projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
mptcp: cleanup accept and poll
[linux-block.git] / net / mptcp / protocol.c
CommitLineData
f870fa0b
MM		 1// SPDX-License-Identifier: GPL-2.0
2/* Multipath TCP
3 *
4 * Copyright (c) 2017 - 2019, Intel Corporation.
5 */
6
7#define pr_fmt(fmt) "MPTCP: " fmt
8
9#include <linux/kernel.h>
10#include <linux/module.h>
11#include <linux/netdevice.h>
7a6a6cbc
PA		 12#include <linux/sched/signal.h>
13#include <linux/atomic.h>
f870fa0b
MM		 14#include <net/sock.h>
15#include <net/inet_common.h>
16#include <net/inet_hashtables.h>
17#include <net/protocol.h>
18#include <net/tcp.h>
3721b9b6 19#include <net/tcp_states.h>
cf7da0d6
PK		 20#if IS_ENABLED(CONFIG_MPTCP_IPV6)
21#include <net/transp_v6.h>
22#endif
f870fa0b 23#include <net/mptcp.h>
e16163b6 24#include <net/xfrm.h>
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
6771bfd9 39struct mptcp_skb_cb {
ab174ad8
PA		 40 u64 map_seq;
41 u64 end_seq;
6771bfd9 42 u32 offset;
b7f653b2 43 u8 has_rxtstamp:1;
6771bfd9
FW		 44};
45
46#define MPTCP_SKB_CB(__skb) ((struct mptcp_skb_cb *)&((__skb)->cb[0]))
47
b7f653b2
FW		 48enum {
49 MPTCP_CMSG_TS = BIT(0),
2c9e7765 50 MPTCP_CMSG_INQ = BIT(1),
b7f653b2
FW		 51};
52
91b6d325 53static struct percpu_counter mptcp_sockets_allocated ____cacheline_aligned_in_smp;
d027236c 54
e16163b6 55static void __mptcp_destroy_sock(struct sock *sk);
d9ca1de8 56static void __mptcp_check_send_data_fin(struct sock *sk);
e16163b6 57
b19bc294
PA		 58DEFINE_PER_CPU(struct mptcp_delegated_action, mptcp_delegated_actions);
59static struct net_device mptcp_napi_dev;
60
2303f994
PK		 61/* If msk has an initial subflow socket, and the MP_CAPABLE handshake has not
62 * completed yet or has failed, return the subflow socket.
63 * Otherwise return NULL.
64 */
1729cf18 65struct socket *__mptcp_nmpc_socket(const struct mptcp_sock *msk)
2303f994 66{
d22f4988 67 if (!msk->subflow || READ_ONCE(msk->can_ack))
2303f994
PK		 68 return NULL;
69
70 return msk->subflow;
71}
72
6f8a612a
FW		 73/* Returns end sequence number of the receiver's advertised window */
74static u64 mptcp_wnd_end(const struct mptcp_sock *msk)
75{
7439d687 76 return READ_ONCE(msk->wnd_end);
6f8a612a
FW		 77}
78
d2f77c53 79static bool mptcp_is_tcpsk(struct sock *sk)
0b4f33de
FW		 80{
81 struct socket *sock = sk->sk_socket;
82
0b4f33de
FW		 83 if (unlikely(sk->sk_prot == &tcp_prot)) {
84 /* we are being invoked after mptcp_accept() has
85 * accepted a non-mp-capable flow: sk is a tcp_sk,
86 * not an mptcp one.
87 *
88 * Hand the socket over to tcp so all further socket ops
89 * bypass mptcp.
90 */
91 sock->ops = &inet_stream_ops;
d2f77c53 92 return true;
0b4f33de
FW		 93#if IS_ENABLED(CONFIG_MPTCP_IPV6)
94 } else if (unlikely(sk->sk_prot == &tcpv6_prot)) {
95 sock->ops = &inet6_stream_ops;
d2f77c53 96 return true;
0b4f33de
FW		 97#endif
98 }
99
d2f77c53 100 return false;
0b4f33de
FW		 101}
102
fa68018d 103static int __mptcp_socket_create(struct mptcp_sock *msk)
2303f994
PK		 104{
105 struct mptcp_subflow_context *subflow;
106 struct sock *sk = (struct sock *)msk;
107 struct socket *ssock;
108 int err;
109
2303f994
PK		 110 err = mptcp_subflow_create_socket(sk, &ssock);
111 if (err)
fa68018d 112 return err;
2303f994 113
8ab183de 114 msk->first = ssock->sk;
2303f994
PK		 115 msk->subflow = ssock;
116 subflow = mptcp_subflow_ctx(ssock->sk);
cec37a6e 117 list_add(&subflow->node, &msk->conn_list);
e16163b6 118 sock_hold(ssock->sk);
2303f994 119 subflow->request_mptcp = 1;
866f26f2 120 mptcp_sock_graft(msk->first, sk->sk_socket);
e1ff9e82 121
fa68018d 122 return 0;
2303f994
PK		 123}
124
ab174ad8
PA		 125static void mptcp_drop(struct sock *sk, struct sk_buff *skb)
126{
127 sk_drops_add(sk, skb);
128 __kfree_skb(skb);
129}
130
6511882c
PA		 131static void mptcp_rmem_charge(struct sock *sk, int size)
132{
133 mptcp_sk(sk)->rmem_fwd_alloc -= size;
134}
135
8268ed4c
PA		 136static bool mptcp_try_coalesce(struct sock *sk, struct sk_buff *to,
137 struct sk_buff *from)
138{
139 bool fragstolen;
140 int delta;
141
142 if (MPTCP_SKB_CB(from)->offset ||
143 !skb_try_coalesce(to, from, &fragstolen, &delta))
144 return false;
145
06242e44
PA		 146 pr_debug("colesced seq %llx into %llx new len %d new end seq %llx",
147 MPTCP_SKB_CB(from)->map_seq, MPTCP_SKB_CB(to)->map_seq,
148 to->len, MPTCP_SKB_CB(from)->end_seq);
ab174ad8 149 MPTCP_SKB_CB(to)->end_seq = MPTCP_SKB_CB(from)->end_seq;
8268ed4c
PA		 150 kfree_skb_partial(from, fragstolen);
151 atomic_add(delta, &sk->sk_rmem_alloc);
6511882c 152 mptcp_rmem_charge(sk, delta);
8268ed4c
PA		 153 return true;
154}
155
ab174ad8
PA		 156static bool mptcp_ooo_try_coalesce(struct mptcp_sock *msk, struct sk_buff *to,
157 struct sk_buff *from)
158{
159 if (MPTCP_SKB_CB(from)->map_seq != MPTCP_SKB_CB(to)->end_seq)
160 return false;
161
162 return mptcp_try_coalesce((struct sock *)msk, to, from);
163}
164
6511882c
PA		 165static void __mptcp_rmem_reclaim(struct sock *sk, int amount)
166{
167 amount >>= SK_MEM_QUANTUM_SHIFT;
168 mptcp_sk(sk)->rmem_fwd_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
169 __sk_mem_reduce_allocated(sk, amount);
170}
171
172static void mptcp_rmem_uncharge(struct sock *sk, int size)
173{
174 struct mptcp_sock *msk = mptcp_sk(sk);
175 int reclaimable;
176
177 msk->rmem_fwd_alloc += size;
178 reclaimable = msk->rmem_fwd_alloc - sk_unused_reserved_mem(sk);
179
180 /* see sk_mem_uncharge() for the rationale behind the following schema */
181 if (unlikely(reclaimable >= SK_RECLAIM_THRESHOLD))
182 __mptcp_rmem_reclaim(sk, SK_RECLAIM_CHUNK);
183}
184
185static void mptcp_rfree(struct sk_buff *skb)
186{
187 unsigned int len = skb->truesize;
188 struct sock *sk = skb->sk;
189
190 atomic_sub(len, &sk->sk_rmem_alloc);
191 mptcp_rmem_uncharge(sk, len);
192}
193
194static void mptcp_set_owner_r(struct sk_buff *skb, struct sock *sk)
195{
196 skb_orphan(skb);
197 skb->sk = sk;
198 skb->destructor = mptcp_rfree;
199 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
200 mptcp_rmem_charge(sk, skb->truesize);
201}
202
ab174ad8
PA		 203/* "inspired" by tcp_data_queue_ofo(), main differences:
204 * - use mptcp seqs
205 * - don't cope with sacks
206 */
207static void mptcp_data_queue_ofo(struct mptcp_sock *msk, struct sk_buff *skb)
208{
209 struct sock *sk = (struct sock *)msk;
210 struct rb_node **p, *parent;
211 u64 seq, end_seq, max_seq;
212 struct sk_buff *skb1;
213
214 seq = MPTCP_SKB_CB(skb)->map_seq;
215 end_seq = MPTCP_SKB_CB(skb)->end_seq;
fa3fe2b1 216 max_seq = READ_ONCE(msk->rcv_wnd_sent);
ab174ad8 217
06242e44
PA		 218 pr_debug("msk=%p seq=%llx limit=%llx empty=%d", msk, seq, max_seq,
219 RB_EMPTY_ROOT(&msk->out_of_order_queue));
fa3fe2b1 220 if (after64(end_seq, max_seq)) {
ab174ad8
PA		 221 /* out of window */
222 mptcp_drop(sk, skb);
fa3fe2b1
FW		 223 pr_debug("oow by %lld, rcv_wnd_sent %llu\n",
224 (unsigned long long)end_seq - (unsigned long)max_seq,
225 (unsigned long long)msk->rcv_wnd_sent);
06242e44 226 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_NODSSWINDOW);
ab174ad8
PA		 227 return;
228 }
229
230 p = &msk->out_of_order_queue.rb_node;
06242e44 231 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOQUEUE);
ab174ad8
PA		 232 if (RB_EMPTY_ROOT(&msk->out_of_order_queue)) {
233 rb_link_node(&skb->rbnode, NULL, p);
234 rb_insert_color(&skb->rbnode, &msk->out_of_order_queue);
235 msk->ooo_last_skb = skb;
236 goto end;
237 }
238
239 /* with 2 subflows, adding at end of ooo queue is quite likely
240 * Use of ooo_last_skb avoids the O(Log(N)) rbtree lookup.
241 */
06242e44
PA		 242 if (mptcp_ooo_try_coalesce(msk, msk->ooo_last_skb, skb)) {
243 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOMERGE);
244 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOQUEUETAIL);
ab174ad8 245 return;
06242e44 246 }
ab174ad8
PA		 247
248 /* Can avoid an rbtree lookup if we are adding skb after ooo_last_skb */
249 if (!before64(seq, MPTCP_SKB_CB(msk->ooo_last_skb)->end_seq)) {
06242e44 250 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOQUEUETAIL);
ab174ad8
PA		 251 parent = &msk->ooo_last_skb->rbnode;
252 p = &parent->rb_right;
253 goto insert;
254 }
255
256 /* Find place to insert this segment. Handle overlaps on the way. */
257 parent = NULL;
258 while (*p) {
259 parent = *p;
260 skb1 = rb_to_skb(parent);
261 if (before64(seq, MPTCP_SKB_CB(skb1)->map_seq)) {
262 p = &parent->rb_left;
263 continue;
264 }
265 if (before64(seq, MPTCP_SKB_CB(skb1)->end_seq)) {
266 if (!after64(end_seq, MPTCP_SKB_CB(skb1)->end_seq)) {
267 /* All the bits are present. Drop. */
268 mptcp_drop(sk, skb);
06242e44 269 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
ab174ad8
PA		 270 return;
271 }
272 if (after64(seq, MPTCP_SKB_CB(skb1)->map_seq)) {
273 /* partial overlap:
274 * | skb |
275 * | skb1 |
276 * continue traversing
277 */
278 } else {
279 /* skb's seq == skb1's seq and skb covers skb1.
280 * Replace skb1 with skb.
281 */
282 rb_replace_node(&skb1->rbnode, &skb->rbnode,
283 &msk->out_of_order_queue);
284 mptcp_drop(sk, skb1);
06242e44 285 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
ab174ad8
PA		 286 goto merge_right;
287 }
288 } else if (mptcp_ooo_try_coalesce(msk, skb1, skb)) {
06242e44 289 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOMERGE);
ab174ad8
PA		 290 return;
291 }
292 p = &parent->rb_right;
293 }
06242e44 294
ab174ad8
PA		 295insert:
296 /* Insert segment into RB tree. */
297 rb_link_node(&skb->rbnode, parent, p);
298 rb_insert_color(&skb->rbnode, &msk->out_of_order_queue);
299
300merge_right:
301 /* Remove other segments covered by skb. */
302 while ((skb1 = skb_rb_next(skb)) != NULL) {
303 if (before64(end_seq, MPTCP_SKB_CB(skb1)->end_seq))
304 break;
305 rb_erase(&skb1->rbnode, &msk->out_of_order_queue);
306 mptcp_drop(sk, skb1);
06242e44 307 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
ab174ad8
PA		 308 }
309 /* If there is no skb after us, we are the last_skb ! */
310 if (!skb1)
311 msk->ooo_last_skb = skb;
312
313end:
314 skb_condense(skb);
6511882c
PA		 315 mptcp_set_owner_r(skb, sk);
316}
317
318static bool mptcp_rmem_schedule(struct sock *sk, struct sock *ssk, int size)
319{
320 struct mptcp_sock *msk = mptcp_sk(sk);
321 int amt, amount;
322
323 if (size < msk->rmem_fwd_alloc)
324 return true;
325
326 amt = sk_mem_pages(size);
327 amount = amt << SK_MEM_QUANTUM_SHIFT;
328 msk->rmem_fwd_alloc += amount;
329 if (!__sk_mem_raise_allocated(sk, size, amt, SK_MEM_RECV)) {
330 if (ssk->sk_forward_alloc < amount) {
331 msk->rmem_fwd_alloc -= amount;
332 return false;
333 }
334
335 ssk->sk_forward_alloc -= amount;
336 }
337 return true;
ab174ad8
PA		 338}
339
340static bool __mptcp_move_skb(struct mptcp_sock *msk, struct sock *ssk,
341 struct sk_buff *skb, unsigned int offset,
342 size_t copy_len)
6771bfd9 343{
ab174ad8 344 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
6771bfd9 345 struct sock *sk = (struct sock *)msk;
4e637c70 346 struct sk_buff *tail;
b7f653b2 347 bool has_rxtstamp;
6771bfd9
FW		 348
349 __skb_unlink(skb, &ssk->sk_receive_queue);
6771bfd9 350
4e637c70
FW		 351 skb_ext_reset(skb);
352 skb_orphan(skb);
ab174ad8 353
9c3f94e1 354 /* try to fetch required memory from subflow */
6511882c
PA		 355 if (!mptcp_rmem_schedule(sk, ssk, skb->truesize))
356 goto drop;
9c3f94e1 357
b7f653b2
FW		 358 has_rxtstamp = TCP_SKB_CB(skb)->has_rxtstamp;
359
ab174ad8
PA		 360 /* the skb map_seq accounts for the skb offset:
361 * mptcp_subflow_get_mapped_dsn() is based on the current tp->copied_seq
362 * value
363 */
364 MPTCP_SKB_CB(skb)->map_seq = mptcp_subflow_get_mapped_dsn(subflow);
365 MPTCP_SKB_CB(skb)->end_seq = MPTCP_SKB_CB(skb)->map_seq + copy_len;
8268ed4c 366 MPTCP_SKB_CB(skb)->offset = offset;
b7f653b2 367 MPTCP_SKB_CB(skb)->has_rxtstamp = has_rxtstamp;
4e637c70 368
ab174ad8
PA		 369 if (MPTCP_SKB_CB(skb)->map_seq == msk->ack_seq) {
370 /* in sequence */
8b0308fe 371 WRITE_ONCE(msk->ack_seq, msk->ack_seq + copy_len);
ab174ad8
PA		 372 tail = skb_peek_tail(&sk->sk_receive_queue);
373 if (tail && mptcp_try_coalesce(sk, tail, skb))
374 return true;
4e637c70 375
6511882c 376 mptcp_set_owner_r(skb, sk);
ab174ad8
PA		 377 __skb_queue_tail(&sk->sk_receive_queue, skb);
378 return true;
379 } else if (after64(MPTCP_SKB_CB(skb)->map_seq, msk->ack_seq)) {
380 mptcp_data_queue_ofo(msk, skb);
381 return false;
382 }
383
384 /* old data, keep it simple and drop the whole pkt, sender
385 * will retransmit as needed, if needed.
386 */
06242e44 387 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
9c3f94e1 388drop:
ab174ad8
PA		 389 mptcp_drop(sk, skb);
390 return false;
6771bfd9
FW		 391}
392
16a9a9da
MM		 393static void mptcp_stop_timer(struct sock *sk)
394{
395 struct inet_connection_sock *icsk = inet_csk(sk);
396
397 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
398 mptcp_sk(sk)->timer_ival = 0;
399}
400
e16163b6
PA		 401static void mptcp_close_wake_up(struct sock *sk)
402{
403 if (sock_flag(sk, SOCK_DEAD))
404 return;
405
406 sk->sk_state_change(sk);
407 if (sk->sk_shutdown == SHUTDOWN_MASK ||
408 sk->sk_state == TCP_CLOSE)
409 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_HUP);
410 else
411 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
412}
413
6e628cd3 414static bool mptcp_pending_data_fin_ack(struct sock *sk)
16a9a9da
MM		 415{
416 struct mptcp_sock *msk = mptcp_sk(sk);
417
6e628cd3
PA		 418 return !__mptcp_check_fallback(msk) &&
419 ((1 << sk->sk_state) &
420 (TCPF_FIN_WAIT1 | TCPF_CLOSING | TCPF_LAST_ACK)) &&
421 msk->write_seq == READ_ONCE(msk->snd_una);
422}
423
424static void mptcp_check_data_fin_ack(struct sock *sk)
425{
426 struct mptcp_sock *msk = mptcp_sk(sk);
16a9a9da
MM		 427
428 /* Look for an acknowledged DATA_FIN */
6e628cd3 429 if (mptcp_pending_data_fin_ack(sk)) {
16a9a9da
MM		 430 WRITE_ONCE(msk->snd_data_fin_enable, 0);
431
432 switch (sk->sk_state) {
433 case TCP_FIN_WAIT1:
434 inet_sk_state_store(sk, TCP_FIN_WAIT2);
16a9a9da
MM		 435 break;
436 case TCP_CLOSING:
16a9a9da
MM		 437 case TCP_LAST_ACK:
438 inet_sk_state_store(sk, TCP_CLOSE);
16a9a9da
MM		 439 break;
440 }
441
e16163b6 442 mptcp_close_wake_up(sk);
16a9a9da
MM		 443 }
444}
445
3721b9b6
MM		 446static bool mptcp_pending_data_fin(struct sock *sk, u64 *seq)
447{
448 struct mptcp_sock *msk = mptcp_sk(sk);
449
450 if (READ_ONCE(msk->rcv_data_fin) &&
451 ((1 << sk->sk_state) &
452 (TCPF_ESTABLISHED | TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2))) {
453 u64 rcv_data_fin_seq = READ_ONCE(msk->rcv_data_fin_seq);
454
455 if (msk->ack_seq == rcv_data_fin_seq) {
456 if (seq)
457 *seq = rcv_data_fin_seq;
458
459 return true;
460 }
461 }
462
463 return false;
464}
465
6477dd39
MM		 466static void mptcp_set_datafin_timeout(const struct sock *sk)
467{
468 struct inet_connection_sock *icsk = inet_csk(sk);
469
470 mptcp_sk(sk)->timer_ival = min(TCP_RTO_MAX,
471 TCP_RTO_MIN << icsk->icsk_retransmits);
472}
473
33d41c9c 474static void __mptcp_set_timeout(struct sock *sk, long tout)
3721b9b6 475{
3721b9b6
MM		 476 mptcp_sk(sk)->timer_ival = tout > 0 ? tout : TCP_RTO_MIN;
477}
478
33d41c9c
PA		 479static long mptcp_timeout_from_subflow(const struct mptcp_subflow_context *subflow)
480{
481 const struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
482
71b7dec2
PA		 483 return inet_csk(ssk)->icsk_pending && !subflow->stale_count ?
