_space.copied << 1) + 16 * advmss;
1929
1930 grow = rcvwin * (msk->rcvq_space.copied - msk->rcvq_space.space);
1931
1932 do_div(grow, msk->rcvq_space.space);
1933 rcvwin += (grow << 1);
1934
1935 rcvmem = SKB_TRUESIZE(advmss + MAX_TCP_HEADER);
1936 while (tcp_win_from_space(sk, rcvmem) < advmss)
1937 rcvmem += 128;
1938
1939 do_div(rcvwin, advmss);
1940 rcvbuf = min_t(u64, rcvwin * rcvmem,
02739545 1941 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_rmem[2]));
a6b118fe
FW		 1942
1943 if (rcvbuf > sk->sk_rcvbuf) {
1944 u32 window_clamp;
1945
1946 window_clamp = tcp_win_from_space(sk, rcvbuf);
1947 WRITE_ONCE(sk->sk_rcvbuf, rcvbuf);
1948
1949 /* Make subflows follow along. If we do not do this, we
1950 * get drops at subflow level if skbs can't be moved to
1951 * the mptcp rx queue fast enough (announced rcv_win can
1952 * exceed ssk->sk_rcvbuf).
1953 */
1954 mptcp_for_each_subflow(msk, subflow) {
1955 struct sock *ssk;
c76c6956 1956 bool slow;
a6b118fe
FW		 1957
1958 ssk = mptcp_subflow_tcp_sock(subflow);
c76c6956 1959 slow = lock_sock_fast(ssk);
a6b118fe
FW		 1960 WRITE_ONCE(ssk->sk_rcvbuf, rcvbuf);
1961 tcp_sk(ssk)->window_clamp = window_clamp;
c76c6956
PA		 1962 tcp_cleanup_rbuf(ssk, 1);
1963 unlock_sock_fast(ssk, slow);
a6b118fe
FW		 1964 }
1965 }
1966 }
1967
1968 msk->rcvq_space.space = msk->rcvq_space.copied;
1969new_measure:
1970 msk->rcvq_space.copied = 0;
1971 msk->rcvq_space.time = mstamp;
1972}
1973
87952603
PA		 1974static void __mptcp_update_rmem(struct sock *sk)
1975{
1976 struct mptcp_sock *msk = mptcp_sk(sk);
1977
1978 if (!msk->rmem_released)
1979 return;
1980
1981 atomic_sub(msk->rmem_released, &sk->sk_rmem_alloc);
6511882c 1982 mptcp_rmem_uncharge(sk, msk->rmem_released);
ce599c51 1983 WRITE_ONCE(msk->rmem_released, 0);
87952603
PA		 1984}
1985
1986static void __mptcp_splice_receive_queue(struct sock *sk)
1987{
1988 struct mptcp_sock *msk = mptcp_sk(sk);
1989
1990 skb_queue_splice_tail_init(&sk->sk_receive_queue, &msk->receive_queue);
1991}
1992
e3859603 1993static bool __mptcp_move_skbs(struct mptcp_sock *msk)
6771bfd9 1994{
87952603 1995 struct sock *sk = (struct sock *)msk;
6771bfd9 1996 unsigned int moved = 0;
87952603 1997 bool ret, done;
d5f49190 1998
6771bfd9
FW		 1999 do {
2000 struct sock *ssk = mptcp_subflow_recv_lookup(msk);
65f49fe7 2001 bool slowpath;
6771bfd9 2002
87952603
PA		 2003 /* we can have data pending in the subflows only if the msk
2004 * receive buffer was full at subflow_data_ready() time,
2005 * that is an unlikely slow path.
2006 */
2007 if (likely(!ssk))
6771bfd9
FW		 2008 break;
2009
65f49fe7 2010 slowpath = lock_sock_fast(ssk);
87952603 2011 mptcp_data_lock(sk);
e3859603 2012 __mptcp_update_rmem(sk);
6771bfd9 2013 done = __mptcp_move_skbs_from_subflow(msk, ssk, &moved);
87952603 2014 mptcp_data_unlock(sk);
499ada50
PA		 2015
2016 if (unlikely(ssk->sk_err))
2017 __mptcp_error_report(sk);
65f49fe7 2018 unlock_sock_fast(ssk, slowpath);
6771bfd9
FW		 2019 } while (!done);
2020
87952603
PA		 2021 /* acquire the data lock only if some input data is pending */
2022 ret = moved > 0;
2023 if (!RB_EMPTY_ROOT(&msk->out_of_order_queue) ||
2024 !skb_queue_empty_lockless(&sk->sk_receive_queue)) {
2025 mptcp_data_lock(sk);
2026 __mptcp_update_rmem(sk);
2027 ret |= __mptcp_ofo_queue(msk);
2028 __mptcp_splice_receive_queue(sk);
2029 mptcp_data_unlock(sk);
ab174ad8 2030 }
87952603
PA		 2031 if (ret)
2032 mptcp_check_data_fin((struct sock *)msk);
2033 return !skb_queue_empty(&msk->receive_queue);
6771bfd9
FW		 2034}
2035
2c9e7765
FW		 2036static unsigned int mptcp_inq_hint(const struct sock *sk)
2037{
2038 const struct mptcp_sock *msk = mptcp_sk(sk);
2039 const struct sk_buff *skb;
2040
2041 skb = skb_peek(&msk->receive_queue);
2042 if (skb) {
2043 u64 hint_val = msk->ack_seq - MPTCP_SKB_CB(skb)->map_seq;
2044
2045 if (hint_val >= INT_MAX)
2046 return INT_MAX;
2047
2048 return (unsigned int)hint_val;
2049 }
2050
2051 if (sk->sk_state == TCP_CLOSE || (sk->sk_shutdown & RCV_SHUTDOWN))
2052 return 1;
2053
2054 return 0;
2055}
2056
f870fa0b 2057static int mptcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
ec095263 2058 int flags, int *addr_len)
f870fa0b
MM		 2059{
2060 struct mptcp_sock *msk = mptcp_sk(sk);
b7f653b2
FW		 2061 struct scm_timestamping_internal tss;
2062 int copied = 0, cmsg_flags = 0;
7a6a6cbc
PA		 2063 int target;
2064 long timeo;
f870fa0b 2065
cb9d80f4
PA		 2066 /* MSG_ERRQUEUE is really a no-op till we support IP_RECVERR */
2067 if (unlikely(flags & MSG_ERRQUEUE))
2068 return inet_recv_error(sk, msg, len, addr_len);
2069
6511882c 2070 lock_sock(sk);
fd897679
PA		 2071 if (unlikely(sk->sk_state == TCP_LISTEN)) {
2072 copied = -ENOTCONN;
2073 goto out_err;
2074 }
2075
ec095263 2076 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
7a6a6cbc
PA		 2077
2078 len = min_t(size_t, len, INT_MAX);
2079 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
2080
2c9e7765
FW		 2081 if (unlikely(msk->recvmsg_inq))
2082 cmsg_flags = MPTCP_CMSG_INQ;
2083
05e3ecea 2084 while (copied < len) {
e3859603 2085 int bytes_read;
7a6a6cbc 2086
b7f653b2 2087 bytes_read = __mptcp_recvmsg_mskq(msk, msg, len - copied, flags, &tss, &cmsg_flags);
6771bfd9
FW		 2088 if (unlikely(bytes_read < 0)) {
2089 if (!copied)
2090 copied = bytes_read;
2091 goto out_err;
2092 }
7a6a6cbc 2093
6771bfd9 2094 copied += bytes_read;
7a6a6cbc 2095
ea4ca586 2096 /* be sure to advertise window change */
e3859603
PA		 2097 mptcp_cleanup_rbuf(msk);
2098
2099 if (skb_queue_empty(&msk->receive_queue) && __mptcp_move_skbs(msk))
2100 continue;
ea4ca586 2101
7a6a6cbc
PA		 2102 /* only the master socket status is relevant here. The exit
2103 * conditions mirror closely tcp_recvmsg()
2104 */
2105 if (copied >= target)
2106 break;
2107
2108 if (copied) {
2109 if (sk->sk_err ||
2110 sk->sk_state == TCP_CLOSE ||
2111 (sk->sk_shutdown & RCV_SHUTDOWN) ||
2112 !timeo ||
2113 signal_pending(current))
2114 break;
2115 } else {
2116 if (sk->sk_err) {
2117 copied = sock_error(sk);
2118 break;
2119 }
2120
87952603
PA		 2121 if (sk->sk_shutdown & RCV_SHUTDOWN) {
2122 /* race breaker: the shutdown could be after the
2123 * previous receive queue check
2124 */
e3859603 2125 if (__mptcp_move_skbs(msk))
87952603 2126 continue;
7a6a6cbc 2127 break;
87952603 2128 }
7a6a6cbc
PA		 2129
2130 if (sk->sk_state == TCP_CLOSE) {
2131 copied = -ENOTCONN;
2132 break;
2133 }
2134
2135 if (!timeo) {
2136 copied = -EAGAIN;
2137 break;
2138 }
2139
2140 if (signal_pending(current)) {
2141 copied = sock_intr_errno(timeo);
2142 break;
2143 }
2144 }
2145
2146 pr_debug("block timeout %ld", timeo);
612f71d7 2147 sk_wait_data(sk, &timeo, NULL);
7a6a6cbc 2148 }
3c90e377 2149
6771bfd9 2150out_err:
b7f653b2
FW		 2151 if (cmsg_flags && copied >= 0) {
2152 if (cmsg_flags & MPTCP_CMSG_TS)
2153 tcp_recv_timestamp(msg, sk, &tss);
2c9e7765
FW		 2154
2155 if (cmsg_flags & MPTCP_CMSG_INQ) {
2156 unsigned int inq = mptcp_inq_hint(sk);
2157
2158 put_cmsg(msg, SOL_TCP, TCP_CM_INQ, sizeof(inq), &inq);
2159 }
b7f653b2
FW		 2160 }
2161
612f71d7
PA		 2162 pr_debug("msk=%p rx queue empty=%d:%d copied=%d",
2163 msk, skb_queue_empty_lockless(&sk->sk_receive_queue),
2164 skb_queue_empty(&msk->receive_queue), copied);
ca4fb892
YL		 2165 if (!(flags & MSG_PEEK))
2166 mptcp_rcv_space_adjust(msk, copied);
a6b118fe 2167
7a6a6cbc 2168 release_sock(sk);
cec37a6e
PK		 2169 return copied;
2170}
2171
b51f9b80
PA		 2172static void mptcp_retransmit_timer(struct timer_list *t)
2173{
2174 struct inet_connection_sock *icsk = from_timer(icsk, t,
2175 icsk_retransmit_timer);
2176 struct sock *sk = &icsk->icsk_inet.sk;
2d6f5a2b 2177 struct mptcp_sock *msk = mptcp_sk(sk);
b51f9b80
PA		 2178
2179 bh_lock_sock(sk);
2180 if (!sock_owned_by_user(sk)) {
2d6f5a2b
PA		 2181 /* we need a process context to retransmit */
2182 if (!test_and_set_bit(MPTCP_WORK_RTX, &msk->flags))
2183 mptcp_schedule_work(sk);
b51f9b80
PA		 2184 } else {
2185 /* delegate our work to tcp_release_cb() */
e9d09bac 2186 __set_bit(MPTCP_RETRANSMIT, &msk->cb_flags);
b51f9b80
PA		 2187 }
2188 bh_unlock_sock(sk);
2189 sock_put(sk);
2190}
2191
e16163b6
PA		 2192static void mptcp_timeout_timer(struct timer_list *t)
2193{
2194 struct sock *sk = from_timer(sk, t, sk_timer);
2195
2196 mptcp_schedule_work(sk);
b6d69fc8 2197 sock_put(sk);
e16163b6
PA		 2198}
2199
3b1d6210
PA		 2200/* Find an idle subflow. Return NULL if there is unacked data at tcp
2201 * level.
2202 *
2203 * A backup subflow is returned only if that is the only kind available.
