projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
mptcp: add CurrEstab MIB counter support
[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
b7f653b2
FW		 39enum {
40 MPTCP_CMSG_TS = BIT(0),
2c9e7765 41 MPTCP_CMSG_INQ = BIT(1),
b7f653b2
FW		 42};
43
91b6d325 44static struct percpu_counter mptcp_sockets_allocated ____cacheline_aligned_in_smp;
d027236c 45
e16163b6 46static void __mptcp_destroy_sock(struct sock *sk);
81c1d029 47static void mptcp_check_send_data_fin(struct sock *sk);
e16163b6 48
b19bc294
PA		 49DEFINE_PER_CPU(struct mptcp_delegated_action, mptcp_delegated_actions);
50static struct net_device mptcp_napi_dev;
51
6f8a612a
FW		 52/* Returns end sequence number of the receiver's advertised window */
53static u64 mptcp_wnd_end(const struct mptcp_sock *msk)
54{
7439d687 55 return READ_ONCE(msk->wnd_end);
6f8a612a
FW		 56}
57
8e2b8a9f 58static const struct proto_ops *mptcp_fallback_tcp_ops(const struct sock *sk)
0b4f33de 59{
0b4f33de 60#if IS_ENABLED(CONFIG_MPTCP_IPV6)
8e2b8a9f
DC		 61 if (sk->sk_prot == &tcpv6_prot)
62 return &inet6_stream_ops;
0b4f33de 63#endif
8e2b8a9f
DC		 64 WARN_ON_ONCE(sk->sk_prot != &tcp_prot);
65 return &inet_stream_ops;
0b4f33de
FW		 66}
67
fa68018d 68static int __mptcp_socket_create(struct mptcp_sock *msk)
2303f994
PK		 69{
70 struct mptcp_subflow_context *subflow;
71 struct sock *sk = (struct sock *)msk;
72 struct socket *ssock;
73 int err;
74
6bc1fe7d 75 err = mptcp_subflow_create_socket(sk, sk->sk_family, &ssock);
2303f994 76 if (err)
fa68018d 77 return err;
2303f994 78
b8dc6d6c 79 msk->scaling_ratio = tcp_sk(ssock->sk)->scaling_ratio;
1b1b43ee 80 WRITE_ONCE(msk->first, ssock->sk);
2303f994 81 subflow = mptcp_subflow_ctx(ssock->sk);
cec37a6e 82 list_add(&subflow->node, &msk->conn_list);
e16163b6 83 sock_hold(ssock->sk);
2303f994 84 subflow->request_mptcp = 1;
6f06b4d4 85 subflow->subflow_id = msk->subflow_id++;
4cf86ae8
PA		 86
87 /* This is the first subflow, always with id 0 */
88 subflow->local_id_valid = 1;
866f26f2 89 mptcp_sock_graft(msk->first, sk->sk_socket);
39880bd8 90 iput(SOCK_INODE(ssock));
e1ff9e82 91
fa68018d 92 return 0;
2303f994
PK		 93}
94
ddb1a072
PA		 95/* If the MPC handshake is not started, returns the first subflow,
96 * eventually allocating it.
97 */
3f326a82 98struct sock *__mptcp_nmpc_sk(struct mptcp_sock *msk)
ddb1a072
PA		 99{
100 struct sock *sk = (struct sock *)msk;
101 int ret;
102
103 if (!((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)))
104 return ERR_PTR(-EINVAL);
105
3f326a82 106 if (!msk->first) {
ddb1a072
PA		 107 ret = __mptcp_socket_create(msk);
108 if (ret)
109 return ERR_PTR(ret);
ddb1a072
PA		 110 }
111
3f326a82 112 return msk->first;
ddb1a072
PA		 113}
114
ab174ad8
PA		 115static void mptcp_drop(struct sock *sk, struct sk_buff *skb)
116{
117 sk_drops_add(sk, skb);
118 __kfree_skb(skb);
119}
120
9531e4a8
ED		 121static void mptcp_rmem_fwd_alloc_add(struct sock *sk, int size)
122{
123 WRITE_ONCE(mptcp_sk(sk)->rmem_fwd_alloc,
124 mptcp_sk(sk)->rmem_fwd_alloc + size);
125}
126
6511882c
PA		 127static void mptcp_rmem_charge(struct sock *sk, int size)
128{
9531e4a8 129 mptcp_rmem_fwd_alloc_add(sk, -size);
6511882c
PA		 130}
131
8268ed4c
PA		 132static bool mptcp_try_coalesce(struct sock *sk, struct sk_buff *to,
133 struct sk_buff *from)
134{
135 bool fragstolen;
136 int delta;
137
138 if (MPTCP_SKB_CB(from)->offset ||
139 !skb_try_coalesce(to, from, &fragstolen, &delta))
140 return false;
141
06242e44
PA		 142 pr_debug("colesced seq %llx into %llx new len %d new end seq %llx",
143 MPTCP_SKB_CB(from)->map_seq, MPTCP_SKB_CB(to)->map_seq,
144 to->len, MPTCP_SKB_CB(from)->end_seq);
ab174ad8 145 MPTCP_SKB_CB(to)->end_seq = MPTCP_SKB_CB(from)->end_seq;
7288ff6e
PA		 146
147 /* note the fwd memory can reach a negative value after accounting
148 * for the delta, but the later skb free will restore a non
149 * negative one
150 */
8268ed4c 151 atomic_add(delta, &sk->sk_rmem_alloc);
6511882c 152 mptcp_rmem_charge(sk, delta);
7288ff6e
PA		 153 kfree_skb_partial(from, fragstolen);
154
8268ed4c
PA		 155 return true;
156}
157
ab174ad8
PA		 158static bool mptcp_ooo_try_coalesce(struct mptcp_sock *msk, struct sk_buff *to,
159 struct sk_buff *from)
160{
161 if (MPTCP_SKB_CB(from)->map_seq != MPTCP_SKB_CB(to)->end_seq)
162 return false;
163
164 return mptcp_try_coalesce((struct sock *)msk, to, from);
165}
166
6511882c
PA		 167static void __mptcp_rmem_reclaim(struct sock *sk, int amount)
168{
100fdd1f 169 amount >>= PAGE_SHIFT;
9531e4a8 170 mptcp_rmem_charge(sk, amount << PAGE_SHIFT);
6511882c
PA		 171 __sk_mem_reduce_allocated(sk, amount);
172}
173
174static void mptcp_rmem_uncharge(struct sock *sk, int size)
175{
176 struct mptcp_sock *msk = mptcp_sk(sk);
177 int reclaimable;
178
9531e4a8 179 mptcp_rmem_fwd_alloc_add(sk, size);
6511882c
PA		 180 reclaimable = msk->rmem_fwd_alloc - sk_unused_reserved_mem(sk);
181
182 /* see sk_mem_uncharge() for the rationale behind the following schema */
d24141fe
PA		 183 if (unlikely(reclaimable >= PAGE_SIZE))
184 __mptcp_rmem_reclaim(sk, reclaimable);
6511882c
PA		 185}
186
187static void mptcp_rfree(struct sk_buff *skb)
188{
189 unsigned int len = skb->truesize;
190 struct sock *sk = skb->sk;
191
192 atomic_sub(len, &sk->sk_rmem_alloc);
193 mptcp_rmem_uncharge(sk, len);
194}
195
36b122ba 196void mptcp_set_owner_r(struct sk_buff *skb, struct sock *sk)
6511882c
PA		 197{
198 skb_orphan(skb);
199 skb->sk = sk;
200 skb->destructor = mptcp_rfree;
201 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
202 mptcp_rmem_charge(sk, skb->truesize);
203}
204
ab174ad8
PA		 205/* "inspired" by tcp_data_queue_ofo(), main differences:
206 * - use mptcp seqs
207 * - don't cope with sacks
208 */
209static void mptcp_data_queue_ofo(struct mptcp_sock *msk, struct sk_buff *skb)
210{
211 struct sock *sk = (struct sock *)msk;
212 struct rb_node **p, *parent;
213 u64 seq, end_seq, max_seq;
214 struct sk_buff *skb1;
215
216 seq = MPTCP_SKB_CB(skb)->map_seq;
217 end_seq = MPTCP_SKB_CB(skb)->end_seq;
f3589be0 218 max_seq = atomic64_read(&msk->rcv_wnd_sent);
ab174ad8 219
06242e44
PA		 220 pr_debug("msk=%p seq=%llx limit=%llx empty=%d", msk, seq, max_seq,
221 RB_EMPTY_ROOT(&msk->out_of_order_queue));
fa3fe2b1 222 if (after64(end_seq, max_seq)) {
ab174ad8
PA		 223 /* out of window */
224 mptcp_drop(sk, skb);
fa3fe2b1
FW		 225 pr_debug("oow by %lld, rcv_wnd_sent %llu\n",
226 (unsigned long long)end_seq - (unsigned long)max_seq,
f3589be0 227 (unsigned long long)atomic64_read(&msk->rcv_wnd_sent));
06242e44 228 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_NODSSWINDOW);
ab174ad8
PA		 229 return;
230 }
231
232 p = &msk->out_of_order_queue.rb_node;
06242e44 233 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOQUEUE);
ab174ad8
PA		 234 if (RB_EMPTY_ROOT(&msk->out_of_order_queue)) {
235 rb_link_node(&skb->rbnode, NULL, p);
236 rb_insert_color(&skb->rbnode, &msk->out_of_order_queue);
237 msk->ooo_last_skb = skb;
238 goto end;
239 }
240
241 /* with 2 subflows, adding at end of ooo queue is quite likely
242 * Use of ooo_last_skb avoids the O(Log(N)) rbtree lookup.
243 */
06242e44
PA		 244 if (mptcp_ooo_try_coalesce(msk, msk->ooo_last_skb, skb)) {
245 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOMERGE);
246 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOQUEUETAIL);
ab174ad8 247 return;
06242e44 248 }
ab174ad8
PA		 249
250 /* Can avoid an rbtree lookup if we are adding skb after ooo_last_skb */
251 if (!before64(seq, MPTCP_SKB_CB(msk->ooo_last_skb)->end_seq)) {
06242e44 252 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOQUEUETAIL);
ab174ad8
PA		 253 parent = &msk->ooo_last_skb->rbnode;
254 p = &parent->rb_right;
255 goto insert;
256 }
257
258 /* Find place to insert this segment. Handle overlaps on the way. */
259 parent = NULL;
260 while (*p) {
261 parent = *p;
262 skb1 = rb_to_skb(parent);
263 if (before64(seq, MPTCP_SKB_CB(skb1)->map_seq)) {
264 p = &parent->rb_left;
265 continue;
266 }
267 if (before64(seq, MPTCP_SKB_CB(skb1)->end_seq)) {
268 if (!after64(end_seq, MPTCP_SKB_CB(skb1)->end_seq)) {
269 /* All the bits are present. Drop. */
270 mptcp_drop(sk, skb);
06242e44 271 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
ab174ad8
PA		 272 return;
273 }
274 if (after64(seq, MPTCP_SKB_CB(skb1)->map_seq)) {
275 /* partial overlap:
276 * | skb |
277 * | skb1 |
278 * continue traversing
279 */
280 } else {
281 /* skb's seq == skb1's seq and skb covers skb1.
282 * Replace skb1 with skb.
283 */
284 rb_replace_node(&skb1->rbnode, &skb->rbnode,
285 &msk->out_of_order_queue);
286 mptcp_drop(sk, skb1);
06242e44 287 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
ab174ad8
PA		 288 goto merge_right;
289 }
290 } else if (mptcp_ooo_try_coalesce(msk, skb1, skb)) {
06242e44 291 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOMERGE);
ab174ad8
PA		 292 return;
293 }
294 p = &parent->rb_right;
295 }
06242e44 296
ab174ad8
PA		 297insert:
298 /* Insert segment into RB tree. */
299 rb_link_node(&skb->rbnode, parent, p);
300 rb_insert_color(&skb->rbnode, &msk->out_of_order_queue);
301
302merge_right:
303 /* Remove other segments covered by skb. */
304 while ((skb1 = skb_rb_next(skb)) != NULL) {
305 if (before64(end_seq, MPTCP_SKB_CB(skb1)->end_seq))
306 break;
307 rb_erase(&skb1->rbnode, &msk->out_of_order_queue);
308 mptcp_drop(sk, skb1);
06242e44 309 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
ab174ad8
PA		 310 }
311 /* If there is no skb after us, we are the last_skb ! */
312 if (!skb1)
313 msk->ooo_last_skb = skb;
314
315end:
316 skb_condense(skb);
6511882c
PA		 317 mptcp_set_owner_r(skb, sk);
318}
319
320static bool mptcp_rmem_schedule(struct sock *sk, struct sock *ssk, int size)
321{
322 struct mptcp_sock *msk = mptcp_sk(sk);
323 int amt, amount;
324
69d93dae 325 if (size <= msk->rmem_fwd_alloc)
6511882c
PA		 326 return true;
327
69d93dae 328 size -= msk->rmem_fwd_alloc;
6511882c 329 amt = sk_mem_pages(size);
100fdd1f 330 amount = amt << PAGE_SHIFT;
4aaa1685
PA		 331 if (!__sk_mem_raise_allocated(sk, size, amt, SK_MEM_RECV))
332 return false;
6511882c 333
9531e4a8 334 mptcp_rmem_fwd_alloc_add(sk, amount);
6511882c 335 return true;
ab174ad8
PA		 336}
337
338static bool __mptcp_move_skb(struct mptcp_sock *msk, struct sock *ssk,
339 struct sk_buff *skb, unsigned int offset,
340 size_t copy_len)
6771bfd9 341{
ab174ad8 342 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
6771bfd9 343 struct sock *sk = (struct sock *)msk;
4e637c70 344 struct sk_buff *tail;
b7f653b2 345 bool has_rxtstamp;
6771bfd9
FW		 346
347 __skb_unlink(skb, &ssk->sk_receive_queue);
6771bfd9 348
4e637c70
FW		 349 skb_ext_reset(skb);
350 skb_orphan(skb);
ab174ad8 351
9c3f94e1 352 /* try to fetch required memory from subflow */
6511882c
PA		 353 if (!mptcp_rmem_schedule(sk, ssk, skb->truesize))
354 goto drop;
9c3f94e1 355
b7f653b2
FW		 356 has_rxtstamp = TCP_SKB_CB(skb)->has_rxtstamp;
357
ab174ad8
PA		 358 /* the skb map_seq accounts for the skb offset:
359 * mptcp_subflow_get_mapped_dsn() is based on the current tp->copied_seq
360 * value
361 */
362 MPTCP_SKB_CB(skb)->map_seq = mptcp_subflow_get_mapped_dsn(subflow);
363 MPTCP_SKB_CB(skb)->end_seq = MPTCP_SKB_CB(skb)->map_seq + copy_len;
8268ed4c 364 MPTCP_SKB_CB(skb)->offset = offset;
b7f653b2 365 MPTCP_SKB_CB(skb)->has_rxtstamp = has_rxtstamp;
4e637c70 366
ab174ad8
PA		 367 if (MPTCP_SKB_CB(skb)->map_seq == msk->ack_seq) {
368 /* in sequence */
38967f42 369 msk->bytes_received += copy_len;
8b0308fe 370 WRITE_ONCE(msk->ack_seq, msk->ack_seq + copy_len);
ab174ad8
PA		 371 tail = skb_peek_tail(&sk->sk_receive_queue);
372 if (tail && mptcp_try_coalesce(sk, tail, skb))
373 return true;
4e637c70 374
6511882c 375 mptcp_set_owner_r(skb, sk);
ab174ad8
PA		 376 __skb_queue_tail(&sk->sk_receive_queue, skb);
377 return true;
378 } else if (after64(MPTCP_SKB_CB(skb)->map_seq, msk->ack_seq)) {
379 mptcp_data_queue_ofo(msk, skb);
380 return false;
381 }
382
383 /* old data, keep it simple and drop the whole pkt, sender
384 * will retransmit as needed, if needed.
385 */
06242e44 386 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
9c3f94e1 387drop:
ab174ad8
PA		 388 mptcp_drop(sk, skb);
389 return false;
6771bfd9
FW		 390}
391
f6909dc1 392static void mptcp_stop_rtx_timer(struct sock *sk)
16a9a9da
MM		 393{
394 struct inet_connection_sock *icsk = inet_csk(sk);
395
396 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
397 mptcp_sk(sk)->timer_ival = 0;
398}
399
e16163b6
PA		 400static void mptcp_close_wake_up(struct sock *sk)
401{
402 if (sock_flag(sk, SOCK_DEAD))
403 return;
404
405 sk->sk_state_change(sk);
406 if (sk->sk_shutdown == SHUTDOWN_MASK ||
407 sk->sk_state == TCP_CLOSE)
408 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_HUP);
409 else
410 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
411}
412
6e628cd3 413static bool mptcp_pending_data_fin_ack(struct sock *sk)
16a9a9da
MM		 414{
415 struct mptcp_sock *msk = mptcp_sk(sk);
416
81c1d029 417 return ((1 << sk->sk_state) &
6e628cd3
PA		 418 (TCPF_FIN_WAIT1 | TCPF_CLOSING | TCPF_LAST_ACK)) &&
419 msk->write_seq == READ_ONCE(msk->snd_una);
420}
421
422static void mptcp_check_data_fin_ack(struct sock *sk)
423{
424 struct mptcp_sock *msk = mptcp_sk(sk);
16a9a9da
MM		 425
426 /* Look for an acknowledged DATA_FIN */
6e628cd3 427 if (mptcp_pending_data_fin_ack(sk)) {
16a9a9da
MM		 428 WRITE_ONCE(msk->snd_data_fin_enable, 0);
429
430 switch (sk->sk_state) {
431 case TCP_FIN_WAIT1:
432 inet_sk_state_store(sk, TCP_FIN_WAIT2);
16a9a9da
MM		 433 break;
434 case TCP_CLOSING:
16a9a9da
MM		 435 case TCP_LAST_ACK:
436 inet_sk_state_store(sk, TCP_CLOSE);
16a9a9da
MM		 437 break;
438 }
439
e16163b6 440 mptcp_close_wake_up(sk);
16a9a9da
MM		 441 }
442}
443
3721b9b6
MM		 444static bool mptcp_pending_data_fin(struct sock *sk, u64 *seq)
445{
446 struct mptcp_sock *msk = mptcp_sk(sk);
447
448 if (READ_ONCE(msk->rcv_data_fin) &&
449 ((1 << sk->sk_state) &
450 (TCPF_ESTABLISHED | TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2))) {
451 u64 rcv_data_fin_seq = READ_ONCE(msk->rcv_data_fin_seq);
452
453 if (msk->ack_seq == rcv_data_fin_seq) {
454 if (seq)
455 *seq = rcv_data_fin_seq;
456
457 return true;
458 }
459 }
460
461 return false;
462}
463
403a40f2 464static void mptcp_set_datafin_timeout(struct sock *sk)
6477dd39
MM		 465{
466 struct inet_connection_sock *icsk = inet_csk(sk);
877d11f0 467 u32 retransmits;
6477dd39 468
877d11f0
MM		 469 retransmits = min_t(u32, icsk->icsk_retransmits,
470 ilog2(TCP_RTO_MAX / TCP_RTO_MIN));
471
472 mptcp_sk(sk)->timer_ival = TCP_RTO_MIN << retransmits;
6477dd39
MM		 473}
474
33d41c9c 475static void __mptcp_set_timeout(struct sock *sk, long tout)
3721b9b6 476{
3721b9b6
MM		 477 mptcp_sk(sk)->timer_ival = tout > 0 ? tout : TCP_RTO_MIN;
478}
479
33d41c9c
PA		 480static long mptcp_timeout_from_subflow(const struct mptcp_subflow_context *subflow)
481{
482 const struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
483
71b7dec2
PA		 484 return inet_csk(ssk)->icsk_pending && !subflow->stale_count ?