484 inet_csk(ssk)->icsk_timeout - jiffies : 0;
33d41c9c
PA		 485}
486
487static void mptcp_set_timeout(struct sock *sk)
488{
489 struct mptcp_subflow_context *subflow;
490 long tout = 0;
491
492 mptcp_for_each_subflow(mptcp_sk(sk), subflow)
493 tout = max(tout, mptcp_timeout_from_subflow(subflow));
494 __mptcp_set_timeout(sk, tout);
495}
496
fd897679
PA		 497static bool tcp_can_send_ack(const struct sock *ssk)
498{
499 return !((1 << inet_sk_state_load(ssk)) &
20bc80b6 500 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_TIME_WAIT | TCPF_CLOSE | TCPF_LISTEN));
fd897679
PA		 501}
502
340fa666
MM		 503void mptcp_subflow_send_ack(struct sock *ssk)
504{
505 bool slow;
506
507 slow = lock_sock_fast(ssk);
508 if (tcp_can_send_ack(ssk))
509 tcp_send_ack(ssk);
510 unlock_sock_fast(ssk, slow);
511}
512
fd897679 513static void mptcp_send_ack(struct mptcp_sock *msk)
7ed90803
PA		 514{
515 struct mptcp_subflow_context *subflow;
516
340fa666
MM		 517 mptcp_for_each_subflow(msk, subflow)
518 mptcp_subflow_send_ack(mptcp_subflow_tcp_sock(subflow));
fd897679
PA		 519}
520
fde56eea 521static void mptcp_subflow_cleanup_rbuf(struct sock *ssk)
fd897679 522{
75e908c3 523 bool slow;
fd897679 524
75e908c3 525 slow = lock_sock_fast(ssk);
fde56eea 526 if (tcp_can_send_ack(ssk))
fd897679 527 tcp_cleanup_rbuf(ssk, 1);
75e908c3 528 unlock_sock_fast(ssk, slow);
fde56eea
PA		 529}
530
531static bool mptcp_subflow_could_cleanup(const struct sock *ssk, bool rx_empty)
532{
533 const struct inet_connection_sock *icsk = inet_csk(ssk);
c4512c63 534 u8 ack_pending = READ_ONCE(icsk->icsk_ack.pending);
fde56eea
PA		 535 const struct tcp_sock *tp = tcp_sk(ssk);
536
537 return (ack_pending & ICSK_ACK_SCHED) &&
538 ((READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->rcv_wup) >
539 READ_ONCE(icsk->icsk_ack.rcv_mss)) ||
540 (rx_empty && ack_pending &
541 (ICSK_ACK_PUSHED2 | ICSK_ACK_PUSHED)));
fd897679
PA		 542}
543
544static void mptcp_cleanup_rbuf(struct mptcp_sock *msk)
545{
e3859603 546 int old_space = READ_ONCE(msk->old_wspace);
fd897679 547 struct mptcp_subflow_context *subflow;
e3859603 548 struct sock *sk = (struct sock *)msk;
fde56eea
PA		 549 int space = __mptcp_space(sk);
550 bool cleanup, rx_empty;
e3859603 551
fde56eea 552 cleanup = (space > 0) && (space >= (old_space << 1));
ce599c51 553 rx_empty = !__mptcp_rmem(sk);
fd897679 554
fde56eea
PA		 555 mptcp_for_each_subflow(msk, subflow) {
556 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
fd897679 557
fde56eea
PA		 558 if (cleanup || mptcp_subflow_could_cleanup(ssk, rx_empty))
559 mptcp_subflow_cleanup_rbuf(ssk);
7ed90803
PA		 560 }
561}
562
563static bool mptcp_check_data_fin(struct sock *sk)
3721b9b6
MM		 564{
565 struct mptcp_sock *msk = mptcp_sk(sk);
566 u64 rcv_data_fin_seq;
7ed90803 567 bool ret = false;
3721b9b6 568
781bf13d 569 if (__mptcp_check_fallback(msk))
7ed90803 570 return ret;
3721b9b6
MM		 571
572 /* Need to ack a DATA_FIN received from a peer while this side
573 * of the connection is in ESTABLISHED, FIN_WAIT1, or FIN_WAIT2.
574 * msk->rcv_data_fin was set when parsing the incoming options
575 * at the subflow level and the msk lock was not held, so this
576 * is the first opportunity to act on the DATA_FIN and change
577 * the msk state.
578 *
579 * If we are caught up to the sequence number of the incoming
580 * DATA_FIN, send the DATA_ACK now and do state transition. If
581 * not caught up, do nothing and let the recv code send DATA_ACK
582 * when catching up.
583 */
584
585 if (mptcp_pending_data_fin(sk, &rcv_data_fin_seq)) {
917944da 586 WRITE_ONCE(msk->ack_seq, msk->ack_seq + 1);
3721b9b6
MM		 587 WRITE_ONCE(msk->rcv_data_fin, 0);
588
589 sk->sk_shutdown |= RCV_SHUTDOWN;
16a9a9da 590 smp_mb__before_atomic(); /* SHUTDOWN must be visible first */
3721b9b6
MM		 591
592 switch (sk->sk_state) {
593 case TCP_ESTABLISHED:
594 inet_sk_state_store(sk, TCP_CLOSE_WAIT);
595 break;
596 case TCP_FIN_WAIT1:
597 inet_sk_state_store(sk, TCP_CLOSING);
598 break;
599 case TCP_FIN_WAIT2:
600 inet_sk_state_store(sk, TCP_CLOSE);
3721b9b6
MM		 601 break;
602 default:
603 /* Other states not expected */
604 WARN_ON_ONCE(1);
605 break;
606 }
607
7ed90803 608 ret = true;
fd897679 609 mptcp_send_ack(msk);
e16163b6 610 mptcp_close_wake_up(sk);
3721b9b6 611 }
7ed90803 612 return ret;
3721b9b6
MM		 613}
614
6771bfd9
FW		 615static bool __mptcp_move_skbs_from_subflow(struct mptcp_sock *msk,
616 struct sock *ssk,
617 unsigned int *bytes)
618{
619 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
600911ff 620 struct sock *sk = (struct sock *)msk;
6771bfd9
FW		 621 unsigned int moved = 0;
622 bool more_data_avail;
623 struct tcp_sock *tp;
624 bool done = false;
13c7ba0c
FW		 625 int sk_rbuf;
626
627 sk_rbuf = READ_ONCE(sk->sk_rcvbuf);
628
629 if (!(sk->sk_userlocks & SOCK_RCVBUF_LOCK)) {
630 int ssk_rbuf = READ_ONCE(ssk->sk_rcvbuf);
631
632 if (unlikely(ssk_rbuf > sk_rbuf)) {
633 WRITE_ONCE(sk->sk_rcvbuf, ssk_rbuf);
634 sk_rbuf = ssk_rbuf;
635 }
636 }
600911ff 637
ab174ad8 638 pr_debug("msk=%p ssk=%p", msk, ssk);
6771bfd9
FW		 639 tp = tcp_sk(ssk);
640 do {
641 u32 map_remaining, offset;
642 u32 seq = tp->copied_seq;
643 struct sk_buff *skb;
644 bool fin;
645
646 /* try to move as much data as available */
647 map_remaining = subflow->map_data_len -
648 mptcp_subflow_get_map_offset(subflow);
649
650 skb = skb_peek(&ssk->sk_receive_queue);
d9fb8c50
PA		 651 if (!skb) {
652 /* if no data is found, a racing workqueue/recvmsg
653 * already processed the new data, stop here or we
654 * can enter an infinite loop
655 */
656 if (!moved)
657 done = true;
6771bfd9 658 break;
d9fb8c50 659 }
6771bfd9 660
e1ff9e82
DC		 661 if (__mptcp_check_fallback(msk)) {
662 /* if we are running under the workqueue, TCP could have
663 * collapsed skbs between dummy map creation and now
664 * be sure to adjust the size
665 */
666 map_remaining = skb->len;
667 subflow->map_data_len = skb->len;
668 }
669
6771bfd9
FW		 670 offset = seq - TCP_SKB_CB(skb)->seq;
671 fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
672 if (fin) {
673 done = true;
674 seq++;
675 }
676
677 if (offset < skb->len) {
678 size_t len = skb->len - offset;
679
680 if (tp->urg_data)
681 done = true;
682
ab174ad8
PA		 683 if (__mptcp_move_skb(msk, ssk, skb, offset, len))
684 moved += len;
6771bfd9 685 seq += len;
6771bfd9
FW		 686
687 if (WARN_ON_ONCE(map_remaining < len))
688 break;
689 } else {
690 WARN_ON_ONCE(!fin);
691 sk_eat_skb(ssk, skb);
692 done = true;
693 }
694
695 WRITE_ONCE(tp->copied_seq, seq);
696 more_data_avail = mptcp_subflow_data_available(ssk);
600911ff 697
13c7ba0c 698 if (atomic_read(&sk->sk_rmem_alloc) > sk_rbuf) {
600911ff
FW		 699 done = true;
700 break;
701 }
6771bfd9
FW		 702 } while (more_data_avail);
703
6719331c 704 *bytes += moved;
6771bfd9
FW		 705 return done;
706}
707
87952603 708static bool __mptcp_ofo_queue(struct mptcp_sock *msk)
ab174ad8
PA		 709{
710 struct sock *sk = (struct sock *)msk;
711 struct sk_buff *skb, *tail;
712 bool moved = false;
713 struct rb_node *p;
714 u64 end_seq;
715
716 p = rb_first(&msk->out_of_order_queue);
06242e44 717 pr_debug("msk=%p empty=%d", msk, RB_EMPTY_ROOT(&msk->out_of_order_queue));
ab174ad8
PA		 718 while (p) {
719 skb = rb_to_skb(p);
720 if (after64(MPTCP_SKB_CB(skb)->map_seq, msk->ack_seq))
721 break;
722
723 p = rb_next(p);
724 rb_erase(&skb->rbnode, &msk->out_of_order_queue);
725
726 if (unlikely(!after64(MPTCP_SKB_CB(skb)->end_seq,
727 msk->ack_seq))) {
728 mptcp_drop(sk, skb);
06242e44 729 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
ab174ad8
PA		 730 continue;
731 }
732
733 end_seq = MPTCP_SKB_CB(skb)->end_seq;
734 tail = skb_peek_tail(&sk->sk_receive_queue);
735 if (!tail || !mptcp_ooo_try_coalesce(msk, tail, skb)) {
736 int delta = msk->ack_seq - MPTCP_SKB_CB(skb)->map_seq;
737
738 /* skip overlapping data, if any */
06242e44
PA		 739 pr_debug("uncoalesced seq=%llx ack seq=%llx delta=%d",
740 MPTCP_SKB_CB(skb)->map_seq, msk->ack_seq,
741 delta);
ab174ad8 742 MPTCP_SKB_CB(skb)->offset += delta;
2c9e7765 743 MPTCP_SKB_CB(skb)->map_seq += delta;
ab174ad8
PA		 744 __skb_queue_tail(&sk->sk_receive_queue, skb);
745 }
746 msk->ack_seq = end_seq;
747 moved = true;
748 }
749 return moved;
750}
751
2e52213c
FW		 752/* In most cases we will be able to lock the mptcp socket. If its already
753 * owned, we need to defer to the work queue to avoid ABBA deadlock.
754 */
99d1055c 755static bool move_skbs_to_msk(struct mptcp_sock *msk, struct sock *ssk)
2e52213c
FW		 756{
757 struct sock *sk = (struct sock *)msk;
758 unsigned int moved = 0;
759
87952603
PA		 760 __mptcp_move_skbs_from_subflow(msk, ssk, &moved);
761 __mptcp_ofo_queue(msk);
499ada50
PA		 762 if (unlikely(ssk->sk_err)) {
763 if (!sock_owned_by_user(sk))
764 __mptcp_error_report(sk);
765 else
766 set_bit(MPTCP_ERROR_REPORT, &msk->flags);
767 }
2e52213c 768
87952603
PA		 769 /* If the moves have caught up with the DATA_FIN sequence number
770 * it's time to ack the DATA_FIN and change socket state, but
771 * this is not a good place to change state. Let the workqueue
772 * do it.
773 */
774 if (mptcp_pending_data_fin(sk, NULL))
775 mptcp_schedule_work(sk);
99d1055c 776 return moved > 0;
2e52213c
FW		 777}
778
779void mptcp_data_ready(struct sock *sk, struct sock *ssk)
101f6f85 780{
6719331c 781 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
101f6f85 782 struct mptcp_sock *msk = mptcp_sk(sk);
13c7ba0c 783 int sk_rbuf, ssk_rbuf;
101f6f85 784
d7b1bfd0
PA		 785 /* The peer can send data while we are shutting down this
786 * subflow at msk destruction time, but we must avoid enqueuing
787 * more data to the msk receive queue
788 */
789 if (unlikely(subflow->disposable))
790 return;
791
13c7ba0c
FW		 792 ssk_rbuf = READ_ONCE(ssk->sk_rcvbuf);
793 sk_rbuf = READ_ONCE(sk->sk_rcvbuf);
794 if (unlikely(ssk_rbuf > sk_rbuf))
795 sk_rbuf = ssk_rbuf;
796
99d1055c 797 /* over limit? can't append more skbs to msk, Also, no need to wake-up*/
ce599c51
PA		 798 if (__mptcp_rmem(sk) > sk_rbuf) {
799 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_RCVPRUNED);
99d1055c 800 return;
ce599c51 801 }
600911ff 802
99d1055c
PA		 803 /* Wake-up the reader only for in-sequence data */
804 mptcp_data_lock(sk);
612f71d7 805 if (move_skbs_to_msk(msk, ssk))
6719331c 806 sk->sk_data_ready(sk);
612f71d7 807
99d1055c 808 mptcp_data_unlock(sk);
101f6f85
FW		 809}
810
78962489 811static bool mptcp_do_flush_join_list(struct mptcp_sock *msk)
ec3edaa7 812{
5cf92bba 813 struct mptcp_subflow_context *subflow;
df00b087 814 bool ret = false;
5cf92bba 815
ec3edaa7 816 if (likely(list_empty(&msk->join_list)))
78962489 817 return false;
ec3edaa7
PK		 818
819 spin_lock_bh(&msk->join_list_lock);
df00b087
FW		 820 list_for_each_entry(subflow, &msk->join_list, node) {
821 u32 sseq = READ_ONCE(subflow->setsockopt_seq);
78962489 822
df00b087
FW		 823 mptcp_propagate_sndbuf((struct sock *)msk, mptcp_subflow_tcp_sock(subflow));
824 if (READ_ONCE(msk->setsockopt_seq) != sseq)
825 ret = true;
826 }
ec3edaa7
PK		 827 list_splice_tail_init(&msk->join_list, &msk->conn_list);
828 spin_unlock_bh(&msk->join_list_lock);
78962489 829
df00b087 830 return ret;
78962489
FW		 831}
832
833void __mptcp_flush_join_list(struct mptcp_sock *msk)
834{
835 if (likely(!mptcp_do_flush_join_list(msk)))
836 return;
837
838 if (!test_and_set_bit(MPTCP_WORK_SYNC_SETSOCKOPT, &msk->flags))
839 mptcp_schedule_work((struct sock *)msk);
840}
841
842static void mptcp_flush_join_list(struct mptcp_sock *msk)
843{
844 bool sync_needed = test_and_clear_bit(MPTCP_WORK_SYNC_SETSOCKOPT, &msk->flags);
845
846 might_sleep();
847
848 if (!mptcp_do_flush_join_list(msk) && !sync_needed)
849 return;
850
851 mptcp_sockopt_sync_all(msk);
ec3edaa7
PK		 852}
853
b51f9b80
PA		 854static bool mptcp_timer_pending(struct sock *sk)
855{
856 return timer_pending(&inet_csk(sk)->icsk_retransmit_timer);
857}
858
859static void mptcp_reset_timer(struct sock *sk)
860{
861 struct inet_connection_sock *icsk = inet_csk(sk);
862 unsigned long tout;
863
e16163b6
PA		 864 /* prevent rescheduling on close */
865 if (unlikely(inet_sk_state_load(sk) == TCP_CLOSE))
866 return;
867
33d41c9c 868 tout = mptcp_sk(sk)->timer_ival;
b51f9b80
PA		 869 sk_reset_timer(sk, &icsk->icsk_retransmit_timer, jiffies + tout);
870}
871
ba8f48f7
PA		 872bool mptcp_schedule_work(struct sock *sk)
873{
874 if (inet_sk_state_load(sk) != TCP_CLOSE &&
875 schedule_work(&mptcp_sk(sk)->work)) {
876 /* each subflow already holds a reference to the sk, and the
877 * workqueue is invoked by a subflow, so sk can't go away here.