2204 */
ff5a0b42 2205static struct sock *mptcp_subflow_get_retrans(struct mptcp_sock *msk)
3b1d6210 2206{
71b7dec2 2207 struct sock *backup = NULL, *pick = NULL;
3b1d6210 2208 struct mptcp_subflow_context *subflow;
71b7dec2 2209 int min_stale_count = INT_MAX;
3b1d6210 2210
109cdeb8 2211 msk_owned_by_me(msk);
3b1d6210 2212
d5f49190 2213 if (__mptcp_check_fallback(msk))
d9ca1de8 2214 return NULL;
d5f49190 2215
3b1d6210
PA		 2216 mptcp_for_each_subflow(msk, subflow) {
2217 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2218
ff5a0b42 2219 if (!__mptcp_subflow_active(subflow))
d5f49190
PA		 2220 continue;
2221
71b7dec2
PA		 2222 /* still data outstanding at TCP level? skip this */
2223 if (!tcp_rtx_and_write_queues_empty(ssk)) {
2224 mptcp_pm_subflow_chk_stale(msk, ssk);
2225 min_stale_count = min_t(int, min_stale_count, subflow->stale_count);
2226 continue;
860975c6 2227 }
3b1d6210
PA		 2228
2229 if (subflow->backup) {
2230 if (!backup)
2231 backup = ssk;
2232 continue;
2233 }
2234
71b7dec2
PA		 2235 if (!pick)
2236 pick = ssk;
3b1d6210
PA		 2237 }
2238
71b7dec2
PA		 2239 if (pick)
2240 return pick;
2241
2242 /* use backup only if there are no progresses anywhere */
2243 return min_stale_count > 1 ? backup : NULL;
3b1d6210
PA		 2244}
2245
17aee05d
FW		 2246static void mptcp_dispose_initial_subflow(struct mptcp_sock *msk)
2247{
2248 if (msk->subflow) {
2249 iput(SOCK_INODE(msk->subflow));
5b825727 2250 WRITE_ONCE(msk->subflow, NULL);
17aee05d
FW		 2251 }
2252}
2253
1e1d9d6f
PA		 2254bool __mptcp_retransmit_pending_data(struct sock *sk)
2255{
2256 struct mptcp_data_frag *cur, *rtx_head;
2257 struct mptcp_sock *msk = mptcp_sk(sk);
2258
00df24f1 2259 if (__mptcp_check_fallback(msk))
1e1d9d6f
PA		 2260 return false;
2261
2262 if (tcp_rtx_and_write_queues_empty(sk))
2263 return false;
2264
2265 /* the closing socket has some data untransmitted and/or unacked:
2266 * some data in the mptcp rtx queue has not really xmitted yet.
2267 * keep it simple and re-inject the whole mptcp level rtx queue
2268 */
2269 mptcp_data_lock(sk);
2270 __mptcp_clean_una_wakeup(sk);
2271 rtx_head = mptcp_rtx_head(sk);
2272 if (!rtx_head) {
2273 mptcp_data_unlock(sk);
2274 return false;
2275 }
2276
0d199e43 2277 msk->recovery_snd_nxt = msk->snd_nxt;
1e1d9d6f
PA		 2278 msk->recovery = true;
2279 mptcp_data_unlock(sk);
2280
2281 msk->first_pending = rtx_head;
1e1d9d6f
PA		 2282 msk->snd_burst = 0;
2283
2284 /* be sure to clear the "sent status" on all re-injected fragments */
2285 list_for_each_entry(cur, &msk->rtx_queue, list) {
2286 if (!cur->already_sent)
2287 break;
2288 cur->already_sent = 0;
2289 }
2290
2291 return true;
2292}
2293
b29fcfb5
PA		 2294/* flags for __mptcp_close_ssk() */
2295#define MPTCP_CF_PUSH BIT(1)
2296#define MPTCP_CF_FASTCLOSE BIT(2)
2297
cec37a6e
PK		 2298/* subflow sockets can be either outgoing (connect) or incoming
2299 * (accept).
2300 *
2301 * Outgoing subflows use in-kernel sockets.
2302 * Incoming subflows do not have their own 'struct socket' allocated,
2303 * so we need to use tcp_close() after detaching them from the mptcp
2304 * parent socket.
2305 */
a141e02e 2306static void __mptcp_close_ssk(struct sock *sk, struct sock *ssk,
b29fcfb5
PA		 2307 struct mptcp_subflow_context *subflow,
2308 unsigned int flags)
cec37a6e 2309{
e0be4931 2310 struct mptcp_sock *msk = mptcp_sk(sk);
63740448
PA		 2311 bool dispose_it, need_push = false;
2312
2313 /* If the first subflow moved to a close state before accept, e.g. due
2314 * to an incoming reset, mptcp either:
2315 * - if either the subflow or the msk are dead, destroy the context
2316 * (the subflow socket is deleted by inet_child_forget) and the msk
2317 * - otherwise do nothing at the moment and take action at accept and/or
2318 * listener shutdown - user-space must be able to accept() the closed
2319 * socket.
2320 */
2321 if (msk->in_accept_queue && msk->first == ssk) {
2322 if (!sock_flag(sk, SOCK_DEAD) && !sock_flag(ssk, SOCK_DEAD))
2323 return;
2324
2325 /* ensure later check in mptcp_worker() will dispose the msk */
2326 sock_set_flag(sk, SOCK_DEAD);
2327 lock_sock_nested(ssk, SINGLE_DEPTH_NESTING);
2328 mptcp_subflow_drop_ctx(ssk);
2329 goto out_release;
2330 }
e0be4931 2331
b29fcfb5
PA		 2332 dispose_it = !msk->subflow || ssk != msk->subflow->sk;
2333 if (dispose_it)
2334 list_del(&subflow->node);
cec37a6e 2335
3f8b2667 2336 lock_sock_nested(ssk, SINGLE_DEPTH_NESTING);
e16163b6 2337
d21f8348
PA		 2338 if (flags & MPTCP_CF_FASTCLOSE) {
2339 /* be sure to force the tcp_disconnect() path,
2340 * to generate the egress reset
2341 */
2342 ssk->sk_lingertime = 0;
2343 sock_set_flag(ssk, SOCK_LINGER);
b29fcfb5 2344 subflow->send_fastclose = 1;
d21f8348 2345 }
b29fcfb5
PA		 2346
2347 need_push = (flags & MPTCP_CF_PUSH) && __mptcp_retransmit_pending_data(sk);
2348 if (!dispose_it) {
c2b2ae39
PA		 2349 /* The MPTCP code never wait on the subflow sockets, TCP-level
2350 * disconnect should never fail
2351 */
2352 WARN_ON_ONCE(tcp_disconnect(ssk, 0));
b29fcfb5
PA		 2353 msk->subflow->state = SS_UNCONNECTED;
2354 mptcp_subflow_ctx_reset(subflow);
2355 release_sock(ssk);
2356
2357 goto out;
2358 }
2359
d7b1bfd0
PA		 2360 subflow->disposable = 1;
2361
e16163b6
PA		 2362 /* if ssk hit tcp_done(), tcp_cleanup_ulp() cleared the related ops
2363 * the ssk has been already destroyed, we just need to release the
2364 * reference owned by msk;
2365 */
2366 if (!inet_csk(ssk)->icsk_ulp_ops) {
b6985b9b 2367 WARN_ON_ONCE(!sock_flag(ssk, SOCK_DEAD));
e16163b6 2368 kfree_rcu(subflow, rcu);
cec37a6e 2369 } else {
d7b1bfd0 2370 /* otherwise tcp will dispose of the ssk and subflow ctx */
2a6a870e
PA		 2371 if (ssk->sk_state == TCP_LISTEN) {
2372 tcp_set_state(ssk, TCP_CLOSE);
2373 mptcp_subflow_queue_clean(sk, ssk);
2374 inet_csk_listen_stop(ssk);
f8c9dfbd 2375 mptcp_event_pm_listener(ssk, MPTCP_EVENT_LISTENER_CLOSED);
2a6a870e 2376 }
0a3f4f1f 2377
e16163b6
PA		 2378 __tcp_close(ssk, 0);
2379
2380 /* close acquired an extra ref */
2381 __sock_put(ssk);
cec37a6e 2382 }
63740448
PA		 2383
2384out_release:
e16163b6 2385 release_sock(ssk);
e16163b6
PA		 2386
2387 sock_put(ssk);
e0be4931 2388
c8fe62f0 2389 if (ssk == msk->first)
1b1b43ee 2390 WRITE_ONCE(msk->first, NULL);
c8fe62f0 2391
b29fcfb5
PA		 2392out:
2393 if (ssk == msk->last_snd)
2394 msk->last_snd = NULL;
1e1d9d6f
PA		 2395
2396 if (need_push)
2397 __mptcp_push_pending(sk, 0);
f870fa0b
MM		 2398}
2399
a141e02e
FW		 2400void mptcp_close_ssk(struct sock *sk, struct sock *ssk,
2401 struct mptcp_subflow_context *subflow)
2402{
b911c97c
FW		 2403 if (sk->sk_state == TCP_ESTABLISHED)
2404 mptcp_event(MPTCP_EVENT_SUB_CLOSED, mptcp_sk(sk), ssk, GFP_KERNEL);
a88c9e49
PA		 2405
2406 /* subflow aborted before reaching the fully_established status
2407 * attempt the creation of the next subflow
2408 */
2409 mptcp_pm_subflow_check_next(mptcp_sk(sk), ssk, subflow);
2410
b29fcfb5 2411 __mptcp_close_ssk(sk, ssk, subflow, MPTCP_CF_PUSH);
a141e02e
FW		 2412}
2413
dc24f8b4
PA		 2414static unsigned int mptcp_sync_mss(struct sock *sk, u32 pmtu)
2415{
2416 return 0;
2417}
2418
b6985b9b 2419static void __mptcp_close_subflow(struct sock *sk)
0e4f35d7
PA		 2420{
2421 struct mptcp_subflow_context *subflow, *tmp;
b6985b9b 2422 struct mptcp_sock *msk = mptcp_sk(sk);
0e4f35d7 2423
3abc05d9
FW		 2424 might_sleep();
2425
5efbf6f7 2426 mptcp_for_each_subflow_safe(msk, subflow, tmp) {
0e4f35d7
PA		 2427 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2428
2429 if (inet_sk_state_load(ssk) != TCP_CLOSE)
2430 continue;
2431
40947e13
FW		 2432 /* 'subflow_data_ready' will re-sched once rx queue is empty */
2433 if (!skb_queue_empty_lockless(&ssk->sk_receive_queue))
2434 continue;
2435
b6985b9b
PA		 2436 mptcp_close_ssk(sk, ssk, subflow);
2437 }
2438
0e4f35d7
PA		 2439}
2440
63740448 2441static bool mptcp_should_close(const struct sock *sk)
e16163b6
PA		 2442{
2443 s32 delta = tcp_jiffies32 - inet_csk(sk)->icsk_mtup.probe_timestamp;
2444 struct mptcp_subflow_context *subflow;
2445
63740448 2446 if (delta >= TCP_TIMEWAIT_LEN || mptcp_sk(sk)->in_accept_queue)
e16163b6
PA		 2447 return true;
2448
2449 /* if all subflows are in closed status don't bother with additional
2450 * timeout
2451 */
2452 mptcp_for_each_subflow(mptcp_sk(sk), subflow) {
2453 if (inet_sk_state_load(mptcp_subflow_tcp_sock(subflow)) !=
2454 TCP_CLOSE)
2455 return false;
2456 }
2457 return true;
2458}
2459
50c504a2
FW		 2460static void mptcp_check_fastclose(struct mptcp_sock *msk)
2461{
2462 struct mptcp_subflow_context *subflow, *tmp;
80638684 2463 struct sock *sk = (struct sock *)msk;
50c504a2
FW		 2464
2465 if (likely(!READ_ONCE(msk->rcv_fastclose)))
2466 return;
2467
2468 mptcp_token_destroy(msk);
2469
5efbf6f7 2470 mptcp_for_each_subflow_safe(msk, subflow, tmp) {
50c504a2 2471 struct sock *tcp_sk = mptcp_subflow_tcp_sock(subflow);
75e908c3 2472 bool slow;
50c504a2 2473
75e908c3 2474 slow = lock_sock_fast(tcp_sk);
50c504a2
FW		 2475 if (tcp_sk->sk_state != TCP_CLOSE) {
2476 tcp_send_active_reset(tcp_sk, GFP_ATOMIC);
2477 tcp_set_state(tcp_sk, TCP_CLOSE);
2478 }
75e908c3 2479 unlock_sock_fast(tcp_sk, slow);
50c504a2
FW		 2480 }
2481
69800e51