485 inet_csk(ssk)->icsk_timeout - jiffies : 0;
33d41c9c
PA		 486}
487
488static void mptcp_set_timeout(struct sock *sk)
489{
490 struct mptcp_subflow_context *subflow;
491 long tout = 0;
492
493 mptcp_for_each_subflow(mptcp_sk(sk), subflow)
494 tout = max(tout, mptcp_timeout_from_subflow(subflow));
495 __mptcp_set_timeout(sk, tout);
496}
497
76a13b31 498static inline bool tcp_can_send_ack(const struct sock *ssk)
fd897679
PA		 499{
500 return !((1 << inet_sk_state_load(ssk)) &
20bc80b6 501 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_TIME_WAIT | TCPF_CLOSE | TCPF_LISTEN));
fd897679
PA		 502}
503
a6574302
MM		 504void __mptcp_subflow_send_ack(struct sock *ssk)
505{
506 if (tcp_can_send_ack(ssk))
507 tcp_send_ack(ssk);
508}
509
f5360e9b 510static void mptcp_subflow_send_ack(struct sock *ssk)
340fa666
MM		 511{
512 bool slow;
513
514 slow = lock_sock_fast(ssk);
a6574302 515 __mptcp_subflow_send_ack(ssk);
340fa666
MM		 516 unlock_sock_fast(ssk, slow);
517}
518
fd897679 519static void mptcp_send_ack(struct mptcp_sock *msk)
7ed90803
PA		 520{
521 struct mptcp_subflow_context *subflow;
522
340fa666
MM		 523 mptcp_for_each_subflow(msk, subflow)
524 mptcp_subflow_send_ack(mptcp_subflow_tcp_sock(subflow));
fd897679
PA		 525}
526
fde56eea 527static void mptcp_subflow_cleanup_rbuf(struct sock *ssk)
fd897679 528{
75e908c3 529 bool slow;
fd897679 530
75e908c3 531 slow = lock_sock_fast(ssk);
fde56eea 532 if (tcp_can_send_ack(ssk))
fd897679 533 tcp_cleanup_rbuf(ssk, 1);
75e908c3 534 unlock_sock_fast(ssk, slow);
fde56eea
PA		 535}
536
537static bool mptcp_subflow_could_cleanup(const struct sock *ssk, bool rx_empty)
538{
539 const struct inet_connection_sock *icsk = inet_csk(ssk);
c4512c63 540 u8 ack_pending = READ_ONCE(icsk->icsk_ack.pending);
fde56eea
PA		 541 const struct tcp_sock *tp = tcp_sk(ssk);
542
543 return (ack_pending & ICSK_ACK_SCHED) &&
544 ((READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->rcv_wup) >
545 READ_ONCE(icsk->icsk_ack.rcv_mss)) ||
546 (rx_empty && ack_pending &
547 (ICSK_ACK_PUSHED2 | ICSK_ACK_PUSHED)));
fd897679
PA		 548}
549
550static void mptcp_cleanup_rbuf(struct mptcp_sock *msk)
551{
e3859603 552 int old_space = READ_ONCE(msk->old_wspace);
fd897679 553 struct mptcp_subflow_context *subflow;
e3859603 554 struct sock *sk = (struct sock *)msk;
fde56eea
PA		 555 int space = __mptcp_space(sk);
556 bool cleanup, rx_empty;
e3859603 557
fde56eea 558 cleanup = (space > 0) && (space >= (old_space << 1));
ce599c51 559 rx_empty = !__mptcp_rmem(sk);
fd897679 560
fde56eea
PA		 561 mptcp_for_each_subflow(msk, subflow) {
562 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
fd897679 563
fde56eea
PA		 564 if (cleanup || mptcp_subflow_could_cleanup(ssk, rx_empty))
565 mptcp_subflow_cleanup_rbuf(ssk);
7ed90803
PA		 566 }
567}
568
569static bool mptcp_check_data_fin(struct sock *sk)
3721b9b6
MM		 570{
571 struct mptcp_sock *msk = mptcp_sk(sk);
572 u64 rcv_data_fin_seq;
7ed90803 573 bool ret = false;
3721b9b6 574
3721b9b6
MM		 575 /* Need to ack a DATA_FIN received from a peer while this side
576 * of the connection is in ESTABLISHED, FIN_WAIT1, or FIN_WAIT2.
577 * msk->rcv_data_fin was set when parsing the incoming options
578 * at the subflow level and the msk lock was not held, so this
579 * is the first opportunity to act on the DATA_FIN and change
580 * the msk state.
581 *
582 * If we are caught up to the sequence number of the incoming
583 * DATA_FIN, send the DATA_ACK now and do state transition. If
584 * not caught up, do nothing and let the recv code send DATA_ACK
585 * when catching up.
586 */
587
588 if (mptcp_pending_data_fin(sk, &rcv_data_fin_seq)) {
917944da 589 WRITE_ONCE(msk->ack_seq, msk->ack_seq + 1);
3721b9b6
MM		 590 WRITE_ONCE(msk->rcv_data_fin, 0);
591
6b9831bf 592 WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | RCV_SHUTDOWN);
16a9a9da 593 smp_mb__before_atomic(); /* SHUTDOWN must be visible first */
3721b9b6
MM		 594
595 switch (sk->sk_state) {
596 case TCP_ESTABLISHED:
597 inet_sk_state_store(sk, TCP_CLOSE_WAIT);
598 break;
599 case TCP_FIN_WAIT1:
600 inet_sk_state_store(sk, TCP_CLOSING);
601 break;
602 case TCP_FIN_WAIT2:
603 inet_sk_state_store(sk, TCP_CLOSE);
3721b9b6
MM		 604 break;
605 default:
606 /* Other states not expected */
607 WARN_ON_ONCE(1);
608 break;
609 }
610
7ed90803 611 ret = true;
81c1d029
PA		 612 if (!__mptcp_check_fallback(msk))
613 mptcp_send_ack(msk);
e16163b6 614 mptcp_close_wake_up(sk);
3721b9b6 615 }
7ed90803 616 return ret;
3721b9b6
MM		 617}
618
6771bfd9
FW		 619static bool __mptcp_move_skbs_from_subflow(struct mptcp_sock *msk,
620 struct sock *ssk,
621 unsigned int *bytes)
622{
623 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
600911ff 624 struct sock *sk = (struct sock *)msk;
6771bfd9
FW		 625 unsigned int moved = 0;
626 bool more_data_avail;
627 struct tcp_sock *tp;
628 bool done = false;
13c7ba0c
FW		 629 int sk_rbuf;
630
631 sk_rbuf = READ_ONCE(sk->sk_rcvbuf);
632
633 if (!(sk->sk_userlocks & SOCK_RCVBUF_LOCK)) {
634 int ssk_rbuf = READ_ONCE(ssk->sk_rcvbuf);
635
636 if (unlikely(ssk_rbuf > sk_rbuf)) {
637 WRITE_ONCE(sk->sk_rcvbuf, ssk_rbuf);
638 sk_rbuf = ssk_rbuf;
639 }
640 }
600911ff 641
ab174ad8 642 pr_debug("msk=%p ssk=%p", msk, ssk);
6771bfd9
FW		 643 tp = tcp_sk(ssk);
644 do {
645 u32 map_remaining, offset;
646 u32 seq = tp->copied_seq;
647 struct sk_buff *skb;
648 bool fin;
649
650 /* try to move as much data as available */
651 map_remaining = subflow->map_data_len -
652 mptcp_subflow_get_map_offset(subflow);
653
654 skb = skb_peek(&ssk->sk_receive_queue);
d9fb8c50 655 if (!skb) {
d89e3ed7
PA		 656 /* With racing move_skbs_to_msk() and __mptcp_move_skbs(),
657 * a different CPU can have already processed the pending
658 * data, stop here or we can enter an infinite loop
d9fb8c50
PA		 659 */
660 if (!moved)
661 done = true;
6771bfd9 662 break;
d9fb8c50 663 }
6771bfd9 664
e1ff9e82 665 if (__mptcp_check_fallback(msk)) {
d89e3ed7
PA		 666 /* Under fallback skbs have no MPTCP extension and TCP could
667 * collapse them between the dummy map creation and the
668 * current dequeue. Be sure to adjust the map size.
e1ff9e82
DC		 669 */
670 map_remaining = skb->len;
671 subflow->map_data_len = skb->len;
672 }
673
6771bfd9
FW		 674 offset = seq - TCP_SKB_CB(skb)->seq;
675 fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
676 if (fin) {
677 done = true;
678 seq++;
679 }
680
681 if (offset < skb->len) {
682 size_t len = skb->len - offset;
683
684 if (tp->urg_data)
685 done = true;
686
ab174ad8
PA		 687 if (__mptcp_move_skb(msk, ssk, skb, offset, len))
688 moved += len;
6771bfd9 689 seq += len;
6771bfd9
FW		 690
691 if (WARN_ON_ONCE(map_remaining < len))
692 break;
693 } else {
694 WARN_ON_ONCE(!fin);
695 sk_eat_skb(ssk, skb);
696 done = true;
697 }
698
699 WRITE_ONCE(tp->copied_seq, seq);
700 more_data_avail = mptcp_subflow_data_available(ssk);
600911ff 701
13c7ba0c 702 if (atomic_read(&sk->sk_rmem_alloc) > sk_rbuf) {
600911ff
FW		 703 done = true;
704 break;
705 }
6771bfd9
FW		 706 } while (more_data_avail);
707
6719331c 708 *bytes += moved;
6771bfd9
FW		 709 return done;
710}
711
87952603 712static bool __mptcp_ofo_queue(struct mptcp_sock *msk)
ab174ad8
PA		 713{
714 struct sock *sk = (struct sock *)msk;
715 struct sk_buff *skb, *tail;
716 bool moved = false;
717 struct rb_node *p;
718 u64 end_seq;
719
720 p = rb_first(&msk->out_of_order_queue);
06242e44 721 pr_debug("msk=%p empty=%d", msk, RB_EMPTY_ROOT(&msk->out_of_order_queue));
ab174ad8
PA		 722 while (p) {
723 skb = rb_to_skb(p);
724 if (after64(MPTCP_SKB_CB(skb)->map_seq, msk->ack_seq))
725 break;
726
727 p = rb_next(p);
728 rb_erase(&skb->rbnode, &msk->out_of_order_queue);
729
730 if (unlikely(!after64(MPTCP_SKB_CB(skb)->end_seq,
731 msk->ack_seq))) {
732 mptcp_drop(sk, skb);
06242e44 733 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
ab174ad8
PA		 734 continue;
735 }
736
737 end_seq = MPTCP_SKB_CB(skb)->end_seq;
738 tail = skb_peek_tail(&sk->sk_receive_queue);
739 if (!tail || !mptcp_ooo_try_coalesce(msk, tail, skb)) {
740 int delta = msk->ack_seq - MPTCP_SKB_CB(skb)->map_seq;
741
742 /* skip overlapping data, if any */
06242e44
PA		 743 pr_debug("uncoalesced seq=%llx ack seq=%llx delta=%d",
744 MPTCP_SKB_CB(skb)->map_seq, msk->ack_seq,
745 delta);
ab174ad8 746 MPTCP_SKB_CB(skb)->offset += delta;
2c9e7765 747 MPTCP_SKB_CB(skb)->map_seq += delta;
ab174ad8
PA		 748 __skb_queue_tail(&sk->sk_receive_queue, skb);
749 }
38967f42 750 msk->bytes_received += end_seq - msk->ack_seq;
ab174ad8
PA		 751 msk->ack_seq = end_seq;
752 moved = true;
753 }
754 return moved;
755}
756
9f1a9881 757static bool __mptcp_subflow_error_report(struct sock *sk, struct sock *ssk)
d5fbeff1 758{
9f1a9881
PA		 759 int err = sock_error(ssk);
760 int ssk_state;
d5fbeff1 761
9f1a9881
PA		 762 if (!err)
763 return false;
d5fbeff1 764
9f1a9881
PA		 765 /* only propagate errors on fallen-back sockets or
766 * on MPC connect
767 */
768 if (sk->sk_state != TCP_SYN_SENT && !__mptcp_check_fallback(mptcp_sk(sk)))
769 return false;
d5fbeff1 770
9f1a9881
PA		 771 /* We need to propagate only transition to CLOSE state.
772 * Orphaned socket will see such state change via
773 * subflow_sched_work_if_closed() and that path will properly
774 * destroy the msk as needed.
775 */
776 ssk_state = inet_sk_state_load(ssk);
777 if (ssk_state == TCP_CLOSE && !sock_flag(sk, SOCK_DEAD))
778 inet_sk_state_store(sk, ssk_state);
779 WRITE_ONCE(sk->sk_err, -err);
780
781 /* This barrier is coupled with smp_rmb() in mptcp_poll() */
782 smp_wmb();
783 sk_error_report(sk);
784 return true;
785}
d5fbeff1 786
9f1a9881
PA		 787void __mptcp_error_report(struct sock *sk)
788{
789 struct mptcp_subflow_context *subflow;
790 struct mptcp_sock *msk = mptcp_sk(sk);
791
792 mptcp_for_each_subflow(msk, subflow)
793 if (__mptcp_subflow_error_report(sk, mptcp_subflow_tcp_sock(subflow)))
794 break;
d5fbeff1
PA		 795}
796
2e52213c
FW		 797/* In most cases we will be able to lock the mptcp socket. If its already
798 * owned, we need to defer to the work queue to avoid ABBA deadlock.
799 */
99d1055c 800static bool move_skbs_to_msk(struct mptcp_sock *msk, struct sock *ssk)
2e52213c
FW		 801{
802 struct sock *sk = (struct sock *)msk;
803 unsigned int moved = 0;
804
87952603
PA		 805 __mptcp_move_skbs_from_subflow(msk, ssk, &moved);
806 __mptcp_ofo_queue(msk);
499ada50
PA		 807 if (unlikely(ssk->sk_err)) {
808 if (!sock_owned_by_user(sk))
809 __mptcp_error_report(sk);
810 else
e9d09bac 811 __set_bit(MPTCP_ERROR_REPORT, &msk->cb_flags);
499ada50 812 }
2e52213c 813
87952603
PA		 814 /* If the moves have caught up with the DATA_FIN sequence number
815 * it's time to ack the DATA_FIN and change socket state, but
816 * this is not a good place to change state. Let the workqueue
817 * do it.
818 */
819 if (mptcp_pending_data_fin(sk, NULL))
820 mptcp_schedule_work(sk);
99d1055c 821 return moved > 0;
2e52213c
FW		 822}
823
824void mptcp_data_ready(struct sock *sk, struct sock *ssk)
101f6f85 825{
6719331c 826 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
101f6f85 827 struct mptcp_sock *msk = mptcp_sk(sk);
13c7ba0c 828 int sk_rbuf, ssk_rbuf;
101f6f85 829
d7b1bfd0
PA		 830 /* The peer can send data while we are shutting down this
831 * subflow at msk destruction time, but we must avoid enqueuing
832 * more data to the msk receive queue
833 */
834 if (unlikely(subflow->disposable))
835 return;
836
13c7ba0c
FW		 837 ssk_rbuf = READ_ONCE(ssk->sk_rcvbuf);
838 sk_rbuf = READ_ONCE(sk->sk_rcvbuf);
839 if (unlikely(ssk_rbuf > sk_rbuf))
840 sk_rbuf = ssk_rbuf;
841
99d1055c 842 /* over limit? can't append more skbs to msk, Also, no need to wake-up*/
ce599c51
PA		 843 if (__mptcp_rmem(sk) > sk_rbuf) {
844 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_RCVPRUNED);
99d1055c 845 return;
ce599c51 846 }
600911ff 847
99d1055c
PA		 848 /* Wake-up the reader only for in-sequence data */
849 mptcp_data_lock(sk);
5684ab1a 850 if (move_skbs_to_msk(msk, ssk) && mptcp_epollin_ready(sk))
6719331c 851 sk->sk_data_ready(sk);
99d1055c 852 mptcp_data_unlock(sk);
101f6f85
FW		 853}
854
55b47ca7
PA		 855static void mptcp_subflow_joined(struct mptcp_sock *msk, struct sock *ssk)
856{
857 mptcp_subflow_ctx(ssk)->map_seq = READ_ONCE(msk->ack_seq);
858 WRITE_ONCE(msk->allow_infinite_fallback, false);
859 mptcp_event(MPTCP_EVENT_SUB_ESTABLISHED, msk, ssk, GFP_ATOMIC);
860}
861
3e501490 862static bool __mptcp_finish_join(struct mptcp_sock *msk, struct sock *ssk)
ec3edaa7 863{
3e501490 864 struct sock *sk = (struct sock *)msk;
5cf92bba 865
3e501490 866 if (sk->sk_state != TCP_ESTABLISHED)
78962489 867 return false;
ec3edaa7 868
3e501490
PA		 869 /* attach to msk socket only after we are sure we will deal with it
870 * at close time
871 */
872 if (sk->sk_socket && !ssk->sk_socket)
873 mptcp_sock_graft(ssk, sk->sk_socket);
78962489 874
6f06b4d4 875 mptcp_subflow_ctx(ssk)->subflow_id = msk->subflow_id++;
3e501490 876 mptcp_sockopt_sync_locked(msk, ssk);
55b47ca7 877 mptcp_subflow_joined(msk, ssk);
27e5ccc2 878 mptcp_stop_tout_timer(sk);
8005184f 879 __mptcp_propagate_sndbuf(sk, ssk);
3e501490 880 return true;
78962489
FW		 881}
882
56a666c4 883static void __mptcp_flush_join_list(struct sock *sk, struct list_head *join_list)
78962489 884{
3e501490
PA		 885 struct mptcp_subflow_context *tmp, *subflow;
886 struct mptcp_sock *msk = mptcp_sk(sk);
78962489 887
56a666c4 888 list_for_each_entry_safe(subflow, tmp, join_list, node) {
3e501490
PA		 889 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
890 bool slow = lock_sock_fast(ssk);
78962489 891
3e501490
PA		 892 list_move_tail(&subflow->node, &msk->conn_list);
893 if (!__mptcp_finish_join(msk, ssk))
894 mptcp_subflow_reset(ssk);
895 unlock_sock_fast(ssk, slow);
896 }
ec3edaa7
PK		 897}
898
f6909dc1 899static bool mptcp_rtx_timer_pending(struct sock *sk)
b51f9b80
PA		 900{
901 return timer_pending(&inet_csk(sk)->icsk_retransmit_timer);
902}
903
f6909dc1 904static void mptcp_reset_rtx_timer(struct sock *sk)
b51f9b80
PA		 905{
906 struct inet_connection_sock *icsk = inet_csk(sk);
907 unsigned long tout;
908
e16163b6
PA		 909 /* prevent rescheduling on close */
910 if (unlikely(inet_sk_state_load(sk) == TCP_CLOSE))
911 return;
912
33d41c9c 913 tout = mptcp_sk(sk)->timer_ival;
b51f9b80
PA		 914 sk_reset_timer(sk, &icsk->icsk_retransmit_timer, jiffies + tout);
915}
916
ba8f48f7
PA		 917bool mptcp_schedule_work(struct sock *sk)
918{
919 if (inet_sk_state_load(sk) != TCP_CLOSE &&
920 schedule_work(&mptcp_sk(sk)->work)) {
921 /* each subflow already holds a reference to the sk, and the
922 * workqueue is invoked by a subflow, so sk can't go away here.