878 */
879 sock_hold(sk);
880 return true;
881 }
882 return false;
883}
884
59832e24
FW		 885void mptcp_subflow_eof(struct sock *sk)
886{
ba8f48f7
PA		 887 if (!test_and_set_bit(MPTCP_WORK_EOF, &mptcp_sk(sk)->flags))
888 mptcp_schedule_work(sk);
59832e24
FW		 889}
890
5969856a
PA		 891static void mptcp_check_for_eof(struct mptcp_sock *msk)
892{
893 struct mptcp_subflow_context *subflow;
894 struct sock *sk = (struct sock *)msk;
895 int receivers = 0;
896
897 mptcp_for_each_subflow(msk, subflow)
898 receivers += !subflow->rx_eof;
e16163b6
PA		 899 if (receivers)
900 return;
5969856a 901
e16163b6 902 if (!(sk->sk_shutdown & RCV_SHUTDOWN)) {
5969856a
PA		 903 /* hopefully temporary hack: propagate shutdown status
904 * to msk, when all subflows agree on it
905 */
906 sk->sk_shutdown |= RCV_SHUTDOWN;
907
908 smp_mb__before_atomic(); /* SHUTDOWN must be visible first */
5969856a
PA		 909 sk->sk_data_ready(sk);
910 }
e16163b6
PA		 911
912 switch (sk->sk_state) {
913 case TCP_ESTABLISHED:
914 inet_sk_state_store(sk, TCP_CLOSE_WAIT);
915 break;
916 case TCP_FIN_WAIT1:
26aa2314
PA		 917 inet_sk_state_store(sk, TCP_CLOSING);
918 break;
919 case TCP_FIN_WAIT2:
e16163b6
PA		 920 inet_sk_state_store(sk, TCP_CLOSE);
921 break;
922 default:
923 return;
924 }
925 mptcp_close_wake_up(sk);
5969856a
PA		 926}
927
7a6a6cbc
PA		 928static struct sock *mptcp_subflow_recv_lookup(const struct mptcp_sock *msk)
929{
930 struct mptcp_subflow_context *subflow;
931 struct sock *sk = (struct sock *)msk;
932
933 sock_owned_by_me(sk);
934
935 mptcp_for_each_subflow(msk, subflow) {
99d1055c 936 if (READ_ONCE(subflow->data_avail))
7a6a6cbc
PA		 937 return mptcp_subflow_tcp_sock(subflow);
938 }
939
940 return NULL;
941}
942
3f8e0aae
PA		 943static bool mptcp_skb_can_collapse_to(u64 write_seq,
944 const struct sk_buff *skb,
945 const struct mptcp_ext *mpext)
57040755
PA		 946{
947 if (!tcp_skb_can_collapse_to(skb))
948 return false;
949
5a369ca6
PA		 950 /* can collapse only if MPTCP level sequence is in order and this
951 * mapping has not been xmitted yet
952 */
953 return mpext && mpext->data_seq + mpext->data_len == write_seq &&
954 !mpext->frozen;
57040755
PA		 955}
956
29249eac
PA		 957/* we can append data to the given data frag if:
958 * - there is space available in the backing page_frag
959 * - the data frag tail matches the current page_frag free offset
960 * - the data frag end sequence number matches the current write seq
961 */
18b683bf
PA		 962static bool mptcp_frag_can_collapse_to(const struct mptcp_sock *msk,
963 const struct page_frag *pfrag,
964 const struct mptcp_data_frag *df)
965{
966 return df && pfrag->page == df->page &&
d9ca1de8 967 pfrag->size - pfrag->offset > 0 &&
29249eac 968 pfrag->offset == (df->offset + df->data_len) &&
18b683bf
PA		 969 df->data_seq + df->data_len == msk->write_seq;
970}
971
6511882c 972static void __mptcp_mem_reclaim_partial(struct sock *sk)
e93da928 973{
6511882c 974 int reclaimable = mptcp_sk(sk)->rmem_fwd_alloc - sk_unused_reserved_mem(sk);
e93da928 975
765ff425 976 lockdep_assert_held_once(&sk->sk_lock.slock);
b5941f06 977
6511882c 978 __mptcp_rmem_reclaim(sk, reclaimable - 1);
1094c6fe
PA		 979 sk_mem_reclaim_partial(sk);
980}
981
724cfd2e
PA		 982static void mptcp_mem_reclaim_partial(struct sock *sk)
983{
724cfd2e 984 mptcp_data_lock(sk);
6511882c 985 __mptcp_mem_reclaim_partial(sk);
724cfd2e
PA		 986 mptcp_data_unlock(sk);
987}
988
d027236c
PA		 989static void dfrag_uncharge(struct sock *sk, int len)
990{
991 sk_mem_uncharge(sk, len);
7948f6cc 992 sk_wmem_queued_add(sk, -len);
d027236c
PA		 993}
994
995static void dfrag_clear(struct sock *sk, struct mptcp_data_frag *dfrag)
18b683bf 996{
d027236c
PA		 997 int len = dfrag->data_len + dfrag->overhead;
998
18b683bf 999 list_del(&dfrag->list);
d027236c 1000 dfrag_uncharge(sk, len);
18b683bf
PA		 1001 put_page(dfrag->page);
1002}
1003
6e628cd3 1004static void __mptcp_clean_una(struct sock *sk)
18b683bf
PA		 1005{
1006 struct mptcp_sock *msk = mptcp_sk(sk);
1007 struct mptcp_data_frag *dtmp, *dfrag;
d027236c 1008 bool cleaned = false;
e1ff9e82
DC		 1009 u64 snd_una;
1010
1011 /* on fallback we just need to ignore snd_una, as this is really
1012 * plain TCP
1013 */
1014 if (__mptcp_check_fallback(msk))
7439d687 1015 msk->snd_una = READ_ONCE(msk->snd_nxt);
6f8a612a 1016
7439d687 1017 snd_una = msk->snd_una;
18b683bf
PA		 1018 list_for_each_entry_safe(dfrag, dtmp, &msk->rtx_queue, list) {
1019 if (after64(dfrag->data_seq + dfrag->data_len, snd_una))
1020 break;
1021
1e1d9d6f
PA		 1022 if (unlikely(dfrag == msk->first_pending)) {
1023 /* in recovery mode can see ack after the current snd head */
1024 if (WARN_ON_ONCE(!msk->recovery))
1025 break;
1026
1027 WRITE_ONCE(msk->first_pending, mptcp_send_next(sk));
1028 }
1029
d027236c
PA		 1030 dfrag_clear(sk, dfrag);
1031 cleaned = true;
1032 }
1033
7948f6cc
FW		 1034 dfrag = mptcp_rtx_head(sk);
1035 if (dfrag && after64(snd_una, dfrag->data_seq)) {
53eb4c38
PA		 1036 u64 delta = snd_una - dfrag->data_seq;
1037
1e1d9d6f
PA		 1038 /* prevent wrap around in recovery mode */
1039 if (unlikely(delta > dfrag->already_sent)) {
1040 if (WARN_ON_ONCE(!msk->recovery))
1041 goto out;
1042 if (WARN_ON_ONCE(delta > dfrag->data_len))
1043 goto out;
1044 dfrag->already_sent += delta - dfrag->already_sent;
1045 }
7948f6cc
FW		 1046
1047 dfrag->data_seq += delta;
53eb4c38 1048 dfrag->offset += delta;
7948f6cc 1049 dfrag->data_len -= delta;
d9ca1de8 1050 dfrag->already_sent -= delta;
7948f6cc
FW		 1051
1052 dfrag_uncharge(sk, delta);
1053 cleaned = true;
1054 }
1055
1e1d9d6f
PA		 1056 /* all retransmitted data acked, recovery completed */
1057 if (unlikely(msk->recovery) && after64(msk->snd_una, msk->recovery_snd_nxt))
1058 msk->recovery = false;
1059
53eb4c38 1060out:
1094c6fe
PA		 1061 if (cleaned && tcp_under_memory_pressure(sk))
1062 __mptcp_mem_reclaim_partial(sk);
95ed690e 1063
3241a9c0
FW		 1064 if (snd_una == READ_ONCE(msk->snd_nxt) &&
1065 snd_una == READ_ONCE(msk->write_seq)) {
33d41c9c 1066 if (mptcp_timer_pending(sk) && !mptcp_data_fin_enabled(msk))
6e628cd3
PA		 1067 mptcp_stop_timer(sk);
1068 } else {
1069 mptcp_reset_timer(sk);
18b683bf
PA		 1070 }
1071}
1072
417789df
PA		 1073static void __mptcp_clean_una_wakeup(struct sock *sk)
1074{
765ff425
PA		 1075 lockdep_assert_held_once(&sk->sk_lock.slock);
1076
417789df
PA		 1077 __mptcp_clean_una(sk);
1078 mptcp_write_space(sk);
1079}
1080
b5941f06
PA		 1081static void mptcp_clean_una_wakeup(struct sock *sk)
1082{
1083 mptcp_data_lock(sk);
1084 __mptcp_clean_una_wakeup(sk);
1085 mptcp_data_unlock(sk);
1086}
1087
724cfd2e 1088static void mptcp_enter_memory_pressure(struct sock *sk)
18b683bf 1089{
d9ca1de8
PA		 1090 struct mptcp_subflow_context *subflow;
1091 struct mptcp_sock *msk = mptcp_sk(sk);
1092 bool first = true;
1093
18b683bf 1094 sk_stream_moderate_sndbuf(sk);
d9ca1de8
PA		 1095 mptcp_for_each_subflow(msk, subflow) {
1096 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1097
1098 if (first)
1099 tcp_enter_memory_pressure(ssk);
1100 sk_stream_moderate_sndbuf(ssk);
1101 first = false;
1102 }
724cfd2e
PA		 1103}
1104
1105/* ensure we get enough memory for the frag hdr, beyond some minimal amount of
1106 * data
1107 */
1108static bool mptcp_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1109{
1110 if (likely(skb_page_frag_refill(32U + sizeof(struct mptcp_data_frag),
1111 pfrag, sk->sk_allocation)))
1112 return true;
1113
1114 mptcp_enter_memory_pressure(sk);
18b683bf
PA		 1115 return false;
1116}
1117
1118static struct mptcp_data_frag *
1119mptcp_carve_data_frag(const struct mptcp_sock *msk, struct page_frag *pfrag,
1120 int orig_offset)
1121{
1122 int offset = ALIGN(orig_offset, sizeof(long));
1123 struct mptcp_data_frag *dfrag;
1124
1125 dfrag = (struct mptcp_data_frag *)(page_to_virt(pfrag->page) + offset);
1126 dfrag->data_len = 0;
1127 dfrag->data_seq = msk->write_seq;
1128 dfrag->overhead = offset - orig_offset + sizeof(struct mptcp_data_frag);
1129 dfrag->offset = offset + sizeof(struct mptcp_data_frag);
d9ca1de8 1130 dfrag->already_sent = 0;
18b683bf
PA		 1131 dfrag->page = pfrag->page;
1132
1133 return dfrag;
1134}
1135
caf971df
PA		 1136struct mptcp_sendmsg_info {
1137 int mss_now;
1138 int size_goal;
d9ca1de8
PA		 1139 u16 limit;
1140 u16 sent;
1141 unsigned int flags;
1094c6fe 1142 bool data_lock_held;
caf971df
PA		 1143};
1144
6f8a612a
FW		 1145static int mptcp_check_allowed_size(struct mptcp_sock *msk, u64 data_seq,
1146 int avail_size)
1147{
1148 u64 window_end = mptcp_wnd_end(msk);
1149
1150 if (__mptcp_check_fallback(msk))
1151 return avail_size;
1152
1153 if (!before64(data_seq + avail_size, window_end)) {
1154 u64 allowed_size = window_end - data_seq;
1155
1156 return min_t(unsigned int, allowed_size, avail_size);
1157 }
1158
1159 return avail_size;
1160}
1161
724cfd2e
PA		 1162static bool __mptcp_add_ext(struct sk_buff *skb, gfp_t gfp)
1163{
1164 struct skb_ext *mpext = __skb_ext_alloc(gfp);
1165
1166 if (!mpext)
1167 return false;
1168 __skb_ext_set(skb, SKB_EXT_MPTCP, mpext);
1169 return true;
1170}
1171
6e628cd3 1172static struct sk_buff *__mptcp_do_alloc_tx_skb(struct sock *sk, gfp_t gfp)
724cfd2e
PA		 1173{
1174 struct sk_buff *skb;
1175
6e628cd3 1176 skb = alloc_skb_fclone(MAX_TCP_HEADER, gfp);
724cfd2e 1177 if (likely(skb)) {
6e628cd3 1178 if (likely(__mptcp_add_ext(skb, gfp))) {
724cfd2e 1179 skb_reserve(skb, MAX_TCP_HEADER);
a52fe46e 1180 skb->ip_summed = CHECKSUM_PARTIAL;
f70cad10 1181 INIT_LIST_HEAD(&skb->tcp_tsorted_anchor);
724cfd2e
PA		 1182 return skb;
1183 }
1184 __kfree_skb(skb);
1185 } else {
1186 mptcp_enter_memory_pressure(sk);
1187 }
1188 return NULL;
1189}
1190
f70cad10 1191static struct sk_buff *__mptcp_alloc_tx_skb(struct sock *sk, struct sock *ssk, gfp_t gfp)
724cfd2e 1192{
724cfd2e
PA		 1193 struct sk_buff *skb;
1194
6e628cd3 1195 skb = __mptcp_do_alloc_tx_skb(sk, gfp);
724cfd2e 1196 if (!skb)
f70cad10 1197 return NULL;
724cfd2e
PA		 1198
1199 if (likely(sk_wmem_schedule(ssk, skb->truesize))) {
f70cad10
PA		 1200 tcp_skb_entail(ssk, skb);
1201 return skb;
724cfd2e
PA		 1202 }
1203 kfree_skb(skb);
f70cad10 1204 return NULL;
724cfd2e
PA		 1205}
1206
f70cad10 1207static struct sk_buff *mptcp_alloc_tx_skb(struct sock *sk, struct sock *ssk, bool data_lock_held)
724cfd2e 1208{
1094c6fe 1209 gfp_t gfp = data_lock_held ? GFP_ATOMIC : sk->sk_allocation;
724cfd2e 1210
1094c6fe
PA		 1211 if (unlikely(tcp_under_memory_pressure(sk))) {
1212 if (data_lock_held)
1213 __mptcp_mem_reclaim_partial(sk);
1214 else
1215 mptcp_mem_reclaim_partial(sk);
1216 }
1217 return __mptcp_alloc_tx_skb(sk, ssk, gfp);
724cfd2e
PA		 1218}
1219
d0cc2987
GT		 1220/* note: this always recompute the csum on the whole skb, even
1221 * if we just appended a single frag. More status info needed
1222 */
1223static void mptcp_update_data_checksum(struct sk_buff *skb, int added)
1224{
1225 struct mptcp_ext *mpext = mptcp_get_ext(skb);
1226 __wsum csum = ~csum_unfold(mpext->csum);
1227 int offset = skb->len - added;
1228
1229 mpext->csum = csum_fold(csum_block_add(csum, skb_checksum(skb, offset, added, 0), offset));
1230}
1231
6d0060f6 1232static int mptcp_sendmsg_frag(struct sock *sk, struct sock *ssk,
d9ca1de8 1233 struct mptcp_data_frag *dfrag,
caf971df 1234 struct mptcp_sendmsg_info *info)
6d0060f6 1235{
d9ca1de8 1236 u64 data_seq = dfrag->data_seq + info->sent;
f70cad10 1237 int offset = dfrag->offset + info->sent;
6d0060f6 1238 struct mptcp_sock *msk = mptcp_sk(sk);
6f8a612a 1239 bool zero_window_probe = false;
6d0060f6 1240 struct mptcp_ext *mpext = NULL;
f70cad10
PA		 1241 bool can_coalesce = false;
1242 bool reuse_skb = true;
1243 struct sk_buff *skb;
1244 size_t copy;
1245 int i;
6d0060f6 1246
e4b61351 1247 pr_debug("msk=%p ssk=%p sending dfrag at seq=%llu len=%u already sent=%u",
d9ca1de8
PA		 1248 msk, ssk, dfrag->data_seq, dfrag->data_len, info->sent);
1249
f70cad10
PA		 1250 if (WARN_ON_ONCE(info->sent > info->limit ||
1251 info->limit > dfrag->data_len))
1252 return 0;
1253
d9ca1de8
PA		 1254 /* compute send limit */
1255 info->mss_now = tcp_send_mss(ssk, &info->size_goal, info->flags);
f70cad10
PA		 1256 copy = info->size_goal;
1257
57040755 1258 skb = tcp_write_queue_tail(ssk);
f70cad10 1259 if (skb && copy > skb->len) {
57040755
PA		 1260 /* Limit the write to the size available in the
1261 * current skb, if any, so that we create at most a new skb.
1262 * Explicitly tells TCP internals to avoid collapsing on later
1263 * queue management operation, to avoid breaking the ext <->
1264 * SSN association set here
1265 */
d9ca1de8 1266 mpext = skb_ext_find(skb, SKB_EXT_MPTCP);
1094c6fe 1267 if (!mptcp_skb_can_collapse_to(data_seq, skb, mpext)) {
57040755 1268 TCP_SKB_CB(skb)->eor = 1;
1094c6fe
PA		 1269 goto alloc_skb;
1270 }
1271
f70cad10
PA		 1272 i = skb_shinfo(skb)->nr_frags;
1273 can_coalesce = skb_can_coalesce(skb, i, dfrag->page, offset);
1274 if (!can_coalesce && i >= sysctl_max_skb_frags) {
1275 tcp_mark_push(tcp_sk(ssk), skb);
1276 goto alloc_skb;
15e6ca97 1277 }
18b683bf 1278
f70cad10
PA		 1279 copy -= skb->len;
1280 } else {
1094c6fe 1281alloc_skb:
f70cad10
PA		 1282 skb = mptcp_alloc_tx_skb(sk, ssk, info->data_lock_held);
1283 if (!skb)
1284 return -ENOMEM;
1285
1286 i = skb_shinfo(skb)->nr_frags;
1287 reuse_skb = false;
1288 mpext = skb_ext_find(skb, SKB_EXT_MPTCP);
1289 }
1094c6fe 1290
6f8a612a 1291 /* Zero window and all data acked? Probe. */
f70cad10
PA		 1292 copy = mptcp_check_allowed_size(msk, data_seq, copy);
1293 if (copy == 0) {
7439d687
PA		 1294 u64 snd_una = READ_ONCE(msk->snd_una);
1295
f70cad10 1296 if (snd_una != msk->snd_nxt) {
27728ba8 1297 tcp_remove_empty_skb(ssk);
6f8a612a 1298 return 0;
f70cad10
PA		 1299 }
1300
6f8a612a 1301 zero_window_probe = true;
7439d687 1302 data_seq = snd_una - 1;
f70cad10 1303 copy = 1;
6f8a612a 1304
f70cad10
PA		 1305 /* all mptcp-level data is acked, no skbs should be present into the
1306 * ssk write queue
1307 */
1308 WARN_ON_ONCE(reuse_skb);
1309 }
d027236c 1310
f70cad10
PA		 1311 copy = min_t(size_t, copy, info->limit - info->sent);
1312 if (!sk_wmem_schedule(ssk, copy)) {
27728ba8 1313 tcp_remove_empty_skb(ssk);
e2223995 1314 return -ENOMEM;
35759383 1315 }
18b683bf 1316
f70cad10
PA		 1317 if (can_coalesce) {
1318 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1319 } else {
1320 get_page(dfrag->page);
1321 skb_fill_page_desc(skb, i, dfrag->page, offset, copy);
57040755
PA		 1322 }
1323
f70cad10
PA		 1324 skb->len += copy;
1325 skb->data_len += copy;
1326 skb->truesize += copy;
1327 sk_wmem_queued_add(ssk, copy);
1328 sk_mem_charge(ssk, copy);
f70cad10
PA		 1329 WRITE_ONCE(tcp_sk(ssk)->write_seq, tcp_sk(ssk)->write_seq + copy);
1330 TCP_SKB_CB(skb)->end_seq += copy;
1331 tcp_skb_pcount_set(skb, 0);
1332
1333 /* on skb reuse we just need to update the DSS len */
1334 if (reuse_skb) {
1335 TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH;
1336 mpext->data_len += copy;
1337 WARN_ON_ONCE(zero_window_probe);
1338 goto out;
724cfd2e 1339 }
6d0060f6
MM		 1340
1341 memset(mpext, 0, sizeof(*mpext));
d9ca1de8 1342 mpext->data_seq = data_seq;
6d0060f6 1343 mpext->subflow_seq = mptcp_subflow_ctx(ssk)->rel_write_seq;
f70cad10 1344 mpext->data_len = copy;
6d0060f6
MM		 1345 mpext->use_map = 1;
1346 mpext->dsn64 = 1;
1347
1348 pr_debug("data_seq=%llu subflow_seq=%u data_len=%u dsn64=%d",
1349 mpext->data_seq, mpext->subflow_seq, mpext->data_len,
1350 mpext->dsn64);
1351
6f8a612a 1352 if (zero_window_probe) {
f70cad10 1353 mptcp_subflow_ctx(ssk)->rel_write_seq += copy;
6f8a612a 1354 mpext->frozen = 1;
d0cc2987 1355 if (READ_ONCE(msk->csum_enabled))
f70cad10 1356 mptcp_update_data_checksum(skb, copy);
6f8a612a 1357 tcp_push_pending_frames(ssk);
d0cc2987 1358 return 0;
6f8a612a 1359 }
57040755 1360out:
d0cc2987 1361 if (READ_ONCE(msk->csum_enabled))
f70cad10
PA		 1362 mptcp_update_data_checksum(skb, copy);
1363 mptcp_subflow_ctx(ssk)->rel_write_seq += copy;
1364 return copy;
6d0060f6
MM		 1365}
1366
d5f49190
PA		 1367#define MPTCP_SEND_BURST_SIZE ((1 << 16) - \
1368 sizeof(struct tcphdr) - \
1369 MAX_TCP_OPTION_SPACE - \
1370 sizeof(struct ipv6hdr) - \
1371 sizeof(struct frag_hdr))
1372
1373struct subflow_send_info {
1374 struct sock *ssk;
3ce0852c 1375 u64 linger_time;
d5f49190
PA		 1376};
1377
ff5a0b42
PA		 1378void mptcp_subflow_set_active(struct mptcp_subflow_context *subflow)
1379{
1380 if (!subflow->stale)
1381 return;
1382
1383 subflow->stale = 0;
fc1b4e3b 1384 MPTCP_INC_STATS(sock_net(mptcp_subflow_tcp_sock(subflow)), MPTCP_MIB_SUBFLOWRECOVER);
ff5a0b42
PA		 1385}
1386
1387bool mptcp_subflow_active(struct mptcp_subflow_context *subflow)
1388{
1389 if (unlikely(subflow->stale)) {
1390 u32 rcv_tstamp = READ_ONCE(tcp_sk(mptcp_subflow_tcp_sock(subflow))->rcv_tstamp);
1391
1392 if (subflow->stale_rcv_tstamp == rcv_tstamp)
1393 return false;
1394
1395 mptcp_subflow_set_active(subflow);
1396 }
1397 return __mptcp_subflow_active(subflow);
1398}
1399
3ce0852c
PA		 1400#define SSK_MODE_ACTIVE 0
1401#define SSK_MODE_BACKUP 1
1402#define SSK_MODE_MAX 2
1403
33d41c9c
PA		 1404/* implement the mptcp packet scheduler;
1405 * returns the subflow that will transmit the next DSS
1406 * additionally updates the rtx timeout
1407 */
5cf92bba 1408static struct sock *mptcp_subflow_get_send(struct mptcp_sock *msk)
f296234c 1409{
3ce0852c 1410 struct subflow_send_info send_info[SSK_MODE_MAX];
f296234c 1411 struct mptcp_subflow_context *subflow;
33d41c9c 1412 struct sock *sk = (struct sock *)msk;
3ce0852c 1413 u32 pace, burst, wmem;
d5f49190
PA		 1414 int i, nr_active = 0;
1415 struct sock *ssk;
3ce0852c 1416 u64 linger_time;
33d41c9c 1417 long tout = 0;
f296234c 1418
33d41c9c 1419 sock_owned_by_me(sk);
f296234c 1420
d5f49190
PA		 1421 if (__mptcp_check_fallback(msk)) {
1422 if (!msk->first)
f296234c 1423 return NULL;
d5f49190
PA		 1424 return sk_stream_memory_free(msk->first) ? msk->first : NULL;
1425 }
1426
1427 /* re-use last subflow, if the burst allow that */
1428 if (msk->last_snd && msk->snd_burst > 0 &&
1429 sk_stream_memory_free(msk->last_snd) &&
33d41c9c
PA		 1430 mptcp_subflow_active(mptcp_subflow_ctx(msk->last_snd))) {
1431 mptcp_set_timeout(sk);
d5f49190 1432 return msk->last_snd;
33d41c9c 1433 }
f296234c 1434
d5f49190 1435 /* pick the subflow with the lower wmem/wspace ratio */
3ce0852c 1436 for (i = 0; i < SSK_MODE_MAX; ++i) {
d5f49190 1437 send_info[i].ssk = NULL;
3ce0852c 1438 send_info[i].linger_time = -1;
d5f49190 1439 }
3ce0852c 1440
d5f49190 1441 mptcp_for_each_subflow(msk, subflow) {
e10a9892 1442 trace_mptcp_subflow_get_send(subflow);
d5f49190
PA		 1443 ssk = mptcp_subflow_tcp_sock(subflow);
1444 if (!mptcp_subflow_active(subflow))
1445 continue;
1446
33d41c9c 1447 tout = max(tout, mptcp_timeout_from_subflow(subflow));
d5f49190 1448 nr_active += !subflow->backup;
3ce0852c
PA		 1449 pace = subflow->avg_pacing_rate;
1450 if (unlikely(!pace)) {
1451 /* init pacing rate from socket */
1452 subflow->avg_pacing_rate = READ_ONCE(ssk->sk_pacing_rate);
1453 pace = subflow->avg_pacing_rate;
1454 if (!pace)
1455 continue;
1456 }
f296234c 1457
3ce0852c
PA		 1458 linger_time = div_u64((u64)READ_ONCE(ssk->sk_wmem_queued) << 32, pace);
1459 if (linger_time < send_info[subflow->backup].linger_time) {
d5f49190 1460 send_info[subflow->backup].ssk = ssk;
3ce0852c 1461 send_info[subflow->backup].linger_time = linger_time;
d5f49190 1462 }
f296234c 1463 }
33d41c9c 1464 __mptcp_set_timeout(sk, tout);
f296234c 1465
d5f49190
PA		 1466 /* pick the best backup if no other subflow is active */
1467 if (!nr_active)
3ce0852c
PA		 1468 send_info[SSK_MODE_ACTIVE].ssk = send_info[SSK_MODE_BACKUP].ssk;
1469
1470 /* According to the blest algorithm, to avoid HoL blocking for the
1471 * faster flow, we need to:
1472 * - estimate the faster flow linger time
1473 * - use the above to estimate the amount of byte transferred
1474 * by the faster flow
1475 * - check that the amount of queued data is greter than the above,
1476 * otherwise do not use the picked, slower, subflow
1477 * We select the subflow with the shorter estimated time to flush
1478 * the queued mem, which basically ensure the above. We just need
1479 * to check that subflow has a non empty cwin.