PA		 2482 /* Mirror the tcp_reset() error propagation */
2483 switch (sk->sk_state) {
2484 case TCP_SYN_SENT:
9ae8e5ad 2485 WRITE_ONCE(sk->sk_err, ECONNREFUSED);
69800e51
PA		 2486 break;
2487 case TCP_CLOSE_WAIT:
9ae8e5ad 2488 WRITE_ONCE(sk->sk_err, EPIPE);
69800e51
PA		 2489 break;
2490 case TCP_CLOSE:
2491 return;
2492 default:
9ae8e5ad 2493 WRITE_ONCE(sk->sk_err, ECONNRESET);
69800e51
PA		 2494 }
2495
50c504a2 2496 inet_sk_state_store(sk, TCP_CLOSE);
6b9831bf 2497 WRITE_ONCE(sk->sk_shutdown, SHUTDOWN_MASK);
50c504a2 2498 smp_mb__before_atomic(); /* SHUTDOWN must be visible first */
50c504a2
FW		 2499 set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags);
2500
69800e51
PA		 2501 /* the calling mptcp_worker will properly destroy the socket */
2502 if (sock_flag(sk, SOCK_DEAD))
2503 return;
2504
2505 sk->sk_state_change(sk);
2506 sk_error_report(sk);
50c504a2
FW		 2507}
2508
2948d0a1 2509static void __mptcp_retrans(struct sock *sk)
80992017 2510{
2948d0a1 2511 struct mptcp_sock *msk = mptcp_sk(sk);
caf971df 2512 struct mptcp_sendmsg_info info = {};
3b1d6210 2513 struct mptcp_data_frag *dfrag;
3b1d6210 2514 size_t copied = 0;
2948d0a1
PA		 2515 struct sock *ssk;
2516 int ret;
3b1d6210 2517
b5941f06 2518 mptcp_clean_una_wakeup(sk);
3241a9c0
FW		 2519
2520 /* first check ssk: need to kick "stale" logic */
2521 ssk = mptcp_subflow_get_retrans(msk);
3b1d6210 2522 dfrag = mptcp_rtx_head(sk);
6477dd39
MM		 2523 if (!dfrag) {
2524 if (mptcp_data_fin_enabled(msk)) {
2525 struct inet_connection_sock *icsk = inet_csk(sk);
2526
2527 icsk->icsk_retransmits++;
2528 mptcp_set_datafin_timeout(sk);
2529 mptcp_send_ack(msk);
2530
2531 goto reset_timer;
2532 }
2533
3241a9c0
FW		 2534 if (!mptcp_send_head(sk))
2535 return;
2536
2537 goto reset_timer;
6477dd39 2538 }
3b1d6210 2539
3b1d6210 2540 if (!ssk)
2948d0a1 2541 goto reset_timer;
3b1d6210
PA		 2542
2543 lock_sock(ssk);
2544
d9ca1de8
PA		 2545 /* limit retransmission to the bytes already sent on some subflows */
2546 info.sent = 0;
4e14867d
PA		 2547 info.limit = READ_ONCE(msk->csum_enabled) ? dfrag->data_len : dfrag->already_sent;
2548 while (info.sent < info.limit) {
d9ca1de8 2549 ret = mptcp_sendmsg_frag(sk, ssk, dfrag, &info);
6f8a612a 2550 if (ret <= 0)
3b1d6210
PA		 2551 break;
2552
fc518953 2553 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_RETRANSSEGS);
3b1d6210 2554 copied += ret;
d9ca1de8 2555 info.sent += ret;
3b1d6210 2556 }
4e14867d
PA		 2557 if (copied) {
2558 dfrag->already_sent = max(dfrag->already_sent, info.sent);
caf971df
PA		 2559 tcp_push(ssk, 0, info.mss_now, tcp_sk(ssk)->nonagle,
2560 info.size_goal);
0530020a 2561 WRITE_ONCE(msk->allow_infinite_fallback, false);
4e14867d 2562 }
3b1d6210 2563
3b1d6210
PA		 2564 release_sock(ssk);
2565
2948d0a1 2566reset_timer:
3241a9c0
FW		 2567 mptcp_check_and_set_pending(sk);
2568
3b1d6210
PA		 2569 if (!mptcp_timer_pending(sk))
2570 mptcp_reset_timer(sk);
2948d0a1
PA		 2571}
2572
76a13b31
GT		 2573/* schedule the timeout timer for the relevant event: either close timeout
2574 * or mp_fail timeout. The close timeout takes precedence on the mp_fail one
2575 */
2576void mptcp_reset_timeout(struct mptcp_sock *msk, unsigned long fail_tout)
2577{
2578 struct sock *sk = (struct sock *)msk;
2579 unsigned long timeout, close_timeout;
2580
2581 if (!fail_tout && !sock_flag(sk, SOCK_DEAD))
2582 return;
2583
2584 close_timeout = inet_csk(sk)->icsk_mtup.probe_timestamp - tcp_jiffies32 + jiffies + TCP_TIMEWAIT_LEN;
2585
2586 /* the close timeout takes precedence on the fail one, and here at least one of
2587 * them is active
2588 */
2589 timeout = sock_flag(sk, SOCK_DEAD) ? close_timeout : fail_tout;
2590
2591 sk_reset_timer(sk, &sk->sk_timer, timeout);
2592}
2593
49fa1919
GT		 2594static void mptcp_mp_fail_no_response(struct mptcp_sock *msk)
2595{
76a13b31 2596 struct sock *ssk = msk->first;
49fa1919
GT		 2597 bool slow;
2598
76a13b31
GT		 2599 if (!ssk)
2600 return;
49fa1919 2601
76a13b31
GT		 2602 pr_debug("MP_FAIL doesn't respond, reset the subflow");
2603
2604 slow = lock_sock_fast(ssk);
2605 mptcp_subflow_reset(ssk);
2606 WRITE_ONCE(mptcp_subflow_ctx(ssk)->fail_tout, 0);
2607 unlock_sock_fast(ssk, slow);
2608
2609 mptcp_reset_timeout(msk, 0);
49fa1919
GT		 2610}
2611
d21f8348
PA		 2612static void mptcp_do_fastclose(struct sock *sk)
2613{
2614 struct mptcp_subflow_context *subflow, *tmp;
2615 struct mptcp_sock *msk = mptcp_sk(sk);
2616
2617 mptcp_for_each_subflow_safe(msk, subflow, tmp)
2618 __mptcp_close_ssk(sk, mptcp_subflow_tcp_sock(subflow),
2619 subflow, MPTCP_CF_FASTCLOSE);
2620}
2621
2948d0a1
PA		 2622static void mptcp_worker(struct work_struct *work)
2623{
2624 struct mptcp_sock *msk = container_of(work, struct mptcp_sock, work);
80638684 2625 struct sock *sk = (struct sock *)msk;
76a13b31 2626 unsigned long fail_tout;
2948d0a1
PA		 2627 int state;
2628
2629 lock_sock(sk);
2630 state = sk->sk_state;
d6a04437 2631 if (unlikely((1 << state) & (TCPF_CLOSE | TCPF_LISTEN)))
2948d0a1
PA		 2632 goto unlock;
2633
2948d0a1
PA		 2634 mptcp_check_fastclose(msk);
2635
86e39e04 2636 mptcp_pm_nl_work(msk);
2948d0a1 2637
81c1d029
PA		 2638 mptcp_check_send_data_fin(sk);
2639 mptcp_check_data_fin_ack(sk);
2948d0a1
PA		 2640 mptcp_check_data_fin(sk);
2641
b6985b9b
PA		 2642 if (test_and_clear_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags))
2643 __mptcp_close_subflow(sk);
2644
2948d0a1
PA		 2645 /* There is no point in keeping around an orphaned sk timedout or
2646 * closed, but we need the msk around to reply to incoming DATA_FIN,
2647 * even if it is orphaned and in FIN_WAIT2 state
2648 */
d21f8348 2649 if (sock_flag(sk, SOCK_DEAD)) {
63740448 2650 if (mptcp_should_close(sk)) {
d21f8348
PA		 2651 inet_sk_state_store(sk, TCP_CLOSE);
2652 mptcp_do_fastclose(sk);
2653 }
2654 if (sk->sk_state == TCP_CLOSE) {
2655 __mptcp_destroy_sock(sk);
2656 goto unlock;
2657 }
2948d0a1
PA		 2658 }
2659
2948d0a1
PA		 2660 if (test_and_clear_bit(MPTCP_WORK_RTX, &msk->flags))
2661 __mptcp_retrans(sk);
3b1d6210 2662
76a13b31
GT		 2663 fail_tout = msk->first ? READ_ONCE(mptcp_subflow_ctx(msk->first)->fail_tout) : 0;
2664 if (fail_tout && time_after(jiffies, fail_tout))
2665 mptcp_mp_fail_no_response(msk);
49fa1919 2666
3b1d6210 2667unlock:
80992017
PA		 2668 release_sock(sk);
2669 sock_put(sk);
2670}
2671
784325e9 2672static int __mptcp_init_sock(struct sock *sk)
f870fa0b 2673{
cec37a6e
PK		 2674 struct mptcp_sock *msk = mptcp_sk(sk);
2675
2676 INIT_LIST_HEAD(&msk->conn_list);
ec3edaa7 2677 INIT_LIST_HEAD(&msk->join_list);
18b683bf 2678 INIT_LIST_HEAD(&msk->rtx_queue);
80992017 2679 INIT_WORK(&msk->work, mptcp_worker);
87952603 2680 __skb_queue_head_init(&msk->receive_queue);
ab174ad8 2681 msk->out_of_order_queue = RB_ROOT;
f0e6a4cf 2682 msk->first_pending = NULL;
6511882c 2683 msk->rmem_fwd_alloc = 0;
ce599c51 2684 WRITE_ONCE(msk->rmem_released, 0);
33d41c9c 2685 msk->timer_ival = TCP_RTO_MIN;
cec37a6e 2686
1b1b43ee 2687 WRITE_ONCE(msk->first, NULL);
dc24f8b4 2688 inet_csk(sk)->icsk_sync_mss = mptcp_sync_mss;
752e9067 2689 WRITE_ONCE(msk->csum_enabled, mptcp_is_checksum_enabled(sock_net(sk)));
0530020a 2690 WRITE_ONCE(msk->allow_infinite_fallback, true);
1e1d9d6f 2691 msk->recovery = false;
8ab183de 2692
1b1c7a0e
PK		 2693 mptcp_pm_data_init(msk);
2694
b51f9b80
PA		 2695 /* re-use the csk retrans timer for MPTCP-level retrans */
2696 timer_setup(&msk->sk.icsk_retransmit_timer, mptcp_retransmit_timer, 0);
e16163b6 2697 timer_setup(&sk->sk_timer, mptcp_timeout_timer, 0);
aa1fbd94 2698
f870fa0b
MM		 2699 return 0;
2700}
2701
b29fcfb5 2702static void mptcp_ca_reset(struct sock *sk)
784325e9 2703{
20b5759f 2704 struct inet_connection_sock *icsk = inet_csk(sk);
b29fcfb5
PA		 2705
2706 tcp_assign_congestion_control(sk);
2707 strcpy(mptcp_sk(sk)->ca_name, icsk->icsk_ca_ops->name);
2708
2709 /* no need to keep a reference to the ops, the name will suffice */
2710 tcp_cleanup_congestion_control(sk);
2711 icsk->icsk_ca_ops = NULL;
2712}
2713
2714static int mptcp_init_sock(struct sock *sk)
2715{
fc518953
FW		 2716 struct net *net = sock_net(sk);
2717 int ret;
18b683bf 2718
b6c08380
GT		 2719 ret = __mptcp_init_sock(sk);
2720 if (ret)
2721 return ret;
2722
fc518953
FW		 2723 if (!mptcp_is_enabled(net))
2724 return -ENOPROTOOPT;
2725
2726 if (unlikely(!net->mib.mptcp_statistics) && !mptcp_mib_alloc(net))
2727 return -ENOMEM;
2728
a5ef058d
PA		 2729 set_bit(SOCK_CUSTOM_SOCKOPT, &sk->sk_socket->flags);
2730
20b5759f
PA		 2731 /* fetch the ca name; do it outside __mptcp_init_sock(), so that clone will
2732 * propagate the correct value
2733 */
b29fcfb5 2734 mptcp_ca_reset(sk);
20b5759f 2735
d027236c 2736 sk_sockets_allocated_inc(sk);
a963853f
GT		 2737 sk->sk_rcvbuf = READ_ONCE(net->ipv4.sysctl_tcp_rmem[1]);
2738 sk->sk_sndbuf = READ_ONCE(net->ipv4.sysctl_tcp_wmem[1]);
d027236c 2739
18b683bf
PA		 2740 return 0;
2741}
2742
2743static void __mptcp_clear_xmit(struct sock *sk)
2744{
2745 struct mptcp_sock *msk = mptcp_sk(sk);
2746 struct mptcp_data_frag *dtmp, *dfrag;
2747
d9ca1de8 2748 WRITE_ONCE(msk->first_pending, NULL);
18b683bf 2749 list_for_each_entry_safe(dfrag, dtmp, &msk->rtx_queue, list)
d027236c 2750 dfrag_clear(sk, dfrag);
784325e9