923 */
924 sock_hold(sk);
925 return true;
926 }
927 return false;
928}
929
7a6a6cbc
PA		 930static struct sock *mptcp_subflow_recv_lookup(const struct mptcp_sock *msk)
931{
932 struct mptcp_subflow_context *subflow;
7a6a6cbc 933
109cdeb8 934 msk_owned_by_me(msk);
7a6a6cbc
PA		 935
936 mptcp_for_each_subflow(msk, subflow) {
99d1055c 937 if (READ_ONCE(subflow->data_avail))
7a6a6cbc
PA		 938 return mptcp_subflow_tcp_sock(subflow);
939 }
940
941 return NULL;
942}
943
3f8e0aae
PA		 944static bool mptcp_skb_can_collapse_to(u64 write_seq,
945 const struct sk_buff *skb,
946 const struct mptcp_ext *mpext)
57040755
PA		 947{
948 if (!tcp_skb_can_collapse_to(skb))
949 return false;
950
5a369ca6
PA		 951 /* can collapse only if MPTCP level sequence is in order and this
952 * mapping has not been xmitted yet
953 */
954 return mpext && mpext->data_seq + mpext->data_len == write_seq &&
955 !mpext->frozen;
57040755
PA		 956}
957
29249eac
PA		 958/* we can append data to the given data frag if:
959 * - there is space available in the backing page_frag
960 * - the data frag tail matches the current page_frag free offset
961 * - the data frag end sequence number matches the current write seq
962 */
18b683bf
PA		 963static bool mptcp_frag_can_collapse_to(const struct mptcp_sock *msk,
964 const struct page_frag *pfrag,
965 const struct mptcp_data_frag *df)
966{
967 return df && pfrag->page == df->page &&
d9ca1de8 968 pfrag->size - pfrag->offset > 0 &&
29249eac 969 pfrag->offset == (df->offset + df->data_len) &&
18b683bf
PA		 970 df->data_seq + df->data_len == msk->write_seq;
971}
972
d027236c
PA		 973static void dfrag_uncharge(struct sock *sk, int len)
974{
975 sk_mem_uncharge(sk, len);
7948f6cc 976 sk_wmem_queued_add(sk, -len);
d027236c
PA		 977}
978
979static void dfrag_clear(struct sock *sk, struct mptcp_data_frag *dfrag)
18b683bf 980{
d027236c
PA		 981 int len = dfrag->data_len + dfrag->overhead;
982
18b683bf 983 list_del(&dfrag->list);
d027236c 984 dfrag_uncharge(sk, len);
18b683bf
PA		 985 put_page(dfrag->page);
986}
987
6e628cd3 988static void __mptcp_clean_una(struct sock *sk)
18b683bf
PA		 989{
990 struct mptcp_sock *msk = mptcp_sk(sk);
991 struct mptcp_data_frag *dtmp, *dfrag;
e1ff9e82
DC		 992 u64 snd_una;
993
7439d687 994 snd_una = msk->snd_una;
18b683bf
PA		 995 list_for_each_entry_safe(dfrag, dtmp, &msk->rtx_queue, list) {
996 if (after64(dfrag->data_seq + dfrag->data_len, snd_una))
997 break;
998
1e1d9d6f
PA		 999 if (unlikely(dfrag == msk->first_pending)) {
1000 /* in recovery mode can see ack after the current snd head */
1001 if (WARN_ON_ONCE(!msk->recovery))
1002 break;
1003
1004 WRITE_ONCE(msk->first_pending, mptcp_send_next(sk));
1005 }
1006
d027236c 1007 dfrag_clear(sk, dfrag);
d027236c
PA		 1008 }
1009
7948f6cc
FW		 1010 dfrag = mptcp_rtx_head(sk);
1011 if (dfrag && after64(snd_una, dfrag->data_seq)) {
53eb4c38
PA		 1012 u64 delta = snd_una - dfrag->data_seq;
1013
1e1d9d6f
PA		 1014 /* prevent wrap around in recovery mode */
1015 if (unlikely(delta > dfrag->already_sent)) {
1016 if (WARN_ON_ONCE(!msk->recovery))
1017 goto out;
1018 if (WARN_ON_ONCE(delta > dfrag->data_len))
1019 goto out;
1020 dfrag->already_sent += delta - dfrag->already_sent;
1021 }
7948f6cc
FW		 1022
1023 dfrag->data_seq += delta;
53eb4c38 1024 dfrag->offset += delta;
7948f6cc 1025 dfrag->data_len -= delta;
d9ca1de8 1026 dfrag->already_sent -= delta;
7948f6cc
FW		 1027
1028 dfrag_uncharge(sk, delta);
7948f6cc
FW		 1029 }
1030
1e1d9d6f
PA		 1031 /* all retransmitted data acked, recovery completed */
1032 if (unlikely(msk->recovery) && after64(msk->snd_una, msk->recovery_snd_nxt))
1033 msk->recovery = false;
1034
53eb4c38 1035out:
3241a9c0
FW		 1036 if (snd_una == READ_ONCE(msk->snd_nxt) &&
1037 snd_una == READ_ONCE(msk->write_seq)) {
f6909dc1
PA		 1038 if (mptcp_rtx_timer_pending(sk) && !mptcp_data_fin_enabled(msk))
1039 mptcp_stop_rtx_timer(sk);
6e628cd3 1040 } else {
f6909dc1 1041 mptcp_reset_rtx_timer(sk);
18b683bf
PA		 1042 }
1043}
1044
417789df
PA		 1045static void __mptcp_clean_una_wakeup(struct sock *sk)
1046{
765ff425
PA		 1047 lockdep_assert_held_once(&sk->sk_lock.slock);
1048
417789df
PA		 1049 __mptcp_clean_una(sk);
1050 mptcp_write_space(sk);
1051}
1052
b5941f06
PA		 1053static void mptcp_clean_una_wakeup(struct sock *sk)
1054{
1055 mptcp_data_lock(sk);
1056 __mptcp_clean_una_wakeup(sk);
1057 mptcp_data_unlock(sk);
1058}
1059
724cfd2e 1060static void mptcp_enter_memory_pressure(struct sock *sk)
18b683bf 1061{
d9ca1de8
PA		 1062 struct mptcp_subflow_context *subflow;
1063 struct mptcp_sock *msk = mptcp_sk(sk);
1064 bool first = true;
1065
d9ca1de8
PA		 1066 mptcp_for_each_subflow(msk, subflow) {
1067 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1068
1069 if (first)
1070 tcp_enter_memory_pressure(ssk);
1071 sk_stream_moderate_sndbuf(ssk);
8005184f 1072
d9ca1de8
PA		 1073 first = false;
1074 }
8005184f 1075 __mptcp_sync_sndbuf(sk);
724cfd2e
PA		 1076}
1077
1078/* ensure we get enough memory for the frag hdr, beyond some minimal amount of
1079 * data
1080 */
1081static bool mptcp_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1082{
1083 if (likely(skb_page_frag_refill(32U + sizeof(struct mptcp_data_frag),
1084 pfrag, sk->sk_allocation)))
1085 return true;
1086
1087 mptcp_enter_memory_pressure(sk);
18b683bf
PA		 1088 return false;
1089}
1090
1091static struct mptcp_data_frag *
1092mptcp_carve_data_frag(const struct mptcp_sock *msk, struct page_frag *pfrag,
1093 int orig_offset)
1094{
1095 int offset = ALIGN(orig_offset, sizeof(long));
1096 struct mptcp_data_frag *dfrag;
1097
1098 dfrag = (struct mptcp_data_frag *)(page_to_virt(pfrag->page) + offset);
1099 dfrag->data_len = 0;
1100 dfrag->data_seq = msk->write_seq;
1101 dfrag->overhead = offset - orig_offset + sizeof(struct mptcp_data_frag);
1102 dfrag->offset = offset + sizeof(struct mptcp_data_frag);
d9ca1de8 1103 dfrag->already_sent = 0;
18b683bf
PA		 1104 dfrag->page = pfrag->page;
1105
1106 return dfrag;
1107}
1108
caf971df
PA		 1109struct mptcp_sendmsg_info {
1110 int mss_now;
1111 int size_goal;
d9ca1de8
PA		 1112 u16 limit;
1113 u16 sent;
1114 unsigned int flags;
1094c6fe 1115 bool data_lock_held;
caf971df
PA		 1116};
1117
b713d006
PA		 1118static int mptcp_check_allowed_size(const struct mptcp_sock *msk, struct sock *ssk,
1119 u64 data_seq, int avail_size)
6f8a612a
FW		 1120{
1121 u64 window_end = mptcp_wnd_end(msk);
b713d006 1122 u64 mptcp_snd_wnd;
6f8a612a
FW		 1123
1124 if (__mptcp_check_fallback(msk))
1125 return avail_size;
1126
b713d006
PA		 1127 mptcp_snd_wnd = window_end - data_seq;
1128 avail_size = min_t(unsigned int, mptcp_snd_wnd, avail_size);
6f8a612a 1129
92be2f52 1130 if (unlikely(tcp_sk(ssk)->snd_wnd < mptcp_snd_wnd)) {
b713d006 1131 tcp_sk(ssk)->snd_wnd = min_t(u64, U32_MAX, mptcp_snd_wnd);
92be2f52
PA		 1132 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_SNDWNDSHARED);
1133 }
6f8a612a
FW		 1134
1135 return avail_size;
1136}
1137
724cfd2e
PA		 1138static bool __mptcp_add_ext(struct sk_buff *skb, gfp_t gfp)
1139{
1140 struct skb_ext *mpext = __skb_ext_alloc(gfp);
1141
1142 if (!mpext)
1143 return false;
1144 __skb_ext_set(skb, SKB_EXT_MPTCP, mpext);
1145 return true;
1146}
1147
6e628cd3 1148static struct sk_buff *__mptcp_do_alloc_tx_skb(struct sock *sk, gfp_t gfp)
724cfd2e
PA		 1149{
1150 struct sk_buff *skb;
1151
6e628cd3 1152 skb = alloc_skb_fclone(MAX_TCP_HEADER, gfp);
724cfd2e 1153 if (likely(skb)) {
6e628cd3 1154 if (likely(__mptcp_add_ext(skb, gfp))) {
724cfd2e 1155 skb_reserve(skb, MAX_TCP_HEADER);
a52fe46e 1156 skb->ip_summed = CHECKSUM_PARTIAL;
f70cad10 1157 INIT_LIST_HEAD(&skb->tcp_tsorted_anchor);
724cfd2e
PA		 1158 return skb;
1159 }
1160 __kfree_skb(skb);
1161 } else {
1162 mptcp_enter_memory_pressure(sk);
1163 }
1164 return NULL;
1165}
1166
f70cad10 1167static struct sk_buff *__mptcp_alloc_tx_skb(struct sock *sk, struct sock *ssk, gfp_t gfp)
724cfd2e 1168{
724cfd2e
PA		 1169 struct sk_buff *skb;
1170
6e628cd3 1171 skb = __mptcp_do_alloc_tx_skb(sk, gfp);
724cfd2e 1172 if (!skb)
f70cad10 1173 return NULL;
724cfd2e
PA		 1174
1175 if (likely(sk_wmem_schedule(ssk, skb->truesize))) {
f70cad10
PA		 1176 tcp_skb_entail(ssk, skb);
1177 return skb;
724cfd2e 1178 }
3ef3905a 1179 tcp_skb_tsorted_anchor_cleanup(skb);
724cfd2e 1180 kfree_skb(skb);
f70cad10 1181 return NULL;
724cfd2e
PA		 1182}
1183
f70cad10 1184static struct sk_buff *mptcp_alloc_tx_skb(struct sock *sk, struct sock *ssk, bool data_lock_held)
724cfd2e 1185{
1094c6fe 1186 gfp_t gfp = data_lock_held ? GFP_ATOMIC : sk->sk_allocation;
724cfd2e 1187
1094c6fe 1188 return __mptcp_alloc_tx_skb(sk, ssk, gfp);
724cfd2e
PA		 1189}
1190
d0cc2987
GT		 1191/* note: this always recompute the csum on the whole skb, even
1192 * if we just appended a single frag. More status info needed
1193 */
1194static void mptcp_update_data_checksum(struct sk_buff *skb, int added)
1195{
1196 struct mptcp_ext *mpext = mptcp_get_ext(skb);
1197 __wsum csum = ~csum_unfold(mpext->csum);
1198 int offset = skb->len - added;
1199
1200 mpext->csum = csum_fold(csum_block_add(csum, skb_checksum(skb, offset, added, 0), offset));
1201}
1202
1e39e5a3
GT		 1203static void mptcp_update_infinite_map(struct mptcp_sock *msk,
1204 struct sock *ssk,
1205 struct mptcp_ext *mpext)
1206{
1207 if (!mpext)
1208 return;
1209
1210 mpext->infinite_map = 1;
1211 mpext->data_len = 0;
1212
104125b8 1213 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPTX);
1e39e5a3
GT		 1214 mptcp_subflow_ctx(ssk)->send_infinite_map = 0;
1215 pr_fallback(msk);
d51991e2 1216 mptcp_do_fallback(ssk);
1e39e5a3
GT		 1217}
1218
9fce92f0
PA		 1219#define MPTCP_MAX_GSO_SIZE (GSO_LEGACY_MAX_SIZE - (MAX_TCP_HEADER + 1))
1220
6d0060f6 1221static int mptcp_sendmsg_frag(struct sock *sk, struct sock *ssk,
d9ca1de8 1222 struct mptcp_data_frag *dfrag,
caf971df 1223 struct mptcp_sendmsg_info *info)
6d0060f6 1224{
d9ca1de8 1225 u64 data_seq = dfrag->data_seq + info->sent;
f70cad10 1226 int offset = dfrag->offset + info->sent;
6d0060f6 1227 struct mptcp_sock *msk = mptcp_sk(sk);
6f8a612a 1228 bool zero_window_probe = false;
6d0060f6 1229 struct mptcp_ext *mpext = NULL;
f70cad10
PA		 1230 bool can_coalesce = false;
1231 bool reuse_skb = true;
1232 struct sk_buff *skb;
1233 size_t copy;
1234 int i;
6d0060f6 1235
e4b61351 1236 pr_debug("msk=%p ssk=%p sending dfrag at seq=%llu len=%u already sent=%u",
d9ca1de8
PA		 1237 msk, ssk, dfrag->data_seq, dfrag->data_len, info->sent);
1238
f70cad10
PA		 1239 if (WARN_ON_ONCE(info->sent > info->limit ||
1240 info->limit > dfrag->data_len))
1241 return 0;
1242
c886d702
PA		 1243 if (unlikely(!__tcp_can_send(ssk)))
1244 return -EAGAIN;
1245
d9ca1de8 1246 /* compute send limit */
9fce92f0
PA		 1247 if (unlikely(ssk->sk_gso_max_size > MPTCP_MAX_GSO_SIZE))
1248 ssk->sk_gso_max_size = MPTCP_MAX_GSO_SIZE;
d9ca1de8 1249 info->mss_now = tcp_send_mss(ssk, &info->size_goal, info->flags);
f70cad10
PA		 1250 copy = info->size_goal;
1251
57040755 1252 skb = tcp_write_queue_tail(ssk);
f70cad10 1253 if (skb && copy > skb->len) {
57040755
PA		 1254 /* Limit the write to the size available in the
1255 * current skb, if any, so that we create at most a new skb.
1256 * Explicitly tells TCP internals to avoid collapsing on later
1257 * queue management operation, to avoid breaking the ext <->
1258 * SSN association set here
1259 */
a16c054b 1260 mpext = mptcp_get_ext(skb);
1094c6fe 1261 if (!mptcp_skb_can_collapse_to(data_seq, skb, mpext)) {
57040755 1262 TCP_SKB_CB(skb)->eor = 1;
1094c6fe
PA		 1263 goto alloc_skb;
1264 }
1265
f70cad10
PA		 1266 i = skb_shinfo(skb)->nr_frags;
1267 can_coalesce = skb_can_coalesce(skb, i, dfrag->page, offset);
657b991a 1268 if (!can_coalesce && i >= READ_ONCE(sysctl_max_skb_frags)) {
f70cad10
PA		 1269 tcp_mark_push(tcp_sk(ssk), skb);
1270 goto alloc_skb;
15e6ca97 1271 }
18b683bf 1272
f70cad10
PA		 1273 copy -= skb->len;
1274 } else {
1094c6fe 1275alloc_skb:
f70cad10
PA		 1276 skb = mptcp_alloc_tx_skb(sk, ssk, info->data_lock_held);
1277 if (!skb)
1278 return -ENOMEM;
1279
1280 i = skb_shinfo(skb)->nr_frags;
1281 reuse_skb = false;
a16c054b 1282 mpext = mptcp_get_ext(skb);
f70cad10 1283 }
1094c6fe 1284
6f8a612a 1285 /* Zero window and all data acked? Probe. */
b713d006 1286 copy = mptcp_check_allowed_size(msk, ssk, data_seq, copy);
f70cad10 1287 if (copy == 0) {
7439d687
PA		 1288 u64 snd_una = READ_ONCE(msk->snd_una);
1289
72377ab2 1290 if (snd_una != msk->snd_nxt || tcp_write_queue_tail(ssk)) {
27728ba8 1291 tcp_remove_empty_skb(ssk);
6f8a612a 1292 return 0;
f70cad10
PA		 1293 }
1294
6f8a612a 1295 zero_window_probe = true;
7439d687 1296 data_seq = snd_una - 1;
f70cad10 1297 copy = 1;
f70cad10 1298 }
d027236c 1299
f70cad10
PA		 1300 copy = min_t(size_t, copy, info->limit - info->sent);
1301 if (!sk_wmem_schedule(ssk, copy)) {
27728ba8 1302 tcp_remove_empty_skb(ssk);
e2223995 1303 return -ENOMEM;
35759383 1304 }
18b683bf 1305
f70cad10
PA		 1306 if (can_coalesce) {
1307 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1308 } else {
1309 get_page(dfrag->page);
1310 skb_fill_page_desc(skb, i, dfrag->page, offset, copy);
57040755
PA		 1311 }
1312
f70cad10
PA		 1313 skb->len += copy;
1314 skb->data_len += copy;
1315 skb->truesize += copy;
1316 sk_wmem_queued_add(ssk, copy);
1317 sk_mem_charge(ssk, copy);
f70cad10
PA		 1318 WRITE_ONCE(tcp_sk(ssk)->write_seq, tcp_sk(ssk)->write_seq + copy);
1319 TCP_SKB_CB(skb)->end_seq += copy;
1320 tcp_skb_pcount_set(skb, 0);
1321
1322 /* on skb reuse we just need to update the DSS len */
1323 if (reuse_skb) {
1324 TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH;
1325 mpext->data_len += copy;
f70cad10 1326 goto out;
724cfd2e 1327 }
6d0060f6
MM		 1328
1329 memset(mpext, 0, sizeof(*mpext));
d9ca1de8 1330 mpext->data_seq = data_seq;
6d0060f6 1331 mpext->subflow_seq = mptcp_subflow_ctx(ssk)->rel_write_seq;
f70cad10 1332 mpext->data_len = copy;
6d0060f6
MM		 1333 mpext->use_map = 1;
1334 mpext->dsn64 = 1;
1335
1336 pr_debug("data_seq=%llu subflow_seq=%u data_len=%u dsn64=%d",
1337 mpext->data_seq, mpext->subflow_seq, mpext->data_len,
1338 mpext->dsn64);
1339
6f8a612a 1340 if (zero_window_probe) {
f70cad10 1341 mptcp_subflow_ctx(ssk)->rel_write_seq += copy;
6f8a612a 1342 mpext->frozen = 1;
d0cc2987 1343 if (READ_ONCE(msk->csum_enabled))
f70cad10 1344 mptcp_update_data_checksum(skb, copy);
6f8a612a 1345 tcp_push_pending_frames(ssk);
d0cc2987 1346 return 0;
6f8a612a 1347 }
57040755 1348out:
d0cc2987 1349 if (READ_ONCE(msk->csum_enabled))
f70cad10 1350 mptcp_update_data_checksum(skb, copy);
1e39e5a3
GT		 1351 if (mptcp_subflow_ctx(ssk)->send_infinite_map)
1352 mptcp_update_infinite_map(msk, ssk, mpext);
0eb4e7ee 1353 trace_mptcp_sendmsg_frag(mpext);
f70cad10
PA		 1354 mptcp_subflow_ctx(ssk)->rel_write_seq += copy;
1355 return copy;
6d0060f6
MM		 1356}
1357
d5f49190
PA		 1358#define MPTCP_SEND_BURST_SIZE ((1 << 16) - \
1359 sizeof(struct tcphdr) - \
1360 MAX_TCP_OPTION_SPACE - \
1361 sizeof(struct ipv6hdr) - \
1362 sizeof(struct frag_hdr))
1363
1364struct subflow_send_info {
1365 struct sock *ssk;
3ce0852c 1366 u64 linger_time;
d5f49190
PA		 1367};
1368
ff5a0b42
PA		 1369void mptcp_subflow_set_active(struct mptcp_subflow_context *subflow)
1370{
1371 if (!subflow->stale)
1372 return;
1373
1374 subflow->stale = 0;
fc1b4e3b 1375 MPTCP_INC_STATS(sock_net(mptcp_subflow_tcp_sock(subflow)), MPTCP_MIB_SUBFLOWRECOVER);
ff5a0b42
PA		 1376}
1377
1378bool mptcp_subflow_active(struct mptcp_subflow_context *subflow)
1379{
1380 if (unlikely(subflow->stale)) {
1381 u32 rcv_tstamp = READ_ONCE(tcp_sk(mptcp_subflow_tcp_sock(subflow))->rcv_tstamp);
1382
1383 if (subflow->stale_rcv_tstamp == rcv_tstamp)
1384 return false;
1385
1386 mptcp_subflow_set_active(subflow);
1387 }
1388 return __mptcp_subflow_active(subflow);
1389}
1390
3ce0852c
PA		 1391#define SSK_MODE_ACTIVE 0
1392#define SSK_MODE_BACKUP 1
1393#define SSK_MODE_MAX 2
1394
33d41c9c
PA		 1395/* implement the mptcp packet scheduler;
1396 * returns the subflow that will transmit the next DSS
1397 * additionally updates the rtx timeout
1398 */
07336a87 1399struct sock *mptcp_subflow_get_send(struct mptcp_sock *msk)
f296234c 1400{
3ce0852c 1401 struct subflow_send_info send_info[SSK_MODE_MAX];
f296234c 1402 struct mptcp_subflow_context *subflow;
33d41c9c 1403 struct sock *sk = (struct sock *)msk;
3ce0852c 1404 u32 pace, burst, wmem;
d5f49190
PA		 1405 int i, nr_active = 0;
1406 struct sock *ssk;
3ce0852c 1407 u64 linger_time;
33d41c9c 1408 long tout = 0;
f296234c 1409
d5f49190 1410 /* pick the subflow with the lower wmem/wspace ratio */
3ce0852c 1411 for (i = 0; i < SSK_MODE_MAX; ++i) {
d5f49190 1412 send_info[i].ssk = NULL;
3ce0852c 1413 send_info[i].linger_time = -1;
d5f49190 1414 }
3ce0852c 1415
d5f49190 1416 mptcp_for_each_subflow(msk, subflow) {
e10a9892 1417 trace_mptcp_subflow_get_send(subflow);
d5f49190
PA		 1418 ssk = mptcp_subflow_tcp_sock(subflow);
1419 if (!mptcp_subflow_active(subflow))
1420 continue;
1421
33d41c9c 1422 tout = max(tout, mptcp_timeout_from_subflow(subflow));
d5f49190 1423 nr_active += !subflow->backup;
3ce0852c
PA		 1424 pace = subflow->avg_pacing_rate;
1425 if (unlikely(!pace)) {
1426 /* init pacing rate from socket */
1427 subflow->avg_pacing_rate = READ_ONCE(ssk->sk_pacing_rate);
1428 pace = subflow->avg_pacing_rate;
1429 if (!pace)
1430 continue;
1431 }
f296234c 1432
3ce0852c
PA		 1433 linger_time = div_u64((u64)READ_ONCE(ssk->sk_wmem_queued) << 32, pace);
1434 if (linger_time < send_info[subflow->backup].linger_time) {
d5f49190 1435 send_info[subflow->backup].ssk = ssk;
3ce0852c 1436 send_info[subflow->backup].linger_time = linger_time;
d5f49190 1437 }
f296234c 1438 }
33d41c9c 1439 __mptcp_set_timeout(sk, tout);
f296234c 1440
d5f49190
PA		 1441 /* pick the best backup if no other subflow is active */
1442 if (!nr_active)
3ce0852c
PA		 1443 send_info[SSK_MODE_ACTIVE].ssk = send_info[SSK_MODE_BACKUP].ssk;
1444
1445 /* According to the blest algorithm, to avoid HoL blocking for the
1446 * faster flow, we need to:
1447 * - estimate the faster flow linger time
1448 * - use the above to estimate the amount of byte transferred
1449 * by the faster flow
1450 * - check that the amount of queued data is greter than the above,
1451 * otherwise do not use the picked, slower, subflow
1452 * We select the subflow with the shorter estimated time to flush
1453 * the queued mem, which basically ensure the above. We just need
1454 * to check that subflow has a non empty cwin.
1455 */
1456 ssk = send_info[SSK_MODE_ACTIVE].ssk;
b713d006 1457 if (!ssk || !sk_stream_memory_free(ssk))
3ce0852c 1458 return NULL;
5cf92bba 1459
b713d006 1460 burst = min_t(int, MPTCP_SEND_BURST_SIZE, mptcp_wnd_end(msk) - msk->snd_nxt);
3ce0852c 1461 wmem = READ_ONCE(ssk->sk_wmem_queued);
ebc1e08f 1462 if (!burst)
b713d006 1463 return ssk;
b713d006 1464
3ce0852c
PA		 1465 subflow = mptcp_subflow_ctx(ssk);
1466 subflow->avg_pacing_rate = div_u64((u64)subflow->avg_pacing_rate * wmem +
1467 READ_ONCE(ssk->sk_pacing_rate) * burst,
1468 burst + wmem);
3ce0852c
PA		 1469 msk->snd_burst = burst;
1470 return ssk;
f296234c
PK		 1471}
1472
b8e0def3 1473static void mptcp_push_release(struct sock *ssk, struct mptcp_sendmsg_info *info)
d9ca1de8 1474{
d9ca1de8
PA		 1475 tcp_push(ssk, 0, info->mss_now, tcp_sk(ssk)->nonagle, info->size_goal);
1476 release_sock(ssk);
1477}
1478
0d199e43
FW		 1479static void mptcp_update_post_push(struct mptcp_sock *msk,
1480 struct mptcp_data_frag *dfrag,
1481 u32 sent)
1482{
1483 u64 snd_nxt_new = dfrag->data_seq;
1484
1485 dfrag->already_sent += sent;
1486
1487 msk->snd_burst -= sent;
0d199e43
FW		 1488
1489 snd_nxt_new += dfrag->already_sent;
1490
1491 /* snd_nxt_new can be smaller than snd_nxt in case mptcp
1492 * is recovering after a failover. In that event, this re-sends
1493 * old segments.