1480 */
1481 ssk = send_info[SSK_MODE_ACTIVE].ssk;
1482 if (!ssk || !sk_stream_memory_free(ssk) || !tcp_sk(ssk)->snd_wnd)
1483 return NULL;
5cf92bba 1484
3ce0852c
PA		 1485 burst = min_t(int, MPTCP_SEND_BURST_SIZE, tcp_sk(ssk)->snd_wnd);
1486 wmem = READ_ONCE(ssk->sk_wmem_queued);
1487 subflow = mptcp_subflow_ctx(ssk);
1488 subflow->avg_pacing_rate = div_u64((u64)subflow->avg_pacing_rate * wmem +
1489 READ_ONCE(ssk->sk_pacing_rate) * burst,
1490 burst + wmem);
1491 msk->last_snd = ssk;
1492 msk->snd_burst = burst;
1493 return ssk;
f296234c
PK		 1494}
1495
b8e0def3 1496static void mptcp_push_release(struct sock *ssk, struct mptcp_sendmsg_info *info)
d9ca1de8 1497{
d9ca1de8
PA		 1498 tcp_push(ssk, 0, info->mss_now, tcp_sk(ssk)->nonagle, info->size_goal);
1499 release_sock(ssk);
1500}
1501
0d199e43
FW		 1502static void mptcp_update_post_push(struct mptcp_sock *msk,
1503 struct mptcp_data_frag *dfrag,
1504 u32 sent)
1505{
1506 u64 snd_nxt_new = dfrag->data_seq;
1507
1508 dfrag->already_sent += sent;
1509
1510 msk->snd_burst -= sent;
0d199e43
FW		 1511
1512 snd_nxt_new += dfrag->already_sent;
1513
1514 /* snd_nxt_new can be smaller than snd_nxt in case mptcp
1515 * is recovering after a failover. In that event, this re-sends
1516 * old segments.
1517 *
1518 * Thus compute snd_nxt_new candidate based on
1519 * the dfrag->data_seq that was sent and the data
1520 * that has been handed to the subflow for transmission
1521 * and skip update in case it was old dfrag.
1522 */
1523 if (likely(after64(snd_nxt_new, msk->snd_nxt)))
1524 msk->snd_nxt = snd_nxt_new;
1525}
1526
8b38217a 1527void mptcp_check_and_set_pending(struct sock *sk)
3241a9c0
FW		 1528{
1529 if (mptcp_send_head(sk) &&
1530 !test_bit(MPTCP_PUSH_PENDING, &mptcp_sk(sk)->flags))
1531 set_bit(MPTCP_PUSH_PENDING, &mptcp_sk(sk)->flags);
1532}
1533
ff5a0b42 1534void __mptcp_push_pending(struct sock *sk, unsigned int flags)
f870fa0b 1535{
d9ca1de8 1536 struct sock *prev_ssk = NULL, *ssk = NULL;
f870fa0b 1537 struct mptcp_sock *msk = mptcp_sk(sk);
caf971df 1538 struct mptcp_sendmsg_info info = {
d9ca1de8 1539 .flags = flags,
caf971df 1540 };
d9ca1de8
PA		 1541 struct mptcp_data_frag *dfrag;
1542 int len, copied = 0;
d9ca1de8
PA		 1543
1544 while ((dfrag = mptcp_send_head(sk))) {
1545 info.sent = dfrag->already_sent;
1546 info.limit = dfrag->data_len;
1547 len = dfrag->data_len - dfrag->already_sent;
1548 while (len > 0) {
1549 int ret = 0;
1550
1551 prev_ssk = ssk;
3d79e375 1552 __mptcp_flush_join_list(msk);
5cf92bba 1553 ssk = mptcp_subflow_get_send(msk);
d9ca1de8 1554
9758f40e
PA		 1555 /* First check. If the ssk has changed since
1556 * the last round, release prev_ssk
d9ca1de8
PA		 1557 */
1558 if (ssk != prev_ssk && prev_ssk)
b8e0def3 1559 mptcp_push_release(prev_ssk, &info);
d9ca1de8
PA		 1560 if (!ssk)
1561 goto out;
1562
9758f40e
PA		 1563 /* Need to lock the new subflow only if different
1564 * from the previous one, otherwise we are still
1565 * helding the relevant lock
1566 */
1567 if (ssk != prev_ssk)
d9ca1de8
PA		 1568 lock_sock(ssk);
1569
1570 ret = mptcp_sendmsg_frag(sk, ssk, dfrag, &info);
1571 if (ret <= 0) {
b8e0def3 1572 mptcp_push_release(ssk, &info);
d9ca1de8
PA		 1573 goto out;
1574 }
1575
1576 info.sent += ret;
d9ca1de8
PA		 1577 copied += ret;
1578 len -= ret;
0d199e43
FW		 1579
1580 mptcp_update_post_push(msk, dfrag, ret);
d9ca1de8
PA		 1581 }
1582 WRITE_ONCE(msk->first_pending, mptcp_send_next(sk));
1583 }
1584
1585 /* at this point we held the socket lock for the last subflow we used */
1586 if (ssk)
b8e0def3 1587 mptcp_push_release(ssk, &info);
d9ca1de8
PA		 1588
1589out:
33d41c9c
PA		 1590 /* ensure the rtx timer is running */
1591 if (!mptcp_timer_pending(sk))
1592 mptcp_reset_timer(sk);
1593 if (copied)
d9ca1de8
PA		 1594 __mptcp_check_send_data_fin(sk);
1595}
1596
6e628cd3
PA		 1597static void __mptcp_subflow_push_pending(struct sock *sk, struct sock *ssk)
1598{
1599 struct mptcp_sock *msk = mptcp_sk(sk);
1094c6fe
PA		 1600 struct mptcp_sendmsg_info info = {
1601 .data_lock_held = true,
1602 };
6e628cd3 1603 struct mptcp_data_frag *dfrag;
b19bc294 1604 struct sock *xmit_ssk;
6e628cd3 1605 int len, copied = 0;
b19bc294 1606 bool first = true;
6e628cd3
PA		 1607
1608 info.flags = 0;
1609 while ((dfrag = mptcp_send_head(sk))) {
1610 info.sent = dfrag->already_sent;
1611 info.limit = dfrag->data_len;
1612 len = dfrag->data_len - dfrag->already_sent;
1613 while (len > 0) {
1614 int ret = 0;
1615
b19bc294
PA		 1616 /* the caller already invoked the packet scheduler,
1617 * check for a different subflow usage only after
1618 * spooling the first chunk of data
1619 */
1620 xmit_ssk = first ? ssk : mptcp_subflow_get_send(mptcp_sk(sk));
1621 if (!xmit_ssk)
1622 goto out;
1623 if (xmit_ssk != ssk) {
bcd97734
PA		 1624 mptcp_subflow_delegate(mptcp_subflow_ctx(xmit_ssk),
1625 MPTCP_DELEGATE_SEND);
b19bc294
PA		 1626 goto out;
1627 }
1628
6e628cd3
PA		 1629 ret = mptcp_sendmsg_frag(sk, ssk, dfrag, &info);
1630 if (ret <= 0)
1631 goto out;
1632
1633 info.sent += ret;
6e628cd3
PA		 1634 copied += ret;
1635 len -= ret;
b19bc294 1636 first = false;
0d199e43
FW		 1637
1638 mptcp_update_post_push(msk, dfrag, ret);
6e628cd3
PA		 1639 }
1640 WRITE_ONCE(msk->first_pending, mptcp_send_next(sk));
1641 }
1642
1643out:
1644 /* __mptcp_alloc_tx_skb could have released some wmem and we are
1645 * not going to flush it via release_sock()
1646 */
6e628cd3 1647 if (copied) {
6e628cd3
PA		 1648 tcp_push(ssk, 0, info.mss_now, tcp_sk(ssk)->nonagle,
1649 info.size_goal);
d09d818e
PA		 1650 if (!mptcp_timer_pending(sk))
1651 mptcp_reset_timer(sk);
1652
6e628cd3
PA		 1653 if (msk->snd_data_fin_enable &&
1654 msk->snd_nxt + 1 == msk->write_seq)
1655 mptcp_schedule_work(sk);
1656 }
1657}
1658
5cf92bba
PA		 1659static void mptcp_set_nospace(struct sock *sk)
1660{
1661 /* enable autotune */
1662 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1663
1664 /* will be cleared on avail space */
1665 set_bit(MPTCP_NOSPACE, &mptcp_sk(sk)->flags);
1666}
1667
d9ca1de8
PA		 1668static int mptcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1669{
1670 struct mptcp_sock *msk = mptcp_sk(sk);
17091708 1671 struct page_frag *pfrag;
6d0060f6 1672 size_t copied = 0;
caf971df 1673 int ret = 0;
6d0060f6 1674 long timeo;
f870fa0b 1675
987858e5
PA		 1676 /* we don't support FASTOPEN yet */
1677 if (msg->msg_flags & MSG_FASTOPEN)
f870fa0b
MM		 1678 return -EOPNOTSUPP;
1679
987858e5
PA		 1680 /* silently ignore everything else */
1681 msg->msg_flags &= MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL;
1682
6511882c 1683 lock_sock(sk);
1954b860
MM		 1684
1685 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1686
1687 if ((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) {
1688 ret = sk_stream_wait_connect(sk, &timeo);
1689 if (ret)
1690 goto out;
1691 }
1692
17091708 1693 pfrag = sk_page_frag(sk);
18b683bf 1694
d9ca1de8 1695 while (msg_data_left(msg)) {
724cfd2e 1696 int total_ts, frag_truesize = 0;
d9ca1de8 1697 struct mptcp_data_frag *dfrag;
d9ca1de8
PA		 1698 bool dfrag_collapsed;
1699 size_t psize, offset;
18b683bf 1700
d9ca1de8
PA		 1701 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN)) {
1702 ret = -EPIPE;
f296234c
PK		 1703 goto out;
1704 }
da51aef5 1705
d9ca1de8
PA		 1706 /* reuse tail pfrag, if possible, or carve a new one from the
1707 * page allocator
1708 */
1709 dfrag = mptcp_pending_tail(sk);
1710 dfrag_collapsed = mptcp_frag_can_collapse_to(msk, pfrag, dfrag);
1711 if (!dfrag_collapsed) {
6e628cd3
PA		 1712 if (!sk_stream_memory_free(sk))
1713 goto wait_for_memory;
1714
d9ca1de8
PA		 1715 if (!mptcp_page_frag_refill(sk, pfrag))
1716 goto wait_for_memory;
1717
1718 dfrag = mptcp_carve_data_frag(msk, pfrag, pfrag->offset);
1719 frag_truesize = dfrag->overhead;
72511aab 1720 }
6d0060f6 1721
d9ca1de8
PA		 1722 /* we do not bound vs wspace, to allow a single packet.
1723 * memory accounting will prevent execessive memory usage
1724 * anyway
d5f49190 1725 */
d9ca1de8
PA		 1726 offset = dfrag->offset + dfrag->data_len;
1727 psize = pfrag->size - offset;
1728 psize = min_t(size_t, psize, msg_data_left(msg));
724cfd2e 1729 total_ts = psize + frag_truesize;
d9ca1de8 1730
6511882c 1731 if (!sk_wmem_schedule(sk, total_ts))
724cfd2e 1732 goto wait_for_memory;
724cfd2e 1733
d9ca1de8
PA		 1734 if (copy_page_from_iter(dfrag->page, offset, psize,
1735 &msg->msg_iter) != psize) {
1736 ret = -EFAULT;
1737 goto out;
72511aab
FW		 1738 }
1739
d9ca1de8 1740 /* data successfully copied into the write queue */
6511882c 1741 sk->sk_forward_alloc -= total_ts;
d9ca1de8
PA		 1742 copied += psize;
1743 dfrag->data_len += psize;
1744 frag_truesize += psize;
1745 pfrag->offset += frag_truesize;
1746 WRITE_ONCE(msk->write_seq, msk->write_seq + psize);
1747
1748 /* charge data on mptcp pending queue to the msk socket
1749 * Note: we charge such data both to sk and ssk
fb529e62 1750 */
d9ca1de8 1751 sk_wmem_queued_add(sk, frag_truesize);
d9ca1de8
PA		 1752 if (!dfrag_collapsed) {
1753 get_page(dfrag->page);
1754 list_add_tail(&dfrag->list, &msk->rtx_queue);
1755 if (!msk->first_pending)
1756 WRITE_ONCE(msk->first_pending, dfrag);
fb529e62 1757 }
e4b61351 1758 pr_debug("msk=%p dfrag at seq=%llu len=%u sent=%u new=%d", msk,
d9ca1de8
PA		 1759 dfrag->data_seq, dfrag->data_len, dfrag->already_sent,
1760 !dfrag_collapsed);
6d0060f6 1761
d9ca1de8 1762 continue;
b51f9b80 1763
d9ca1de8 1764wait_for_memory:
5cf92bba 1765 mptcp_set_nospace(sk);
c2e6048f 1766 __mptcp_push_pending(sk, msg->msg_flags);
d9ca1de8
PA		 1767 ret = sk_stream_wait_memory(sk, &timeo);
1768 if (ret)
1769 goto out;
57040755 1770 }
6d0060f6 1771
13e16037 1772 if (copied)
c2e6048f 1773 __mptcp_push_pending(sk, msg->msg_flags);
d9ca1de8 1774
1954b860 1775out:
cec37a6e 1776 release_sock(sk);
8555c6bf 1777 return copied ? : ret;
f870fa0b
MM		 1778}
1779
6771bfd9
FW		 1780static int __mptcp_recvmsg_mskq(struct mptcp_sock *msk,
1781 struct msghdr *msg,
b7f653b2
FW		 1782 size_t len, int flags,
1783 struct scm_timestamping_internal *tss,
1784 int *cmsg_flags)
6771bfd9 1785{
ca4fb892 1786 struct sk_buff *skb, *tmp;
6771bfd9
FW		 1787 int copied = 0;
1788
ca4fb892 1789 skb_queue_walk_safe(&msk->receive_queue, skb, tmp) {
6771bfd9
FW		 1790 u32 offset = MPTCP_SKB_CB(skb)->offset;
1791 u32 data_len = skb->len - offset;
1792 u32 count = min_t(size_t, len - copied, data_len);
1793 int err;
1794
d976092c
PA		 1795 if (!(flags & MSG_TRUNC)) {
1796 err = skb_copy_datagram_msg(skb, offset, msg, count);
1797 if (unlikely(err < 0)) {
1798 if (!copied)
1799 return err;
1800 break;
1801 }
6771bfd9
FW		 1802 }
1803
b7f653b2
FW		 1804 if (MPTCP_SKB_CB(skb)->has_rxtstamp) {
1805 tcp_update_recv_tstamps(skb, tss);
1806 *cmsg_flags |= MPTCP_CMSG_TS;
1807 }
1808
6771bfd9
FW		 1809 copied += count;
1810
1811 if (count < data_len) {
2c9e7765 1812 if (!(flags & MSG_PEEK)) {
ca4fb892 1813 MPTCP_SKB_CB(skb)->offset += count;
2c9e7765
FW		 1814 MPTCP_SKB_CB(skb)->map_seq += count;
1815 }
6771bfd9
FW		 1816 break;
1817 }
1818
ca4fb892
YL		 1819 if (!(flags & MSG_PEEK)) {
1820 /* we will bulk release the skb memory later */
1821 skb->destructor = NULL;
ce599c51 1822 WRITE_ONCE(msk->rmem_released, msk->rmem_released + skb->truesize);
ca4fb892
YL		 1823 __skb_unlink(skb, &msk->receive_queue);
1824 __kfree_skb(skb);
1825 }
6771bfd9
FW		 1826
1827 if (copied >= len)
1828 break;
1829 }
1830
1831 return copied;
1832}
1833
a6b118fe
FW		 1834/* receive buffer autotuning. See tcp_rcv_space_adjust for more information.
1835 *
1836 * Only difference: Use highest rtt estimate of the subflows in use.
1837 */
1838static void mptcp_rcv_space_adjust(struct mptcp_sock *msk, int copied)
1839{
1840 struct mptcp_subflow_context *subflow;
1841 struct sock *sk = (struct sock *)msk;
1842 u32 time, advmss = 1;
1843 u64 rtt_us, mstamp;
1844
1845 sock_owned_by_me(sk);
1846
1847 if (copied <= 0)
1848 return;
1849
1850 msk->rcvq_space.copied += copied;
1851
1852 mstamp = div_u64(tcp_clock_ns(), NSEC_PER_USEC);
1853 time = tcp_stamp_us_delta(mstamp, msk->rcvq_space.time);
1854
1855 rtt_us = msk->rcvq_space.rtt_us;
1856 if (rtt_us && time < (rtt_us >> 3))
1857 return;
1858
1859 rtt_us = 0;
1860 mptcp_for_each_subflow(msk, subflow) {
1861 const struct tcp_sock *tp;
1862 u64 sf_rtt_us;
1863 u32 sf_advmss;
1864
1865 tp = tcp_sk(mptcp_subflow_tcp_sock(subflow));
1866
1867 sf_rtt_us = READ_ONCE(tp->rcv_rtt_est.rtt_us);
1868 sf_advmss = READ_ONCE(tp->advmss);
1869
1870 rtt_us = max(sf_rtt_us, rtt_us);
1871 advmss = max(sf_advmss, advmss);
1872 }
1873
1874 msk->rcvq_space.rtt_us = rtt_us;
1875 if (time < (rtt_us >> 3) || rtt_us == 0)
1876 return;
1877
1878 if (msk->rcvq_space.copied <= msk->rcvq_space.space)
1879 goto new_measure;
1880
1881 if (sock_net(sk)->ipv4.sysctl_tcp_moderate_rcvbuf &&
1882 !(sk->sk_userlocks & SOCK_RCVBUF_LOCK)) {
1883 int rcvmem, rcvbuf;
1884 u64 rcvwin, grow;
1885
1886 rcvwin = ((u64)msk->rcvq_space.copied << 1) + 16 * advmss;
1887
1888 grow = rcvwin * (msk->rcvq_space.copied - msk->rcvq_space.space);
1889
1890 do_div(grow, msk->rcvq_space.space);
1891 rcvwin += (grow << 1);
1892
1893 rcvmem = SKB_TRUESIZE(advmss + MAX_TCP_HEADER);
1894 while (tcp_win_from_space(sk, rcvmem) < advmss)
1895 rcvmem += 128;
1896
1897 do_div(rcvwin, advmss);
1898 rcvbuf = min_t(u64, rcvwin * rcvmem,
1899 sock_net(sk)->ipv4.sysctl_tcp_rmem[2]);
1900
1901 if (rcvbuf > sk->sk_rcvbuf) {
1902 u32 window_clamp;
1903
1904 window_clamp = tcp_win_from_space(sk, rcvbuf);
1905 WRITE_ONCE(sk->sk_rcvbuf, rcvbuf);
1906
1907 /* Make subflows follow along. If we do not do this, we
1908 * get drops at subflow level if skbs can't be moved to
1909 * the mptcp rx queue fast enough (announced rcv_win can
1910 * exceed ssk->sk_rcvbuf).