MB		 2751}
2752
30e51b92 2753void mptcp_cancel_work(struct sock *sk)
80992017
PA		 2754{
2755 struct mptcp_sock *msk = mptcp_sk(sk);
2756
b2771d24 2757 if (cancel_work_sync(&msk->work))
e16163b6 2758 __sock_put(sk);
80992017
PA		 2759}
2760
d0876b22 2761void mptcp_subflow_shutdown(struct sock *sk, struct sock *ssk, int how)
21498490
PK		 2762{
2763 lock_sock(ssk);
2764
2765 switch (ssk->sk_state) {
2766 case TCP_LISTEN:
2767 if (!(how & RCV_SHUTDOWN))
2768 break;
df561f66 2769 fallthrough;
21498490 2770 case TCP_SYN_SENT:
c2b2ae39 2771 WARN_ON_ONCE(tcp_disconnect(ssk, O_NONBLOCK));
21498490
PK		 2772 break;
2773 default:
43b54c6e
MM		 2774 if (__mptcp_check_fallback(mptcp_sk(sk))) {
2775 pr_debug("Fallback");
2776 ssk->sk_shutdown |= how;
2777 tcp_shutdown(ssk, how);
81c1d029
PA		 2778
2779 /* simulate the data_fin ack reception to let the state
2780 * machine move forward
2781 */
2782 WRITE_ONCE(mptcp_sk(sk)->snd_una, mptcp_sk(sk)->snd_nxt);
2783 mptcp_schedule_work(sk);
43b54c6e
MM		 2784 } else {
2785 pr_debug("Sending DATA_FIN on subflow %p", ssk);
43b54c6e 2786 tcp_send_ack(ssk);
6477dd39
MM		 2787 if (!mptcp_timer_pending(sk))
2788 mptcp_reset_timer(sk);
43b54c6e 2789 }
21498490
PK		 2790 break;
2791 }
2792
21498490
PK		 2793 release_sock(ssk);
2794}
2795
6920b851
MM		 2796static const unsigned char new_state[16] = {
2797 /* current state: new state: action: */
2798 [0 /* (Invalid) */] = TCP_CLOSE,
2799 [TCP_ESTABLISHED] = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2800 [TCP_SYN_SENT] = TCP_CLOSE,
2801 [TCP_SYN_RECV] = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2802 [TCP_FIN_WAIT1] = TCP_FIN_WAIT1,
2803 [TCP_FIN_WAIT2] = TCP_FIN_WAIT2,
2804 [TCP_TIME_WAIT] = TCP_CLOSE, /* should not happen ! */
2805 [TCP_CLOSE] = TCP_CLOSE,
2806 [TCP_CLOSE_WAIT] = TCP_LAST_ACK | TCP_ACTION_FIN,
2807 [TCP_LAST_ACK] = TCP_LAST_ACK,
2808 [TCP_LISTEN] = TCP_CLOSE,
2809 [TCP_CLOSING] = TCP_CLOSING,
2810 [TCP_NEW_SYN_RECV] = TCP_CLOSE, /* should not happen ! */
2811};
2812
2813static int mptcp_close_state(struct sock *sk)
2814{
2815 int next = (int)new_state[sk->sk_state];
2816 int ns = next & TCP_STATE_MASK;
2817
2818 inet_sk_state_store(sk, ns);
2819
2820 return next & TCP_ACTION_FIN;
2821}
2822
81c1d029 2823static void mptcp_check_send_data_fin(struct sock *sk)
f870fa0b 2824{
e16163b6 2825 struct mptcp_subflow_context *subflow;
f870fa0b
MM		 2826 struct mptcp_sock *msk = mptcp_sk(sk);
2827
e16163b6
PA		 2828 pr_debug("msk=%p snd_data_fin_enable=%d pending=%d snd_nxt=%llu write_seq=%llu",
2829 msk, msk->snd_data_fin_enable, !!mptcp_send_head(sk),
2830 msk->snd_nxt, msk->write_seq);
43b54c6e 2831
e16163b6
PA		 2832 /* we still need to enqueue subflows or not really shutting down,
2833 * skip this
2834 */
2835 if (!msk->snd_data_fin_enable || msk->snd_nxt + 1 != msk->write_seq ||
2836 mptcp_send_head(sk))
2837 return;
2838
2839 WRITE_ONCE(msk->snd_nxt, msk->write_seq);
2840
e16163b6
PA		 2841 mptcp_for_each_subflow(msk, subflow) {
2842 struct sock *tcp_sk = mptcp_subflow_tcp_sock(subflow);
43b54c6e 2843
e16163b6 2844 mptcp_subflow_shutdown(sk, tcp_sk, SEND_SHUTDOWN);
43b54c6e 2845 }
e16163b6 2846}
2c22c06c 2847
e16163b6
PA		 2848static void __mptcp_wr_shutdown(struct sock *sk)
2849{
2850 struct mptcp_sock *msk = mptcp_sk(sk);
43b54c6e 2851
e16163b6
PA		 2852 pr_debug("msk=%p snd_data_fin_enable=%d shutdown=%x state=%d pending=%d",
2853 msk, msk->snd_data_fin_enable, sk->sk_shutdown, sk->sk_state,
2854 !!mptcp_send_head(sk));
2855
2856 /* will be ignored by fallback sockets */
2857 WRITE_ONCE(msk->write_seq, msk->write_seq + 1);
2858 WRITE_ONCE(msk->snd_data_fin_enable, 1);
2859
81c1d029 2860 mptcp_check_send_data_fin(sk);
e16163b6
PA		 2861}
2862
2863static void __mptcp_destroy_sock(struct sock *sk)
2864{
e16163b6 2865 struct mptcp_sock *msk = mptcp_sk(sk);
e16163b6
PA		 2866
2867 pr_debug("msk=%p", msk);
f870fa0b 2868
3abc05d9
FW		 2869 might_sleep();
2870
bcf3cf93 2871 mptcp_stop_timer(sk);
e16163b6
PA		 2872 sk_stop_timer(sk, &sk->sk_timer);
2873 msk->pm.status = 0;
b2c5b614 2874
e16163b6 2875 sk->sk_prot->destroy(sk);
80992017 2876
6511882c 2877 WARN_ON_ONCE(msk->rmem_fwd_alloc);
87952603 2878 WARN_ON_ONCE(msk->rmem_released);
e16163b6
PA		 2879 sk_stream_kill_queues(sk);
2880 xfrm_sk_free_policy(sk);
aa1fbd94 2881
e16163b6
PA		 2882 sock_put(sk);
2883}
2884
63740448
PA		 2885void __mptcp_unaccepted_force_close(struct sock *sk)
2886{
2887 sock_set_flag(sk, SOCK_DEAD);
2888 inet_sk_state_store(sk, TCP_CLOSE);
2889 mptcp_do_fastclose(sk);
2890 __mptcp_destroy_sock(sk);
2891}
2892
d21f8348
PA		 2893static __poll_t mptcp_check_readable(struct mptcp_sock *msk)
2894{
2895 /* Concurrent splices from sk_receive_queue into receive_queue will
2896 * always show at least one non-empty queue when checked in this order.
2897 */
2898 if (skb_queue_empty_lockless(&((struct sock *)msk)->sk_receive_queue) &&
2899 skb_queue_empty_lockless(&msk->receive_queue))
2900 return 0;
2901
2902 return EPOLLIN | EPOLLRDNORM;
2903}
2904
c558246e
MD		 2905static void mptcp_listen_inuse_dec(struct sock *sk)
2906{
2907 if (inet_sk_state_load(sk) == TCP_LISTEN)
2908 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
2909}
2910
26d3e21c 2911bool __mptcp_close(struct sock *sk, long timeout)
e16163b6
PA		 2912{
2913 struct mptcp_subflow_context *subflow;
76a13b31 2914 struct mptcp_sock *msk = mptcp_sk(sk);
e16163b6 2915 bool do_cancel_work = false;
d4e85922 2916 int subflows_alive = 0;
e16163b6 2917
6b9831bf 2918 WRITE_ONCE(sk->sk_shutdown, SHUTDOWN_MASK);
e16163b6
PA		 2919
2920 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE)) {
c558246e 2921 mptcp_listen_inuse_dec(sk);
e16163b6
PA		 2922 inet_sk_state_store(sk, TCP_CLOSE);
2923 goto cleanup;
2924 }
6771bfd9 2925
8d547809
PA		 2926 if (mptcp_check_readable(msk) || timeout < 0) {
2927 /* If the msk has read data, or the caller explicitly ask it,
2928 * do the MPTCP equivalent of TCP reset, aka MPTCP fastclose
2929 */
d21f8348
PA		 2930 inet_sk_state_store(sk, TCP_CLOSE);
2931 mptcp_do_fastclose(sk);
8d547809 2932 timeout = 0;
d21f8348 2933 } else if (mptcp_close_state(sk)) {
e16163b6 2934 __mptcp_wr_shutdown(sk);
d21f8348 2935 }
e16163b6
PA		 2936
2937 sk_stream_wait_close(sk, timeout);
2938
2939cleanup:
2940 /* orphan all the subflows */
2941 inet_csk(sk)->icsk_mtup.probe_timestamp = tcp_jiffies32;
76a13b31 2942 mptcp_for_each_subflow(msk, subflow) {
e16163b6 2943 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
49054556 2944 bool slow = lock_sock_fast_nested(ssk);
e16163b6 2945
d4e85922
PA		 2946 subflows_alive += ssk->sk_state != TCP_CLOSE;
2947
76a13b31
GT		 2948 /* since the close timeout takes precedence on the fail one,
2949 * cancel the latter
2950 */
2951 if (ssk == msk->first)
2952 subflow->fail_tout = 0;
2953
fe948001
MD		 2954 /* detach from the parent socket, but allow data_ready to
2955 * push incoming data into the mptcp stack, to properly ack it
2956 */
2957 ssk->sk_socket = NULL;
2958 ssk->sk_wq = NULL;
e16163b6 2959 unlock_sock_fast(ssk, slow);
e16163b6
PA		 2960 }
2961 sock_orphan(sk);
2962
d4e85922
PA		 2963 /* all the subflows are closed, only timeout can change the msk
2964 * state, let's not keep resources busy for no reasons
2965 */
2966 if (subflows_alive == 0)
2967 inet_sk_state_store(sk, TCP_CLOSE);
2968
e16163b6
PA		 2969 sock_hold(sk);
2970 pr_debug("msk=%p state=%d", sk, sk->sk_state);
00df24f1 2971 if (msk->token)
76a13b31 2972 mptcp_event(MPTCP_EVENT_CLOSED, msk, NULL, GFP_KERNEL);
b29fcfb5 2973
e16163b6
PA		 2974 if (sk->sk_state == TCP_CLOSE) {
2975 __mptcp_destroy_sock(sk);
2976 do_cancel_work = true;
2977 } else {
76a13b31 2978 mptcp_reset_timeout(msk, 0);
e16163b6 2979 }
26d3e21c
MD		 2980
2981 return do_cancel_work;
2982}
2983
2984static void mptcp_close(struct sock *sk, long timeout)
2985{
2986 bool do_cancel_work;
2987
2988 lock_sock(sk);
2989
2990 do_cancel_work = __mptcp_close(sk, timeout);
e16163b6
PA		 2991 release_sock(sk);
2992 if (do_cancel_work)
2993 mptcp_cancel_work(sk);
b911c97c 2994
e16163b6 2995 sock_put(sk);
f870fa0b
MM		 2996}
2997
7e8b88ec 2998static void mptcp_copy_inaddrs(struct sock *msk, const struct sock *ssk)
cf7da0d6
PK		 2999{
3000#if IS_ENABLED(CONFIG_MPTCP_IPV6)
3001 const struct ipv6_pinfo *ssk6 = inet6_sk(ssk);
3002 struct ipv6_pinfo *msk6 = inet6_sk(msk);
3003
3004 msk->sk_v6_daddr = ssk->sk_v6_daddr;
3005 msk->sk_v6_rcv_saddr = ssk->sk_v6_rcv_saddr;
3006
3007 if (msk6 && ssk6) {
3008 msk6->saddr = ssk6->saddr;
3009 msk6->flow_label = ssk6->flow_label;
3010 }
3011#endif
3012
3013 inet_sk(msk)->inet_num = inet_sk(ssk)->inet_num;
3014 inet_sk(msk)->inet_dport = inet_sk(ssk)->inet_dport;
3015 inet_sk(msk)->inet_sport = inet_sk(ssk)->inet_sport;
3016 inet_sk(msk)->inet_daddr = inet_sk(ssk)->inet_daddr;
3017 inet_sk(msk)->inet_saddr = inet_sk(ssk)->inet_saddr;
3018 inet_sk(msk)->inet_rcv_saddr = inet_sk(ssk)->inet_rcv_saddr;
3019}
3020
18b683bf
PA		 3021static int mptcp_disconnect(struct sock *sk, int flags)
3022{
76e2a55d
PA		 3023 struct mptcp_sock *msk = mptcp_sk(sk);
3024
0ad529d9
PA		 3025 /* Deny disconnect if other threads are blocked in sk_wait_event()
3026 * or inet_wait_for_connect().