1494 *
1495 * Thus compute snd_nxt_new candidate based on
1496 * the dfrag->data_seq that was sent and the data
1497 * that has been handed to the subflow for transmission
1498 * and skip update in case it was old dfrag.
1499 */
38967f42
PA		 1500 if (likely(after64(snd_nxt_new, msk->snd_nxt))) {
1501 msk->bytes_sent += snd_nxt_new - msk->snd_nxt;
0d199e43 1502 msk->snd_nxt = snd_nxt_new;
38967f42 1503 }
0d199e43
FW		 1504}
1505
8b38217a 1506void mptcp_check_and_set_pending(struct sock *sk)
3241a9c0 1507{
e9d09bac
PA		 1508 if (mptcp_send_head(sk))
1509 mptcp_sk(sk)->push_pending |= BIT(MPTCP_PUSH_PENDING);
3241a9c0
FW		 1510}
1511
c5b4297d
GT		 1512static int __subflow_push_pending(struct sock *sk, struct sock *ssk,
1513 struct mptcp_sendmsg_info *info)
f870fa0b
MM		 1514{
1515 struct mptcp_sock *msk = mptcp_sk(sk);
d9ca1de8 1516 struct mptcp_data_frag *dfrag;
c5b4297d 1517 int len, copied = 0, err = 0;
d9ca1de8
PA		 1518
1519 while ((dfrag = mptcp_send_head(sk))) {
c5b4297d
GT		 1520 info->sent = dfrag->already_sent;
1521 info->limit = dfrag->data_len;
d9ca1de8
PA		 1522 len = dfrag->data_len - dfrag->already_sent;
1523 while (len > 0) {
1524 int ret = 0;
1525
c5b4297d 1526 ret = mptcp_sendmsg_frag(sk, ssk, dfrag, info);
d9ca1de8 1527 if (ret <= 0) {
c5b4297d 1528 err = copied ? : ret;
d9ca1de8
PA		 1529 goto out;
1530 }
1531
c5b4297d
GT		 1532 info->sent += ret;
1533 copied += ret;
d9ca1de8 1534 len -= ret;
0d199e43
FW		 1535
1536 mptcp_update_post_push(msk, dfrag, ret);
d9ca1de8
PA		 1537 }
1538 WRITE_ONCE(msk->first_pending, mptcp_send_next(sk));
c5b4297d
GT		 1539
1540 if (msk->snd_burst <= 0 ||
1541 !sk_stream_memory_free(ssk) ||
1542 !mptcp_subflow_active(mptcp_subflow_ctx(ssk))) {
1543 err = copied;
1544 goto out;
1545 }
1546 mptcp_set_timeout(sk);
1547 }
1548 err = copied;
1549
1550out:
1551 return err;
1552}
1553
1554void __mptcp_push_pending(struct sock *sk, unsigned int flags)
1555{
1556 struct sock *prev_ssk = NULL, *ssk = NULL;
1557 struct mptcp_sock *msk = mptcp_sk(sk);
1558 struct mptcp_sendmsg_info info = {
1559 .flags = flags,
1560 };
1561 bool do_check_data_fin = false;
0fa1b378 1562 int push_count = 1;
c5b4297d 1563
0fa1b378
GT		 1564 while (mptcp_send_head(sk) && (push_count > 0)) {
1565 struct mptcp_subflow_context *subflow;
c5b4297d
GT		 1566 int ret = 0;
1567
0fa1b378
GT		 1568 if (mptcp_sched_get_send(msk))
1569 break;
c5b4297d 1570
0fa1b378
GT		 1571 push_count = 0;
1572
1573 mptcp_for_each_subflow(msk, subflow) {
1574 if (READ_ONCE(subflow->scheduled)) {
1575 mptcp_subflow_set_scheduled(subflow, false);
1576
1577 prev_ssk = ssk;
1578 ssk = mptcp_subflow_tcp_sock(subflow);
1579 if (ssk != prev_ssk) {
1580 /* First check. If the ssk has changed since
1581 * the last round, release prev_ssk
1582 */
1583 if (prev_ssk)
1584 mptcp_push_release(prev_ssk, &info);
1585
1586 /* Need to lock the new subflow only if different
1587 * from the previous one, otherwise we are still
1588 * helding the relevant lock
1589 */
1590 lock_sock(ssk);
1591 }
1592
1593 push_count++;
1594
1595 ret = __subflow_push_pending(sk, ssk, &info);
1596 if (ret <= 0) {
1597 if (ret != -EAGAIN ||
1598 (1 << ssk->sk_state) &
1599 (TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2 | TCPF_CLOSE))
1600 push_count--;
1601 continue;
1602 }
1603 do_check_data_fin = true;
1604 }
c5b4297d 1605 }
d9ca1de8
PA		 1606 }
1607
1608 /* at this point we held the socket lock for the last subflow we used */
1609 if (ssk)
b8e0def3 1610 mptcp_push_release(ssk, &info);
d9ca1de8 1611
33d41c9c 1612 /* ensure the rtx timer is running */
f6909dc1
PA		 1613 if (!mptcp_rtx_timer_pending(sk))
1614 mptcp_reset_rtx_timer(sk);
0522b424 1615 if (do_check_data_fin)
81c1d029 1616 mptcp_check_send_data_fin(sk);
d9ca1de8
PA		 1617}
1618
73a0052a 1619static void __mptcp_subflow_push_pending(struct sock *sk, struct sock *ssk, bool first)
6e628cd3
PA		 1620{
1621 struct mptcp_sock *msk = mptcp_sk(sk);
1094c6fe
PA		 1622 struct mptcp_sendmsg_info info = {
1623 .data_lock_held = true,
1624 };
0fa1b378 1625 bool keep_pushing = true;
b19bc294 1626 struct sock *xmit_ssk;
c5b4297d 1627 int copied = 0;
6e628cd3
PA		 1628
1629 info.flags = 0;
0fa1b378
GT		 1630 while (mptcp_send_head(sk) && keep_pushing) {
1631 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
c5b4297d 1632 int ret = 0;
b19bc294 1633
c5b4297d
GT		 1634 /* check for a different subflow usage only after
1635 * spooling the first chunk of data
1636 */
0fa1b378
GT		 1637 if (first) {
1638 mptcp_subflow_set_scheduled(subflow, false);
1639 ret = __subflow_push_pending(sk, ssk, &info);
1640 first = false;
1641 if (ret <= 0)
1642 break;
1643 copied += ret;
1644 continue;
1645 }
1646
1647 if (mptcp_sched_get_send(msk))
c5b4297d 1648 goto out;
0fa1b378
GT		 1649
1650 if (READ_ONCE(subflow->scheduled)) {
1651 mptcp_subflow_set_scheduled(subflow, false);
1652 ret = __subflow_push_pending(sk, ssk, &info);
1653 if (ret <= 0)
1654 keep_pushing = false;
1655 copied += ret;
6e628cd3 1656 }
c5b4297d 1657
0fa1b378
GT		 1658 mptcp_for_each_subflow(msk, subflow) {
1659 if (READ_ONCE(subflow->scheduled)) {
1660 xmit_ssk = mptcp_subflow_tcp_sock(subflow);
1661 if (xmit_ssk != ssk) {
1662 mptcp_subflow_delegate(subflow,
1663 MPTCP_DELEGATE_SEND);
1664 keep_pushing = false;
1665 }
1666 }
1667 }
6e628cd3
PA		 1668 }
1669
1670out:
1671 /* __mptcp_alloc_tx_skb could have released some wmem and we are
1672 * not going to flush it via release_sock()
1673 */
6e628cd3 1674 if (copied) {
6e628cd3
PA		 1675 tcp_push(ssk, 0, info.mss_now, tcp_sk(ssk)->nonagle,
1676 info.size_goal);
f6909dc1
PA		 1677 if (!mptcp_rtx_timer_pending(sk))
1678 mptcp_reset_rtx_timer(sk);
d09d818e 1679
6e628cd3
PA		 1680 if (msk->snd_data_fin_enable &&
1681 msk->snd_nxt + 1 == msk->write_seq)
1682 mptcp_schedule_work(sk);
1683 }
1684}
1685
5cf92bba
PA		 1686static void mptcp_set_nospace(struct sock *sk)
1687{
1688 /* enable autotune */
1689 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1690
1691 /* will be cleared on avail space */
1692 set_bit(MPTCP_NOSPACE, &mptcp_sk(sk)->flags);
1693}
1694
7d803344
PA		 1695static int mptcp_disconnect(struct sock *sk, int flags);
1696
a2702a07 1697static int mptcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg,
fa9e5746
PA		 1698 size_t len, int *copied_syn)
1699{
1700 unsigned int saved_flags = msg->msg_flags;
1701 struct mptcp_sock *msk = mptcp_sk(sk);
a2702a07 1702 struct sock *ssk;
fa9e5746
PA		 1703 int ret;
1704
a2702a07
PA		 1705 /* on flags based fastopen the mptcp is supposed to create the
1706 * first subflow right now. Otherwise we are in the defer_connect
1707 * path, and the first subflow must be already present.
1708 * Since the defer_connect flag is cleared after the first succsful
1709 * fastopen attempt, no need to check for additional subflow status.
1710 */
ddb1a072 1711 if (msg->msg_flags & MSG_FASTOPEN) {
3f326a82
PA		 1712 ssk = __mptcp_nmpc_sk(msk);
1713 if (IS_ERR(ssk))
1714 return PTR_ERR(ssk);
ddb1a072 1715 }
a2702a07
PA		 1716 if (!msk->first)
1717 return -EINVAL;
1718
1719 ssk = msk->first;
1720
fa9e5746
PA		 1721 lock_sock(ssk);
1722 msg->msg_flags |= MSG_DONTWAIT;
7d803344 1723 msk->fastopening = 1;
fa9e5746 1724 ret = tcp_sendmsg_fastopen(ssk, msg, copied_syn, len, NULL);
7d803344 1725 msk->fastopening = 0;
fa9e5746
PA		 1726 msg->msg_flags = saved_flags;
1727 release_sock(ssk);
1728
1729 /* do the blocking bits of inet_stream_connect outside the ssk socket lock */
1730 if (ret == -EINPROGRESS && !(msg->msg_flags & MSG_DONTWAIT)) {
1731 ret = __inet_stream_connect(sk->sk_socket, msg->msg_name,
1732 msg->msg_namelen, msg->msg_flags, 1);
1733
1734 /* Keep the same behaviour of plain TCP: zero the copied bytes in
1735 * case of any error, except timeout or signal
1736 */
1737 if (ret && ret != -EINPROGRESS && ret != -ERESTARTSYS && ret != -EINTR)
1738 *copied_syn = 0;
7d803344 1739 } else if (ret && ret != -EINPROGRESS) {
c2b2ae39
PA		 1740 /* The disconnect() op called by tcp_sendmsg_fastopen()/
1741 * __inet_stream_connect() can fail, due to looking check,
1742 * see mptcp_disconnect().
1743 * Attempt it again outside the problematic scope.
1744 */
1745 if (!mptcp_disconnect(sk, 0))
1746 sk->sk_socket->state = SS_UNCONNECTED;
fa9e5746 1747 }
08e39c0d 1748 inet_clear_bit(DEFER_CONNECT, sk);
fa9e5746
PA		 1749
1750 return ret;
1751}
1752
0ffe8e74
PA		 1753static int do_copy_data_nocache(struct sock *sk, int copy,
1754 struct iov_iter *from, char *to)
1755{
1756 if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1757 if (!copy_from_iter_full_nocache(to, copy, from))
1758 return -EFAULT;
1759 } else if (!copy_from_iter_full(to, copy, from)) {
1760 return -EFAULT;
1761 }
1762 return 0;
1763}
1764
d9ca1de8
PA		 1765static int mptcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1766{
1767 struct mptcp_sock *msk = mptcp_sk(sk);
17091708 1768 struct page_frag *pfrag;
6d0060f6 1769 size_t copied = 0;
caf971df 1770 int ret = 0;
6d0060f6 1771 long timeo;
f870fa0b 1772
987858e5 1773 /* silently ignore everything else */
1e777f39 1774 msg->msg_flags &= MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL | MSG_FASTOPEN;
987858e5 1775
6511882c 1776 lock_sock(sk);
1954b860 1777
08e39c0d
ED		 1778 if (unlikely(inet_test_bit(DEFER_CONNECT, sk) ||
1779 msg->msg_flags & MSG_FASTOPEN)) {
d98a82a6
DS		 1780 int copied_syn = 0;
1781
a2702a07 1782 ret = mptcp_sendmsg_fastopen(sk, msg, len, &copied_syn);
d98a82a6 1783 copied += copied_syn;
fa9e5746 1784 if (ret == -EINPROGRESS && copied_syn > 0)
d98a82a6 1785 goto out;
fa9e5746 1786 else if (ret)
69800e51 1787 goto do_error;
d98a82a6
DS		 1788 }
1789
1954b860
MM		 1790 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1791
1792 if ((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) {
1793 ret = sk_stream_wait_connect(sk, &timeo);
1794 if (ret)
69800e51 1795 goto do_error;
1954b860
MM		 1796 }
1797
69800e51
PA		 1798 ret = -EPIPE;
1799 if (unlikely(sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN)))
1800 goto do_error;
1801
17091708 1802 pfrag = sk_page_frag(sk);
18b683bf 1803
d9ca1de8 1804 while (msg_data_left(msg)) {
724cfd2e 1805 int total_ts, frag_truesize = 0;
d9ca1de8 1806 struct mptcp_data_frag *dfrag;
d9ca1de8
PA		 1807 bool dfrag_collapsed;
1808 size_t psize, offset;
18b683bf 1809
d9ca1de8
PA		 1810 /* reuse tail pfrag, if possible, or carve a new one from the
1811 * page allocator
1812 */
1813 dfrag = mptcp_pending_tail(sk);
1814 dfrag_collapsed = mptcp_frag_can_collapse_to(msk, pfrag, dfrag);
1815 if (!dfrag_collapsed) {
6e628cd3
PA		 1816 if (!sk_stream_memory_free(sk))
1817 goto wait_for_memory;
1818
d9ca1de8
PA		 1819 if (!mptcp_page_frag_refill(sk, pfrag))
1820 goto wait_for_memory;
1821
1822 dfrag = mptcp_carve_data_frag(msk, pfrag, pfrag->offset);
1823 frag_truesize = dfrag->overhead;
72511aab 1824 }
6d0060f6 1825
d9ca1de8
PA		 1826 /* we do not bound vs wspace, to allow a single packet.
1827 * memory accounting will prevent execessive memory usage
1828 * anyway
d5f49190 1829 */
d9ca1de8
PA		 1830 offset = dfrag->offset + dfrag->data_len;
1831 psize = pfrag->size - offset;
1832 psize = min_t(size_t, psize, msg_data_left(msg));
724cfd2e 1833 total_ts = psize + frag_truesize;
d9ca1de8 1834
6511882c 1835 if (!sk_wmem_schedule(sk, total_ts))
724cfd2e 1836 goto wait_for_memory;
724cfd2e 1837
0ffe8e74
PA		 1838 ret = do_copy_data_nocache(sk, psize, &msg->msg_iter,
1839 page_address(dfrag->page) + offset);
1840 if (ret)
69800e51 1841 goto do_error;
72511aab 1842
d9ca1de8 1843 /* data successfully copied into the write queue */
5e6300e7 1844 sk_forward_alloc_add(sk, -total_ts);
d9ca1de8
PA		 1845 copied += psize;
1846 dfrag->data_len += psize;
1847 frag_truesize += psize;
1848 pfrag->offset += frag_truesize;
1849 WRITE_ONCE(msk->write_seq, msk->write_seq + psize);
1850
1851 /* charge data on mptcp pending queue to the msk socket
1852 * Note: we charge such data both to sk and ssk
fb529e62 1853 */
d9ca1de8 1854 sk_wmem_queued_add(sk, frag_truesize);
d9ca1de8
PA		 1855 if (!dfrag_collapsed) {
1856 get_page(dfrag->page);
1857 list_add_tail(&dfrag->list, &msk->rtx_queue);
1858 if (!msk->first_pending)
1859 WRITE_ONCE(msk->first_pending, dfrag);
fb529e62 1860 }
e4b61351 1861 pr_debug("msk=%p dfrag at seq=%llu len=%u sent=%u new=%d", msk,
d9ca1de8
PA		 1862 dfrag->data_seq, dfrag->data_len, dfrag->already_sent,
1863 !dfrag_collapsed);
6d0060f6 1864
d9ca1de8 1865 continue;
b51f9b80 1866
d9ca1de8 1867wait_for_memory:
5cf92bba 1868 mptcp_set_nospace(sk);
c2e6048f 1869 __mptcp_push_pending(sk, msg->msg_flags);
d9ca1de8
PA		 1870 ret = sk_stream_wait_memory(sk, &timeo);
1871 if (ret)
69800e51 1872 goto do_error;
57040755 1873 }
6d0060f6 1874
13e16037 1875 if (copied)
c2e6048f 1876 __mptcp_push_pending(sk, msg->msg_flags);
d9ca1de8 1877
1954b860 1878out:
cec37a6e 1879 release_sock(sk);
69800e51
PA		 1880 return copied;
1881
1882do_error:
1883 if (copied)
1884 goto out;
1885
1886 copied = sk_stream_error(sk, msg->msg_flags, ret);
1887 goto out;
f870fa0b
MM		 1888}
1889
6771bfd9
FW		 1890static int __mptcp_recvmsg_mskq(struct mptcp_sock *msk,
1891 struct msghdr *msg,
b7f653b2
FW		 1892 size_t len, int flags,
1893 struct scm_timestamping_internal *tss,
1894 int *cmsg_flags)
6771bfd9 1895{
ca4fb892 1896 struct sk_buff *skb, *tmp;
6771bfd9
FW		 1897 int copied = 0;
1898
ca4fb892 1899 skb_queue_walk_safe(&msk->receive_queue, skb, tmp) {
6771bfd9
FW		 1900 u32 offset = MPTCP_SKB_CB(skb)->offset;
1901 u32 data_len = skb->len - offset;
1902 u32 count = min_t(size_t, len - copied, data_len);
1903 int err;
1904
d976092c
PA		 1905 if (!(flags & MSG_TRUNC)) {
1906 err = skb_copy_datagram_msg(skb, offset, msg, count);
1907 if (unlikely(err < 0)) {
1908 if (!copied)
1909 return err;
1910 break;
1911 }
6771bfd9
FW		 1912 }
1913
b7f653b2
FW		 1914 if (MPTCP_SKB_CB(skb)->has_rxtstamp) {
1915 tcp_update_recv_tstamps(skb, tss);
1916 *cmsg_flags |= MPTCP_CMSG_TS;
1917 }
1918
6771bfd9
FW		 1919 copied += count;
1920
1921 if (count < data_len) {
2c9e7765 1922 if (!(flags & MSG_PEEK)) {
ca4fb892 1923 MPTCP_SKB_CB(skb)->offset += count;
2c9e7765 1924 MPTCP_SKB_CB(skb)->map_seq += count;
5684ab1a 1925 msk->bytes_consumed += count;
2c9e7765 1926 }
6771bfd9
FW		 1927 break;
1928 }
1929
ca4fb892
YL		 1930 if (!(flags & MSG_PEEK)) {
1931 /* we will bulk release the skb memory later */
1932 skb->destructor = NULL;
ce599c51 1933 WRITE_ONCE(msk->rmem_released, msk->rmem_released + skb->truesize);
ca4fb892
YL		 1934 __skb_unlink(skb, &msk->receive_queue);
1935 __kfree_skb(skb);
5684ab1a 1936 msk->bytes_consumed += count;
ca4fb892 1937 }
6771bfd9
FW		 1938
1939 if (copied >= len)
1940 break;
1941 }
1942
1943 return copied;
1944}
1945
a6b118fe
FW		 1946/* receive buffer autotuning. See tcp_rcv_space_adjust for more information.
1947 *
1948 * Only difference: Use highest rtt estimate of the subflows in use.