1911 */
1912 mptcp_for_each_subflow(msk, subflow) {
1913 struct sock *ssk;
c76c6956 1914 bool slow;
a6b118fe
FW		 1915
1916 ssk = mptcp_subflow_tcp_sock(subflow);
c76c6956 1917 slow = lock_sock_fast(ssk);
a6b118fe
FW		 1918 WRITE_ONCE(ssk->sk_rcvbuf, rcvbuf);
1919 tcp_sk(ssk)->window_clamp = window_clamp;
c76c6956
PA		 1920 tcp_cleanup_rbuf(ssk, 1);
1921 unlock_sock_fast(ssk, slow);
a6b118fe
FW		 1922 }
1923 }
1924 }
1925
1926 msk->rcvq_space.space = msk->rcvq_space.copied;
1927new_measure:
1928 msk->rcvq_space.copied = 0;
1929 msk->rcvq_space.time = mstamp;
1930}
1931
87952603
PA		 1932static void __mptcp_update_rmem(struct sock *sk)
1933{
1934 struct mptcp_sock *msk = mptcp_sk(sk);
1935
1936 if (!msk->rmem_released)
1937 return;
1938
1939 atomic_sub(msk->rmem_released, &sk->sk_rmem_alloc);
6511882c 1940 mptcp_rmem_uncharge(sk, msk->rmem_released);
ce599c51 1941 WRITE_ONCE(msk->rmem_released, 0);
87952603
PA		 1942}
1943
1944static void __mptcp_splice_receive_queue(struct sock *sk)
1945{
1946 struct mptcp_sock *msk = mptcp_sk(sk);
1947
1948 skb_queue_splice_tail_init(&sk->sk_receive_queue, &msk->receive_queue);
1949}
1950
e3859603 1951static bool __mptcp_move_skbs(struct mptcp_sock *msk)
6771bfd9 1952{
87952603 1953 struct sock *sk = (struct sock *)msk;
6771bfd9 1954 unsigned int moved = 0;
87952603 1955 bool ret, done;
d5f49190 1956
78962489 1957 mptcp_flush_join_list(msk);
6771bfd9
FW		 1958 do {
1959 struct sock *ssk = mptcp_subflow_recv_lookup(msk);
65f49fe7 1960 bool slowpath;
6771bfd9 1961
87952603
PA		 1962 /* we can have data pending in the subflows only if the msk
1963 * receive buffer was full at subflow_data_ready() time,
1964 * that is an unlikely slow path.
1965 */
1966 if (likely(!ssk))
6771bfd9
FW		 1967 break;
1968
65f49fe7 1969 slowpath = lock_sock_fast(ssk);
87952603 1970 mptcp_data_lock(sk);
e3859603 1971 __mptcp_update_rmem(sk);
6771bfd9 1972 done = __mptcp_move_skbs_from_subflow(msk, ssk, &moved);
87952603 1973 mptcp_data_unlock(sk);
499ada50
PA		 1974
1975 if (unlikely(ssk->sk_err))
1976 __mptcp_error_report(sk);
65f49fe7 1977 unlock_sock_fast(ssk, slowpath);
6771bfd9
FW		 1978 } while (!done);
1979
87952603
PA		 1980 /* acquire the data lock only if some input data is pending */
1981 ret = moved > 0;
1982 if (!RB_EMPTY_ROOT(&msk->out_of_order_queue) ||
1983 !skb_queue_empty_lockless(&sk->sk_receive_queue)) {
1984 mptcp_data_lock(sk);
1985 __mptcp_update_rmem(sk);
1986 ret |= __mptcp_ofo_queue(msk);
1987 __mptcp_splice_receive_queue(sk);
1988 mptcp_data_unlock(sk);
ab174ad8 1989 }
87952603
PA		 1990 if (ret)
1991 mptcp_check_data_fin((struct sock *)msk);
1992 return !skb_queue_empty(&msk->receive_queue);
6771bfd9
FW		 1993}
1994
2c9e7765
FW		 1995static unsigned int mptcp_inq_hint(const struct sock *sk)
1996{
1997 const struct mptcp_sock *msk = mptcp_sk(sk);
1998 const struct sk_buff *skb;
1999
2000 skb = skb_peek(&msk->receive_queue);
2001 if (skb) {
2002 u64 hint_val = msk->ack_seq - MPTCP_SKB_CB(skb)->map_seq;
2003
2004 if (hint_val >= INT_MAX)
2005 return INT_MAX;
2006
2007 return (unsigned int)hint_val;
2008 }
2009
2010 if (sk->sk_state == TCP_CLOSE || (sk->sk_shutdown & RCV_SHUTDOWN))
2011 return 1;
2012
2013 return 0;
2014}
2015
f870fa0b
MM		 2016static int mptcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
2017 int nonblock, int flags, int *addr_len)
2018{
2019 struct mptcp_sock *msk = mptcp_sk(sk);
b7f653b2
FW		 2020 struct scm_timestamping_internal tss;
2021 int copied = 0, cmsg_flags = 0;
7a6a6cbc
PA		 2022 int target;
2023 long timeo;
f870fa0b 2024
cb9d80f4
PA		 2025 /* MSG_ERRQUEUE is really a no-op till we support IP_RECVERR */
2026 if (unlikely(flags & MSG_ERRQUEUE))
2027 return inet_recv_error(sk, msg, len, addr_len);
2028
6511882c 2029 lock_sock(sk);
fd897679
PA		 2030 if (unlikely(sk->sk_state == TCP_LISTEN)) {
2031 copied = -ENOTCONN;
2032 goto out_err;
2033 }
2034
7a6a6cbc
PA		 2035 timeo = sock_rcvtimeo(sk, nonblock);
2036
2037 len = min_t(size_t, len, INT_MAX);
2038 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
2039
2c9e7765
FW		 2040 if (unlikely(msk->recvmsg_inq))
2041 cmsg_flags = MPTCP_CMSG_INQ;
2042
05e3ecea 2043 while (copied < len) {
e3859603 2044 int bytes_read;
7a6a6cbc 2045
b7f653b2 2046 bytes_read = __mptcp_recvmsg_mskq(msk, msg, len - copied, flags, &tss, &cmsg_flags);
6771bfd9
FW		 2047 if (unlikely(bytes_read < 0)) {
2048 if (!copied)
2049 copied = bytes_read;
2050 goto out_err;
2051 }
7a6a6cbc 2052
6771bfd9 2053 copied += bytes_read;
7a6a6cbc 2054
ea4ca586 2055 /* be sure to advertise window change */
e3859603
PA		 2056 mptcp_cleanup_rbuf(msk);
2057
2058 if (skb_queue_empty(&msk->receive_queue) && __mptcp_move_skbs(msk))
2059 continue;
ea4ca586 2060
7a6a6cbc
PA		 2061 /* only the master socket status is relevant here. The exit
2062 * conditions mirror closely tcp_recvmsg()
2063 */
2064 if (copied >= target)
2065 break;
2066
2067 if (copied) {
2068 if (sk->sk_err ||
2069 sk->sk_state == TCP_CLOSE ||
2070 (sk->sk_shutdown & RCV_SHUTDOWN) ||
2071 !timeo ||
2072 signal_pending(current))
2073 break;
2074 } else {
2075 if (sk->sk_err) {
2076 copied = sock_error(sk);
2077 break;
2078 }
2079
5969856a
PA		 2080 if (test_and_clear_bit(MPTCP_WORK_EOF, &msk->flags))
2081 mptcp_check_for_eof(msk);
2082
87952603
PA		 2083 if (sk->sk_shutdown & RCV_SHUTDOWN) {
2084 /* race breaker: the shutdown could be after the
2085 * previous receive queue check
2086 */
e3859603 2087 if (__mptcp_move_skbs(msk))
87952603 2088 continue;
7a6a6cbc 2089 break;
87952603 2090 }
7a6a6cbc
PA		 2091
2092 if (sk->sk_state == TCP_CLOSE) {
2093 copied = -ENOTCONN;
2094 break;
2095 }
2096
2097 if (!timeo) {
2098 copied = -EAGAIN;
2099 break;
2100 }
2101
2102 if (signal_pending(current)) {
2103 copied = sock_intr_errno(timeo);
2104 break;
2105 }
2106 }
2107
2108 pr_debug("block timeout %ld", timeo);
612f71d7 2109 sk_wait_data(sk, &timeo, NULL);
7a6a6cbc 2110 }
3c90e377 2111
6771bfd9 2112out_err:
b7f653b2
FW		 2113 if (cmsg_flags && copied >= 0) {
2114 if (cmsg_flags & MPTCP_CMSG_TS)
2115 tcp_recv_timestamp(msg, sk, &tss);
2c9e7765
FW		 2116
2117 if (cmsg_flags & MPTCP_CMSG_INQ) {
2118 unsigned int inq = mptcp_inq_hint(sk);
2119
2120 put_cmsg(msg, SOL_TCP, TCP_CM_INQ, sizeof(inq), &inq);
2121 }
b7f653b2
FW		 2122 }
2123
612f71d7
PA		 2124 pr_debug("msk=%p rx queue empty=%d:%d copied=%d",
2125 msk, skb_queue_empty_lockless(&sk->sk_receive_queue),
2126 skb_queue_empty(&msk->receive_queue), copied);
ca4fb892
YL		 2127 if (!(flags & MSG_PEEK))
2128 mptcp_rcv_space_adjust(msk, copied);
a6b118fe 2129
7a6a6cbc 2130 release_sock(sk);
cec37a6e
PK		 2131 return copied;
2132}
2133
b51f9b80
PA		 2134static void mptcp_retransmit_timer(struct timer_list *t)
2135{
2136 struct inet_connection_sock *icsk = from_timer(icsk, t,
2137 icsk_retransmit_timer);
2138 struct sock *sk = &icsk->icsk_inet.sk;
2d6f5a2b 2139 struct mptcp_sock *msk = mptcp_sk(sk);
b51f9b80
PA		 2140
2141 bh_lock_sock(sk);
2142 if (!sock_owned_by_user(sk)) {
2d6f5a2b
PA		 2143 /* we need a process context to retransmit */
2144 if (!test_and_set_bit(MPTCP_WORK_RTX, &msk->flags))
2145 mptcp_schedule_work(sk);
b51f9b80
PA		 2146 } else {
2147 /* delegate our work to tcp_release_cb() */
2d6f5a2b 2148 set_bit(MPTCP_RETRANSMIT, &msk->flags);
b51f9b80
PA		 2149 }
2150 bh_unlock_sock(sk);
2151 sock_put(sk);
2152}
2153
e16163b6
PA		 2154static void mptcp_timeout_timer(struct timer_list *t)
2155{
2156 struct sock *sk = from_timer(sk, t, sk_timer);
2157
2158 mptcp_schedule_work(sk);
b6d69fc8 2159 sock_put(sk);
e16163b6
PA		 2160}
2161
3b1d6210
PA		 2162/* Find an idle subflow. Return NULL if there is unacked data at tcp
2163 * level.
2164 *
2165 * A backup subflow is returned only if that is the only kind available.
2166 */
ff5a0b42 2167static struct sock *mptcp_subflow_get_retrans(struct mptcp_sock *msk)
3b1d6210 2168{
71b7dec2 2169 struct sock *backup = NULL, *pick = NULL;
3b1d6210 2170 struct mptcp_subflow_context *subflow;
71b7dec2 2171 int min_stale_count = INT_MAX;
3b1d6210
PA		 2172
2173 sock_owned_by_me((const struct sock *)msk);
2174
d5f49190 2175 if (__mptcp_check_fallback(msk))
d9ca1de8 2176 return NULL;
d5f49190 2177
3b1d6210
PA		 2178 mptcp_for_each_subflow(msk, subflow) {
2179 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2180
ff5a0b42 2181 if (!__mptcp_subflow_active(subflow))
d5f49190
PA		 2182 continue;
2183
71b7dec2
PA		 2184 /* still data outstanding at TCP level? skip this */
2185 if (!tcp_rtx_and_write_queues_empty(ssk)) {
2186 mptcp_pm_subflow_chk_stale(msk, ssk);
2187 min_stale_count = min_t(int, min_stale_count, subflow->stale_count);
2188 continue;
860975c6 2189 }
3b1d6210
PA		 2190
2191 if (subflow->backup) {
2192 if (!backup)
2193 backup = ssk;
2194 continue;
2195 }
2196
71b7dec2
PA		 2197 if (!pick)
2198 pick = ssk;
3b1d6210
PA		 2199 }
2200
71b7dec2
PA		 2201 if (pick)
2202 return pick;
2203
2204 /* use backup only if there are no progresses anywhere */
2205 return min_stale_count > 1 ? backup : NULL;
3b1d6210
PA		 2206}
2207
17aee05d
FW		 2208static void mptcp_dispose_initial_subflow(struct mptcp_sock *msk)
2209{
2210 if (msk->subflow) {
2211 iput(SOCK_INODE(msk->subflow));
2212 msk->subflow = NULL;
2213 }
2214}
2215
1e1d9d6f
PA		 2216bool __mptcp_retransmit_pending_data(struct sock *sk)
2217{
2218 struct mptcp_data_frag *cur, *rtx_head;
2219 struct mptcp_sock *msk = mptcp_sk(sk);
2220
2221 if (__mptcp_check_fallback(mptcp_sk(sk)))
2222 return false;
2223
2224 if (tcp_rtx_and_write_queues_empty(sk))
2225 return false;
2226
2227 /* the closing socket has some data untransmitted and/or unacked:
2228 * some data in the mptcp rtx queue has not really xmitted yet.
2229 * keep it simple and re-inject the whole mptcp level rtx queue
2230 */
2231 mptcp_data_lock(sk);
2232 __mptcp_clean_una_wakeup(sk);
2233 rtx_head = mptcp_rtx_head(sk);
2234 if (!rtx_head) {
2235 mptcp_data_unlock(sk);
2236 return false;
2237 }
2238
0d199e43 2239 msk->recovery_snd_nxt = msk->snd_nxt;
1e1d9d6f
PA		 2240 msk->recovery = true;
2241 mptcp_data_unlock(sk);
2242
2243 msk->first_pending = rtx_head;
1e1d9d6f
PA		 2244 msk->snd_burst = 0;
2245
2246 /* be sure to clear the "sent status" on all re-injected fragments */
2247 list_for_each_entry(cur, &msk->rtx_queue, list) {
2248 if (!cur->already_sent)
2249 break;
2250 cur->already_sent = 0;
2251 }
2252
2253 return true;
2254}
2255
b29fcfb5
PA		 2256/* flags for __mptcp_close_ssk() */
2257#define MPTCP_CF_PUSH BIT(1)
2258#define MPTCP_CF_FASTCLOSE BIT(2)
2259
cec37a6e
PK		 2260/* subflow sockets can be either outgoing (connect) or incoming
2261 * (accept).
2262 *
2263 * Outgoing subflows use in-kernel sockets.
2264 * Incoming subflows do not have their own 'struct socket' allocated,
2265 * so we need to use tcp_close() after detaching them from the mptcp
2266 * parent socket.
2267 */
a141e02e 2268static void __mptcp_close_ssk(struct sock *sk, struct sock *ssk,
b29fcfb5
PA		 2269 struct mptcp_subflow_context *subflow,
2270 unsigned int flags)
cec37a6e 2271{
e0be4931 2272 struct mptcp_sock *msk = mptcp_sk(sk);
b29fcfb5 2273 bool need_push, dispose_it;
e0be4931 2274
b29fcfb5
PA		 2275 dispose_it = !msk->subflow || ssk != msk->subflow->sk;
2276 if (dispose_it)
2277 list_del(&subflow->node);
cec37a6e 2278
3f8b2667 2279 lock_sock_nested(ssk, SINGLE_DEPTH_NESTING);
e16163b6 2280
b29fcfb5
PA		 2281 if (flags & MPTCP_CF_FASTCLOSE)
2282 subflow->send_fastclose = 1;
2283
2284 need_push = (flags & MPTCP_CF_PUSH) && __mptcp_retransmit_pending_data(sk);
2285 if (!dispose_it) {
2286 tcp_disconnect(ssk, 0);
2287 msk->subflow->state = SS_UNCONNECTED;
2288 mptcp_subflow_ctx_reset(subflow);
2289 release_sock(ssk);
2290
2291 goto out;
2292 }
2293
e16163b6
PA		 2294 /* if we are invoked by the msk cleanup code, the subflow is
2295 * already orphaned
2296 */
866f26f2 2297 if (ssk->sk_socket)
e16163b6 2298 sock_orphan(ssk);
e16163b6 2299
d7b1bfd0
PA		 2300 subflow->disposable = 1;
2301
e16163b6
PA		 2302 /* if ssk hit tcp_done(), tcp_cleanup_ulp() cleared the related ops
2303 * the ssk has been already destroyed, we just need to release the
2304 * reference owned by msk;
2305 */
2306 if (!inet_csk(ssk)->icsk_ulp_ops) {
2307 kfree_rcu(subflow, rcu);
cec37a6e 2308 } else {
d7b1bfd0 2309 /* otherwise tcp will dispose of the ssk and subflow ctx */
e16163b6
PA		 2310 __tcp_close(ssk, 0);
2311
2312 /* close acquired an extra ref */
2313 __sock_put(ssk);
cec37a6e 2314 }
e16163b6 2315 release_sock(ssk);
e16163b6
PA		 2316
2317 sock_put(ssk);
e0be4931 2318
c8fe62f0
FW		 2319 if (ssk == msk->first)
2320 msk->first = NULL;
2321
b29fcfb5
PA		 2322out:
2323 if (ssk == msk->last_snd)
2324 msk->last_snd = NULL;
1e1d9d6f
PA		 2325
2326 if (need_push)
2327 __mptcp_push_pending(sk, 0);
f870fa0b
MM		 2328}
2329
a141e02e
FW		 2330void mptcp_close_ssk(struct sock *sk, struct sock *ssk,
2331 struct mptcp_subflow_context *subflow)
2332{
b911c97c
FW		 2333 if (sk->sk_state == TCP_ESTABLISHED)
2334 mptcp_event(MPTCP_EVENT_SUB_CLOSED, mptcp_sk(sk), ssk, GFP_KERNEL);
b29fcfb5 2335 __mptcp_close_ssk(sk, ssk, subflow, MPTCP_CF_PUSH);
a141e02e
FW		 2336}
2337
dc24f8b4
PA		 2338static unsigned int mptcp_sync_mss(struct sock *sk, u32 pmtu)
2339{
2340 return 0;
2341}
2342
0e4f35d7
PA		 2343static void __mptcp_close_subflow(struct mptcp_sock *msk)
2344{
2345 struct mptcp_subflow_context *subflow, *tmp;
2346
3abc05d9
FW		 2347 might_sleep();
2348
0e4f35d7
PA		 2349 list_for_each_entry_safe(subflow, tmp, &msk->conn_list, node) {
2350 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2351
2352 if (inet_sk_state_load(ssk) != TCP_CLOSE)
2353 continue;
2354
40947e13
FW		 2355 /* 'subflow_data_ready' will re-sched once rx queue is empty */
2356 if (!skb_queue_empty_lockless(&ssk->sk_receive_queue))
2357 continue;
2358
a141e02e 2359 mptcp_close_ssk((struct sock *)msk, ssk, subflow);
0e4f35d7
PA		 2360 }
2361}
2362
e16163b6
PA		 2363static bool mptcp_check_close_timeout(const struct sock *sk)
2364{
2365 s32 delta = tcp_jiffies32 - inet_csk(sk)->icsk_mtup.probe_timestamp;
2366 struct mptcp_subflow_context *subflow;
2367
2368 if (delta >= TCP_TIMEWAIT_LEN)
2369 return true;
2370
2371 /* if all subflows are in closed status don't bother with additional
2372 * timeout
2373 */
2374 mptcp_for_each_subflow(mptcp_sk(sk), subflow) {
2375 if (inet_sk_state_load(mptcp_subflow_tcp_sock(subflow)) !=
2376 TCP_CLOSE)
2377 return false;
2378 }
2379 return true;
2380}
2381
50c504a2
FW		 2382static void mptcp_check_fastclose(struct mptcp_sock *msk)
2383{
2384 struct mptcp_subflow_context *subflow, *tmp;
2385 struct sock *sk = &msk->sk.icsk_inet.sk;