3027 */
3028 if (sk->sk_wait_pending)
3029 return -EBUSY;
3030
7d803344
PA		 3031 /* We are on the fastopen error path. We can't call straight into the
3032 * subflows cleanup code due to lock nesting (we are already under
c2b2ae39 3033 * msk->firstsocket lock).
7d803344
PA		 3034 */
3035 if (msk->fastopening)
c2b2ae39 3036 return -EBUSY;
7d803344 3037
c558246e 3038 mptcp_listen_inuse_dec(sk);
b29fcfb5
PA		 3039 inet_sk_state_store(sk, TCP_CLOSE);
3040
bcf3cf93 3041 mptcp_stop_timer(sk);
b29fcfb5
PA		 3042 sk_stop_timer(sk, &sk->sk_timer);
3043
00df24f1
GT		 3044 if (msk->token)
3045 mptcp_event(MPTCP_EVENT_CLOSED, msk, NULL, GFP_KERNEL);
b29fcfb5 3046
c0bf3c6a
PA		 3047 /* msk->subflow is still intact, the following will not free the first
3048 * subflow
3049 */
3050 mptcp_destroy_common(msk, MPTCP_CF_FASTCLOSE);
b29fcfb5 3051 msk->last_snd = NULL;
e9d09bac
PA		 3052 WRITE_ONCE(msk->flags, 0);
3053 msk->cb_flags = 0;
3054 msk->push_pending = 0;
b29fcfb5
PA		 3055 msk->recovery = false;
3056 msk->can_ack = false;
3057 msk->fully_established = false;
3058 msk->rcv_data_fin = false;
3059 msk->snd_data_fin_enable = false;
3060 msk->rcv_fastclose = false;
3061 msk->use_64bit_ack = false;
3062 WRITE_ONCE(msk->csum_enabled, mptcp_is_checksum_enabled(sock_net(sk)));
3063 mptcp_pm_data_reset(msk);
3064 mptcp_ca_reset(sk);
3065
6b9831bf 3066 WRITE_ONCE(sk->sk_shutdown, 0);
b29fcfb5 3067 sk_error_report(sk);
42c556fe 3068 return 0;
18b683bf
PA		 3069}
3070
b0519de8
FW		 3071#if IS_ENABLED(CONFIG_MPTCP_IPV6)
3072static struct ipv6_pinfo *mptcp_inet6_sk(const struct sock *sk)
3073{
3074 unsigned int offset = sizeof(struct mptcp6_sock) - sizeof(struct ipv6_pinfo);
3075
3076 return (struct ipv6_pinfo *)(((u8 *)sk) + offset);
3077}
3078#endif
3079
7e8b88ec
PA		 3080struct sock *mptcp_sk_clone_init(const struct sock *sk,
3081 const struct mptcp_options_received *mp_opt,
3082 struct sock *ssk,
3083 struct request_sock *req)
b0519de8 3084{
58b09919 3085 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
b0519de8 3086 struct sock *nsk = sk_clone_lock(sk, GFP_ATOMIC);
58b09919 3087 struct mptcp_sock *msk;
b0519de8
FW		 3088
3089 if (!nsk)
3090 return NULL;
3091
3092#if IS_ENABLED(CONFIG_MPTCP_IPV6)
3093 if (nsk->sk_family == AF_INET6)
3094 inet_sk(nsk)->pinet6 = mptcp_inet6_sk(nsk);
3095#endif
3096
0ad529d9 3097 nsk->sk_wait_pending = 0;
58b09919
PA		 3098 __mptcp_init_sock(nsk);
3099
3100 msk = mptcp_sk(nsk);
3101 msk->local_key = subflow_req->local_key;
3102 msk->token = subflow_req->token;
5b825727 3103 WRITE_ONCE(msk->subflow, NULL);
b6985b9b 3104 msk->in_accept_queue = 1;
b93df08c 3105 WRITE_ONCE(msk->fully_established, false);
74c7dfbe 3106 if (mp_opt->suboptions & OPTION_MPTCP_CSUMREQD)
06251181 3107 WRITE_ONCE(msk->csum_enabled, true);
58b09919 3108
58b09919 3109 msk->write_seq = subflow_req->idsn + 1;
eaa2ffab 3110 msk->snd_nxt = msk->write_seq;
7439d687
PA		 3111 msk->snd_una = msk->write_seq;
3112 msk->wnd_end = msk->snd_nxt + req->rsk_rcv_wnd;
df00b087 3113 msk->setsockopt_seq = mptcp_sk(sk)->setsockopt_seq;
6f8a612a 3114
5e20087d 3115 sock_reset_flag(nsk, SOCK_RCU_FREE);
0c148460 3116 security_inet_csk_clone(nsk, req);
7e8b88ec
PA		 3117
3118 /* this can't race with mptcp_close(), as the msk is
3119 * not yet exposted to user-space
3120 */
3121 inet_sk_state_store(nsk, TCP_ESTABLISHED);
3122
3123 /* The msk maintain a ref to each subflow in the connections list */
3124 WRITE_ONCE(msk->first, ssk);
3125 list_add(&mptcp_subflow_ctx(ssk)->node, &msk->conn_list);
3126 sock_hold(ssk);
3127
3128 /* new mpc subflow takes ownership of the newly
3129 * created mptcp socket
3130 */
3131 mptcp_token_accept(subflow_req, msk);
3132
3133 /* set msk addresses early to ensure mptcp_pm_get_local_id()
3134 * uses the correct data
3135 */
3136 mptcp_copy_inaddrs(nsk, ssk);
3137 mptcp_propagate_sndbuf(nsk, ssk);
3138
3139 mptcp_rcv_space_init(msk, ssk);
58b09919
PA		 3140 bh_unlock_sock(nsk);
3141
b6985b9b 3142 /* note: the newly allocated socket refcount is 2 now */
b0519de8
FW		 3143 return nsk;
3144}
3145
a6b118fe
FW		 3146void mptcp_rcv_space_init(struct mptcp_sock *msk, const struct sock *ssk)
3147{
3148 const struct tcp_sock *tp = tcp_sk(ssk);
3149
3150 msk->rcvq_space.copied = 0;
3151 msk->rcvq_space.rtt_us = 0;
3152
3153 msk->rcvq_space.time = tp->tcp_mstamp;
3154
3155 /* initial rcv_space offering made to peer */
3156 msk->rcvq_space.space = min_t(u32, tp->rcv_wnd,
3157 TCP_INIT_CWND * tp->advmss);
3158 if (msk->rcvq_space.space == 0)
3159 msk->rcvq_space.space = TCP_INIT_CWND * TCP_MSS_DEFAULT;
6f8a612a 3160
7439d687 3161 WRITE_ONCE(msk->wnd_end, msk->snd_nxt + tcp_sk(ssk)->snd_wnd);
a6b118fe
FW		 3162}
3163
cf7da0d6
PK		 3164static struct sock *mptcp_accept(struct sock *sk, int flags, int *err,
3165 bool kern)
3166{
3167 struct mptcp_sock *msk = mptcp_sk(sk);
3168 struct socket *listener;
3169 struct sock *newsk;
3170
5b825727 3171 listener = READ_ONCE(msk->subflow);
cf7da0d6
PK		 3172 if (WARN_ON_ONCE(!listener)) {
3173 *err = -EINVAL;
3174 return NULL;
3175 }
3176
3177 pr_debug("msk=%p, listener=%p", msk, mptcp_subflow_ctx(listener->sk));
3178 newsk = inet_csk_accept(listener->sk, flags, err, kern);
3179 if (!newsk)
3180 return NULL;
3181
3182 pr_debug("msk=%p, subflow is mptcp=%d", msk, sk_is_mptcp(newsk));
cf7da0d6
PK		 3183 if (sk_is_mptcp(newsk)) {
3184 struct mptcp_subflow_context *subflow;
3185 struct sock *new_mptcp_sock;
cf7da0d6
PK		 3186
3187 subflow = mptcp_subflow_ctx(newsk);
58b09919 3188 new_mptcp_sock = subflow->conn;
cf7da0d6 3189
58b09919
PA		 3190 /* is_mptcp should be false if subflow->conn is missing, see
3191 * subflow_syn_recv_sock()
3192 */
3193 if (WARN_ON_ONCE(!new_mptcp_sock)) {
3194 tcp_sk(newsk)->is_mptcp = 0;
d6692b3b 3195 goto out;
cf7da0d6
PK		 3196 }
3197
cf7da0d6 3198 newsk = new_mptcp_sock;
0397c6d8 3199 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPCAPABLEPASSIVEACK);
fc518953
FW		 3200 } else {
3201 MPTCP_INC_STATS(sock_net(sk),
3202 MPTCP_MIB_MPCAPABLEPASSIVEFALLBACK);
cf7da0d6
PK		 3203 }
3204
d6692b3b
FW		 3205out:
3206 newsk->sk_kern_sock = kern;
cf7da0d6
PK		 3207 return newsk;
3208}
3209
c0bf3c6a 3210void mptcp_destroy_common(struct mptcp_sock *msk, unsigned int flags)
5c8c1640 3211{
c0bf3c6a 3212 struct mptcp_subflow_context *subflow, *tmp;
87952603
PA		 3213 struct sock *sk = (struct sock *)msk;
3214
6e628cd3
PA		 3215 __mptcp_clear_xmit(sk);
3216
c0bf3c6a 3217 /* join list will be eventually flushed (with rst) at sock lock release time */
5efbf6f7 3218 mptcp_for_each_subflow_safe(msk, subflow, tmp)
c0bf3c6a
PA		 3219 __mptcp_close_ssk(sk, mptcp_subflow_tcp_sock(subflow), subflow, flags);
3220
87952603 3221 /* move to sk_receive_queue, sk_stream_kill_queues will purge it */
b29fcfb5 3222 mptcp_data_lock(sk);
87952603 3223 skb_queue_splice_tail_init(&msk->receive_queue, &sk->sk_receive_queue);
6511882c 3224 __skb_queue_purge(&sk->sk_receive_queue);
5c8c1640 3225 skb_rbtree_purge(&msk->out_of_order_queue);
b29fcfb5 3226 mptcp_data_unlock(sk);
6511882c
PA		 3227
3228 /* move all the rx fwd alloc into the sk_mem_reclaim_final in
3229 * inet_sock_destruct() will dispose it
3230 */
3231 sk->sk_forward_alloc += msk->rmem_fwd_alloc;
3232 msk->rmem_fwd_alloc = 0;
5c8c1640
GT		 3233 mptcp_token_destroy(msk);
3234 mptcp_pm_free_anno_list(msk);
4638de5a 3235 mptcp_free_local_addr_list(msk);
5c8c1640
GT		 3236}
3237
79c0949e
PK		 3238static void mptcp_destroy(struct sock *sk)
3239{
c9fd9c5f
FW		 3240 struct mptcp_sock *msk = mptcp_sk(sk);
3241
c0bf3c6a
PA		 3242 /* clears msk->subflow, allowing the following to close
3243 * even the initial subflow
3244 */
3245 mptcp_dispose_initial_subflow(msk);
3246 mptcp_destroy_common(msk, 0);
d027236c 3247 sk_sockets_allocated_dec(sk);
79c0949e
PK		 3248}
3249
6e628cd3
PA		 3250void __mptcp_data_acked(struct sock *sk)
3251{
3252 if (!sock_owned_by_user(sk))
3253 __mptcp_clean_una(sk);
3254 else
e9d09bac 3255 __set_bit(MPTCP_CLEAN_UNA, &mptcp_sk(sk)->cb_flags);
6e628cd3
PA		 3256
3257 if (mptcp_pending_data_fin_ack(sk))
3258 mptcp_schedule_work(sk);
3259}
3260
219d0499 3261void __mptcp_check_push(struct sock *sk, struct sock *ssk)
6e628cd3
PA		 3262{
3263 if (!mptcp_send_head(sk))
3264 return;
3265
73a0052a
GT		 3266 if (!sock_owned_by_user(sk))
3267 __mptcp_subflow_push_pending(sk, ssk, false);
3268 else
e9d09bac 3269 __set_bit(MPTCP_PUSH_PENDING, &mptcp_sk(sk)->cb_flags);
6e628cd3
PA		 3270}
3271
e9d09bac
PA		 3272#define MPTCP_FLAGS_PROCESS_CTX_NEED (BIT(MPTCP_PUSH_PENDING) | \
3273 BIT(MPTCP_RETRANSMIT) | \
3274 BIT(MPTCP_FLUSH_JOIN_LIST))
3275