1949 */
1950static void mptcp_rcv_space_adjust(struct mptcp_sock *msk, int copied)
1951{
1952 struct mptcp_subflow_context *subflow;
1953 struct sock *sk = (struct sock *)msk;
b8dc6d6c 1954 u8 scaling_ratio = U8_MAX;
a6b118fe
FW		 1955 u32 time, advmss = 1;
1956 u64 rtt_us, mstamp;
1957
109cdeb8 1958 msk_owned_by_me(msk);
a6b118fe
FW		 1959
1960 if (copied <= 0)
1961 return;
1962
1963 msk->rcvq_space.copied += copied;
1964
1965 mstamp = div_u64(tcp_clock_ns(), NSEC_PER_USEC);
1966 time = tcp_stamp_us_delta(mstamp, msk->rcvq_space.time);
1967
1968 rtt_us = msk->rcvq_space.rtt_us;
1969 if (rtt_us && time < (rtt_us >> 3))
1970 return;
1971
1972 rtt_us = 0;
1973 mptcp_for_each_subflow(msk, subflow) {
1974 const struct tcp_sock *tp;
1975 u64 sf_rtt_us;
1976 u32 sf_advmss;
1977
1978 tp = tcp_sk(mptcp_subflow_tcp_sock(subflow));
1979
1980 sf_rtt_us = READ_ONCE(tp->rcv_rtt_est.rtt_us);
1981 sf_advmss = READ_ONCE(tp->advmss);
1982
1983 rtt_us = max(sf_rtt_us, rtt_us);
1984 advmss = max(sf_advmss, advmss);
b8dc6d6c 1985 scaling_ratio = min(tp->scaling_ratio, scaling_ratio);
a6b118fe
FW		 1986 }
1987
1988 msk->rcvq_space.rtt_us = rtt_us;
b8dc6d6c 1989 msk->scaling_ratio = scaling_ratio;
a6b118fe
FW		 1990 if (time < (rtt_us >> 3) || rtt_us == 0)
1991 return;
1992
1993 if (msk->rcvq_space.copied <= msk->rcvq_space.space)
1994 goto new_measure;
1995
78047648 1996 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_moderate_rcvbuf) &&
a6b118fe 1997 !(sk->sk_userlocks & SOCK_RCVBUF_LOCK)) {
a6b118fe 1998 u64 rcvwin, grow;
b8dc6d6c 1999 int rcvbuf;
a6b118fe
FW		 2000
2001 rcvwin = ((u64)msk->rcvq_space.copied << 1) + 16 * advmss;
2002
2003 grow = rcvwin * (msk->rcvq_space.copied - msk->rcvq_space.space);
2004
2005 do_div(grow, msk->rcvq_space.space);
2006 rcvwin += (grow << 1);
2007
b8dc6d6c 2008 rcvbuf = min_t(u64, __tcp_space_from_win(scaling_ratio, rcvwin),
02739545 2009 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_rmem[2]));
a6b118fe
FW		 2010
2011 if (rcvbuf > sk->sk_rcvbuf) {
2012 u32 window_clamp;
2013
b8dc6d6c 2014 window_clamp = __tcp_win_from_space(scaling_ratio, rcvbuf);
a6b118fe
FW		 2015 WRITE_ONCE(sk->sk_rcvbuf, rcvbuf);
2016
2017 /* Make subflows follow along. If we do not do this, we
2018 * get drops at subflow level if skbs can't be moved to
2019 * the mptcp rx queue fast enough (announced rcv_win can
2020 * exceed ssk->sk_rcvbuf).
2021 */
2022 mptcp_for_each_subflow(msk, subflow) {
2023 struct sock *ssk;
c76c6956 2024 bool slow;
a6b118fe
FW		 2025
2026 ssk = mptcp_subflow_tcp_sock(subflow);
c76c6956 2027 slow = lock_sock_fast(ssk);
a6b118fe
FW		 2028 WRITE_ONCE(ssk->sk_rcvbuf, rcvbuf);
2029 tcp_sk(ssk)->window_clamp = window_clamp;
c76c6956
PA		 2030 tcp_cleanup_rbuf(ssk, 1);
2031 unlock_sock_fast(ssk, slow);
a6b118fe
FW		 2032 }
2033 }
2034 }
2035
2036 msk->rcvq_space.space = msk->rcvq_space.copied;
2037new_measure:
2038 msk->rcvq_space.copied = 0;
2039 msk->rcvq_space.time = mstamp;
2040}
2041
87952603
PA		 2042static void __mptcp_update_rmem(struct sock *sk)
2043{
2044 struct mptcp_sock *msk = mptcp_sk(sk);
2045
2046 if (!msk->rmem_released)
2047 return;
2048
2049 atomic_sub(msk->rmem_released, &sk->sk_rmem_alloc);
6511882c 2050 mptcp_rmem_uncharge(sk, msk->rmem_released);
ce599c51 2051 WRITE_ONCE(msk->rmem_released, 0);
87952603
PA		 2052}
2053
2054static void __mptcp_splice_receive_queue(struct sock *sk)
2055{
2056 struct mptcp_sock *msk = mptcp_sk(sk);
2057
2058 skb_queue_splice_tail_init(&sk->sk_receive_queue, &msk->receive_queue);
2059}
2060
e3859603 2061static bool __mptcp_move_skbs(struct mptcp_sock *msk)
6771bfd9 2062{
87952603 2063 struct sock *sk = (struct sock *)msk;
6771bfd9 2064 unsigned int moved = 0;
87952603 2065 bool ret, done;
d5f49190 2066
6771bfd9
FW		 2067 do {
2068 struct sock *ssk = mptcp_subflow_recv_lookup(msk);
65f49fe7 2069 bool slowpath;
6771bfd9 2070
87952603
PA		 2071 /* we can have data pending in the subflows only if the msk
2072 * receive buffer was full at subflow_data_ready() time,
2073 * that is an unlikely slow path.
2074 */
2075 if (likely(!ssk))
6771bfd9
FW		 2076 break;
2077
65f49fe7 2078 slowpath = lock_sock_fast(ssk);
87952603 2079 mptcp_data_lock(sk);
e3859603 2080 __mptcp_update_rmem(sk);
6771bfd9 2081 done = __mptcp_move_skbs_from_subflow(msk, ssk, &moved);
87952603 2082 mptcp_data_unlock(sk);
499ada50
PA		 2083
2084 if (unlikely(ssk->sk_err))
2085 __mptcp_error_report(sk);
65f49fe7 2086 unlock_sock_fast(ssk, slowpath);
6771bfd9
FW		 2087 } while (!done);
2088
87952603
PA		 2089 /* acquire the data lock only if some input data is pending */
2090 ret = moved > 0;
2091 if (!RB_EMPTY_ROOT(&msk->out_of_order_queue) ||
2092 !skb_queue_empty_lockless(&sk->sk_receive_queue)) {
2093 mptcp_data_lock(sk);
2094 __mptcp_update_rmem(sk);
2095 ret |= __mptcp_ofo_queue(msk);
2096 __mptcp_splice_receive_queue(sk);
2097 mptcp_data_unlock(sk);
ab174ad8 2098 }
87952603
PA		 2099 if (ret)
2100 mptcp_check_data_fin((struct sock *)msk);
2101 return !skb_queue_empty(&msk->receive_queue);
6771bfd9
FW		 2102}
2103
2c9e7765
FW		 2104static unsigned int mptcp_inq_hint(const struct sock *sk)
2105{
2106 const struct mptcp_sock *msk = mptcp_sk(sk);
2107 const struct sk_buff *skb;
2108
2109 skb = skb_peek(&msk->receive_queue);
2110 if (skb) {
2111 u64 hint_val = msk->ack_seq - MPTCP_SKB_CB(skb)->map_seq;
2112
2113 if (hint_val >= INT_MAX)
2114 return INT_MAX;
2115
2116 return (unsigned int)hint_val;
2117 }
2118
2119 if (sk->sk_state == TCP_CLOSE || (sk->sk_shutdown & RCV_SHUTDOWN))
2120 return 1;
2121
2122 return 0;
2123}
2124
f870fa0b 2125static int mptcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
ec095263 2126 int flags, int *addr_len)
f870fa0b
MM		 2127{
2128 struct mptcp_sock *msk = mptcp_sk(sk);
b7f653b2
FW		 2129 struct scm_timestamping_internal tss;
2130 int copied = 0, cmsg_flags = 0;
7a6a6cbc
PA		 2131 int target;
2132 long timeo;
f870fa0b 2133
cb9d80f4
PA		 2134 /* MSG_ERRQUEUE is really a no-op till we support IP_RECVERR */
2135 if (unlikely(flags & MSG_ERRQUEUE))
2136 return inet_recv_error(sk, msg, len, addr_len);
2137
6511882c 2138 lock_sock(sk);
fd897679
PA		 2139 if (unlikely(sk->sk_state == TCP_LISTEN)) {
2140 copied = -ENOTCONN;
2141 goto out_err;
2142 }
2143
ec095263 2144 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
7a6a6cbc
PA		 2145
2146 len = min_t(size_t, len, INT_MAX);
2147 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
2148
2c9e7765
FW		 2149 if (unlikely(msk->recvmsg_inq))
2150 cmsg_flags = MPTCP_CMSG_INQ;
2151
05e3ecea 2152 while (copied < len) {
e3859603 2153 int bytes_read;
7a6a6cbc 2154
b7f653b2 2155 bytes_read = __mptcp_recvmsg_mskq(msk, msg, len - copied, flags, &tss, &cmsg_flags);
6771bfd9
FW		 2156 if (unlikely(bytes_read < 0)) {
2157 if (!copied)
2158 copied = bytes_read;
2159 goto out_err;
2160 }
7a6a6cbc 2161
6771bfd9 2162 copied += bytes_read;
7a6a6cbc 2163
ea4ca586 2164 /* be sure to advertise window change */
e3859603
PA		 2165 mptcp_cleanup_rbuf(msk);
2166
2167 if (skb_queue_empty(&msk->receive_queue) && __mptcp_move_skbs(msk))
2168 continue;
ea4ca586 2169
7a6a6cbc
PA		 2170 /* only the master socket status is relevant here. The exit
2171 * conditions mirror closely tcp_recvmsg()
2172 */
2173 if (copied >= target)
2174 break;
2175
2176 if (copied) {
2177 if (sk->sk_err ||
2178 sk->sk_state == TCP_CLOSE ||
2179 (sk->sk_shutdown & RCV_SHUTDOWN) ||
2180 !timeo ||
2181 signal_pending(current))
2182 break;
2183 } else {
2184 if (sk->sk_err) {
2185 copied = sock_error(sk);
2186 break;
2187 }
2188
87952603
PA		 2189 if (sk->sk_shutdown & RCV_SHUTDOWN) {
2190 /* race breaker: the shutdown could be after the
2191 * previous receive queue check
2192 */
e3859603 2193 if (__mptcp_move_skbs(msk))
87952603 2194 continue;
7a6a6cbc 2195 break;
87952603 2196 }
7a6a6cbc
PA		 2197
2198 if (sk->sk_state == TCP_CLOSE) {
2199 copied = -ENOTCONN;
2200 break;
2201 }
2202
2203 if (!timeo) {
2204 copied = -EAGAIN;
2205 break;
2206 }
2207
2208 if (signal_pending(current)) {
2209 copied = sock_intr_errno(timeo);
2210 break;
2211 }
2212 }
2213
2214 pr_debug("block timeout %ld", timeo);
612f71d7 2215 sk_wait_data(sk, &timeo, NULL);
7a6a6cbc 2216 }
3c90e377 2217
6771bfd9 2218out_err:
b7f653b2
FW		 2219 if (cmsg_flags && copied >= 0) {
2220 if (cmsg_flags & MPTCP_CMSG_TS)
2221 tcp_recv_timestamp(msg, sk, &tss);
2c9e7765
FW		 2222
2223 if (cmsg_flags & MPTCP_CMSG_INQ) {
2224 unsigned int inq = mptcp_inq_hint(sk);
2225
2226 put_cmsg(msg, SOL_TCP, TCP_CM_INQ, sizeof(inq), &inq);
2227 }
b7f653b2
FW		 2228 }
2229
612f71d7
PA		 2230 pr_debug("msk=%p rx queue empty=%d:%d copied=%d",
2231 msk, skb_queue_empty_lockless(&sk->sk_receive_queue),
2232 skb_queue_empty(&msk->receive_queue), copied);
ca4fb892
YL		 2233 if (!(flags & MSG_PEEK))
2234 mptcp_rcv_space_adjust(msk, copied);
a6b118fe 2235
7a6a6cbc 2236 release_sock(sk);
cec37a6e
PK		 2237 return copied;
2238}
2239
b51f9b80
PA		 2240static void mptcp_retransmit_timer(struct timer_list *t)
2241{
2242 struct inet_connection_sock *icsk = from_timer(icsk, t,
2243 icsk_retransmit_timer);
2244 struct sock *sk = &icsk->icsk_inet.sk;
2d6f5a2b 2245 struct mptcp_sock *msk = mptcp_sk(sk);
b51f9b80
PA		 2246
2247 bh_lock_sock(sk);
2248 if (!sock_owned_by_user(sk)) {
2d6f5a2b
PA		 2249 /* we need a process context to retransmit */
2250 if (!test_and_set_bit(MPTCP_WORK_RTX, &msk->flags))
2251 mptcp_schedule_work(sk);
b51f9b80
PA		 2252 } else {
2253 /* delegate our work to tcp_release_cb() */
e9d09bac 2254 __set_bit(MPTCP_RETRANSMIT, &msk->cb_flags);
b51f9b80
PA		 2255 }
2256 bh_unlock_sock(sk);
2257 sock_put(sk);
2258}
2259
f6909dc1 2260static void mptcp_tout_timer(struct timer_list *t)
e16163b6
PA		 2261{
2262 struct sock *sk = from_timer(sk, t, sk_timer);
2263
2264 mptcp_schedule_work(sk);
b6d69fc8 2265 sock_put(sk);
e16163b6
PA		 2266}
2267
3b1d6210
PA		 2268/* Find an idle subflow. Return NULL if there is unacked data at tcp
2269 * level.
2270 *
2271 * A backup subflow is returned only if that is the only kind available.
2272 */
07336a87 2273struct sock *mptcp_subflow_get_retrans(struct mptcp_sock *msk)
3b1d6210 2274{
71b7dec2 2275 struct sock *backup = NULL, *pick = NULL;
3b1d6210 2276 struct mptcp_subflow_context *subflow;
71b7dec2 2277 int min_stale_count = INT_MAX;
3b1d6210 2278
3b1d6210
PA		 2279 mptcp_for_each_subflow(msk, subflow) {
2280 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2281
ff5a0b42 2282 if (!__mptcp_subflow_active(subflow))
d5f49190
PA		 2283 continue;
2284
71b7dec2
PA		 2285 /* still data outstanding at TCP level? skip this */
2286 if (!tcp_rtx_and_write_queues_empty(ssk)) {
2287 mptcp_pm_subflow_chk_stale(msk, ssk);
2288 min_stale_count = min_t(int, min_stale_count, subflow->stale_count);
2289 continue;
860975c6 2290 }
3b1d6210
PA		 2291
2292 if (subflow->backup) {
2293 if (!backup)
2294 backup = ssk;
2295 continue;
2296 }
2297
71b7dec2
PA		 2298 if (!pick)
2299 pick = ssk;
3b1d6210
PA		 2300 }
2301
71b7dec2
PA		 2302 if (pick)
2303 return pick;
2304
2305 /* use backup only if there are no progresses anywhere */
2306 return min_stale_count > 1 ? backup : NULL;
3b1d6210
PA		 2307}
2308
1e1d9d6f
PA		 2309bool __mptcp_retransmit_pending_data(struct sock *sk)
2310{
2311 struct mptcp_data_frag *cur, *rtx_head;
2312 struct mptcp_sock *msk = mptcp_sk(sk);
2313
00df24f1 2314 if (__mptcp_check_fallback(msk))
1e1d9d6f
PA		 2315 return false;
2316
2317 if (tcp_rtx_and_write_queues_empty(sk))
2318 return false;
2319
2320 /* the closing socket has some data untransmitted and/or unacked:
2321 * some data in the mptcp rtx queue has not really xmitted yet.
2322 * keep it simple and re-inject the whole mptcp level rtx queue
2323 */
2324 mptcp_data_lock(sk);
2325 __mptcp_clean_una_wakeup(sk);
2326 rtx_head = mptcp_rtx_head(sk);
2327 if (!rtx_head) {
2328 mptcp_data_unlock(sk);
2329 return false;
2330 }
2331
0d199e43 2332 msk->recovery_snd_nxt = msk->snd_nxt;
1e1d9d6f
PA		 2333 msk->recovery = true;
2334 mptcp_data_unlock(sk);
2335
2336 msk->first_pending = rtx_head;
1e1d9d6f
PA		 2337 msk->snd_burst = 0;
2338
2339 /* be sure to clear the "sent status" on all re-injected fragments */
2340 list_for_each_entry(cur, &msk->rtx_queue, list) {
2341 if (!cur->already_sent)
2342 break;
2343 cur->already_sent = 0;
2344 }
2345
2346 return true;
2347}
2348
b29fcfb5
PA		 2349/* flags for __mptcp_close_ssk() */
2350#define MPTCP_CF_PUSH BIT(1)
2351#define MPTCP_CF_FASTCLOSE BIT(2)
2352
14c56686
GT		 2353/* be sure to send a reset only if the caller asked for it, also
2354 * clean completely the subflow status when the subflow reaches
2355 * TCP_CLOSE state
2356 */
2357static void __mptcp_subflow_disconnect(struct sock *ssk,
2358 struct mptcp_subflow_context *subflow,
2359 unsigned int flags)
2360{
2361 if (((1 << ssk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)) ||
2362 (flags & MPTCP_CF_FASTCLOSE)) {
2363 /* The MPTCP code never wait on the subflow sockets, TCP-level
2364 * disconnect should never fail
2365 */
2366 WARN_ON_ONCE(tcp_disconnect(ssk, 0));
2367 mptcp_subflow_ctx_reset(subflow);
2368 } else {
2369 tcp_shutdown(ssk, SEND_SHUTDOWN);
2370 }
2371}
2372
cec37a6e
PK		 2373/* subflow sockets can be either outgoing (connect) or incoming
2374 * (accept).
2375 *
2376 * Outgoing subflows use in-kernel sockets.
2377 * Incoming subflows do not have their own 'struct socket' allocated,
2378 * so we need to use tcp_close() after detaching them from the mptcp
2379 * parent socket.
2380 */
a141e02e 2381static void __mptcp_close_ssk(struct sock *sk, struct sock *ssk,
b29fcfb5
PA		 2382 struct mptcp_subflow_context *subflow,
2383 unsigned int flags)
cec37a6e 2384{
e0be4931 2385 struct mptcp_sock *msk = mptcp_sk(sk);
63740448
PA		 2386 bool dispose_it, need_push = false;
2387
2388 /* If the first subflow moved to a close state before accept, e.g. due
27e5ccc2
PA		 2389 * to an incoming reset or listener shutdown, the subflow socket is
2390 * already deleted by inet_child_forget() and the mptcp socket can't
2391 * survive too.
63740448 2392 */
27e5ccc2
PA		 2393 if (msk->in_accept_queue && msk->first == ssk &&
2394 (sock_flag(sk, SOCK_DEAD) || sock_flag(ssk, SOCK_DEAD))) {
63740448
PA		 2395 /* ensure later check in mptcp_worker() will dispose the msk */
2396 sock_set_flag(sk, SOCK_DEAD);
d866ae9a 2397 mptcp_set_close_tout(sk, tcp_jiffies32 - (mptcp_close_timeout(sk) + 1));
63740448
PA		 2398 lock_sock_nested(ssk, SINGLE_DEPTH_NESTING);
2399 mptcp_subflow_drop_ctx(ssk);
2400 goto out_release;
2401 }
e0be4931 2402
39880bd8 2403 dispose_it = msk->free_first || ssk != msk->first;
b29fcfb5
PA		 2404 if (dispose_it)
2405 list_del(&subflow->node);
cec37a6e 2406
3f8b2667 2407 lock_sock_nested(ssk, SINGLE_DEPTH_NESTING);
e16163b6 2408
ff18f9ef 2409 if ((flags & MPTCP_CF_FASTCLOSE) && !__mptcp_check_fallback(msk)) {
14c56686 2410 /* be sure to force the tcp_close path
d21f8348
PA		 2411 * to generate the egress reset
2412 */
2413 ssk->sk_lingertime = 0;
2414 sock_set_flag(ssk, SOCK_LINGER);
b29fcfb5 2415 subflow->send_fastclose = 1;
d21f8348 2416 }
b29fcfb5
PA		 2417
2418 need_push = (flags & MPTCP_CF_PUSH) && __mptcp_retransmit_pending_data(sk);
2419 if (!dispose_it) {
14c56686 2420 __mptcp_subflow_disconnect(ssk, subflow, flags);
b29fcfb5
PA		 2421 release_sock(ssk);
2422
2423 goto out;
2424 }
2425
d7b1bfd0
PA		 2426 subflow->disposable = 1;
2427
e16163b6
PA		 2428 /* if ssk hit tcp_done(), tcp_cleanup_ulp() cleared the related ops
2429 * the ssk has been already destroyed, we just need to release the
2430 * reference owned by msk;
2431 */
2432 if (!inet_csk(ssk)->icsk_ulp_ops) {
b6985b9b 2433 WARN_ON_ONCE(!sock_flag(ssk, SOCK_DEAD));
e16163b6 2434 kfree_rcu(subflow, rcu);
cec37a6e 2435 } else {
d7b1bfd0 2436 /* otherwise tcp will dispose of the ssk and subflow ctx */
e16163b6
PA		 2437 __tcp_close(ssk, 0);
2438
2439 /* close acquired an extra ref */
2440 __sock_put(ssk);
cec37a6e 2441 }
63740448
PA		 2442
2443out_release:
9f1a9881 2444 __mptcp_subflow_error_report(sk, ssk);
e16163b6 2445 release_sock(ssk);
e16163b6
PA		 2446
2447 sock_put(ssk);
e0be4931 2448
c8fe62f0 2449 if (ssk == msk->first)
1b1b43ee 2450 WRITE_ONCE(msk->first, NULL);
c8fe62f0 2451
b29fcfb5 2452out:
8005184f 2453 __mptcp_sync_sndbuf(sk);
1e1d9d6f
PA		 2454 if (need_push)
2455 __mptcp_push_pending(sk, 0);
27e5ccc2
PA		 2456
2457 /* Catch every 'all subflows closed' scenario, including peers silently
2458 * closing them, e.g. due to timeout.
2459 * For established sockets, allow an additional timeout before closing,
2460 * as the protocol can still create more subflows.