2386
2387 if (likely(!READ_ONCE(msk->rcv_fastclose)))
2388 return;
2389
2390 mptcp_token_destroy(msk);
2391
2392 list_for_each_entry_safe(subflow, tmp, &msk->conn_list, node) {
2393 struct sock *tcp_sk = mptcp_subflow_tcp_sock(subflow);
75e908c3 2394 bool slow;
50c504a2 2395
75e908c3 2396 slow = lock_sock_fast(tcp_sk);
50c504a2
FW		 2397 if (tcp_sk->sk_state != TCP_CLOSE) {
2398 tcp_send_active_reset(tcp_sk, GFP_ATOMIC);
2399 tcp_set_state(tcp_sk, TCP_CLOSE);
2400 }
75e908c3 2401 unlock_sock_fast(tcp_sk, slow);
50c504a2
FW		 2402 }
2403
2404 inet_sk_state_store(sk, TCP_CLOSE);
2405 sk->sk_shutdown = SHUTDOWN_MASK;
2406 smp_mb__before_atomic(); /* SHUTDOWN must be visible first */
50c504a2
FW		 2407 set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags);
2408
2409 mptcp_close_wake_up(sk);
2410}
2411
2948d0a1 2412static void __mptcp_retrans(struct sock *sk)
80992017 2413{
2948d0a1 2414 struct mptcp_sock *msk = mptcp_sk(sk);
caf971df 2415 struct mptcp_sendmsg_info info = {};
3b1d6210 2416 struct mptcp_data_frag *dfrag;
3b1d6210 2417 size_t copied = 0;
2948d0a1
PA		 2418 struct sock *ssk;
2419 int ret;
3b1d6210 2420
b5941f06 2421 mptcp_clean_una_wakeup(sk);
3241a9c0
FW		 2422
2423 /* first check ssk: need to kick "stale" logic */
2424 ssk = mptcp_subflow_get_retrans(msk);
3b1d6210 2425 dfrag = mptcp_rtx_head(sk);
6477dd39
MM		 2426 if (!dfrag) {
2427 if (mptcp_data_fin_enabled(msk)) {
2428 struct inet_connection_sock *icsk = inet_csk(sk);
2429
2430 icsk->icsk_retransmits++;
2431 mptcp_set_datafin_timeout(sk);
2432 mptcp_send_ack(msk);
2433
2434 goto reset_timer;
2435 }
2436
3241a9c0
FW		 2437 if (!mptcp_send_head(sk))
2438 return;
2439
2440 goto reset_timer;
6477dd39 2441 }
3b1d6210 2442
3b1d6210 2443 if (!ssk)
2948d0a1 2444 goto reset_timer;
3b1d6210
PA		 2445
2446 lock_sock(ssk);
2447
d9ca1de8
PA		 2448 /* limit retransmission to the bytes already sent on some subflows */
2449 info.sent = 0;
4e14867d
PA		 2450 info.limit = READ_ONCE(msk->csum_enabled) ? dfrag->data_len : dfrag->already_sent;
2451 while (info.sent < info.limit) {
d9ca1de8 2452 ret = mptcp_sendmsg_frag(sk, ssk, dfrag, &info);
6f8a612a 2453 if (ret <= 0)
3b1d6210
PA		 2454 break;
2455
fc518953 2456 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_RETRANSSEGS);
3b1d6210 2457 copied += ret;
d9ca1de8 2458 info.sent += ret;
3b1d6210 2459 }
4e14867d
PA		 2460 if (copied) {
2461 dfrag->already_sent = max(dfrag->already_sent, info.sent);
caf971df
PA		 2462 tcp_push(ssk, 0, info.mss_now, tcp_sk(ssk)->nonagle,
2463 info.size_goal);
4e14867d 2464 }
3b1d6210 2465
3b1d6210
PA		 2466 release_sock(ssk);
2467
2948d0a1 2468reset_timer:
3241a9c0
FW		 2469 mptcp_check_and_set_pending(sk);
2470
3b1d6210
PA		 2471 if (!mptcp_timer_pending(sk))
2472 mptcp_reset_timer(sk);
2948d0a1
PA		 2473}
2474
2475static void mptcp_worker(struct work_struct *work)
2476{
2477 struct mptcp_sock *msk = container_of(work, struct mptcp_sock, work);
2478 struct sock *sk = &msk->sk.icsk_inet.sk;
2479 int state;
2480
2481 lock_sock(sk);
2482 state = sk->sk_state;
2483 if (unlikely(state == TCP_CLOSE))
2484 goto unlock;
2485
2486 mptcp_check_data_fin_ack(sk);
78962489 2487 mptcp_flush_join_list(msk);
2948d0a1
PA		 2488
2489 mptcp_check_fastclose(msk);
2490
2491 if (msk->pm.status)
2492 mptcp_pm_nl_work(msk);
2493
2494 if (test_and_clear_bit(MPTCP_WORK_EOF, &msk->flags))
2495 mptcp_check_for_eof(msk);
2496
2497 __mptcp_check_send_data_fin(sk);
2498 mptcp_check_data_fin(sk);
2499
2500 /* There is no point in keeping around an orphaned sk timedout or
2501 * closed, but we need the msk around to reply to incoming DATA_FIN,
2502 * even if it is orphaned and in FIN_WAIT2 state
2503 */
2504 if (sock_flag(sk, SOCK_DEAD) &&
2505 (mptcp_check_close_timeout(sk) || sk->sk_state == TCP_CLOSE)) {
2506 inet_sk_state_store(sk, TCP_CLOSE);
2507 __mptcp_destroy_sock(sk);
2508 goto unlock;
2509 }
2510
2511 if (test_and_clear_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags))
2512 __mptcp_close_subflow(msk);
2513
2514 if (test_and_clear_bit(MPTCP_WORK_RTX, &msk->flags))
2515 __mptcp_retrans(sk);
3b1d6210
PA		 2516
2517unlock:
80992017
PA		 2518 release_sock(sk);
2519 sock_put(sk);
2520}
2521
784325e9 2522static int __mptcp_init_sock(struct sock *sk)
f870fa0b 2523{
cec37a6e
PK		 2524 struct mptcp_sock *msk = mptcp_sk(sk);
2525
ec3edaa7
PK		 2526 spin_lock_init(&msk->join_list_lock);
2527
cec37a6e 2528 INIT_LIST_HEAD(&msk->conn_list);
ec3edaa7 2529 INIT_LIST_HEAD(&msk->join_list);
18b683bf 2530 INIT_LIST_HEAD(&msk->rtx_queue);
80992017 2531 INIT_WORK(&msk->work, mptcp_worker);
87952603 2532 __skb_queue_head_init(&msk->receive_queue);
ab174ad8 2533 msk->out_of_order_queue = RB_ROOT;
f0e6a4cf 2534 msk->first_pending = NULL;
6511882c 2535 msk->rmem_fwd_alloc = 0;
ce599c51 2536 WRITE_ONCE(msk->rmem_released, 0);
33d41c9c 2537 msk->timer_ival = TCP_RTO_MIN;
cec37a6e 2538
8ab183de 2539 msk->first = NULL;
dc24f8b4 2540 inet_csk(sk)->icsk_sync_mss = mptcp_sync_mss;
752e9067 2541 WRITE_ONCE(msk->csum_enabled, mptcp_is_checksum_enabled(sock_net(sk)));
1e1d9d6f 2542 msk->recovery = false;
8ab183de 2543
1b1c7a0e
PK		 2544 mptcp_pm_data_init(msk);
2545
b51f9b80
PA		 2546 /* re-use the csk retrans timer for MPTCP-level retrans */
2547 timer_setup(&msk->sk.icsk_retransmit_timer, mptcp_retransmit_timer, 0);
e16163b6 2548 timer_setup(&sk->sk_timer, mptcp_timeout_timer, 0);
aa1fbd94 2549
f870fa0b
MM		 2550 return 0;
2551}
2552
b29fcfb5 2553static void mptcp_ca_reset(struct sock *sk)
784325e9 2554{
20b5759f 2555 struct inet_connection_sock *icsk = inet_csk(sk);
b29fcfb5
PA		 2556
2557 tcp_assign_congestion_control(sk);
2558 strcpy(mptcp_sk(sk)->ca_name, icsk->icsk_ca_ops->name);
2559
2560 /* no need to keep a reference to the ops, the name will suffice */
2561 tcp_cleanup_congestion_control(sk);
2562 icsk->icsk_ca_ops = NULL;
2563}
2564
2565static int mptcp_init_sock(struct sock *sk)
2566{
fc518953
FW		 2567 struct net *net = sock_net(sk);
2568 int ret;
18b683bf 2569
b6c08380
GT		 2570 ret = __mptcp_init_sock(sk);
2571 if (ret)
2572 return ret;
2573
fc518953
FW		 2574 if (!mptcp_is_enabled(net))
2575 return -ENOPROTOOPT;
2576
2577 if (unlikely(!net->mib.mptcp_statistics) && !mptcp_mib_alloc(net))
2578 return -ENOMEM;
2579
fa68018d
PA		 2580 ret = __mptcp_socket_create(mptcp_sk(sk));
2581 if (ret)
2582 return ret;
2583
20b5759f
PA		 2584 /* fetch the ca name; do it outside __mptcp_init_sock(), so that clone will
2585 * propagate the correct value
2586 */
b29fcfb5 2587 mptcp_ca_reset(sk);
20b5759f 2588
d027236c 2589 sk_sockets_allocated_inc(sk);
a6b118fe 2590 sk->sk_rcvbuf = sock_net(sk)->ipv4.sysctl_tcp_rmem[1];
da51aef5 2591 sk->sk_sndbuf = sock_net(sk)->ipv4.sysctl_tcp_wmem[1];
d027236c 2592
18b683bf
PA		 2593 return 0;
2594}
2595
2596static void __mptcp_clear_xmit(struct sock *sk)
2597{
2598 struct mptcp_sock *msk = mptcp_sk(sk);
2599 struct mptcp_data_frag *dtmp, *dfrag;
2600
d9ca1de8 2601 WRITE_ONCE(msk->first_pending, NULL);
18b683bf 2602 list_for_each_entry_safe(dfrag, dtmp, &msk->rtx_queue, list)
d027236c 2603 dfrag_clear(sk, dfrag);
784325e9
MB		 2604}
2605
80992017
PA		 2606static void mptcp_cancel_work(struct sock *sk)
2607{
2608 struct mptcp_sock *msk = mptcp_sk(sk);
2609
b2771d24 2610 if (cancel_work_sync(&msk->work))
e16163b6 2611 __sock_put(sk);
80992017
PA		 2612}
2613
d0876b22 2614void mptcp_subflow_shutdown(struct sock *sk, struct sock *ssk, int how)
21498490
PK		 2615{
2616 lock_sock(ssk);
2617
2618 switch (ssk->sk_state) {
2619 case TCP_LISTEN:
2620 if (!(how & RCV_SHUTDOWN))
2621 break;
df561f66 2622 fallthrough;
21498490
PK		 2623 case TCP_SYN_SENT:
2624 tcp_disconnect(ssk, O_NONBLOCK);
2625 break;
2626 default:
43b54c6e
MM		 2627 if (__mptcp_check_fallback(mptcp_sk(sk))) {
2628 pr_debug("Fallback");
2629 ssk->sk_shutdown |= how;
2630 tcp_shutdown(ssk, how);
2631 } else {
2632 pr_debug("Sending DATA_FIN on subflow %p", ssk);
43b54c6e 2633 tcp_send_ack(ssk);
6477dd39
MM		 2634 if (!mptcp_timer_pending(sk))
2635 mptcp_reset_timer(sk);
43b54c6e 2636 }
21498490
PK		 2637 break;
2638 }
2639
21498490
PK		 2640 release_sock(ssk);
2641}
2642
6920b851
MM		 2643static const unsigned char new_state[16] = {
2644 /* current state: new state: action: */
2645 [0 /* (Invalid) */] = TCP_CLOSE,
2646 [TCP_ESTABLISHED] = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2647 [TCP_SYN_SENT] = TCP_CLOSE,
2648 [TCP_SYN_RECV] = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2649 [TCP_FIN_WAIT1] = TCP_FIN_WAIT1,
2650 [TCP_FIN_WAIT2] = TCP_FIN_WAIT2,
2651 [TCP_TIME_WAIT] = TCP_CLOSE, /* should not happen ! */
2652 [TCP_CLOSE] = TCP_CLOSE,
2653 [TCP_CLOSE_WAIT] = TCP_LAST_ACK | TCP_ACTION_FIN,
2654 [TCP_LAST_ACK] = TCP_LAST_ACK,
2655 [TCP_LISTEN] = TCP_CLOSE,
2656 [TCP_CLOSING] = TCP_CLOSING,
2657 [TCP_NEW_SYN_RECV] = TCP_CLOSE, /* should not happen ! */
2658};
2659
2660static int mptcp_close_state(struct sock *sk)
2661{
2662 int next = (int)new_state[sk->sk_state];
2663 int ns = next & TCP_STATE_MASK;
2664
2665 inet_sk_state_store(sk, ns);
2666
2667 return next & TCP_ACTION_FIN;
2668}
2669
e16163b6 2670static void __mptcp_check_send_data_fin(struct sock *sk)
f870fa0b 2671{
e16163b6 2672 struct mptcp_subflow_context *subflow;
f870fa0b
MM		 2673 struct mptcp_sock *msk = mptcp_sk(sk);
2674
e16163b6
PA		 2675 pr_debug("msk=%p snd_data_fin_enable=%d pending=%d snd_nxt=%llu write_seq=%llu",
2676 msk, msk->snd_data_fin_enable, !!mptcp_send_head(sk),
2677 msk->snd_nxt, msk->write_seq);
43b54c6e 2678
e16163b6
PA		 2679 /* we still need to enqueue subflows or not really shutting down,
2680 * skip this
2681 */
2682 if (!msk->snd_data_fin_enable || msk->snd_nxt + 1 != msk->write_seq ||
2683 mptcp_send_head(sk))
2684 return;
2685
2686 WRITE_ONCE(msk->snd_nxt, msk->write_seq);
2687
26aa2314
PA		 2688 /* fallback socket will not get data_fin/ack, can move to the next
2689 * state now
2690 */
2691 if (__mptcp_check_fallback(msk)) {
58cd405b 2692 WRITE_ONCE(msk->snd_una, msk->write_seq);
26aa2314
PA		 2693 if ((1 << sk->sk_state) & (TCPF_CLOSING | TCPF_LAST_ACK)) {
2694 inet_sk_state_store(sk, TCP_CLOSE);
2695 mptcp_close_wake_up(sk);
2696 } else if (sk->sk_state == TCP_FIN_WAIT1) {
2697 inet_sk_state_store(sk, TCP_FIN_WAIT2);
2698 }
43b54c6e
MM		 2699 }
2700
78962489 2701 mptcp_flush_join_list(msk);
e16163b6
PA		 2702 mptcp_for_each_subflow(msk, subflow) {
2703 struct sock *tcp_sk = mptcp_subflow_tcp_sock(subflow);
43b54c6e 2704
e16163b6 2705 mptcp_subflow_shutdown(sk, tcp_sk, SEND_SHUTDOWN);
43b54c6e 2706 }
e16163b6 2707}
2c22c06c 2708
e16163b6
PA		 2709static void __mptcp_wr_shutdown(struct sock *sk)
2710{
2711 struct mptcp_sock *msk = mptcp_sk(sk);
43b54c6e 2712
e16163b6
PA		 2713 pr_debug("msk=%p snd_data_fin_enable=%d shutdown=%x state=%d pending=%d",
2714 msk, msk->snd_data_fin_enable, sk->sk_shutdown, sk->sk_state,
2715 !!mptcp_send_head(sk));
2716
2717 /* will be ignored by fallback sockets */
2718 WRITE_ONCE(msk->write_seq, msk->write_seq + 1);
2719 WRITE_ONCE(msk->snd_data_fin_enable, 1);
2720
2721 __mptcp_check_send_data_fin(sk);
2722}
2723
2724static void __mptcp_destroy_sock(struct sock *sk)
2725{
2726 struct mptcp_subflow_context *subflow, *tmp;
2727 struct mptcp_sock *msk = mptcp_sk(sk);
2728 LIST_HEAD(conn_list);
2729
2730 pr_debug("msk=%p", msk);
f870fa0b 2731
3abc05d9
FW		 2732 might_sleep();
2733
10f6d46c
PA		 2734 /* be sure to always acquire the join list lock, to sync vs
2735 * mptcp_finish_join().
2736 */
2737 spin_lock_bh(&msk->join_list_lock);
2738 list_splice_tail_init(&msk->join_list, &msk->conn_list);
2739 spin_unlock_bh(&msk->join_list_lock);
b2c5b614
FW		 2740 list_splice_init(&msk->conn_list, &conn_list);
2741
6e628cd3 2742 sk_stop_timer(sk, &msk->sk.icsk_retransmit_timer);
e16163b6
PA		 2743 sk_stop_timer(sk, &sk->sk_timer);
2744 msk->pm.status = 0;
b2c5b614 2745
b29fcfb5
PA		 2746 /* clears msk->subflow, allowing the following loop to close
2747 * even the initial subflow
2748 */
2749 mptcp_dispose_initial_subflow(msk);
b2c5b614 2750 list_for_each_entry_safe(subflow, tmp, &conn_list, node) {
cec37a6e 2751 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
b29fcfb5 2752 __mptcp_close_ssk(sk, ssk, subflow, 0);
f870fa0b
MM		 2753 }
2754
e16163b6 2755 sk->sk_prot->destroy(sk);
80992017 2756
6511882c 2757 WARN_ON_ONCE(msk->rmem_fwd_alloc);
87952603 2758 WARN_ON_ONCE(msk->rmem_released);
e16163b6
PA		 2759 sk_stream_kill_queues(sk);
2760 xfrm_sk_free_policy(sk);
aa1fbd94 2761
e16163b6
PA		 2762 sk_refcnt_debug_release(sk);
2763 sock_put(sk);
2764}
2765
2766static void mptcp_close(struct sock *sk, long timeout)
2767{
2768 struct mptcp_subflow_context *subflow;
2769 bool do_cancel_work = false;
2770
2771 lock_sock(sk);
2772 sk->sk_shutdown = SHUTDOWN_MASK;
2773
2774 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE)) {
2775 inet_sk_state_store(sk, TCP_CLOSE);
2776 goto cleanup;
2777 }
6771bfd9 2778
e16163b6
PA		 2779 if (mptcp_close_state(sk))
2780 __mptcp_wr_shutdown(sk);
2781
2782 sk_stream_wait_close(sk, timeout);
2783
2784cleanup:
2785 /* orphan all the subflows */
2786 inet_csk(sk)->icsk_mtup.probe_timestamp = tcp_jiffies32;
44227915 2787 mptcp_for_each_subflow(mptcp_sk(sk), subflow) {
e16163b6 2788 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
49054556 2789 bool slow = lock_sock_fast_nested(ssk);
e16163b6 2790
e16163b6
PA		 2791 sock_orphan(ssk);
2792 unlock_sock_fast(ssk, slow);
e16163b6
PA		 2793 }
2794 sock_orphan(sk);
2795
2796 sock_hold(sk);
2797 pr_debug("msk=%p state=%d", sk, sk->sk_state);
b29fcfb5
PA		 2798 if (mptcp_sk(sk)->token)
2799 mptcp_event(MPTCP_EVENT_CLOSED, mptcp_sk(sk), NULL, GFP_KERNEL);
2800
e16163b6
PA		 2801 if (sk->sk_state == TCP_CLOSE) {
2802 __mptcp_destroy_sock(sk);
2803 do_cancel_work = true;
2804 } else {
2805 sk_reset_timer(sk, &sk->sk_timer, jiffies + TCP_TIMEWAIT_LEN);
2806 }
2807 release_sock(sk);
2808 if (do_cancel_work)
2809 mptcp_cancel_work(sk);
b911c97c 2810
e16163b6 2811 sock_put(sk);
f870fa0b
MM		 2812}
2813
cf7da0d6
PK		 2814static void mptcp_copy_inaddrs(struct sock *msk, const struct sock *ssk)
2815{
2816#if IS_ENABLED(CONFIG_MPTCP_IPV6)
2817 const struct ipv6_pinfo *ssk6 = inet6_sk(ssk);
2818 struct ipv6_pinfo *msk6 = inet6_sk(msk);
2819
2820 msk->sk_v6_daddr = ssk->sk_v6_daddr;
2821 msk->sk_v6_rcv_saddr = ssk->sk_v6_rcv_saddr;
2822
2823 if (msk6 && ssk6) {
2824 msk6->saddr = ssk6->saddr;
2825 msk6->flow_label = ssk6->flow_label;
2826 }
2827#endif
2828
2829 inet_sk(msk)->inet_num = inet_sk(ssk)->inet_num;
2830 inet_sk(msk)->inet_dport = inet_sk(ssk)->inet_dport;
2831 inet_sk(msk)->inet_sport = inet_sk(ssk)->inet_sport;