e93da928 3276/* processes deferred events and flush wmem */
14c441b5 3277static void mptcp_release_cb(struct sock *sk)
e9d09bac 3278 __must_hold(&sk->sk_lock.slock)
14c441b5 3279{
e9d09bac
PA		 3280 struct mptcp_sock *msk = mptcp_sk(sk);
3281
c2e6048f 3282 for (;;) {
e9d09bac
PA		 3283 unsigned long flags = (msk->cb_flags & MPTCP_FLAGS_PROCESS_CTX_NEED) |
3284 msk->push_pending;
56a666c4
PA		 3285 struct list_head join_list;
3286
c2e6048f
PA		 3287 if (!flags)
3288 break;
3289
56a666c4
PA		 3290 INIT_LIST_HEAD(&join_list);
3291 list_splice_init(&msk->join_list, &join_list);
3292
c2e6048f 3293 /* the following actions acquire the subflow socket lock
6e628cd3
PA		 3294 *
3295 * 1) can't be invoked in atomic scope
3296 * 2) must avoid ABBA deadlock with msk socket spinlock: the RX
3297 * datapath acquires the msk socket spinlock while helding
3298 * the subflow socket lock
3299 */
e9d09bac
PA		 3300 msk->push_pending = 0;
3301 msk->cb_flags &= ~flags;
6e628cd3 3302 spin_unlock_bh(&sk->sk_lock.slock);
56a666c4 3303
3e501490 3304 if (flags & BIT(MPTCP_FLUSH_JOIN_LIST))
56a666c4 3305 __mptcp_flush_join_list(sk, &join_list);
2e5de7e0 3306 if (flags & BIT(MPTCP_PUSH_PENDING))
c2e6048f 3307 __mptcp_push_pending(sk, 0);
2d6f5a2b
PA		 3308 if (flags & BIT(MPTCP_RETRANSMIT))
3309 __mptcp_retrans(sk);
c2e6048f
PA		 3310
3311 cond_resched();
6e628cd3
PA		 3312 spin_lock_bh(&sk->sk_lock.slock);
3313 }
c2e6048f 3314
e9d09bac 3315 if (__test_and_clear_bit(MPTCP_CLEAN_UNA, &msk->cb_flags))
417789df 3316 __mptcp_clean_una_wakeup(sk);
65a569b0
PA		 3317 if (unlikely(&msk->cb_flags)) {
3318 /* be sure to set the current sk state before tacking actions
3319 * depending on sk_state, that is processing MPTCP_ERROR_REPORT
3320 */
3321 if (__test_and_clear_bit(MPTCP_CONNECTED, &msk->cb_flags))
3322 __mptcp_set_connected(sk);
3323 if (__test_and_clear_bit(MPTCP_ERROR_REPORT, &msk->cb_flags))
3324 __mptcp_error_report(sk);
43f5b111
PA		 3325 if (__test_and_clear_bit(MPTCP_RESET_SCHEDULER, &msk->cb_flags))
3326 msk->last_snd = NULL;
65a569b0 3327 }
6e628cd3 3328
87952603 3329 __mptcp_update_rmem(sk);
14c441b5
PA		 3330}
3331
bcd97734
PA		 3332/* MP_JOIN client subflow must wait for 4th ack before sending any data:
3333 * TCP can't schedule delack timer before the subflow is fully established.
3334 * MPTCP uses the delack timer to do 3rd ack retransmissions
3335 */
3336static void schedule_3rdack_retransmission(struct sock *ssk)
3337{
3338 struct inet_connection_sock *icsk = inet_csk(ssk);
3339 struct tcp_sock *tp = tcp_sk(ssk);
3340 unsigned long timeout;
3341
3342 if (mptcp_subflow_ctx(ssk)->fully_established)
3343 return;
3344
3345 /* reschedule with a timeout above RTT, as we must look only for drop */
3346 if (tp->srtt_us)
3347 timeout = usecs_to_jiffies(tp->srtt_us >> (3 - 1));
3348 else
3349 timeout = TCP_TIMEOUT_INIT;
3350 timeout += jiffies;
3351
3352 WARN_ON_ONCE(icsk->icsk_ack.pending & ICSK_ACK_TIMER);
3353 icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
3354 icsk->icsk_ack.timeout = timeout;
3355 sk_reset_timer(ssk, &icsk->icsk_delack_timer, timeout);
3356}
3357
b19bc294
PA		 3358void mptcp_subflow_process_delegated(struct sock *ssk)
3359{
3360 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
3361 struct sock *sk = subflow->conn;
3362
bcd97734
PA		 3363 if (test_bit(MPTCP_DELEGATE_SEND, &subflow->delegated_status)) {
3364 mptcp_data_lock(sk);
3365 if (!sock_owned_by_user(sk))
73a0052a 3366 __mptcp_subflow_push_pending(sk, ssk, true);
bcd97734 3367 else
e9d09bac 3368 __set_bit(MPTCP_PUSH_PENDING, &mptcp_sk(sk)->cb_flags);
bcd97734
PA		 3369 mptcp_data_unlock(sk);
3370 mptcp_subflow_delegated_done(subflow, MPTCP_DELEGATE_SEND);
3371 }
3372 if (test_bit(MPTCP_DELEGATE_ACK, &subflow->delegated_status)) {
3373 schedule_3rdack_retransmission(ssk);
3374 mptcp_subflow_delegated_done(subflow, MPTCP_DELEGATE_ACK);
3375 }
b19bc294
PA		 3376}
3377
2c5ebd00
PA		 3378static int mptcp_hash(struct sock *sk)
3379{
3380 /* should never be called,
3381 * we hash the TCP subflows not the master socket
3382 */
3383 WARN_ON_ONCE(1);
3384 return 0;
3385}
3386
3387static void mptcp_unhash(struct sock *sk)
3388{
3389 /* called from sk_common_release(), but nothing to do here */
3390}
3391
cec37a6e 3392static int mptcp_get_port(struct sock *sk, unsigned short snum)
f870fa0b
MM		 3393{
3394 struct mptcp_sock *msk = mptcp_sk(sk);
cec37a6e 3395 struct socket *ssock;
f870fa0b 3396
61761231 3397 ssock = msk->subflow;
cec37a6e
PK		 3398 pr_debug("msk=%p, subflow=%p", msk, ssock);
3399 if (WARN_ON_ONCE(!ssock))
3400 return -EINVAL;
f870fa0b 3401
cec37a6e
PK		 3402 return inet_csk_get_port(ssock->sk, snum);
3403}
f870fa0b 3404
cec37a6e
PK		 3405void mptcp_finish_connect(struct sock *ssk)
3406{
3407 struct mptcp_subflow_context *subflow;
3408 struct mptcp_sock *msk;
3409 struct sock *sk;
f870fa0b 3410
cec37a6e 3411 subflow = mptcp_subflow_ctx(ssk);
cec37a6e
PK		 3412 sk = subflow->conn;
3413 msk = mptcp_sk(sk);
3414
648ef4b8
MM		 3415 pr_debug("msk=%p, token=%u", sk, subflow->token);
3416
b3ea6b27 3417 subflow->map_seq = subflow->iasn;
648ef4b8 3418 subflow->map_subflow_seq = 1;
6d0060f6 3419
cec37a6e
PK		 3420 /* the socket is not connected yet, no msk/subflow ops can access/race
3421 * accessing the field below
3422 */
cec37a6e 3423 WRITE_ONCE(msk->local_key, subflow->local_key);
6d0060f6 3424 WRITE_ONCE(msk->write_seq, subflow->idsn + 1);
eaa2ffab 3425 WRITE_ONCE(msk->snd_nxt, msk->write_seq);
7439d687 3426 WRITE_ONCE(msk->snd_una, msk->write_seq);
1b1c7a0e 3427
6c714f1b 3428 mptcp_pm_new_connection(msk, ssk, 0);
a6b118fe
FW		 3429
3430 mptcp_rcv_space_init(msk, ssk);
f870fa0b
MM		 3431}
3432
866f26f2 3433void mptcp_sock_graft(struct sock *sk, struct socket *parent)
cf7da0d6
PK		 3434{
3435 write_lock_bh(&sk->sk_callback_lock);
3436 rcu_assign_pointer(sk->sk_wq, &parent->wq);
3437 sk_set_socket(sk, parent);
3438 sk->sk_uid = SOCK_INODE(parent)->i_uid;
3439 write_unlock_bh(&sk->sk_callback_lock);
3440}
3441
e16163b6 3442bool mptcp_finish_join(struct sock *ssk)
f296234c 3443{
e16163b6 3444 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
f296234c
PK		 3445 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
3446 struct sock *parent = (void *)msk;
3e501490 3447 bool ret = true;
f296234c
PK		 3448
3449 pr_debug("msk=%p, subflow=%p", msk, subflow);
3450
3451 /* mptcp socket already closing? */
dc87efdb
FW		 3452 if (!mptcp_is_fully_established(parent)) {
3453 subflow->reset_reason = MPTCP_RST_EMPTCP;
f296234c 3454 return false;
dc87efdb 3455 }
f296234c 3456
55b47ca7
PA		 3457 /* active subflow, already present inside the conn_list */
3458 if (!list_empty(&subflow->node)) {
3459 mptcp_subflow_joined(msk, ssk);
3460 return true;
3461 }
f296234c 3462
3e501490
PA		 3463 if (!mptcp_pm_allow_new_subflow(msk))
3464 goto err_prohibited;
10f6d46c 3465
55b47ca7 3466 /* If we can't acquire msk socket lock here, let the release callback
3e501490 3467 * handle it
10f6d46c 3468 */
3e501490
PA		 3469 mptcp_data_lock(parent);
3470 if (!sock_owned_by_user(parent)) {
3471 ret = __mptcp_finish_join(msk, ssk);
3472 if (ret) {
3473 sock_hold(ssk);
3474 list_add_tail(&subflow->node, &msk->conn_list);
3475 }
3476 } else {
e16163b6 3477 sock_hold(ssk);
3e501490 3478 list_add_tail(&subflow->node, &msk->join_list);
e9d09bac 3479 __set_bit(MPTCP_FLUSH_JOIN_LIST, &msk->cb_flags);
e16163b6 3480 }
3e501490
PA		 3481 mptcp_data_unlock(parent);
3482
dc87efdb 3483 if (!ret) {
3e501490 3484err_prohibited:
dc87efdb 3485 subflow->reset_reason = MPTCP_RST_EPROHIBIT;
10f6d46c 3486 return false;
dc87efdb 3487 }
10f6d46c 3488
10f6d46c 3489 return true;
f296234c
PK		 3490}
3491
76e2a55d
PA		 3492static void mptcp_shutdown(struct sock *sk, int how)
3493{
3494 pr_debug("sk=%p, how=%d", sk, how);
3495
3496 if ((how & SEND_SHUTDOWN) && mptcp_close_state(sk))
3497 __mptcp_wr_shutdown(sk);
3498}
3499
6511882c
PA		 3500static int mptcp_forward_alloc_get(const struct sock *sk)
3501{
3502 return sk->sk_forward_alloc + mptcp_sk(sk)->rmem_fwd_alloc;
3503}
3504
644807e3
FW		 3505static int mptcp_ioctl_outq(const struct mptcp_sock *msk, u64 v)
3506{
3507 const struct sock *sk = (void *)msk;
3508 u64 delta;
3509
3510 if (sk->sk_state == TCP_LISTEN)
3511 return -EINVAL;
3512
3513 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
3514 return 0;
3515
3516 delta = msk->write_seq - v;
07c2c7a3
PA		 3517 if (__mptcp_check_fallback(msk) && msk->first) {
3518 struct tcp_sock *tp = tcp_sk(msk->first);
3519
3520 /* the first subflow is disconnected after close - see
3521 * __mptcp_close_ssk(). tcp_disconnect() moves the write_seq
3522 * so ignore that status, too.