2461 */
2462 if (list_is_singular(&msk->conn_list) && msk->first &&
2463 inet_sk_state_load(msk->first) == TCP_CLOSE) {
2464 if (sk->sk_state != TCP_ESTABLISHED ||
2465 msk->in_accept_queue || sock_flag(sk, SOCK_DEAD)) {
2466 inet_sk_state_store(sk, TCP_CLOSE);
2467 mptcp_close_wake_up(sk);
2468 } else {
2469 mptcp_start_tout_timer(sk);
2470 }
2471 }
f870fa0b
MM		 2472}
2473
a141e02e
FW		 2474void mptcp_close_ssk(struct sock *sk, struct sock *ssk,
2475 struct mptcp_subflow_context *subflow)
2476{
b911c97c
FW		 2477 if (sk->sk_state == TCP_ESTABLISHED)
2478 mptcp_event(MPTCP_EVENT_SUB_CLOSED, mptcp_sk(sk), ssk, GFP_KERNEL);
a88c9e49
PA		 2479
2480 /* subflow aborted before reaching the fully_established status
2481 * attempt the creation of the next subflow
2482 */
74cbb0c6 2483 mptcp_pm_subflow_check_next(mptcp_sk(sk), subflow);
a88c9e49 2484
b29fcfb5 2485 __mptcp_close_ssk(sk, ssk, subflow, MPTCP_CF_PUSH);
a141e02e
FW		 2486}
2487
dc24f8b4
PA		 2488static unsigned int mptcp_sync_mss(struct sock *sk, u32 pmtu)
2489{
2490 return 0;
2491}
2492
b6985b9b 2493static void __mptcp_close_subflow(struct sock *sk)
0e4f35d7
PA		 2494{
2495 struct mptcp_subflow_context *subflow, *tmp;
b6985b9b 2496 struct mptcp_sock *msk = mptcp_sk(sk);
0e4f35d7 2497
3abc05d9
FW		 2498 might_sleep();
2499
5efbf6f7 2500 mptcp_for_each_subflow_safe(msk, subflow, tmp) {
0e4f35d7
PA		 2501 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2502
2503 if (inet_sk_state_load(ssk) != TCP_CLOSE)
2504 continue;
2505
40947e13
FW		 2506 /* 'subflow_data_ready' will re-sched once rx queue is empty */
2507 if (!skb_queue_empty_lockless(&ssk->sk_receive_queue))
2508 continue;
2509
b6985b9b
PA		 2510 mptcp_close_ssk(sk, ssk, subflow);
2511 }
2512
0e4f35d7
PA		 2513}
2514
27e5ccc2 2515static bool mptcp_close_tout_expired(const struct sock *sk)
e16163b6 2516{
27e5ccc2
PA		 2517 if (!inet_csk(sk)->icsk_mtup.probe_timestamp ||
2518 sk->sk_state == TCP_CLOSE)
2519 return false;
e16163b6 2520
27e5ccc2 2521 return time_after32(tcp_jiffies32,
d866ae9a 2522 inet_csk(sk)->icsk_mtup.probe_timestamp + mptcp_close_timeout(sk));
e16163b6
PA		 2523}
2524
50c504a2
FW		 2525static void mptcp_check_fastclose(struct mptcp_sock *msk)
2526{
2527 struct mptcp_subflow_context *subflow, *tmp;
80638684 2528 struct sock *sk = (struct sock *)msk;
50c504a2
FW		 2529
2530 if (likely(!READ_ONCE(msk->rcv_fastclose)))
2531 return;
2532
2533 mptcp_token_destroy(msk);
2534
5efbf6f7 2535 mptcp_for_each_subflow_safe(msk, subflow, tmp) {
50c504a2 2536 struct sock *tcp_sk = mptcp_subflow_tcp_sock(subflow);
75e908c3 2537 bool slow;
50c504a2 2538
75e908c3 2539 slow = lock_sock_fast(tcp_sk);
50c504a2
FW		 2540 if (tcp_sk->sk_state != TCP_CLOSE) {
2541 tcp_send_active_reset(tcp_sk, GFP_ATOMIC);
2542 tcp_set_state(tcp_sk, TCP_CLOSE);
2543 }
75e908c3 2544 unlock_sock_fast(tcp_sk, slow);
50c504a2
FW		 2545 }
2546
69800e51
PA		 2547 /* Mirror the tcp_reset() error propagation */
2548 switch (sk->sk_state) {
2549 case TCP_SYN_SENT:
9ae8e5ad 2550 WRITE_ONCE(sk->sk_err, ECONNREFUSED);
69800e51
PA		 2551 break;
2552 case TCP_CLOSE_WAIT:
9ae8e5ad 2553 WRITE_ONCE(sk->sk_err, EPIPE);
69800e51
PA		 2554 break;
2555 case TCP_CLOSE:
2556 return;
2557 default:
9ae8e5ad 2558 WRITE_ONCE(sk->sk_err, ECONNRESET);
69800e51
PA		 2559 }
2560
50c504a2 2561 inet_sk_state_store(sk, TCP_CLOSE);
6b9831bf 2562 WRITE_ONCE(sk->sk_shutdown, SHUTDOWN_MASK);
50c504a2 2563 smp_mb__before_atomic(); /* SHUTDOWN must be visible first */
50c504a2
FW		 2564 set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags);
2565
69800e51
PA		 2566 /* the calling mptcp_worker will properly destroy the socket */
2567 if (sock_flag(sk, SOCK_DEAD))
2568 return;
2569
2570 sk->sk_state_change(sk);
2571 sk_error_report(sk);
50c504a2
FW		 2572}
2573
2948d0a1 2574static void __mptcp_retrans(struct sock *sk)
80992017 2575{
2948d0a1 2576 struct mptcp_sock *msk = mptcp_sk(sk);
ee2708ae 2577 struct mptcp_subflow_context *subflow;
caf971df 2578 struct mptcp_sendmsg_info info = {};
3b1d6210 2579 struct mptcp_data_frag *dfrag;
2948d0a1 2580 struct sock *ssk;
ee2708ae
GT		 2581 int ret, err;
2582 u16 len = 0;
3b1d6210 2583
b5941f06 2584 mptcp_clean_una_wakeup(sk);
3241a9c0
FW		 2585
2586 /* first check ssk: need to kick "stale" logic */
ee2708ae 2587 err = mptcp_sched_get_retrans(msk);
3b1d6210 2588 dfrag = mptcp_rtx_head(sk);
6477dd39
MM		 2589 if (!dfrag) {
2590 if (mptcp_data_fin_enabled(msk)) {
2591 struct inet_connection_sock *icsk = inet_csk(sk);
2592
2593 icsk->icsk_retransmits++;
2594 mptcp_set_datafin_timeout(sk);
2595 mptcp_send_ack(msk);
2596
2597 goto reset_timer;
2598 }
2599
3241a9c0
FW		 2600 if (!mptcp_send_head(sk))
2601 return;
2602
2603 goto reset_timer;
6477dd39 2604 }
3b1d6210 2605
ee2708ae 2606 if (err)
2948d0a1 2607 goto reset_timer;
3b1d6210 2608
ee2708ae
GT		 2609 mptcp_for_each_subflow(msk, subflow) {
2610 if (READ_ONCE(subflow->scheduled)) {
2611 u16 copied = 0;
3b1d6210 2612
ee2708ae 2613 mptcp_subflow_set_scheduled(subflow, false);
3b1d6210 2614
ee2708ae
GT		 2615 ssk = mptcp_subflow_tcp_sock(subflow);
2616
2617 lock_sock(ssk);
2618
2619 /* limit retransmission to the bytes already sent on some subflows */
2620 info.sent = 0;
2621 info.limit = READ_ONCE(msk->csum_enabled) ? dfrag->data_len :
2622 dfrag->already_sent;
2623 while (info.sent < info.limit) {
2624 ret = mptcp_sendmsg_frag(sk, ssk, dfrag, &info);
2625 if (ret <= 0)
2626 break;
2627
2628 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_RETRANSSEGS);
2629 copied += ret;
2630 info.sent += ret;
2631 }
2632 if (copied) {
2633 len = max(copied, len);
2634 tcp_push(ssk, 0, info.mss_now, tcp_sk(ssk)->nonagle,
2635 info.size_goal);
2636 WRITE_ONCE(msk->allow_infinite_fallback, false);
2637 }
2638
2639 release_sock(ssk);
2640 }
4e14867d 2641 }
3b1d6210 2642
ee2708ae
GT		 2643 msk->bytes_retrans += len;
2644 dfrag->already_sent = max(dfrag->already_sent, len);
3b1d6210 2645
2948d0a1 2646reset_timer:
3241a9c0
FW		 2647 mptcp_check_and_set_pending(sk);
2648
f6909dc1
PA		 2649 if (!mptcp_rtx_timer_pending(sk))
2650 mptcp_reset_rtx_timer(sk);
2948d0a1
PA		 2651}
2652
76a13b31
GT		 2653/* schedule the timeout timer for the relevant event: either close timeout
2654 * or mp_fail timeout. The close timeout takes precedence on the mp_fail one
2655 */
f6909dc1 2656void mptcp_reset_tout_timer(struct mptcp_sock *msk, unsigned long fail_tout)
76a13b31
GT		 2657{
2658 struct sock *sk = (struct sock *)msk;
2659 unsigned long timeout, close_timeout;
2660
27e5ccc2 2661 if (!fail_tout && !inet_csk(sk)->icsk_mtup.probe_timestamp)
76a13b31
GT		 2662 return;
2663
27e5ccc2 2664 close_timeout = inet_csk(sk)->icsk_mtup.probe_timestamp - tcp_jiffies32 + jiffies +
d866ae9a 2665 mptcp_close_timeout(sk);
76a13b31
GT		 2666
2667 /* the close timeout takes precedence on the fail one, and here at least one of
2668 * them is active
2669 */
27e5ccc2 2670 timeout = inet_csk(sk)->icsk_mtup.probe_timestamp ? close_timeout : fail_tout;
76a13b31
GT		 2671
2672 sk_reset_timer(sk, &sk->sk_timer, timeout);
2673}
2674
49fa1919
GT		 2675static void mptcp_mp_fail_no_response(struct mptcp_sock *msk)
2676{
76a13b31 2677 struct sock *ssk = msk->first;
49fa1919
GT		 2678 bool slow;
2679
76a13b31
GT		 2680 if (!ssk)
2681 return;
49fa1919 2682
76a13b31
GT		 2683 pr_debug("MP_FAIL doesn't respond, reset the subflow");
2684
2685 slow = lock_sock_fast(ssk);
2686 mptcp_subflow_reset(ssk);
2687 WRITE_ONCE(mptcp_subflow_ctx(ssk)->fail_tout, 0);
2688 unlock_sock_fast(ssk, slow);
49fa1919
GT		 2689}
2690
d21f8348
PA		 2691static void mptcp_do_fastclose(struct sock *sk)
2692{
2693 struct mptcp_subflow_context *subflow, *tmp;
2694 struct mptcp_sock *msk = mptcp_sk(sk);
2695
bbd49d11 2696 inet_sk_state_store(sk, TCP_CLOSE);
d21f8348
PA		 2697 mptcp_for_each_subflow_safe(msk, subflow, tmp)
2698 __mptcp_close_ssk(sk, mptcp_subflow_tcp_sock(subflow),
2699 subflow, MPTCP_CF_FASTCLOSE);
2700}
2701
2948d0a1
PA		 2702static void mptcp_worker(struct work_struct *work)
2703{
2704 struct mptcp_sock *msk = container_of(work, struct mptcp_sock, work);
80638684 2705 struct sock *sk = (struct sock *)msk;
76a13b31 2706 unsigned long fail_tout;
2948d0a1
PA		 2707 int state;
2708
2709 lock_sock(sk);
2710 state = sk->sk_state;
d6a04437 2711 if (unlikely((1 << state) & (TCPF_CLOSE | TCPF_LISTEN)))
2948d0a1
PA		 2712 goto unlock;
2713
2948d0a1
PA		 2714 mptcp_check_fastclose(msk);
2715
86e39e04 2716 mptcp_pm_nl_work(msk);
2948d0a1 2717
81c1d029
PA		 2718 mptcp_check_send_data_fin(sk);
2719 mptcp_check_data_fin_ack(sk);
2948d0a1
PA		 2720 mptcp_check_data_fin(sk);
2721
b6985b9b
PA		 2722 if (test_and_clear_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags))
2723 __mptcp_close_subflow(sk);
2724
27e5ccc2
PA		 2725 if (mptcp_close_tout_expired(sk)) {
2726 mptcp_do_fastclose(sk);
2727 mptcp_close_wake_up(sk);
2728 }
bbd49d11 2729
27e5ccc2
PA		 2730 if (sock_flag(sk, SOCK_DEAD) && sk->sk_state == TCP_CLOSE) {
2731 __mptcp_destroy_sock(sk);
2732 goto unlock;
2948d0a1
PA		 2733 }
2734
2948d0a1
PA		 2735 if (test_and_clear_bit(MPTCP_WORK_RTX, &msk->flags))
2736 __mptcp_retrans(sk);
3b1d6210 2737
76a13b31
GT		 2738 fail_tout = msk->first ? READ_ONCE(mptcp_subflow_ctx(msk->first)->fail_tout) : 0;
2739 if (fail_tout && time_after(jiffies, fail_tout))
2740 mptcp_mp_fail_no_response(msk);
49fa1919 2741
3b1d6210 2742unlock:
80992017
PA		 2743 release_sock(sk);
2744 sock_put(sk);
2745}
2746
e2636917 2747static void __mptcp_init_sock(struct sock *sk)
f870fa0b 2748{
cec37a6e
PK		 2749 struct mptcp_sock *msk = mptcp_sk(sk);
2750
2751 INIT_LIST_HEAD(&msk->conn_list);
ec3edaa7 2752 INIT_LIST_HEAD(&msk->join_list);
18b683bf 2753 INIT_LIST_HEAD(&msk->rtx_queue);
80992017 2754 INIT_WORK(&msk->work, mptcp_worker);
87952603 2755 __skb_queue_head_init(&msk->receive_queue);
ab174ad8 2756 msk->out_of_order_queue = RB_ROOT;
f0e6a4cf 2757 msk->first_pending = NULL;
6511882c 2758 msk->rmem_fwd_alloc = 0;
ce599c51 2759 WRITE_ONCE(msk->rmem_released, 0);
33d41c9c 2760 msk->timer_ival = TCP_RTO_MIN;
5684ab1a 2761 msk->scaling_ratio = TCP_DEFAULT_SCALING_RATIO;
cec37a6e 2762
1b1b43ee 2763 WRITE_ONCE(msk->first, NULL);
dc24f8b4 2764 inet_csk(sk)->icsk_sync_mss = mptcp_sync_mss;
752e9067 2765 WRITE_ONCE(msk->csum_enabled, mptcp_is_checksum_enabled(sock_net(sk)));
0530020a 2766 WRITE_ONCE(msk->allow_infinite_fallback, true);
1e1d9d6f 2767 msk->recovery = false;
6f06b4d4 2768 msk->subflow_id = 1;
8ab183de 2769
1b1c7a0e
PK		 2770 mptcp_pm_data_init(msk);
2771
b51f9b80
PA		 2772 /* re-use the csk retrans timer for MPTCP-level retrans */
2773 timer_setup(&msk->sk.icsk_retransmit_timer, mptcp_retransmit_timer, 0);
f6909dc1 2774 timer_setup(&sk->sk_timer, mptcp_tout_timer, 0);
f870fa0b
MM		 2775}
2776
b29fcfb5 2777static void mptcp_ca_reset(struct sock *sk)
784325e9 2778{
20b5759f 2779 struct inet_connection_sock *icsk = inet_csk(sk);
b29fcfb5
PA		 2780
2781 tcp_assign_congestion_control(sk);
2782 strcpy(mptcp_sk(sk)->ca_name, icsk->icsk_ca_ops->name);
2783
2784 /* no need to keep a reference to the ops, the name will suffice */
2785 tcp_cleanup_congestion_control(sk);
2786 icsk->icsk_ca_ops = NULL;
2787}
2788
2789static int mptcp_init_sock(struct sock *sk)
2790{
fc518953 2791 struct net *net = sock_net(sk);
1730b2b2 2792 int ret;
18b683bf 2793
e2636917 2794 __mptcp_init_sock(sk);
b6c08380 2795
fc518953
FW		 2796 if (!mptcp_is_enabled(net))
2797 return -ENOPROTOOPT;
2798
2799 if (unlikely(!net->mib.mptcp_statistics) && !mptcp_mib_alloc(net))
2800 return -ENOMEM;
2801
1730b2b2
GT		 2802 ret = mptcp_init_sched(mptcp_sk(sk),
2803 mptcp_sched_find(mptcp_get_scheduler(net)));
2804 if (ret)
2805 return ret;
2806
a5ef058d
PA		 2807 set_bit(SOCK_CUSTOM_SOCKOPT, &sk->sk_socket->flags);
2808
20b5759f
PA		 2809 /* fetch the ca name; do it outside __mptcp_init_sock(), so that clone will
2810 * propagate the correct value
2811 */
b29fcfb5 2812 mptcp_ca_reset(sk);
20b5759f 2813
d027236c 2814 sk_sockets_allocated_inc(sk);
a963853f
GT		 2815 sk->sk_rcvbuf = READ_ONCE(net->ipv4.sysctl_tcp_rmem[1]);
2816 sk->sk_sndbuf = READ_ONCE(net->ipv4.sysctl_tcp_wmem[1]);
d027236c 2817
18b683bf
PA		 2818 return 0;
2819}
2820
2821static void __mptcp_clear_xmit(struct sock *sk)
2822{
2823 struct mptcp_sock *msk = mptcp_sk(sk);
2824 struct mptcp_data_frag *dtmp, *dfrag;
2825
d9ca1de8 2826 WRITE_ONCE(msk->first_pending, NULL);
18b683bf 2827 list_for_each_entry_safe(dfrag, dtmp, &msk->rtx_queue, list)
d027236c 2828 dfrag_clear(sk, dfrag);
784325e9
MB		 2829}
2830
30e51b92 2831void mptcp_cancel_work(struct sock *sk)
80992017
PA		 2832{
2833 struct mptcp_sock *msk = mptcp_sk(sk);
2834
b2771d24 2835 if (cancel_work_sync(&msk->work))
e16163b6 2836 __sock_put(sk);
80992017
PA		 2837}
2838
d0876b22 2839void mptcp_subflow_shutdown(struct sock *sk, struct sock *ssk, int how)
21498490
PK		 2840{
2841 lock_sock(ssk);
2842
2843 switch (ssk->sk_state) {
2844 case TCP_LISTEN:
2845 if (!(how & RCV_SHUTDOWN))
2846 break;
df561f66 2847 fallthrough;
21498490 2848 case TCP_SYN_SENT:
c2b2ae39 2849 WARN_ON_ONCE(tcp_disconnect(ssk, O_NONBLOCK));
21498490
PK		 2850 break;
2851 default:
43b54c6e
MM		 2852 if (__mptcp_check_fallback(mptcp_sk(sk))) {
2853 pr_debug("Fallback");
2854 ssk->sk_shutdown |= how;
2855 tcp_shutdown(ssk, how);
81c1d029
PA		 2856
2857 /* simulate the data_fin ack reception to let the state
2858 * machine move forward
2859 */
2860 WRITE_ONCE(mptcp_sk(sk)->snd_una, mptcp_sk(sk)->snd_nxt);
2861 mptcp_schedule_work(sk);
43b54c6e
MM		 2862 } else {
2863 pr_debug("Sending DATA_FIN on subflow %p", ssk);
43b54c6e 2864 tcp_send_ack(ssk);
f6909dc1
PA		 2865 if (!mptcp_rtx_timer_pending(sk))
2866 mptcp_reset_rtx_timer(sk);
43b54c6e 2867 }
21498490
PK		 2868 break;
2869 }
2870
21498490
PK		 2871 release_sock(ssk);
2872}
2873
d9cd27b8
GT		 2874void mptcp_set_state(struct sock *sk, int state)
2875{
2876 int oldstate = sk->sk_state;
2877
2878 switch (state) {
2879 case TCP_ESTABLISHED:
2880 if (oldstate != TCP_ESTABLISHED)
2881 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_CURRESTAB);
2882 break;
2883
2884 default:
2885 if (oldstate == TCP_ESTABLISHED)
2886 MPTCP_DEC_STATS(sock_net(sk), MPTCP_MIB_CURRESTAB);
2887 }
2888
2889 inet_sk_state_store(sk, state);
2890}
2891
6920b851
MM		 2892static const unsigned char new_state[16] = {
2893 /* current state: new state: action: */
2894 [0 /* (Invalid) */] = TCP_CLOSE,
2895 [TCP_ESTABLISHED] = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2896 [TCP_SYN_SENT] = TCP_CLOSE,
2897 [TCP_SYN_RECV] = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2898 [TCP_FIN_WAIT1] = TCP_FIN_WAIT1,
2899 [TCP_FIN_WAIT2] = TCP_FIN_WAIT2,
2900 [TCP_TIME_WAIT] = TCP_CLOSE, /* should not happen ! */
2901 [TCP_CLOSE] = TCP_CLOSE,
2902 [TCP_CLOSE_WAIT] = TCP_LAST_ACK | TCP_ACTION_FIN,
2903 [TCP_LAST_ACK] = TCP_LAST_ACK,
2904 [TCP_LISTEN] = TCP_CLOSE,
2905 [TCP_CLOSING] = TCP_CLOSING,
2906 [TCP_NEW_SYN_RECV] = TCP_CLOSE, /* should not happen ! */
2907};
2908
2909static int mptcp_close_state(struct sock *sk)
2910{
2911 int next = (int)new_state[sk->sk_state];
2912 int ns = next & TCP_STATE_MASK;
2913
2914 inet_sk_state_store(sk, ns);
2915
2916 return next & TCP_ACTION_FIN;
2917}
2918
81c1d029 2919static void mptcp_check_send_data_fin(struct sock *sk)
f870fa0b 2920{
e16163b6 2921 struct mptcp_subflow_context *subflow;
f870fa0b
MM		 2922 struct mptcp_sock *msk = mptcp_sk(sk);
2923
e16163b6
PA		 2924 pr_debug("msk=%p snd_data_fin_enable=%d pending=%d snd_nxt=%llu write_seq=%llu",
2925 msk, msk->snd_data_fin_enable, !!mptcp_send_head(sk),
2926 msk->snd_nxt, msk->write_seq);
43b54c6e 2927
e16163b6
PA		 2928 /* we still need to enqueue subflows or not really shutting down,
2929 * skip this
2930 */
2931 if (!msk->snd_data_fin_enable || msk->snd_nxt + 1 != msk->write_seq ||
2932 mptcp_send_head(sk))
2933 return;
2934
2935 WRITE_ONCE(msk->snd_nxt, msk->write_seq);
2936
e16163b6
PA		 2937 mptcp_for_each_subflow(msk, subflow) {