2832 inet_sk(msk)->inet_daddr = inet_sk(ssk)->inet_daddr;
2833 inet_sk(msk)->inet_saddr = inet_sk(ssk)->inet_saddr;
2834 inet_sk(msk)->inet_rcv_saddr = inet_sk(ssk)->inet_rcv_saddr;
2835}
2836
18b683bf
PA		 2837static int mptcp_disconnect(struct sock *sk, int flags)
2838{
76e2a55d
PA		 2839 struct mptcp_subflow_context *subflow;
2840 struct mptcp_sock *msk = mptcp_sk(sk);
2841
78962489
FW		 2842 mptcp_do_flush_join_list(msk);
2843
b29fcfb5
PA		 2844 inet_sk_state_store(sk, TCP_CLOSE);
2845
13a9499e
PA		 2846 mptcp_for_each_subflow(msk, subflow) {
2847 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2848
b29fcfb5 2849 __mptcp_close_ssk(sk, ssk, subflow, MPTCP_CF_FASTCLOSE);
13a9499e 2850 }
b29fcfb5
PA		 2851
2852 sk_stop_timer(sk, &msk->sk.icsk_retransmit_timer);
2853 sk_stop_timer(sk, &sk->sk_timer);
2854
2855 if (mptcp_sk(sk)->token)
2856 mptcp_event(MPTCP_EVENT_CLOSED, mptcp_sk(sk), NULL, GFP_KERNEL);
2857
2858 mptcp_destroy_common(msk);
2859 msk->last_snd = NULL;
2860 msk->flags = 0;
2861 msk->recovery = false;
2862 msk->can_ack = false;
2863 msk->fully_established = false;
2864 msk->rcv_data_fin = false;
2865 msk->snd_data_fin_enable = false;
2866 msk->rcv_fastclose = false;
2867 msk->use_64bit_ack = false;
2868 WRITE_ONCE(msk->csum_enabled, mptcp_is_checksum_enabled(sock_net(sk)));
2869 mptcp_pm_data_reset(msk);
2870 mptcp_ca_reset(sk);
2871
2872 sk->sk_shutdown = 0;
2873 sk_error_report(sk);
42c556fe 2874 return 0;
18b683bf
PA		 2875}
2876
b0519de8
FW		 2877#if IS_ENABLED(CONFIG_MPTCP_IPV6)
2878static struct ipv6_pinfo *mptcp_inet6_sk(const struct sock *sk)
2879{
2880 unsigned int offset = sizeof(struct mptcp6_sock) - sizeof(struct ipv6_pinfo);
2881
2882 return (struct ipv6_pinfo *)(((u8 *)sk) + offset);
2883}
2884#endif
2885
fca5c82c 2886struct sock *mptcp_sk_clone(const struct sock *sk,
cfde141e 2887 const struct mptcp_options_received *mp_opt,
fca5c82c 2888 struct request_sock *req)
b0519de8 2889{
58b09919 2890 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
b0519de8 2891 struct sock *nsk = sk_clone_lock(sk, GFP_ATOMIC);
58b09919
PA		 2892 struct mptcp_sock *msk;
2893 u64 ack_seq;
b0519de8
FW		 2894
2895 if (!nsk)
2896 return NULL;
2897
2898#if IS_ENABLED(CONFIG_MPTCP_IPV6)
2899 if (nsk->sk_family == AF_INET6)
2900 inet_sk(nsk)->pinet6 = mptcp_inet6_sk(nsk);
2901#endif
2902
58b09919
PA		 2903 __mptcp_init_sock(nsk);
2904
2905 msk = mptcp_sk(nsk);
2906 msk->local_key = subflow_req->local_key;
2907 msk->token = subflow_req->token;
2908 msk->subflow = NULL;
b93df08c 2909 WRITE_ONCE(msk->fully_established, false);
74c7dfbe 2910 if (mp_opt->suboptions & OPTION_MPTCP_CSUMREQD)
06251181 2911 WRITE_ONCE(msk->csum_enabled, true);
58b09919 2912
58b09919 2913 msk->write_seq = subflow_req->idsn + 1;
eaa2ffab 2914 msk->snd_nxt = msk->write_seq;
7439d687
PA		 2915 msk->snd_una = msk->write_seq;
2916 msk->wnd_end = msk->snd_nxt + req->rsk_rcv_wnd;
df00b087 2917 msk->setsockopt_seq = mptcp_sk(sk)->setsockopt_seq;
6f8a612a 2918
74c7dfbe 2919 if (mp_opt->suboptions & OPTIONS_MPTCP_MPC) {
58b09919 2920 msk->can_ack = true;
cfde141e 2921 msk->remote_key = mp_opt->sndr_key;
58b09919
PA		 2922 mptcp_crypto_key_sha(msk->remote_key, NULL, &ack_seq);
2923 ack_seq++;
917944da 2924 WRITE_ONCE(msk->ack_seq, ack_seq);
fa3fe2b1 2925 WRITE_ONCE(msk->rcv_wnd_sent, ack_seq);
58b09919 2926 }
7f20d5fc 2927
5e20087d 2928 sock_reset_flag(nsk, SOCK_RCU_FREE);
7f20d5fc
PA		 2929 /* will be fully established after successful MPC subflow creation */
2930 inet_sk_state_store(nsk, TCP_SYN_RECV);
0c148460
PA		 2931
2932 security_inet_csk_clone(nsk, req);
58b09919
PA		 2933 bh_unlock_sock(nsk);
2934
2935 /* keep a single reference */
2936 __sock_put(nsk);
b0519de8
FW		 2937 return nsk;
2938}
2939
a6b118fe
FW		 2940void mptcp_rcv_space_init(struct mptcp_sock *msk, const struct sock *ssk)
2941{
2942 const struct tcp_sock *tp = tcp_sk(ssk);
2943
2944 msk->rcvq_space.copied = 0;
2945 msk->rcvq_space.rtt_us = 0;
2946
2947 msk->rcvq_space.time = tp->tcp_mstamp;
2948
2949 /* initial rcv_space offering made to peer */
2950 msk->rcvq_space.space = min_t(u32, tp->rcv_wnd,
2951 TCP_INIT_CWND * tp->advmss);
2952 if (msk->rcvq_space.space == 0)
2953 msk->rcvq_space.space = TCP_INIT_CWND * TCP_MSS_DEFAULT;
6f8a612a 2954
7439d687 2955 WRITE_ONCE(msk->wnd_end, msk->snd_nxt + tcp_sk(ssk)->snd_wnd);
a6b118fe
FW		 2956}
2957
cf7da0d6
PK		 2958static struct sock *mptcp_accept(struct sock *sk, int flags, int *err,
2959 bool kern)
2960{
2961 struct mptcp_sock *msk = mptcp_sk(sk);
2962 struct socket *listener;
2963 struct sock *newsk;
2964
2965 listener = __mptcp_nmpc_socket(msk);
2966 if (WARN_ON_ONCE(!listener)) {
2967 *err = -EINVAL;
2968 return NULL;
2969 }
2970
2971 pr_debug("msk=%p, listener=%p", msk, mptcp_subflow_ctx(listener->sk));
2972 newsk = inet_csk_accept(listener->sk, flags, err, kern);
2973 if (!newsk)
2974 return NULL;
2975
2976 pr_debug("msk=%p, subflow is mptcp=%d", msk, sk_is_mptcp(newsk));
cf7da0d6
PK		 2977 if (sk_is_mptcp(newsk)) {
2978 struct mptcp_subflow_context *subflow;
2979 struct sock *new_mptcp_sock;
cf7da0d6
PK		 2980
2981 subflow = mptcp_subflow_ctx(newsk);
58b09919 2982 new_mptcp_sock = subflow->conn;
cf7da0d6 2983
58b09919
PA		 2984 /* is_mptcp should be false if subflow->conn is missing, see
2985 * subflow_syn_recv_sock()
2986 */
2987 if (WARN_ON_ONCE(!new_mptcp_sock)) {
2988 tcp_sk(newsk)->is_mptcp = 0;
d6692b3b 2989 goto out;
cf7da0d6
PK		 2990 }
2991
58b09919
PA		 2992 /* acquire the 2nd reference for the owning socket */
2993 sock_hold(new_mptcp_sock);
cf7da0d6 2994 newsk = new_mptcp_sock;
0397c6d8 2995 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPCAPABLEPASSIVEACK);
fc518953
FW		 2996 } else {
2997 MPTCP_INC_STATS(sock_net(sk),
2998 MPTCP_MIB_MPCAPABLEPASSIVEFALLBACK);
cf7da0d6
PK		 2999 }
3000
d6692b3b
FW		 3001out:
3002 newsk->sk_kern_sock = kern;
cf7da0d6
PK		 3003 return newsk;
3004}
3005
5c8c1640
GT		 3006void mptcp_destroy_common(struct mptcp_sock *msk)
3007{
87952603
PA		 3008 struct sock *sk = (struct sock *)msk;
3009
6e628cd3
PA		 3010 __mptcp_clear_xmit(sk);
3011
87952603 3012 /* move to sk_receive_queue, sk_stream_kill_queues will purge it */
b29fcfb5 3013 mptcp_data_lock(sk);
87952603 3014 skb_queue_splice_tail_init(&msk->receive_queue, &sk->sk_receive_queue);
6511882c 3015 __skb_queue_purge(&sk->sk_receive_queue);
5c8c1640 3016 skb_rbtree_purge(&msk->out_of_order_queue);
b29fcfb5 3017 mptcp_data_unlock(sk);
6511882c
PA		 3018
3019 /* move all the rx fwd alloc into the sk_mem_reclaim_final in
3020 * inet_sock_destruct() will dispose it
3021 */
3022 sk->sk_forward_alloc += msk->rmem_fwd_alloc;
3023 msk->rmem_fwd_alloc = 0;
5c8c1640
GT		 3024 mptcp_token_destroy(msk);
3025 mptcp_pm_free_anno_list(msk);
3026}
3027
79c0949e
PK		 3028static void mptcp_destroy(struct sock *sk)
3029{
c9fd9c5f
FW		 3030 struct mptcp_sock *msk = mptcp_sk(sk);
3031
5c8c1640 3032 mptcp_destroy_common(msk);
d027236c 3033 sk_sockets_allocated_dec(sk);
79c0949e
PK		 3034}
3035
6e628cd3
PA		 3036void __mptcp_data_acked(struct sock *sk)
3037{
3038 if (!sock_owned_by_user(sk))
3039 __mptcp_clean_una(sk);
3040 else
3041 set_bit(MPTCP_CLEAN_UNA, &mptcp_sk(sk)->flags);
3042
3043 if (mptcp_pending_data_fin_ack(sk))
3044 mptcp_schedule_work(sk);
3045}
3046
219d0499 3047void __mptcp_check_push(struct sock *sk, struct sock *ssk)
6e628cd3
PA		 3048{
3049 if (!mptcp_send_head(sk))
3050 return;
3051
40dc9416 3052 if (!sock_owned_by_user(sk)) {
b19bc294
PA		 3053 struct sock *xmit_ssk = mptcp_subflow_get_send(mptcp_sk(sk));
3054
3055 if (xmit_ssk == ssk)
40dc9416 3056 __mptcp_subflow_push_pending(sk, ssk);
b19bc294 3057 else if (xmit_ssk)
bcd97734 3058 mptcp_subflow_delegate(mptcp_subflow_ctx(xmit_ssk), MPTCP_DELEGATE_SEND);
40dc9416 3059 } else {
6e628cd3 3060 set_bit(MPTCP_PUSH_PENDING, &mptcp_sk(sk)->flags);
40dc9416 3061 }
6e628cd3
PA		 3062}
3063
e93da928 3064/* processes deferred events and flush wmem */
14c441b5
PA		 3065static void mptcp_release_cb(struct sock *sk)
3066{
c2e6048f 3067 for (;;) {
2d6f5a2b
PA		 3068 unsigned long flags = 0;
3069
c2e6048f 3070 if (test_and_clear_bit(MPTCP_PUSH_PENDING, &mptcp_sk(sk)->flags))
2e5de7e0 3071 flags |= BIT(MPTCP_PUSH_PENDING);
2d6f5a2b
PA		 3072 if (test_and_clear_bit(MPTCP_RETRANSMIT, &mptcp_sk(sk)->flags))
3073 flags |= BIT(MPTCP_RETRANSMIT);
c2e6048f
PA		 3074 if (!flags)
3075 break;
3076
3077 /* the following actions acquire the subflow socket lock
6e628cd3
PA		 3078 *
3079 * 1) can't be invoked in atomic scope
3080 * 2) must avoid ABBA deadlock with msk socket spinlock: the RX
3081 * datapath acquires the msk socket spinlock while helding
3082 * the subflow socket lock
3083 */
3084
3085 spin_unlock_bh(&sk->sk_lock.slock);
2e5de7e0 3086 if (flags & BIT(MPTCP_PUSH_PENDING))
c2e6048f 3087 __mptcp_push_pending(sk, 0);
2d6f5a2b
PA		 3088 if (flags & BIT(MPTCP_RETRANSMIT))
3089 __mptcp_retrans(sk);
c2e6048f
PA		 3090
3091 cond_resched();
6e628cd3
PA		 3092 spin_lock_bh(&sk->sk_lock.slock);
3093 }
c2e6048f 3094
490274b4
PA		 3095 /* be sure to set the current sk state before tacking actions
3096 * depending on sk_state
3097 */
3098 if (test_and_clear_bit(MPTCP_CONNECTED, &mptcp_sk(sk)->flags))
3099 __mptcp_set_connected(sk);
c2e6048f 3100 if (test_and_clear_bit(MPTCP_CLEAN_UNA, &mptcp_sk(sk)->flags))
417789df 3101 __mptcp_clean_una_wakeup(sk);
15cc1045
PA		 3102 if (test_and_clear_bit(MPTCP_ERROR_REPORT, &mptcp_sk(sk)->flags))
3103 __mptcp_error_report(sk);
6e628cd3 3104
87952603 3105 __mptcp_update_rmem(sk);
14c441b5
PA		 3106}
3107
bcd97734
PA		 3108/* MP_JOIN client subflow must wait for 4th ack before sending any data:
3109 * TCP can't schedule delack timer before the subflow is fully established.
3110 * MPTCP uses the delack timer to do 3rd ack retransmissions
3111 */
3112static void schedule_3rdack_retransmission(struct sock *ssk)
3113{
3114 struct inet_connection_sock *icsk = inet_csk(ssk);
3115 struct tcp_sock *tp = tcp_sk(ssk);
3116 unsigned long timeout;
3117
3118 if (mptcp_subflow_ctx(ssk)->fully_established)
3119 return;
3120
3121 /* reschedule with a timeout above RTT, as we must look only for drop */
3122 if (tp->srtt_us)
3123 timeout = usecs_to_jiffies(tp->srtt_us >> (3 - 1));
3124 else
3125 timeout = TCP_TIMEOUT_INIT;
3126 timeout += jiffies;
3127
3128 WARN_ON_ONCE(icsk->icsk_ack.pending & ICSK_ACK_TIMER);
3129 icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
3130 icsk->icsk_ack.timeout = timeout;
3131 sk_reset_timer(ssk, &icsk->icsk_delack_timer, timeout);
3132}
3133
b19bc294
PA		 3134void mptcp_subflow_process_delegated(struct sock *ssk)
3135{
3136 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
3137 struct sock *sk = subflow->conn;
3138
bcd97734
PA		 3139 if (test_bit(MPTCP_DELEGATE_SEND, &subflow->delegated_status)) {
3140 mptcp_data_lock(sk);
3141 if (!sock_owned_by_user(sk))
3142 __mptcp_subflow_push_pending(sk, ssk);
3143 else
3144 set_bit(MPTCP_PUSH_PENDING, &mptcp_sk(sk)->flags);
3145 mptcp_data_unlock(sk);
3146 mptcp_subflow_delegated_done(subflow, MPTCP_DELEGATE_SEND);
3147 }
3148 if (test_bit(MPTCP_DELEGATE_ACK, &subflow->delegated_status)) {
3149 schedule_3rdack_retransmission(ssk);
3150 mptcp_subflow_delegated_done(subflow, MPTCP_DELEGATE_ACK);
3151 }
b19bc294
PA		 3152}
3153
2c5ebd00
PA		 3154static int mptcp_hash(struct sock *sk)
3155{
3156 /* should never be called,
3157 * we hash the TCP subflows not the master socket
3158 */
3159 WARN_ON_ONCE(1);
3160 return 0;
3161}
3162
3163static void mptcp_unhash(struct sock *sk)
3164{
3165 /* called from sk_common_release(), but nothing to do here */
3166}
3167
cec37a6e 3168static int mptcp_get_port(struct sock *sk, unsigned short snum)
f870fa0b
MM		 3169{
3170 struct mptcp_sock *msk = mptcp_sk(sk);
cec37a6e 3171 struct socket *ssock;
f870fa0b 3172
cec37a6e
PK		 3173 ssock = __mptcp_nmpc_socket(msk);
3174 pr_debug("msk=%p, subflow=%p", msk, ssock);
3175 if (WARN_ON_ONCE(!ssock))
3176 return -EINVAL;
f870fa0b 3177
cec37a6e
PK		 3178 return inet_csk_get_port(ssock->sk, snum);
3179}
f870fa0b 3180
cec37a6e
PK		 3181void mptcp_finish_connect(struct sock *ssk)
3182{
3183 struct mptcp_subflow_context *subflow;
3184 struct mptcp_sock *msk;
3185 struct sock *sk;
6d0060f6 3186 u64 ack_seq;
f870fa0b 3187
cec37a6e 3188 subflow = mptcp_subflow_ctx(ssk);
cec37a6e
PK		 3189 sk = subflow->conn;
3190 msk = mptcp_sk(sk);
3191
648ef4b8
MM		 3192 pr_debug("msk=%p, token=%u", sk, subflow->token);
3193
6d0060f6
MM		 3194 mptcp_crypto_key_sha(subflow->remote_key, NULL, &ack_seq);
3195 ack_seq++;
648ef4b8
MM		 3196 subflow->map_seq = ack_seq;
3197 subflow->map_subflow_seq = 1;
6d0060f6 3198
cec37a6e
PK		 3199 /* the socket is not connected yet, no msk/subflow ops can access/race
3200 * accessing the field below
3201 */
3202 WRITE_ONCE(msk->remote_key, subflow->remote_key);
3203 WRITE_ONCE(msk->local_key, subflow->local_key);
6d0060f6 3204 WRITE_ONCE(msk->write_seq, subflow->idsn + 1);
eaa2ffab 3205 WRITE_ONCE(msk->snd_nxt, msk->write_seq);
6d0060f6 3206 WRITE_ONCE(msk->ack_seq, ack_seq);
fa3fe2b1 3207 WRITE_ONCE(msk->rcv_wnd_sent, ack_seq);
d22f4988 3208 WRITE_ONCE(msk->can_ack, 1);
7439d687 3209 WRITE_ONCE(msk->snd_una, msk->write_seq);
1b1c7a0e 3210
6c714f1b 3211 mptcp_pm_new_connection(msk, ssk, 0);
a6b118fe
FW		 3212
3213 mptcp_rcv_space_init(msk, ssk);
f870fa0b
MM		 3214}
3215
866f26f2 3216void mptcp_sock_graft(struct sock *sk, struct socket *parent)
cf7da0d6
PK		 3217{
3218 write_lock_bh(&sk->sk_callback_lock);
3219 rcu_assign_pointer(sk->sk_wq, &parent->wq);
3220 sk_set_socket(sk, parent);
3221 sk->sk_uid = SOCK_INODE(parent)->i_uid;
3222 write_unlock_bh(&sk->sk_callback_lock);
3223}
3224
e16163b6 3225bool mptcp_finish_join(struct sock *ssk)
f296234c 3226{
e16163b6 3227 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
f296234c
PK		 3228 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
3229 struct sock *parent = (void *)msk;
3230 struct socket *parent_sock;
ec3edaa7 3231 bool ret;
f296234c
PK		 3232
3233 pr_debug("msk=%p, subflow=%p", msk, subflow);
3234
3235 /* mptcp socket already closing? */
dc87efdb
FW		 3236 if (!mptcp_is_fully_established(parent)) {
3237 subflow->reset_reason = MPTCP_RST_EMPTCP;
f296234c 3238 return false;
dc87efdb 3239 }
f296234c
PK		 3240
3241 if (!msk->pm.server_side)
b911c97c 3242 goto out;
f296234c 3243
dc87efdb
FW		 3244 if (!mptcp_pm_allow_new_subflow(msk)) {
3245 subflow->reset_reason = MPTCP_RST_EPROHIBIT;
10f6d46c 3246 return false;
dc87efdb 3247 }
10f6d46c
PA		 3248
3249 /* active connections are already on conn_list, and we can't acquire
3250 * msk lock here.