3523 */
3524 if (!((1 << msk->first->sk_state) &
3525 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE)))
3526 delta += READ_ONCE(tp->write_seq) - tp->snd_una;
3527 }
644807e3
FW		 3528 if (delta > INT_MAX)
3529 delta = INT_MAX;
3530
3531 return (int)delta;
3532}
3533
3534static int mptcp_ioctl(struct sock *sk, int cmd, unsigned long arg)
3535{
3536 struct mptcp_sock *msk = mptcp_sk(sk);
3537 bool slow;
3538 int answ;
3539
3540 switch (cmd) {
3541 case SIOCINQ:
3542 if (sk->sk_state == TCP_LISTEN)
3543 return -EINVAL;
3544
3545 lock_sock(sk);
3546 __mptcp_move_skbs(msk);
3547 answ = mptcp_inq_hint(sk);
3548 release_sock(sk);
3549 break;
3550 case SIOCOUTQ:
3551 slow = lock_sock_fast(sk);
3552 answ = mptcp_ioctl_outq(msk, READ_ONCE(msk->snd_una));
3553 unlock_sock_fast(sk, slow);
3554 break;
3555 case SIOCOUTQNSD:
3556 slow = lock_sock_fast(sk);
3557 answ = mptcp_ioctl_outq(msk, msk->snd_nxt);
3558 unlock_sock_fast(sk, slow);
3559 break;
3560 default:
3561 return -ENOIOCTLCMD;
3562 }
3563
3564 return put_user(answ, (int __user *)arg);
3565}
3566
54f1944e
PA		 3567static void mptcp_subflow_early_fallback(struct mptcp_sock *msk,
3568 struct mptcp_subflow_context *subflow)
3569{
3570 subflow->request_mptcp = 0;
3571 __mptcp_do_fallback(msk);
3572}
3573
3574static int mptcp_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
3575{
3576 struct mptcp_subflow_context *subflow;
3577 struct mptcp_sock *msk = mptcp_sk(sk);
3578 struct socket *ssock;
3579 int err = -EINVAL;
3580
3581 ssock = __mptcp_nmpc_socket(msk);
ddb1a072
PA		 3582 if (IS_ERR(ssock))
3583 return PTR_ERR(ssock);
54f1944e
PA		 3584
3585 mptcp_token_destroy(msk);
3586 inet_sk_state_store(sk, TCP_SYN_SENT);
3587 subflow = mptcp_subflow_ctx(ssock->sk);
3588#ifdef CONFIG_TCP_MD5SIG
3589 /* no MPTCP if MD5SIG is enabled on this socket or we may run out of
3590 * TCP option space.
3591 */
3592 if (rcu_access_pointer(tcp_sk(ssock->sk)->md5sig_info))
3593 mptcp_subflow_early_fallback(msk, subflow);
3594#endif
3595 if (subflow->request_mptcp && mptcp_token_new_connect(ssock->sk)) {
3596 MPTCP_INC_STATS(sock_net(ssock->sk), MPTCP_MIB_TOKENFALLBACKINIT);
3597 mptcp_subflow_early_fallback(msk, subflow);
3598 }
3599 if (likely(!__mptcp_check_fallback(msk)))
3600 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPCAPABLEACTIVE);
3601
3602 /* if reaching here via the fastopen/sendmsg path, the caller already
3603 * acquired the subflow socket lock, too.
3604 */
7d803344 3605 if (msk->fastopening)
786fc124 3606 err = __inet_stream_connect(ssock, uaddr, addr_len, O_NONBLOCK, 1);
54f1944e 3607 else
786fc124 3608 err = inet_stream_connect(ssock, uaddr, addr_len, O_NONBLOCK);
54f1944e
PA		 3609 inet_sk(sk)->defer_connect = inet_sk(ssock->sk)->defer_connect;
3610
3611 /* on successful connect, the msk state will be moved to established by
3612 * subflow_finish_connect()
3613 */
3614 if (unlikely(err && err != -EINPROGRESS)) {
3615 inet_sk_state_store(sk, inet_sk_state_load(ssock->sk));
3616 return err;
3617 }
3618
3619 mptcp_copy_inaddrs(sk, ssock->sk);
3620
786fc124
PA		 3621 /* silence EINPROGRESS and let the caller inet_stream_connect
3622 * handle the connection in progress
54f1944e 3623 */
786fc124 3624 return 0;
54f1944e
PA		 3625}
3626
f870fa0b
MM		 3627static struct proto mptcp_prot = {
3628 .name = "MPTCP",
3629 .owner = THIS_MODULE,
3630 .init = mptcp_init_sock,
54f1944e 3631 .connect = mptcp_connect,
18b683bf 3632 .disconnect = mptcp_disconnect,
f870fa0b 3633 .close = mptcp_close,
cf7da0d6 3634 .accept = mptcp_accept,
717e79c8
PK		 3635 .setsockopt = mptcp_setsockopt,
3636 .getsockopt = mptcp_getsockopt,
76e2a55d 3637 .shutdown = mptcp_shutdown,
79c0949e 3638 .destroy = mptcp_destroy,
f870fa0b 3639 .sendmsg = mptcp_sendmsg,
644807e3 3640 .ioctl = mptcp_ioctl,
f870fa0b 3641 .recvmsg = mptcp_recvmsg,
14c441b5 3642 .release_cb = mptcp_release_cb,
2c5ebd00
PA		 3643 .hash = mptcp_hash,
3644 .unhash = mptcp_unhash,
cec37a6e 3645 .get_port = mptcp_get_port,
6511882c 3646 .forward_alloc_get = mptcp_forward_alloc_get,
d027236c 3647 .sockets_allocated = &mptcp_sockets_allocated,
0defbb0a 3648
d027236c 3649 .memory_allocated = &tcp_memory_allocated,
0defbb0a
ED		 3650 .per_cpu_fw_alloc = &tcp_memory_per_cpu_fw_alloc,
3651
d027236c 3652 .memory_pressure = &tcp_memory_pressure,
d027236c 3653 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_tcp_wmem),
989ef49b 3654 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_tcp_rmem),
d027236c 3655 .sysctl_mem = sysctl_tcp_mem,
f870fa0b 3656 .obj_size = sizeof(struct mptcp_sock),
2c5ebd00 3657 .slab_flags = SLAB_TYPESAFE_BY_RCU,
f870fa0b
MM		 3658 .no_autobind = true,
3659};
3660
2303f994
PK		 3661static int mptcp_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3662{
3663 struct mptcp_sock *msk = mptcp_sk(sock->sk);
3664 struct socket *ssock;
cf7da0d6 3665 int err;
2303f994
PK		 3666
3667 lock_sock(sock->sk);
fa68018d 3668 ssock = __mptcp_nmpc_socket(msk);
ddb1a072
PA		 3669 if (IS_ERR(ssock)) {
3670 err = PTR_ERR(ssock);
2303f994
PK		 3671 goto unlock;
3672 }
3673
3674 err = ssock->ops->bind(ssock, uaddr, addr_len);
cf7da0d6
PK		 3675 if (!err)
3676 mptcp_copy_inaddrs(sock->sk, ssock->sk);
2303f994
PK		 3677
3678unlock:
3679 release_sock(sock->sk);
3680 return err;
3681}
3682
cf7da0d6
PK		 3683static int mptcp_listen(struct socket *sock, int backlog)
3684{
3685 struct mptcp_sock *msk = mptcp_sk(sock->sk);
cfdcfeed 3686 struct sock *sk = sock->sk;
cf7da0d6
PK		 3687 struct socket *ssock;
3688 int err;
3689
3690 pr_debug("msk=%p", msk);
3691
cfdcfeed 3692 lock_sock(sk);
fa68018d 3693 ssock = __mptcp_nmpc_socket(msk);
ddb1a072
PA		 3694 if (IS_ERR(ssock)) {
3695 err = PTR_ERR(ssock);
cf7da0d6
PK		 3696 goto unlock;
3697 }
3698
fa68018d 3699 mptcp_token_destroy(msk);
cfdcfeed
MD		 3700 inet_sk_state_store(sk, TCP_LISTEN);
3701 sock_set_flag(sk, SOCK_RCU_FREE);
5e20087d 3702
cf7da0d6 3703 err = ssock->ops->listen(ssock, backlog);
cfdcfeed 3704 inet_sk_state_store(sk, inet_sk_state_load(ssock->sk));
c558246e
MD		 3705 if (!err) {
3706 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
cfdcfeed 3707 mptcp_copy_inaddrs(sk, ssock->sk);
c558246e 3708 }
cf7da0d6 3709
f8c9dfbd
GT		 3710 mptcp_event_pm_listener(ssock->sk, MPTCP_EVENT_LISTENER_CREATED);
3711
cf7da0d6 3712unlock:
cfdcfeed 3713 release_sock(sk);
cf7da0d6
PK		 3714 return err;
3715}
3716
cf7da0d6
PK		 3717static int mptcp_stream_accept(struct socket *sock, struct socket *newsock,
3718 int flags, bool kern)
3719{
3720 struct mptcp_sock *msk = mptcp_sk(sock->sk);
3721 struct socket *ssock;
3722 int err;
3723
3724 pr_debug("msk=%p", msk);
3725
5b825727 3726 /* Buggy applications can call accept on socket states other then LISTEN
61761231
PA		 3727 * but no need to allocate the first subflow just to error out.