2938 struct sock *tcp_sk = mptcp_subflow_tcp_sock(subflow);
43b54c6e 2939
e16163b6 2940 mptcp_subflow_shutdown(sk, tcp_sk, SEND_SHUTDOWN);
43b54c6e 2941 }
e16163b6 2942}
2c22c06c 2943
e16163b6
PA		 2944static void __mptcp_wr_shutdown(struct sock *sk)
2945{
2946 struct mptcp_sock *msk = mptcp_sk(sk);
43b54c6e 2947
e16163b6
PA		 2948 pr_debug("msk=%p snd_data_fin_enable=%d shutdown=%x state=%d pending=%d",
2949 msk, msk->snd_data_fin_enable, sk->sk_shutdown, sk->sk_state,
2950 !!mptcp_send_head(sk));
2951
2952 /* will be ignored by fallback sockets */
2953 WRITE_ONCE(msk->write_seq, msk->write_seq + 1);
2954 WRITE_ONCE(msk->snd_data_fin_enable, 1);
2955
81c1d029 2956 mptcp_check_send_data_fin(sk);
e16163b6
PA		 2957}
2958
2959static void __mptcp_destroy_sock(struct sock *sk)
2960{
e16163b6 2961 struct mptcp_sock *msk = mptcp_sk(sk);
e16163b6
PA		 2962
2963 pr_debug("msk=%p", msk);
f870fa0b 2964
3abc05d9
FW		 2965 might_sleep();
2966
f6909dc1 2967 mptcp_stop_rtx_timer(sk);
e16163b6
PA		 2968 sk_stop_timer(sk, &sk->sk_timer);
2969 msk->pm.status = 0;
1730b2b2 2970 mptcp_release_sched(msk);
b2c5b614 2971
e16163b6 2972 sk->sk_prot->destroy(sk);
80992017 2973
6511882c 2974 WARN_ON_ONCE(msk->rmem_fwd_alloc);
87952603 2975 WARN_ON_ONCE(msk->rmem_released);
e16163b6
PA		 2976 sk_stream_kill_queues(sk);
2977 xfrm_sk_free_policy(sk);
aa1fbd94 2978
e16163b6
PA		 2979 sock_put(sk);
2980}
2981
63740448
PA		 2982void __mptcp_unaccepted_force_close(struct sock *sk)
2983{
2984 sock_set_flag(sk, SOCK_DEAD);
63740448
PA		 2985 mptcp_do_fastclose(sk);
2986 __mptcp_destroy_sock(sk);
2987}
2988
5684ab1a 2989static __poll_t mptcp_check_readable(struct sock *sk)
d21f8348 2990{
5684ab1a 2991 return mptcp_epollin_ready(sk) ? EPOLLIN | EPOLLRDNORM : 0;
d21f8348
PA		 2992}
2993
57fc0f1c 2994static void mptcp_check_listen_stop(struct sock *sk)
c558246e 2995{
57fc0f1c
PA		 2996 struct sock *ssk;
2997
2998 if (inet_sk_state_load(sk) != TCP_LISTEN)
2999 return;
3000
3001 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
3002 ssk = mptcp_sk(sk)->first;
3003 if (WARN_ON_ONCE(!ssk || inet_sk_state_load(ssk) != TCP_LISTEN))
3004 return;
3005
3006 lock_sock_nested(ssk, SINGLE_DEPTH_NESTING);
3fffa15b 3007 tcp_set_state(ssk, TCP_CLOSE);
57fc0f1c
PA		 3008 mptcp_subflow_queue_clean(sk, ssk);
3009 inet_csk_listen_stop(ssk);
3010 mptcp_event_pm_listener(ssk, MPTCP_EVENT_LISTENER_CLOSED);
57fc0f1c 3011 release_sock(ssk);
c558246e
MD		 3012}
3013
26d3e21c 3014bool __mptcp_close(struct sock *sk, long timeout)
e16163b6
PA		 3015{
3016 struct mptcp_subflow_context *subflow;
76a13b31 3017 struct mptcp_sock *msk = mptcp_sk(sk);
e16163b6 3018 bool do_cancel_work = false;
d4e85922 3019 int subflows_alive = 0;
e16163b6 3020
6b9831bf 3021 WRITE_ONCE(sk->sk_shutdown, SHUTDOWN_MASK);
e16163b6
PA		 3022
3023 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE)) {
57fc0f1c 3024 mptcp_check_listen_stop(sk);
e16163b6
PA		 3025 inet_sk_state_store(sk, TCP_CLOSE);
3026 goto cleanup;
3027 }
6771bfd9 3028
5684ab1a 3029 if (mptcp_data_avail(msk) || timeout < 0) {
8d547809
PA		 3030 /* If the msk has read data, or the caller explicitly ask it,
3031 * do the MPTCP equivalent of TCP reset, aka MPTCP fastclose
3032 */
d21f8348 3033 mptcp_do_fastclose(sk);
8d547809 3034 timeout = 0;
d21f8348 3035 } else if (mptcp_close_state(sk)) {
e16163b6 3036 __mptcp_wr_shutdown(sk);
d21f8348 3037 }
e16163b6
PA		 3038
3039 sk_stream_wait_close(sk, timeout);
3040
3041cleanup:
3042 /* orphan all the subflows */
76a13b31 3043 mptcp_for_each_subflow(msk, subflow) {
e16163b6 3044 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
49054556 3045 bool slow = lock_sock_fast_nested(ssk);
e16163b6 3046
d4e85922
PA		 3047 subflows_alive += ssk->sk_state != TCP_CLOSE;
3048
76a13b31
GT		 3049 /* since the close timeout takes precedence on the fail one,
3050 * cancel the latter
3051 */
3052 if (ssk == msk->first)
3053 subflow->fail_tout = 0;
3054
fe948001
MD		 3055 /* detach from the parent socket, but allow data_ready to
3056 * push incoming data into the mptcp stack, to properly ack it
3057 */
3058 ssk->sk_socket = NULL;
3059 ssk->sk_wq = NULL;
e16163b6 3060 unlock_sock_fast(ssk, slow);
e16163b6
PA		 3061 }
3062 sock_orphan(sk);
3063
d4e85922
PA		 3064 /* all the subflows are closed, only timeout can change the msk
3065 * state, let's not keep resources busy for no reasons
3066 */
3067 if (subflows_alive == 0)
3068 inet_sk_state_store(sk, TCP_CLOSE);
3069
e16163b6
PA		 3070 sock_hold(sk);
3071 pr_debug("msk=%p state=%d", sk, sk->sk_state);
00df24f1 3072 if (msk->token)
76a13b31 3073 mptcp_event(MPTCP_EVENT_CLOSED, msk, NULL, GFP_KERNEL);
b29fcfb5 3074
e16163b6
PA		 3075 if (sk->sk_state == TCP_CLOSE) {
3076 __mptcp_destroy_sock(sk);
3077 do_cancel_work = true;
3078 } else {
27e5ccc2 3079 mptcp_start_tout_timer(sk);
e16163b6 3080 }
26d3e21c
MD		 3081
3082 return do_cancel_work;
3083}
3084
3085static void mptcp_close(struct sock *sk, long timeout)
3086{
3087 bool do_cancel_work;
3088
3089 lock_sock(sk);
3090
3091 do_cancel_work = __mptcp_close(sk, timeout);
e16163b6
PA		 3092 release_sock(sk);
3093 if (do_cancel_work)
3094 mptcp_cancel_work(sk);
b911c97c 3095
e16163b6 3096 sock_put(sk);
f870fa0b
MM		 3097}
3098
7e8b88ec 3099static void mptcp_copy_inaddrs(struct sock *msk, const struct sock *ssk)
cf7da0d6
PK		 3100{
3101#if IS_ENABLED(CONFIG_MPTCP_IPV6)
3102 const struct ipv6_pinfo *ssk6 = inet6_sk(ssk);
3103 struct ipv6_pinfo *msk6 = inet6_sk(msk);
3104
3105 msk->sk_v6_daddr = ssk->sk_v6_daddr;
3106 msk->sk_v6_rcv_saddr = ssk->sk_v6_rcv_saddr;
3107
3108 if (msk6 && ssk6) {
3109 msk6->saddr = ssk6->saddr;
3110 msk6->flow_label = ssk6->flow_label;
3111 }
3112#endif
3113
3114 inet_sk(msk)->inet_num = inet_sk(ssk)->inet_num;
3115 inet_sk(msk)->inet_dport = inet_sk(ssk)->inet_dport;
3116 inet_sk(msk)->inet_sport = inet_sk(ssk)->inet_sport;
3117 inet_sk(msk)->inet_daddr = inet_sk(ssk)->inet_daddr;
3118 inet_sk(msk)->inet_saddr = inet_sk(ssk)->inet_saddr;
3119 inet_sk(msk)->inet_rcv_saddr = inet_sk(ssk)->inet_rcv_saddr;
3120}
3121
18b683bf
PA		 3122static int mptcp_disconnect(struct sock *sk, int flags)
3123{
76e2a55d
PA		 3124 struct mptcp_sock *msk = mptcp_sk(sk);
3125
7d803344
PA		 3126 /* We are on the fastopen error path. We can't call straight into the
3127 * subflows cleanup code due to lock nesting (we are already under
c2b2ae39 3128 * msk->firstsocket lock).
7d803344
PA		 3129 */
3130 if (msk->fastopening)
c2b2ae39 3131 return -EBUSY;
7d803344 3132
57fc0f1c 3133 mptcp_check_listen_stop(sk);
b29fcfb5
PA		 3134 inet_sk_state_store(sk, TCP_CLOSE);
3135
f6909dc1 3136 mptcp_stop_rtx_timer(sk);
27e5ccc2 3137 mptcp_stop_tout_timer(sk);
b29fcfb5 3138
00df24f1
GT		 3139 if (msk->token)
3140 mptcp_event(MPTCP_EVENT_CLOSED, msk, NULL, GFP_KERNEL);
b29fcfb5 3141
c0bf3c6a
PA		 3142 /* msk->subflow is still intact, the following will not free the first
3143 * subflow
3144 */
3145 mptcp_destroy_common(msk, MPTCP_CF_FASTCLOSE);
e9d09bac
PA		 3146 WRITE_ONCE(msk->flags, 0);
3147 msk->cb_flags = 0;
3148 msk->push_pending = 0;
b29fcfb5
PA		 3149 msk->recovery = false;
3150 msk->can_ack = false;
3151 msk->fully_established = false;
3152 msk->rcv_data_fin = false;
3153 msk->snd_data_fin_enable = false;
3154 msk->rcv_fastclose = false;
3155 msk->use_64bit_ack = false;
5684ab1a 3156 msk->bytes_consumed = 0;
b29fcfb5
PA		 3157 WRITE_ONCE(msk->csum_enabled, mptcp_is_checksum_enabled(sock_net(sk)));
3158 mptcp_pm_data_reset(msk);
3159 mptcp_ca_reset(sk);
38967f42
PA		 3160 msk->bytes_acked = 0;
3161 msk->bytes_received = 0;
3162 msk->bytes_sent = 0;
3163 msk->bytes_retrans = 0;
b29fcfb5 3164
6b9831bf 3165 WRITE_ONCE(sk->sk_shutdown, 0);
b29fcfb5 3166 sk_error_report(sk);
42c556fe 3167 return 0;
18b683bf
PA		 3168}
3169
b0519de8
FW		 3170#if IS_ENABLED(CONFIG_MPTCP_IPV6)
3171static struct ipv6_pinfo *mptcp_inet6_sk(const struct sock *sk)
3172{
3173 unsigned int offset = sizeof(struct mptcp6_sock) - sizeof(struct ipv6_pinfo);
3174
3175 return (struct ipv6_pinfo *)(((u8 *)sk) + offset);
3176}
3177#endif
3178
7e8b88ec
PA		 3179struct sock *mptcp_sk_clone_init(const struct sock *sk,
3180 const struct mptcp_options_received *mp_opt,
3181 struct sock *ssk,
3182 struct request_sock *req)
b0519de8 3183{
58b09919 3184 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
b0519de8 3185 struct sock *nsk = sk_clone_lock(sk, GFP_ATOMIC);
58b09919 3186 struct mptcp_sock *msk;
b0519de8
FW		 3187
3188 if (!nsk)
3189 return NULL;
3190
3191#if IS_ENABLED(CONFIG_MPTCP_IPV6)
3192 if (nsk->sk_family == AF_INET6)
3193 inet_sk(nsk)->pinet6 = mptcp_inet6_sk(nsk);
3194#endif
3195
58b09919
PA		 3196 __mptcp_init_sock(nsk);
3197
3198 msk = mptcp_sk(nsk);
3199 msk->local_key = subflow_req->local_key;
3200 msk->token = subflow_req->token;
b6985b9b 3201 msk->in_accept_queue = 1;
b93df08c 3202 WRITE_ONCE(msk->fully_established, false);
74c7dfbe 3203 if (mp_opt->suboptions & OPTION_MPTCP_CSUMREQD)
06251181 3204 WRITE_ONCE(msk->csum_enabled, true);
58b09919 3205
58b09919 3206 msk->write_seq = subflow_req->idsn + 1;
eaa2ffab 3207 msk->snd_nxt = msk->write_seq;
7439d687
PA		 3208 msk->snd_una = msk->write_seq;
3209 msk->wnd_end = msk->snd_nxt + req->rsk_rcv_wnd;
df00b087 3210 msk->setsockopt_seq = mptcp_sk(sk)->setsockopt_seq;
1730b2b2 3211 mptcp_init_sched(msk, mptcp_sk(sk)->sched);
6f8a612a 3212
6f06b4d4
PA		 3213 /* passive msk is created after the first/MPC subflow */
3214 msk->subflow_id = 2;
3215
5e20087d 3216 sock_reset_flag(nsk, SOCK_RCU_FREE);
0c148460 3217 security_inet_csk_clone(nsk, req);
7e8b88ec
PA		 3218
3219 /* this can't race with mptcp_close(), as the msk is
3220 * not yet exposted to user-space
3221 */
3222 inet_sk_state_store(nsk, TCP_ESTABLISHED);
3223
3224 /* The msk maintain a ref to each subflow in the connections list */
3225 WRITE_ONCE(msk->first, ssk);
3226 list_add(&mptcp_subflow_ctx(ssk)->node, &msk->conn_list);
3227 sock_hold(ssk);
3228
3229 /* new mpc subflow takes ownership of the newly
3230 * created mptcp socket
3231 */
3232 mptcp_token_accept(subflow_req, msk);
3233
3234 /* set msk addresses early to ensure mptcp_pm_get_local_id()
3235 * uses the correct data
3236 */
3237 mptcp_copy_inaddrs(nsk, ssk);
8005184f 3238 __mptcp_propagate_sndbuf(nsk, ssk);
7e8b88ec
PA		 3239
3240 mptcp_rcv_space_init(msk, ssk);
58b09919
PA		 3241 bh_unlock_sock(nsk);
3242
b6985b9b 3243 /* note: the newly allocated socket refcount is 2 now */
b0519de8
FW		 3244 return nsk;
3245}
3246
a6b118fe
FW		 3247void mptcp_rcv_space_init(struct mptcp_sock *msk, const struct sock *ssk)
3248{
3249 const struct tcp_sock *tp = tcp_sk(ssk);
3250
3251 msk->rcvq_space.copied = 0;
3252 msk->rcvq_space.rtt_us = 0;
3253
3254 msk->rcvq_space.time = tp->tcp_mstamp;
3255
3256 /* initial rcv_space offering made to peer */
3257 msk->rcvq_space.space = min_t(u32, tp->rcv_wnd,
3258 TCP_INIT_CWND * tp->advmss);
3259 if (msk->rcvq_space.space == 0)
3260 msk->rcvq_space.space = TCP_INIT_CWND * TCP_MSS_DEFAULT;
6f8a612a 3261
7439d687 3262 WRITE_ONCE(msk->wnd_end, msk->snd_nxt + tcp_sk(ssk)->snd_wnd);
a6b118fe
FW		 3263}
3264
c0bf3c6a 3265void mptcp_destroy_common(struct mptcp_sock *msk, unsigned int flags)
5c8c1640 3266{
c0bf3c6a 3267 struct mptcp_subflow_context *subflow, *tmp;
87952603
PA		 3268 struct sock *sk = (struct sock *)msk;
3269
6e628cd3
PA		 3270 __mptcp_clear_xmit(sk);
3271
c0bf3c6a 3272 /* join list will be eventually flushed (with rst) at sock lock release time */
5efbf6f7 3273 mptcp_for_each_subflow_safe(msk, subflow, tmp)
c0bf3c6a
PA		 3274 __mptcp_close_ssk(sk, mptcp_subflow_tcp_sock(subflow), subflow, flags);
3275
87952603 3276 /* move to sk_receive_queue, sk_stream_kill_queues will purge it */
b29fcfb5 3277 mptcp_data_lock(sk);
87952603 3278 skb_queue_splice_tail_init(&msk->receive_queue, &sk->sk_receive_queue);
6511882c 3279 __skb_queue_purge(&sk->sk_receive_queue);
5c8c1640 3280 skb_rbtree_purge(&msk->out_of_order_queue);
b29fcfb5 3281 mptcp_data_unlock(sk);
6511882c
PA		 3282
3283 /* move all the rx fwd alloc into the sk_mem_reclaim_final in
3284 * inet_sock_destruct() will dispose it
3285 */
5e6300e7 3286 sk_forward_alloc_add(sk, msk->rmem_fwd_alloc);
9531e4a8 3287 WRITE_ONCE(msk->rmem_fwd_alloc, 0);
5c8c1640
GT		 3288 mptcp_token_destroy(msk);
3289 mptcp_pm_free_anno_list(msk);
4638de5a 3290 mptcp_free_local_addr_list(msk);
5c8c1640
GT		 3291}
3292
79c0949e
PK		 3293static void mptcp_destroy(struct sock *sk)
3294{
c9fd9c5f
FW		 3295 struct mptcp_sock *msk = mptcp_sk(sk);
3296
39880bd8
PA		 3297 /* allow the following to close even the initial subflow */
3298 msk->free_first = 1;
c0bf3c6a 3299 mptcp_destroy_common(msk, 0);
d027236c 3300 sk_sockets_allocated_dec(sk);
79c0949e
PK		 3301}
3302
6e628cd3
PA		 3303void __mptcp_data_acked(struct sock *sk)
3304{
3305 if (!sock_owned_by_user(sk))
3306 __mptcp_clean_una(sk);
3307 else
e9d09bac 3308 __set_bit(MPTCP_CLEAN_UNA, &mptcp_sk(sk)->cb_flags);
6e628cd3
PA		 3309
3310 if (mptcp_pending_data_fin_ack(sk))
3311 mptcp_schedule_work(sk);
3312}
3313
219d0499 3314void __mptcp_check_push(struct sock *sk, struct sock *ssk)
6e628cd3
PA		 3315{
3316 if (!mptcp_send_head(sk))
3317 return;
3318
73a0052a
GT		 3319 if (!sock_owned_by_user(sk))
3320 __mptcp_subflow_push_pending(sk, ssk, false);
3321 else
e9d09bac 3322 __set_bit(MPTCP_PUSH_PENDING, &mptcp_sk(sk)->cb_flags);
6e628cd3
PA		 3323}
3324
e9d09bac
PA		 3325#define MPTCP_FLAGS_PROCESS_CTX_NEED (BIT(MPTCP_PUSH_PENDING) | \
3326 BIT(MPTCP_RETRANSMIT) | \
3327 BIT(MPTCP_FLUSH_JOIN_LIST))
3328
e93da928 3329/* processes deferred events and flush wmem */
14c441b5 3330static void mptcp_release_cb(struct sock *sk)
e9d09bac 3331 __must_hold(&sk->sk_lock.slock)
14c441b5 3332{
e9d09bac
PA		 3333 struct mptcp_sock *msk = mptcp_sk(sk);
3334
c2e6048f 3335 for (;;) {
e9d09bac
PA		 3336 unsigned long flags = (msk->cb_flags & MPTCP_FLAGS_PROCESS_CTX_NEED) |
3337 msk->push_pending;
56a666c4
PA		 3338 struct list_head join_list;
3339
c2e6048f
PA		 3340 if (!flags)
3341 break;
3342
56a666c4
PA		 3343 INIT_LIST_HEAD(&join_list);
3344 list_splice_init(&msk->join_list, &join_list);
3345
c2e6048f 3346 /* the following actions acquire the subflow socket lock
6e628cd3
PA		 3347 *
3348 * 1) can't be invoked in atomic scope
3349 * 2) must avoid ABBA deadlock with msk socket spinlock: the RX
3350 * datapath acquires the msk socket spinlock while helding
3351 * the subflow socket lock
3352 */
e9d09bac
PA		 3353 msk->push_pending = 0;
3354 msk->cb_flags &= ~flags;
6e628cd3 3355 spin_unlock_bh(&sk->sk_lock.slock);
56a666c4 3356
3e501490 3357 if (flags & BIT(MPTCP_FLUSH_JOIN_LIST))
56a666c4 3358 __mptcp_flush_join_list(sk, &join_list);
2e5de7e0 3359 if (flags & BIT(MPTCP_PUSH_PENDING))
c2e6048f 3360 __mptcp_push_pending(sk, 0);
2d6f5a2b
PA		 3361 if (flags & BIT(MPTCP_RETRANSMIT))
3362 __mptcp_retrans(sk);
c2e6048f
PA		 3363
3364 cond_resched();
6e628cd3
PA		 3365 spin_lock_bh(&sk->sk_lock.slock);
3366 }
c2e6048f 3367
e9d09bac 3368 if (__test_and_clear_bit(MPTCP_CLEAN_UNA, &msk->cb_flags))
417789df 3369 __mptcp_clean_una_wakeup(sk);
17ebf8a4 3370 if (unlikely(msk->cb_flags)) {
4fd19a30 3371 /* be sure to sync the msk state before taking actions
d109a776
PA		 3372 * depending on sk_state (MPTCP_ERROR_REPORT)
3373 * On sk release avoid actions depending on the first subflow
65a569b0 3374 */
4fd19a30
PA		 3375 if (__test_and_clear_bit(MPTCP_SYNC_STATE, &msk->cb_flags) && msk->first)
3376 __mptcp_sync_state(sk, msk->pending_state);
65a569b0
PA		 3377 if (__test_and_clear_bit(MPTCP_ERROR_REPORT, &msk->cb_flags))
3378 __mptcp_error_report(sk);
8005184f
PA		 3379 if (__test_and_clear_bit(MPTCP_SYNC_SNDBUF, &msk->cb_flags))
3380 __mptcp_sync_sndbuf(sk);
65a569b0 3381 }
6e628cd3 3382
87952603 3383 __mptcp_update_rmem(sk);
14c441b5
PA		 3384}
3385
bcd97734
PA		 3386/* MP_JOIN client subflow must wait for 4th ack before sending any data:
3387 * TCP can't schedule delack timer before the subflow is fully established.