3251 * use the join list lock as synchronization point and double-check
e16163b6 3252 * msk status to avoid racing with __mptcp_destroy_sock()
10f6d46c
PA		 3253 */
3254 spin_lock_bh(&msk->join_list_lock);
3255 ret = inet_sk_state_load(parent) == TCP_ESTABLISHED;
e16163b6 3256 if (ret && !WARN_ON_ONCE(!list_empty(&subflow->node))) {
10f6d46c 3257 list_add_tail(&subflow->node, &msk->join_list);
e16163b6
PA		 3258 sock_hold(ssk);
3259 }
10f6d46c 3260 spin_unlock_bh(&msk->join_list_lock);
dc87efdb
FW		 3261 if (!ret) {
3262 subflow->reset_reason = MPTCP_RST_EPROHIBIT;
10f6d46c 3263 return false;
dc87efdb 3264 }
10f6d46c
PA		 3265
3266 /* attach to msk socket only after we are sure he will deal with us
3267 * at close time
3268 */
f296234c 3269 parent_sock = READ_ONCE(parent->sk_socket);
e16163b6
PA		 3270 if (parent_sock && !ssk->sk_socket)
3271 mptcp_sock_graft(ssk, parent_sock);
917944da 3272 subflow->map_seq = READ_ONCE(msk->ack_seq);
b911c97c
FW		 3273out:
3274 mptcp_event(MPTCP_EVENT_SUB_ESTABLISHED, msk, ssk, GFP_ATOMIC);
10f6d46c 3275 return true;
f296234c
PK		 3276}
3277
76e2a55d
PA		 3278static void mptcp_shutdown(struct sock *sk, int how)
3279{
3280 pr_debug("sk=%p, how=%d", sk, how);
3281
3282 if ((how & SEND_SHUTDOWN) && mptcp_close_state(sk))
3283 __mptcp_wr_shutdown(sk);
3284}
3285
6511882c
PA		 3286static int mptcp_forward_alloc_get(const struct sock *sk)
3287{
3288 return sk->sk_forward_alloc + mptcp_sk(sk)->rmem_fwd_alloc;
3289}
3290
644807e3
FW		 3291static int mptcp_ioctl_outq(const struct mptcp_sock *msk, u64 v)
3292{
3293 const struct sock *sk = (void *)msk;
3294 u64 delta;
3295
3296 if (sk->sk_state == TCP_LISTEN)
3297 return -EINVAL;
3298
3299 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
3300 return 0;
3301
3302 delta = msk->write_seq - v;
3303 if (delta > INT_MAX)
3304 delta = INT_MAX;
3305
3306 return (int)delta;
3307}
3308
3309static int mptcp_ioctl(struct sock *sk, int cmd, unsigned long arg)
3310{
3311 struct mptcp_sock *msk = mptcp_sk(sk);
3312 bool slow;
3313 int answ;
3314
3315 switch (cmd) {
3316 case SIOCINQ:
3317 if (sk->sk_state == TCP_LISTEN)
3318 return -EINVAL;
3319
3320 lock_sock(sk);
3321 __mptcp_move_skbs(msk);
3322 answ = mptcp_inq_hint(sk);
3323 release_sock(sk);
3324 break;
3325 case SIOCOUTQ:
3326 slow = lock_sock_fast(sk);
3327 answ = mptcp_ioctl_outq(msk, READ_ONCE(msk->snd_una));
3328 unlock_sock_fast(sk, slow);
3329 break;
3330 case SIOCOUTQNSD:
3331 slow = lock_sock_fast(sk);
3332 answ = mptcp_ioctl_outq(msk, msk->snd_nxt);
3333 unlock_sock_fast(sk, slow);
3334 break;
3335 default:
3336 return -ENOIOCTLCMD;
3337 }
3338
3339 return put_user(answ, (int __user *)arg);
3340}
3341
f870fa0b
MM		 3342static struct proto mptcp_prot = {
3343 .name = "MPTCP",
3344 .owner = THIS_MODULE,
3345 .init = mptcp_init_sock,
18b683bf 3346 .disconnect = mptcp_disconnect,
f870fa0b 3347 .close = mptcp_close,
cf7da0d6 3348 .accept = mptcp_accept,
717e79c8
PK		 3349 .setsockopt = mptcp_setsockopt,
3350 .getsockopt = mptcp_getsockopt,
76e2a55d 3351 .shutdown = mptcp_shutdown,
79c0949e 3352 .destroy = mptcp_destroy,
f870fa0b 3353 .sendmsg = mptcp_sendmsg,
644807e3 3354 .ioctl = mptcp_ioctl,
f870fa0b 3355 .recvmsg = mptcp_recvmsg,
14c441b5 3356 .release_cb = mptcp_release_cb,
2c5ebd00
PA		 3357 .hash = mptcp_hash,
3358 .unhash = mptcp_unhash,
cec37a6e 3359 .get_port = mptcp_get_port,
6511882c 3360 .forward_alloc_get = mptcp_forward_alloc_get,
d027236c
PA		 3361 .sockets_allocated = &mptcp_sockets_allocated,
3362 .memory_allocated = &tcp_memory_allocated,
3363 .memory_pressure = &tcp_memory_pressure,
d027236c 3364 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_tcp_wmem),
989ef49b 3365 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_tcp_rmem),
d027236c 3366 .sysctl_mem = sysctl_tcp_mem,
f870fa0b 3367 .obj_size = sizeof(struct mptcp_sock),
2c5ebd00 3368 .slab_flags = SLAB_TYPESAFE_BY_RCU,
f870fa0b
MM		 3369 .no_autobind = true,
3370};
3371
2303f994
PK		 3372static int mptcp_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3373{
3374 struct mptcp_sock *msk = mptcp_sk(sock->sk);
3375 struct socket *ssock;
cf7da0d6 3376 int err;
2303f994
PK		 3377
3378 lock_sock(sock->sk);
fa68018d
PA		 3379 ssock = __mptcp_nmpc_socket(msk);
3380 if (!ssock) {
3381 err = -EINVAL;
2303f994
PK		 3382 goto unlock;
3383 }
3384
3385 err = ssock->ops->bind(ssock, uaddr, addr_len);
cf7da0d6
PK		 3386 if (!err)
3387 mptcp_copy_inaddrs(sock->sk, ssock->sk);
2303f994
PK		 3388
3389unlock:
3390 release_sock(sock->sk);
3391 return err;
3392}
3393
0235d075
PA		 3394static void mptcp_subflow_early_fallback(struct mptcp_sock *msk,
3395 struct mptcp_subflow_context *subflow)
3396{
3397 subflow->request_mptcp = 0;
3398 __mptcp_do_fallback(msk);
3399}
3400
2303f994
PK		 3401static int mptcp_stream_connect(struct socket *sock, struct sockaddr *uaddr,
3402 int addr_len, int flags)
3403{
3404 struct mptcp_sock *msk = mptcp_sk(sock->sk);
2c5ebd00 3405 struct mptcp_subflow_context *subflow;
2303f994
PK		 3406 struct socket *ssock;
3407 int err;
3408
3409 lock_sock(sock->sk);
41be81a8
PA		 3410 if (sock->state != SS_UNCONNECTED && msk->subflow) {
3411 /* pending connection or invalid state, let existing subflow
3412 * cope with that
3413 */
3414 ssock = msk->subflow;
3415 goto do_connect;
3416 }
3417
fa68018d
PA		 3418 ssock = __mptcp_nmpc_socket(msk);
3419 if (!ssock) {
3420 err = -EINVAL;
2303f994
PK		 3421 goto unlock;
3422 }
3423
fa68018d
PA		 3424 mptcp_token_destroy(msk);
3425 inet_sk_state_store(sock->sk, TCP_SYN_SENT);
2c5ebd00 3426 subflow = mptcp_subflow_ctx(ssock->sk);
cf7da0d6
PK		 3427#ifdef CONFIG_TCP_MD5SIG
3428 /* no MPTCP if MD5SIG is enabled on this socket or we may run out of
3429 * TCP option space.
3430 */
3431 if (rcu_access_pointer(tcp_sk(ssock->sk)->md5sig_info))
0235d075 3432 mptcp_subflow_early_fallback(msk, subflow);
cf7da0d6 3433#endif
a16195e3
PA		 3434 if (subflow->request_mptcp && mptcp_token_new_connect(ssock->sk)) {
3435 MPTCP_INC_STATS(sock_net(ssock->sk), MPTCP_MIB_TOKENFALLBACKINIT);
0235d075 3436 mptcp_subflow_early_fallback(msk, subflow);
a16195e3 3437 }
5695eb88
PA		 3438 if (likely(!__mptcp_check_fallback(msk)))
3439 MPTCP_INC_STATS(sock_net(sock->sk), MPTCP_MIB_MPCAPABLEACTIVE);
cf7da0d6 3440
41be81a8 3441do_connect:
2303f994 3442 err = ssock->ops->connect(ssock, uaddr, addr_len, flags);
41be81a8
PA		 3443 sock->state = ssock->state;
3444
3445 /* on successful connect, the msk state will be moved to established by
3446 * subflow_finish_connect()
3447 */
367fe04e 3448 if (!err || err == -EINPROGRESS)
41be81a8
PA		 3449 mptcp_copy_inaddrs(sock->sk, ssock->sk);
3450 else
3451 inet_sk_state_store(sock->sk, inet_sk_state_load(ssock->sk));
2303f994
PK		 3452
3453unlock:
3454 release_sock(sock->sk);
3455 return err;
3456}
3457
cf7da0d6
PK		 3458static int mptcp_listen(struct socket *sock, int backlog)
3459{
3460 struct mptcp_sock *msk = mptcp_sk(sock->sk);
3461 struct socket *ssock;
3462 int err;
3463
3464 pr_debug("msk=%p", msk);
3465
3466 lock_sock(sock->sk);
fa68018d
PA		 3467 ssock = __mptcp_nmpc_socket(msk);
3468 if (!ssock) {
3469 err = -EINVAL;
cf7da0d6
PK		 3470 goto unlock;
3471 }
3472
fa68018d
PA		 3473 mptcp_token_destroy(msk);
3474 inet_sk_state_store(sock->sk, TCP_LISTEN);
5e20087d
FW		 3475 sock_set_flag(sock->sk, SOCK_RCU_FREE);
3476
cf7da0d6
PK		 3477 err = ssock->ops->listen(ssock, backlog);
3478 inet_sk_state_store(sock->sk, inet_sk_state_load(ssock->sk));
3479 if (!err)
3480 mptcp_copy_inaddrs(sock->sk, ssock->sk);
3481
3482unlock:
3483 release_sock(sock->sk);
3484 return err;
3485}
3486
cf7da0d6
PK		 3487static int mptcp_stream_accept(struct socket *sock, struct socket *newsock,
3488 int flags, bool kern)
3489{
3490 struct mptcp_sock *msk = mptcp_sk(sock->sk);
3491 struct socket *ssock;
3492 int err;
3493
3494 pr_debug("msk=%p", msk);
3495
cf7da0d6
PK		 3496 ssock = __mptcp_nmpc_socket(msk);
3497 if (!ssock)
71ba088c 3498 return -EINVAL;
cf7da0d6
PK		 3499
3500 err = ssock->ops->accept(sock, newsock, flags, kern);
d2f77c53 3501 if (err == 0 && !mptcp_is_tcpsk(newsock->sk)) {
cf7da0d6
PK		 3502 struct mptcp_sock *msk = mptcp_sk(newsock->sk);
3503 struct mptcp_subflow_context *subflow;
0397c6d8 3504 struct sock *newsk = newsock->sk;
0397c6d8 3505
4d54cc32 3506 lock_sock(newsk);
5b950ff4
PA		 3507
3508 /* PM/worker can now acquire the first subflow socket
3509 * lock without racing with listener queue cleanup,
3510 * we can notify it, if needed.
c8fe62f0
FW		 3511 *
3512 * Even if remote has reset the initial subflow by now
3513 * the refcnt is still at least one.
5b950ff4
PA		 3514 */
3515 subflow = mptcp_subflow_ctx(msk->first);
3516 list_add(&subflow->node, &msk->conn_list);
3517 sock_hold(msk->first);
3518 if (mptcp_is_fully_established(newsk))
6c714f1b 3519 mptcp_pm_fully_established(msk, msk->first, GFP_KERNEL);
5b950ff4 3520
0397c6d8
PA		 3521 mptcp_copy_inaddrs(newsk, msk->first);
3522 mptcp_rcv_space_init(msk, msk->first);
5cf92bba 3523 mptcp_propagate_sndbuf(newsk, msk->first);
cf7da0d6
PK		 3524
3525 /* set ssk->sk_socket of accept()ed flows to mptcp socket.
3526 * This is needed so NOSPACE flag can be set from tcp stack.
3527 */
78962489 3528 mptcp_flush_join_list(msk);
190f8b06 3529 mptcp_for_each_subflow(msk, subflow) {
cf7da0d6
PK		 3530 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
3531
3532 if (!ssk->sk_socket)
3533 mptcp_sock_graft(ssk, newsock);
3534 }
4d54cc32 3535 release_sock(newsk);
cf7da0d6
PK		 3536 }
3537
cf7da0d6 3538 return err;
cf7da0d6
PK		 3539}
3540
8a05661b
PA		 3541static __poll_t mptcp_check_readable(struct mptcp_sock *msk)
3542{
612f71d7
PA		 3543 /* Concurrent splices from sk_receive_queue into receive_queue will
3544 * always show at least one non-empty queue when checked in this order.
3545 */
3546 if (skb_queue_empty_lockless(&((struct sock *)msk)->sk_receive_queue) &&
3547 skb_queue_empty_lockless(&msk->receive_queue))
3548 return 0;
3549
3550 return EPOLLIN | EPOLLRDNORM;
8a05661b
PA		 3551}
3552
8edf0864
FW		 3553static __poll_t mptcp_check_writeable(struct mptcp_sock *msk)
3554{
3555 struct sock *sk = (struct sock *)msk;
8edf0864
FW		 3556
3557 if (unlikely(sk->sk_shutdown & SEND_SHUTDOWN))
dd913410 3558 return EPOLLOUT | EPOLLWRNORM;
8edf0864
FW		 3559
3560 if (sk_stream_is_writeable(sk))
3561 return EPOLLOUT | EPOLLWRNORM;
3562
5cf92bba 3563 mptcp_set_nospace(sk);
6e628cd3
PA		 3564 smp_mb__after_atomic(); /* msk->flags is changed by write_space cb */
3565 if (sk_stream_is_writeable(sk))
3566 return EPOLLOUT | EPOLLWRNORM;
8edf0864 3567
6e628cd3 3568 return 0;
8edf0864
FW		 3569}
3570
2303f994
PK		 3571static __poll_t mptcp_poll(struct file *file, struct socket *sock,
3572 struct poll_table_struct *wait)
3573{
1891c4a0 3574 struct sock *sk = sock->sk;
8ab183de 3575 struct mptcp_sock *msk;
2303f994 3576 __poll_t mask = 0;
8a05661b 3577 int state;
2303f994 3578
1891c4a0 3579 msk = mptcp_sk(sk);
1891c4a0 3580 sock_poll_wait(file, sock, wait);
1891c4a0 3581
8a05661b 3582 state = inet_sk_state_load(sk);
6719331c 3583 pr_debug("msk=%p state=%d flags=%lx", msk, state, msk->flags);
71ba088c
PA		 3584 if (state == TCP_LISTEN) {
3585 if (WARN_ON_ONCE(!msk->subflow || !msk->subflow->sk))
3586 return 0;
3587
3588 return inet_csk_listen_poll(msk->subflow->sk);
3589 }
8a05661b
PA		 3590
3591 if (state != TCP_SYN_SENT && state != TCP_SYN_RECV) {
3592 mask |= mptcp_check_readable(msk);
8edf0864 3593 mask |= mptcp_check_writeable(msk);
8a05661b 3594 }
dd913410
PA		 3595 if (sk->sk_shutdown == SHUTDOWN_MASK || state == TCP_CLOSE)
3596 mask |= EPOLLHUP;
1891c4a0
FW		 3597 if (sk->sk_shutdown & RCV_SHUTDOWN)
3598 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
3599
15cc1045
PA		 3600 /* This barrier is coupled with smp_wmb() in tcp_reset() */
3601 smp_rmb();
3602 if (sk->sk_err)
3603 mask |= EPOLLERR;
3604
2303f994
PK		 3605 return mask;
3606}
3607
e42f1ac6
FW		 3608static const struct proto_ops mptcp_stream_ops = {
3609 .family = PF_INET,
3610 .owner = THIS_MODULE,
0a3cc579 3611 .release = inet_release,
e42f1ac6
FW		 3612 .bind = mptcp_bind,
3613 .connect = mptcp_stream_connect,
3614 .socketpair = sock_no_socketpair,
3615 .accept = mptcp_stream_accept,
d2f77c53 3616 .getname = inet_getname,
e42f1ac6
FW		 3617 .poll = mptcp_poll,
3618 .ioctl = inet_ioctl,
3619 .gettstamp = sock_gettstamp,
3620 .listen = mptcp_listen,
76e2a55d 3621 .shutdown = inet_shutdown,
e42f1ac6
FW		 3622 .setsockopt = sock_common_setsockopt,
3623 .getsockopt = sock_common_getsockopt,
3624 .sendmsg = inet_sendmsg,
3625 .recvmsg = inet_recvmsg,
3626 .mmap = sock_no_mmap,
3627 .sendpage = inet_sendpage,
e42f1ac6 3628};
2303f994 3629
f870fa0b
MM		 3630static struct inet_protosw mptcp_protosw = {
3631 .type = SOCK_STREAM,
3632 .protocol = IPPROTO_MPTCP,
3633 .prot = &mptcp_prot,
2303f994
PK		 3634 .ops = &mptcp_stream_ops,
3635 .flags = INET_PROTOSW_ICSK,
f870fa0b
MM		 3636};
3637
b19bc294
PA		 3638static int mptcp_napi_poll(struct napi_struct *napi, int budget)
3639{
3640 struct mptcp_delegated_action *delegated;
3641 struct mptcp_subflow_context *subflow;
3642 int work_done = 0;
3643
3644 delegated = container_of(napi, struct mptcp_delegated_action, napi);
3645 while ((subflow = mptcp_subflow_delegated_next(delegated)) != NULL) {
3646 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
3647
3648 bh_lock_sock_nested(ssk);
3649 if (!sock_owned_by_user(ssk) &&
3650 mptcp_subflow_has_delegated_action(subflow))
3651 mptcp_subflow_process_delegated(ssk);
3652 /* ... elsewhere tcp_release_cb_override already processed
3653 * the action or will do at next release_sock().
3654 * In both case must dequeue the subflow here - on the same
3655 * CPU that scheduled it.
3656 */
3657 bh_unlock_sock(ssk);
3658 sock_put(ssk);
3659
3660 if (++work_done == budget)
3661 return budget;
3662 }
3663
3664 /* always provide a 0 'work_done' argument, so that napi_complete_done
3665 * will not try accessing the NULL napi->dev ptr
3666 */
3667 napi_complete_done(napi, 0);
3668 return work_done;
3669}
3670
d39dceca 3671void __init mptcp_proto_init(void)
f870fa0b 3672{
b19bc294
PA		 3673 struct mptcp_delegated_action *delegated;
3674 int cpu;
3675
2303f994 3676 mptcp_prot.h.hashinfo = tcp_prot.h.hashinfo;
2303f994 3677
d027236c
PA		 3678 if (percpu_counter_init(&mptcp_sockets_allocated, 0, GFP_KERNEL))
3679 panic("Failed to allocate MPTCP pcpu counter\n");
3680
b19bc294
PA		 3681 init_dummy_netdev(&mptcp_napi_dev);
3682 for_each_possible_cpu(cpu) {
3683 delegated = per_cpu_ptr(&mptcp_delegated_actions, cpu);
3684 INIT_LIST_HEAD(&delegated->head);
3685 netif_tx_napi_add(&mptcp_napi_dev, &delegated->napi, mptcp_napi_poll,
3686 NAPI_POLL_WEIGHT);
3687 napi_enable(&delegated->napi);
3688 }
3689
2303f994 3690 mptcp_subflow_init();
1b1c7a0e 3691 mptcp_pm_init();
2c5ebd00 3692 mptcp_token_init();
2303f994 3693
f870fa0b
MM		 3694 if (proto_register(&mptcp_prot, 1) != 0)
3695 panic("Failed to register MPTCP proto.\n");
3696
3697 inet_register_protosw(&mptcp_protosw);
6771bfd9
FW		 3698
3699 BUILD_BUG_ON(sizeof(struct mptcp_skb_cb) > sizeof_field(struct sk_buff, cb));
f870fa0b
MM		 3700}
3701
3702#if IS_ENABLED(CONFIG_MPTCP_IPV6)
e42f1ac6
FW		 3703static const struct proto_ops mptcp_v6_stream_ops = {
3704 .family = PF_INET6,
3705 .owner = THIS_MODULE,
0a3cc579 3706 .release = inet6_release,
e42f1ac6
FW		 3707 .bind = mptcp_bind,
3708 .connect = mptcp_stream_connect,
3709 .socketpair = sock_no_socketpair,
3710 .accept = mptcp_stream_accept,
d2f77c53 3711 .getname = inet6_getname,
e42f1ac6
FW		 3712 .poll = mptcp_poll,
3713 .ioctl = inet6_ioctl,
3714 .gettstamp = sock_gettstamp,
3715 .listen = mptcp_listen,
76e2a55d 3716 .shutdown = inet_shutdown,
e42f1ac6
FW		 3717 .setsockopt = sock_common_setsockopt,
3718 .getsockopt = sock_common_getsockopt,
3719 .sendmsg = inet6_sendmsg,
3720 .recvmsg = inet6_recvmsg,
3721 .mmap = sock_no_mmap,
3722 .sendpage = inet_sendpage,
3723#ifdef CONFIG_COMPAT
3986912f 3724 .compat_ioctl = inet6_compat_ioctl,
e42f1ac6
FW		 3725#endif
3726};
3727
f870fa0b
MM		 3728static struct proto mptcp_v6_prot;
3729
79c0949e
PK		 3730static void mptcp_v6_destroy(struct sock *sk)
3731{
3732 mptcp_destroy(sk);
3733 inet6_destroy_sock(sk);
3734}
3735
f870fa0b
MM		 3736static struct inet_protosw mptcp_v6_protosw = {
3737 .type = SOCK_STREAM,
3738 .protocol = IPPROTO_MPTCP,
3739 .prot = &mptcp_v6_prot,
2303f994 3740 .ops = &mptcp_v6_stream_ops,
f870fa0b
MM		 3741 .flags = INET_PROTOSW_ICSK,
3742};
3743
d39dceca 3744int __init mptcp_proto_v6_init(void)
f870fa0b
MM		 3745{
3746 int err;
3747
3748 mptcp_v6_prot = mptcp_prot;
3749 strcpy(mptcp_v6_prot.name, "MPTCPv6");
3750 mptcp_v6_prot.slab = NULL;
79c0949e 3751 mptcp_v6_prot.destroy = mptcp_v6_destroy;
b0519de8 3752 mptcp_v6_prot.obj_size = sizeof(struct mptcp6_sock);
f870fa0b
MM		 3753
3754 err = proto_register(&mptcp_v6_prot, 1);
3755 if (err)
3756 return err;
3757
3758 err = inet6_register_protosw(&mptcp_v6_protosw);
3759 if (err)
3760 proto_unregister(&mptcp_v6_prot);
3761
3762 return err;
3763}
3764#endif
Linux 6.x block layer and io_uring tree(s)