3728 */
5b825727 3729 ssock = READ_ONCE(msk->subflow);
cf7da0d6 3730 if (!ssock)
71ba088c 3731 return -EINVAL;
cf7da0d6
PK		 3732
3733 err = ssock->ops->accept(sock, newsock, flags, kern);
d2f77c53 3734 if (err == 0 && !mptcp_is_tcpsk(newsock->sk)) {
cf7da0d6
PK		 3735 struct mptcp_sock *msk = mptcp_sk(newsock->sk);
3736 struct mptcp_subflow_context *subflow;
0397c6d8 3737 struct sock *newsk = newsock->sk;
0397c6d8 3738
a5ef058d 3739 set_bit(SOCK_CUSTOM_SOCKOPT, &newsock->flags);
b6985b9b 3740 msk->in_accept_queue = 0;
a5ef058d 3741
4d54cc32 3742 lock_sock(newsk);
5b950ff4 3743
cf7da0d6
PK		 3744 /* set ssk->sk_socket of accept()ed flows to mptcp socket.
3745 * This is needed so NOSPACE flag can be set from tcp stack.
3746 */
190f8b06 3747 mptcp_for_each_subflow(msk, subflow) {
cf7da0d6
PK		 3748 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
3749
3750 if (!ssk->sk_socket)
3751 mptcp_sock_graft(ssk, newsock);
3752 }
63740448
PA		 3753
3754 /* Do late cleanup for the first subflow as necessary. Also
3755 * deal with bad peers not doing a complete shutdown.
3756 */
3757 if (msk->first &&
3758 unlikely(inet_sk_state_load(msk->first) == TCP_CLOSE)) {
3759 __mptcp_close_ssk(newsk, msk->first,
3760 mptcp_subflow_ctx(msk->first), 0);
3761 if (unlikely(list_empty(&msk->conn_list)))
3762 inet_sk_state_store(newsk, TCP_CLOSE);
3763 }
3764
4d54cc32 3765 release_sock(newsk);
cf7da0d6
PK		 3766 }
3767
cf7da0d6 3768 return err;
cf7da0d6
PK		 3769}
3770
8edf0864
FW		 3771static __poll_t mptcp_check_writeable(struct mptcp_sock *msk)
3772{
3773 struct sock *sk = (struct sock *)msk;
8edf0864 3774
8edf0864
FW		 3775 if (sk_stream_is_writeable(sk))
3776 return EPOLLOUT | EPOLLWRNORM;
3777
5cf92bba 3778 mptcp_set_nospace(sk);
6e628cd3
PA		 3779 smp_mb__after_atomic(); /* msk->flags is changed by write_space cb */
3780 if (sk_stream_is_writeable(sk))
3781 return EPOLLOUT | EPOLLWRNORM;
8edf0864 3782
6e628cd3 3783 return 0;
8edf0864
FW		 3784}
3785
2303f994
PK		 3786static __poll_t mptcp_poll(struct file *file, struct socket *sock,
3787 struct poll_table_struct *wait)
3788{
1891c4a0 3789 struct sock *sk = sock->sk;
8ab183de 3790 struct mptcp_sock *msk;
2303f994 3791 __poll_t mask = 0;
6b9831bf 3792 u8 shutdown;
8a05661b 3793 int state;
2303f994 3794
1891c4a0 3795 msk = mptcp_sk(sk);
1891c4a0 3796 sock_poll_wait(file, sock, wait);
1891c4a0 3797
8a05661b 3798 state = inet_sk_state_load(sk);
6719331c 3799 pr_debug("msk=%p state=%d flags=%lx", msk, state, msk->flags);
71ba088c 3800 if (state == TCP_LISTEN) {
5b825727
PA		 3801 struct socket *ssock = READ_ONCE(msk->subflow);
3802
3803 if (WARN_ON_ONCE(!ssock || !ssock->sk))
71ba088c
PA		 3804 return 0;
3805
5b825727 3806 return inet_csk_listen_poll(ssock->sk);
71ba088c 3807 }
8a05661b 3808
6b9831bf
PA		 3809 shutdown = READ_ONCE(sk->sk_shutdown);
3810 if (shutdown == SHUTDOWN_MASK || state == TCP_CLOSE)
3811 mask |= EPOLLHUP;
3812 if (shutdown & RCV_SHUTDOWN)
3813 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
3814
8a05661b
PA		 3815 if (state != TCP_SYN_SENT && state != TCP_SYN_RECV) {
3816 mask |= mptcp_check_readable(msk);
6b9831bf
PA		 3817 if (shutdown & SEND_SHUTDOWN)
3818 mask |= EPOLLOUT | EPOLLWRNORM;
3819 else
3820 mask |= mptcp_check_writeable(msk);
a42cf9d1
BH		 3821 } else if (state == TCP_SYN_SENT && inet_sk(sk)->defer_connect) {
3822 /* cf tcp_poll() note about TFO */
3823 mask |= EPOLLOUT | EPOLLWRNORM;
8a05661b 3824 }
1891c4a0 3825
d89e3ed7 3826 /* This barrier is coupled with smp_wmb() in __mptcp_error_report() */
15cc1045 3827 smp_rmb();
9ae8e5ad 3828 if (READ_ONCE(sk->sk_err))
15cc1045
PA		 3829 mask |= EPOLLERR;
3830
2303f994
PK		 3831 return mask;
3832}
3833
e42f1ac6
FW		 3834static const struct proto_ops mptcp_stream_ops = {
3835 .family = PF_INET,
3836 .owner = THIS_MODULE,
0a3cc579 3837 .release = inet_release,
e42f1ac6 3838 .bind = mptcp_bind,
786fc124 3839 .connect = inet_stream_connect,
e42f1ac6
FW		 3840 .socketpair = sock_no_socketpair,
3841 .accept = mptcp_stream_accept,
d2f77c53 3842 .getname = inet_getname,
e42f1ac6
FW		 3843 .poll = mptcp_poll,
3844 .ioctl = inet_ioctl,
3845 .gettstamp = sock_gettstamp,
3846 .listen = mptcp_listen,
76e2a55d 3847 .shutdown = inet_shutdown,
e42f1ac6
FW		 3848 .setsockopt = sock_common_setsockopt,
3849 .getsockopt = sock_common_getsockopt,
3850 .sendmsg = inet_sendmsg,
3851 .recvmsg = inet_recvmsg,
3852 .mmap = sock_no_mmap,
3853 .sendpage = inet_sendpage,
e42f1ac6 3854};
2303f994 3855
f870fa0b
MM		 3856static struct inet_protosw mptcp_protosw = {
3857 .type = SOCK_STREAM,
3858 .protocol = IPPROTO_MPTCP,
3859 .prot = &mptcp_prot,
2303f994
PK		 3860 .ops = &mptcp_stream_ops,
3861 .flags = INET_PROTOSW_ICSK,
f870fa0b
MM		 3862};
3863
b19bc294
PA		 3864static int mptcp_napi_poll(struct napi_struct *napi, int budget)
3865{
3866 struct mptcp_delegated_action *delegated;
3867 struct mptcp_subflow_context *subflow;
3868 int work_done = 0;
3869
3870 delegated = container_of(napi, struct mptcp_delegated_action, napi);
3871 while ((subflow = mptcp_subflow_delegated_next(delegated)) != NULL) {
3872 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
3873
3874 bh_lock_sock_nested(ssk);
3875 if (!sock_owned_by_user(ssk) &&
3876 mptcp_subflow_has_delegated_action(subflow))
3877 mptcp_subflow_process_delegated(ssk);
3878 /* ... elsewhere tcp_release_cb_override already processed
3879 * the action or will do at next release_sock().
3880 * In both case must dequeue the subflow here - on the same
3881 * CPU that scheduled it.
3882 */
3883 bh_unlock_sock(ssk);
3884 sock_put(ssk);
3885
3886 if (++work_done == budget)
3887 return budget;
3888 }
3889
3890 /* always provide a 0 'work_done' argument, so that napi_complete_done
3891 * will not try accessing the NULL napi->dev ptr
3892 */
3893 napi_complete_done(napi, 0);
3894 return work_done;
3895}
3896
d39dceca 3897void __init mptcp_proto_init(void)
f870fa0b 3898{
b19bc294
PA		 3899 struct mptcp_delegated_action *delegated;
3900 int cpu;
3901
2303f994 3902 mptcp_prot.h.hashinfo = tcp_prot.h.hashinfo;
2303f994 3903
d027236c
PA		 3904 if (percpu_counter_init(&mptcp_sockets_allocated, 0, GFP_KERNEL))
3905 panic("Failed to allocate MPTCP pcpu counter\n");
3906
b19bc294
PA		 3907 init_dummy_netdev(&mptcp_napi_dev);
3908 for_each_possible_cpu(cpu) {
3909 delegated = per_cpu_ptr(&mptcp_delegated_actions, cpu);
3910 INIT_LIST_HEAD(&delegated->head);
16d083e2
JK		 3911 netif_napi_add_tx(&mptcp_napi_dev, &delegated->napi,
3912 mptcp_napi_poll);
b19bc294
PA		 3913 napi_enable(&delegated->napi);
3914 }
3915
2303f994 3916 mptcp_subflow_init();
1b1c7a0e 3917 mptcp_pm_init();
2c5ebd00 3918 mptcp_token_init();
2303f994 3919
f870fa0b
MM		 3920 if (proto_register(&mptcp_prot, 1) != 0)
3921 panic("Failed to register MPTCP proto.\n");
3922
3923 inet_register_protosw(&mptcp_protosw);
6771bfd9
FW		 3924
3925 BUILD_BUG_ON(sizeof(struct mptcp_skb_cb) > sizeof_field(struct sk_buff, cb));
f870fa0b
MM		 3926}
3927
3928#if IS_ENABLED(CONFIG_MPTCP_IPV6)
e42f1ac6
FW		 3929static const struct proto_ops mptcp_v6_stream_ops = {
3930 .family = PF_INET6,
3931 .owner = THIS_MODULE,
0a3cc579 3932 .release = inet6_release,
e42f1ac6 3933 .bind = mptcp_bind,
786fc124 3934 .connect = inet_stream_connect,
e42f1ac6
FW		 3935 .socketpair = sock_no_socketpair,
3936 .accept = mptcp_stream_accept,
d2f77c53 3937 .getname = inet6_getname,
e42f1ac6
FW		 3938 .poll = mptcp_poll,
3939 .ioctl = inet6_ioctl,
3940 .gettstamp = sock_gettstamp,
3941 .listen = mptcp_listen,
76e2a55d 3942 .shutdown = inet_shutdown,
e42f1ac6
FW		 3943 .setsockopt = sock_common_setsockopt,
3944 .getsockopt = sock_common_getsockopt,
3945 .sendmsg = inet6_sendmsg,
3946 .recvmsg = inet6_recvmsg,
3947 .mmap = sock_no_mmap,
3948 .sendpage = inet_sendpage,
3949#ifdef CONFIG_COMPAT
3986912f 3950 .compat_ioctl = inet6_compat_ioctl,
e42f1ac6
FW		 3951#endif
3952};
3953
f870fa0b
MM		 3954static struct proto mptcp_v6_prot;
3955
3956static struct inet_protosw mptcp_v6_protosw = {
3957 .type = SOCK_STREAM,
3958 .protocol = IPPROTO_MPTCP,
3959 .prot = &mptcp_v6_prot,
2303f994 3960 .ops = &mptcp_v6_stream_ops,
f870fa0b
MM		 3961 .flags = INET_PROTOSW_ICSK,
3962};
3963
d39dceca 3964int __init mptcp_proto_v6_init(void)
f870fa0b
MM		 3965{
3966 int err;
3967
3968 mptcp_v6_prot = mptcp_prot;
3969 strcpy(mptcp_v6_prot.name, "MPTCPv6");
3970 mptcp_v6_prot.slab = NULL;
b0519de8 3971 mptcp_v6_prot.obj_size = sizeof(struct mptcp6_sock);
f870fa0b
MM		 3972
3973 err = proto_register(&mptcp_v6_prot, 1);
3974 if (err)
3975 return err;
3976
3977 err = inet6_register_protosw(&mptcp_v6_protosw);
3978 if (err)
3979 proto_unregister(&mptcp_v6_prot);
3980
3981 return err;
3982}
3983#endif
Linux 6.x block layer and io_uring tree(s)