3388 * MPTCP uses the delack timer to do 3rd ack retransmissions
3389 */
3390static void schedule_3rdack_retransmission(struct sock *ssk)
3391{
3392 struct inet_connection_sock *icsk = inet_csk(ssk);
3393 struct tcp_sock *tp = tcp_sk(ssk);
3394 unsigned long timeout;
3395
3396 if (mptcp_subflow_ctx(ssk)->fully_established)
3397 return;
3398
3399 /* reschedule with a timeout above RTT, as we must look only for drop */
3400 if (tp->srtt_us)
3401 timeout = usecs_to_jiffies(tp->srtt_us >> (3 - 1));
3402 else
3403 timeout = TCP_TIMEOUT_INIT;
3404 timeout += jiffies;
3405
3406 WARN_ON_ONCE(icsk->icsk_ack.pending & ICSK_ACK_TIMER);
3407 icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
3408 icsk->icsk_ack.timeout = timeout;
3409 sk_reset_timer(ssk, &icsk->icsk_delack_timer, timeout);
3410}
3411
a5efdbce 3412void mptcp_subflow_process_delegated(struct sock *ssk, long status)
b19bc294
PA		 3413{
3414 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
3415 struct sock *sk = subflow->conn;
3416
a5efdbce 3417 if (status & BIT(MPTCP_DELEGATE_SEND)) {
bcd97734
PA		 3418 mptcp_data_lock(sk);
3419 if (!sock_owned_by_user(sk))
73a0052a 3420 __mptcp_subflow_push_pending(sk, ssk, true);
bcd97734 3421 else
e9d09bac 3422 __set_bit(MPTCP_PUSH_PENDING, &mptcp_sk(sk)->cb_flags);
bcd97734 3423 mptcp_data_unlock(sk);
bcd97734 3424 }
8005184f
PA		 3425 if (status & BIT(MPTCP_DELEGATE_SNDBUF)) {
3426 mptcp_data_lock(sk);
3427 if (!sock_owned_by_user(sk))
3428 __mptcp_sync_sndbuf(sk);
3429 else
3430 __set_bit(MPTCP_SYNC_SNDBUF, &mptcp_sk(sk)->cb_flags);
3431 mptcp_data_unlock(sk);
3432 }
a5efdbce 3433 if (status & BIT(MPTCP_DELEGATE_ACK))
bcd97734 3434 schedule_3rdack_retransmission(ssk);
b19bc294
PA		 3435}
3436
2c5ebd00
PA		 3437static int mptcp_hash(struct sock *sk)
3438{
3439 /* should never be called,
3440 * we hash the TCP subflows not the master socket
3441 */
3442 WARN_ON_ONCE(1);
3443 return 0;
3444}
3445
3446static void mptcp_unhash(struct sock *sk)
3447{
3448 /* called from sk_common_release(), but nothing to do here */
3449}
3450
cec37a6e 3451static int mptcp_get_port(struct sock *sk, unsigned short snum)
f870fa0b
MM		 3452{
3453 struct mptcp_sock *msk = mptcp_sk(sk);
f870fa0b 3454
cfb63e50
PA		 3455 pr_debug("msk=%p, ssk=%p", msk, msk->first);
3456 if (WARN_ON_ONCE(!msk->first))
cec37a6e 3457 return -EINVAL;
f870fa0b 3458
cfb63e50 3459 return inet_csk_get_port(msk->first, snum);
cec37a6e 3460}
f870fa0b 3461
cec37a6e
PK		 3462void mptcp_finish_connect(struct sock *ssk)
3463{
3464 struct mptcp_subflow_context *subflow;
3465 struct mptcp_sock *msk;
3466 struct sock *sk;
f870fa0b 3467
cec37a6e 3468 subflow = mptcp_subflow_ctx(ssk);
cec37a6e
PK		 3469 sk = subflow->conn;
3470 msk = mptcp_sk(sk);
3471
648ef4b8
MM		 3472 pr_debug("msk=%p, token=%u", sk, subflow->token);
3473
b3ea6b27 3474 subflow->map_seq = subflow->iasn;
648ef4b8 3475 subflow->map_subflow_seq = 1;
6d0060f6 3476
cec37a6e
PK		 3477 /* the socket is not connected yet, no msk/subflow ops can access/race
3478 * accessing the field below
3479 */
cec37a6e 3480 WRITE_ONCE(msk->local_key, subflow->local_key);
6d0060f6 3481 WRITE_ONCE(msk->write_seq, subflow->idsn + 1);
eaa2ffab 3482 WRITE_ONCE(msk->snd_nxt, msk->write_seq);
7439d687 3483 WRITE_ONCE(msk->snd_una, msk->write_seq);
1b1c7a0e 3484
6c714f1b 3485 mptcp_pm_new_connection(msk, ssk, 0);
a6b118fe
FW		 3486
3487 mptcp_rcv_space_init(msk, ssk);
f870fa0b
MM		 3488}
3489
866f26f2 3490void mptcp_sock_graft(struct sock *sk, struct socket *parent)
cf7da0d6
PK		 3491{
3492 write_lock_bh(&sk->sk_callback_lock);
3493 rcu_assign_pointer(sk->sk_wq, &parent->wq);
3494 sk_set_socket(sk, parent);
3495 sk->sk_uid = SOCK_INODE(parent)->i_uid;
3496 write_unlock_bh(&sk->sk_callback_lock);
3497}
3498
e16163b6 3499bool mptcp_finish_join(struct sock *ssk)
f296234c 3500{
e16163b6 3501 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
f296234c
PK		 3502 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
3503 struct sock *parent = (void *)msk;
3e501490 3504 bool ret = true;
f296234c
PK		 3505
3506 pr_debug("msk=%p, subflow=%p", msk, subflow);
3507
3508 /* mptcp socket already closing? */
dc87efdb
FW		 3509 if (!mptcp_is_fully_established(parent)) {
3510 subflow->reset_reason = MPTCP_RST_EMPTCP;
f296234c 3511 return false;
dc87efdb 3512 }
f296234c 3513
55b47ca7
PA		 3514 /* active subflow, already present inside the conn_list */
3515 if (!list_empty(&subflow->node)) {
3516 mptcp_subflow_joined(msk, ssk);
8005184f 3517 mptcp_propagate_sndbuf(parent, ssk);
55b47ca7
PA		 3518 return true;
3519 }
f296234c 3520
3e501490
PA		 3521 if (!mptcp_pm_allow_new_subflow(msk))
3522 goto err_prohibited;
10f6d46c 3523
55b47ca7 3524 /* If we can't acquire msk socket lock here, let the release callback
3e501490 3525 * handle it
10f6d46c 3526 */
3e501490
PA		 3527 mptcp_data_lock(parent);
3528 if (!sock_owned_by_user(parent)) {
3529 ret = __mptcp_finish_join(msk, ssk);
3530 if (ret) {
3531 sock_hold(ssk);
3532 list_add_tail(&subflow->node, &msk->conn_list);
3533 }
3534 } else {
e16163b6 3535 sock_hold(ssk);
3e501490 3536 list_add_tail(&subflow->node, &msk->join_list);
e9d09bac 3537 __set_bit(MPTCP_FLUSH_JOIN_LIST, &msk->cb_flags);
e16163b6 3538 }
3e501490
PA		 3539 mptcp_data_unlock(parent);
3540
dc87efdb 3541 if (!ret) {
3e501490 3542err_prohibited:
dc87efdb 3543 subflow->reset_reason = MPTCP_RST_EPROHIBIT;
10f6d46c 3544 return false;
dc87efdb 3545 }
10f6d46c 3546
10f6d46c 3547 return true;
f296234c
PK		 3548}
3549
76e2a55d
PA		 3550static void mptcp_shutdown(struct sock *sk, int how)
3551{
3552 pr_debug("sk=%p, how=%d", sk, how);
3553
3554 if ((how & SEND_SHUTDOWN) && mptcp_close_state(sk))
3555 __mptcp_wr_shutdown(sk);
3556}
3557
6511882c
PA		 3558static int mptcp_forward_alloc_get(const struct sock *sk)
3559{
9531e4a8
ED		 3560 return READ_ONCE(sk->sk_forward_alloc) +
3561 READ_ONCE(mptcp_sk(sk)->rmem_fwd_alloc);
6511882c
PA		 3562}
3563
644807e3
FW		 3564static int mptcp_ioctl_outq(const struct mptcp_sock *msk, u64 v)
3565{
3566 const struct sock *sk = (void *)msk;
3567 u64 delta;
3568
3569 if (sk->sk_state == TCP_LISTEN)
3570 return -EINVAL;
3571
3572 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
3573 return 0;
3574
3575 delta = msk->write_seq - v;
07c2c7a3
PA		 3576 if (__mptcp_check_fallback(msk) && msk->first) {
3577 struct tcp_sock *tp = tcp_sk(msk->first);
3578
3579 /* the first subflow is disconnected after close - see
3580 * __mptcp_close_ssk(). tcp_disconnect() moves the write_seq
3581 * so ignore that status, too.
3582 */
3583 if (!((1 << msk->first->sk_state) &
3584 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE)))
3585 delta += READ_ONCE(tp->write_seq) - tp->snd_una;
3586 }
644807e3
FW		 3587 if (delta > INT_MAX)
3588 delta = INT_MAX;
3589
3590 return (int)delta;
3591}
3592
e1d001fa 3593static int mptcp_ioctl(struct sock *sk, int cmd, int *karg)
644807e3
FW		 3594{
3595 struct mptcp_sock *msk = mptcp_sk(sk);
3596 bool slow;
644807e3
FW		 3597
3598 switch (cmd) {
3599 case SIOCINQ:
3600 if (sk->sk_state == TCP_LISTEN)
3601 return -EINVAL;
3602
3603 lock_sock(sk);
3604 __mptcp_move_skbs(msk);
e1d001fa 3605 *karg = mptcp_inq_hint(sk);
644807e3
FW		 3606 release_sock(sk);
3607 break;
3608 case SIOCOUTQ:
3609 slow = lock_sock_fast(sk);
e1d001fa 3610 *karg = mptcp_ioctl_outq(msk, READ_ONCE(msk->snd_una));
644807e3
FW		 3611 unlock_sock_fast(sk, slow);
3612 break;
3613 case SIOCOUTQNSD:
3614 slow = lock_sock_fast(sk);
e1d001fa 3615 *karg = mptcp_ioctl_outq(msk, msk->snd_nxt);
644807e3
FW		 3616 unlock_sock_fast(sk, slow);
3617 break;
3618 default:
3619 return -ENOIOCTLCMD;
3620 }
3621
e1d001fa 3622 return 0;
644807e3
FW		 3623}
3624
54f1944e
PA		 3625static void mptcp_subflow_early_fallback(struct mptcp_sock *msk,
3626 struct mptcp_subflow_context *subflow)
3627{
3628 subflow->request_mptcp = 0;
3629 __mptcp_do_fallback(msk);
3630}
3631
3632static int mptcp_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
3633{
3634 struct mptcp_subflow_context *subflow;
3635 struct mptcp_sock *msk = mptcp_sk(sk);
54f1944e 3636 int err = -EINVAL;
ccae357c 3637 struct sock *ssk;
54f1944e 3638
3f326a82
PA		 3639 ssk = __mptcp_nmpc_sk(msk);
3640 if (IS_ERR(ssk))
3641 return PTR_ERR(ssk);
54f1944e 3642
54f1944e 3643 inet_sk_state_store(sk, TCP_SYN_SENT);
ccae357c 3644 subflow = mptcp_subflow_ctx(ssk);
54f1944e
PA		 3645#ifdef CONFIG_TCP_MD5SIG
3646 /* no MPTCP if MD5SIG is enabled on this socket or we may run out of
3647 * TCP option space.
3648 */
ccae357c 3649 if (rcu_access_pointer(tcp_sk(ssk)->md5sig_info))
54f1944e
PA		 3650 mptcp_subflow_early_fallback(msk, subflow);
3651#endif
ccae357c
PA		 3652 if (subflow->request_mptcp && mptcp_token_new_connect(ssk)) {
3653 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_TOKENFALLBACKINIT);
54f1944e
PA		 3654 mptcp_subflow_early_fallback(msk, subflow);
3655 }
3656 if (likely(!__mptcp_check_fallback(msk)))
3657 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPCAPABLEACTIVE);
3658
3659 /* if reaching here via the fastopen/sendmsg path, the caller already
3660 * acquired the subflow socket lock, too.
3661 */
ccae357c
PA		 3662 if (!msk->fastopening)
3663 lock_sock(ssk);
3664
3665 /* the following mirrors closely a very small chunk of code from
3666 * __inet_stream_connect()
3667 */
3668 if (ssk->sk_state != TCP_CLOSE)
3669 goto out;
3670
3671 if (BPF_CGROUP_PRE_CONNECT_ENABLED(ssk)) {
3672 err = ssk->sk_prot->pre_connect(ssk, uaddr, addr_len);
3673 if (err)
3674 goto out;
3675 }
3676
3677 err = ssk->sk_prot->connect(ssk, uaddr, addr_len);
3678 if (err < 0)
3679 goto out;
3680
08e39c0d 3681 inet_assign_bit(DEFER_CONNECT, sk, inet_test_bit(DEFER_CONNECT, ssk));
ccae357c
PA		 3682
3683out:
3684 if (!msk->fastopening)
3685 release_sock(ssk);
54f1944e
PA		 3686
3687 /* on successful connect, the msk state will be moved to established by
3688 * subflow_finish_connect()
3689 */
ccae357c 3690 if (unlikely(err)) {
131a6277
PA		 3691 /* avoid leaving a dangling token in an unconnected socket */
3692 mptcp_token_destroy(msk);
ccae357c 3693 inet_sk_state_store(sk, TCP_CLOSE);
54f1944e
PA		 3694 return err;
3695 }
3696
ccae357c 3697 mptcp_copy_inaddrs(sk, ssk);
786fc124 3698 return 0;
54f1944e
PA		 3699}
3700
f870fa0b
MM		 3701static struct proto mptcp_prot = {
3702 .name = "MPTCP",
3703 .owner = THIS_MODULE,
3704 .init = mptcp_init_sock,
54f1944e 3705 .connect = mptcp_connect,
18b683bf 3706 .disconnect = mptcp_disconnect,
f870fa0b 3707 .close = mptcp_close,
717e79c8
PK		 3708 .setsockopt = mptcp_setsockopt,
3709 .getsockopt = mptcp_getsockopt,
76e2a55d 3710 .shutdown = mptcp_shutdown,
79c0949e 3711 .destroy = mptcp_destroy,
f870fa0b 3712 .sendmsg = mptcp_sendmsg,
644807e3 3713 .ioctl = mptcp_ioctl,
f870fa0b 3714 .recvmsg = mptcp_recvmsg,
14c441b5 3715 .release_cb = mptcp_release_cb,
2c5ebd00
PA		 3716 .hash = mptcp_hash,
3717 .unhash = mptcp_unhash,
cec37a6e 3718 .get_port = mptcp_get_port,
6511882c 3719 .forward_alloc_get = mptcp_forward_alloc_get,
d027236c 3720 .sockets_allocated = &mptcp_sockets_allocated,
0defbb0a 3721
d027236c 3722 .memory_allocated = &tcp_memory_allocated,
0defbb0a
ED		 3723 .per_cpu_fw_alloc = &tcp_memory_per_cpu_fw_alloc,
3724
d027236c 3725 .memory_pressure = &tcp_memory_pressure,
d027236c 3726 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_tcp_wmem),
989ef49b 3727 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_tcp_rmem),
d027236c 3728 .sysctl_mem = sysctl_tcp_mem,
f870fa0b 3729 .obj_size = sizeof(struct mptcp_sock),
2c5ebd00 3730 .slab_flags = SLAB_TYPESAFE_BY_RCU,
f870fa0b
MM		 3731 .no_autobind = true,
3732};
3733
2303f994
PK		 3734static int mptcp_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3735{
3736 struct mptcp_sock *msk = mptcp_sk(sock->sk);
8cf2ebdc 3737 struct sock *ssk, *sk = sock->sk;
8cf2ebdc 3738 int err = -EINVAL;
2303f994 3739
8cf2ebdc 3740 lock_sock(sk);
3f326a82
PA		 3741 ssk = __mptcp_nmpc_sk(msk);
3742 if (IS_ERR(ssk)) {
3743 err = PTR_ERR(ssk);
2303f994
PK		 3744 goto unlock;
3745 }
3746
8cf2ebdc
PA		 3747 if (sk->sk_family == AF_INET)
3748 err = inet_bind_sk(ssk, uaddr, addr_len);
3749#if IS_ENABLED(CONFIG_MPTCP_IPV6)
3750 else if (sk->sk_family == AF_INET6)
3751 err = inet6_bind_sk(ssk, uaddr, addr_len);
3752#endif
cf7da0d6 3753 if (!err)
8cf2ebdc 3754 mptcp_copy_inaddrs(sk, ssk);
2303f994
PK		 3755
3756unlock:
8cf2ebdc 3757 release_sock(sk);
2303f994
PK		 3758 return err;
3759}
3760
cf7da0d6
PK		 3761static int mptcp_listen(struct socket *sock, int backlog)
3762{
3763 struct mptcp_sock *msk = mptcp_sk(sock->sk);
cfdcfeed 3764 struct sock *sk = sock->sk;
40f56d0c 3765 struct sock *ssk;
cf7da0d6
PK		 3766 int err;
3767
3768 pr_debug("msk=%p", msk);
3769
cfdcfeed 3770 lock_sock(sk);
0226436a
PA		 3771
3772 err = -EINVAL;
3773 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
3774 goto unlock;
3775
3f326a82
PA		 3776 ssk = __mptcp_nmpc_sk(msk);
3777 if (IS_ERR(ssk)) {
3778 err = PTR_ERR(ssk);
cf7da0d6
PK		 3779 goto unlock;
3780 }
3781
cfdcfeed
MD		 3782 inet_sk_state_store(sk, TCP_LISTEN);
3783 sock_set_flag(sk, SOCK_RCU_FREE);
5e20087d 3784
40f56d0c
PA		 3785 lock_sock(ssk);
3786 err = __inet_listen_sk(ssk, backlog);
3787 release_sock(ssk);
3788 inet_sk_state_store(sk, inet_sk_state_load(ssk));
3789
c558246e
MD		 3790 if (!err) {
3791 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
40f56d0c
PA		 3792 mptcp_copy_inaddrs(sk, ssk);
3793 mptcp_event_pm_listener(ssk, MPTCP_EVENT_LISTENER_CREATED);
c558246e 3794 }
cf7da0d6
PK		 3795
3796unlock:
cfdcfeed 3797 release_sock(sk);
cf7da0d6
PK		 3798 return err;
3799}
3800
cf7da0d6
PK		 3801static int mptcp_stream_accept(struct socket *sock, struct socket *newsock,
3802 int flags, bool kern)
3803{
3804 struct mptcp_sock *msk = mptcp_sk(sock->sk);
1f6610b9 3805 struct sock *ssk, *newsk;
cf7da0d6
PK		 3806 int err;
3807
3808 pr_debug("msk=%p", msk);
3809
5b825727 3810 /* Buggy applications can call accept on socket states other then LISTEN
61761231
PA		 3811 * but no need to allocate the first subflow just to error out.
3812 */
1f6610b9
PA		 3813 ssk = READ_ONCE(msk->first);
3814 if (!ssk)
71ba088c 3815 return -EINVAL;
cf7da0d6 3816
8e2b8a9f
DC		 3817 pr_debug("ssk=%p, listener=%p", ssk, mptcp_subflow_ctx(ssk));
3818 newsk = inet_csk_accept(ssk, flags, &err, kern);
e76c8ef5
PA		 3819 if (!newsk)
3820 return err;
3821
8e2b8a9f
DC		 3822 pr_debug("newsk=%p, subflow is mptcp=%d", newsk, sk_is_mptcp(newsk));
3823 if (sk_is_mptcp(newsk)) {
cf7da0d6 3824 struct mptcp_subflow_context *subflow;
8e2b8a9f
DC		 3825 struct sock *new_mptcp_sock;
3826
3827 subflow = mptcp_subflow_ctx(newsk);
3828 new_mptcp_sock = subflow->conn;
3829
3830 /* is_mptcp should be false if subflow->conn is missing, see
3831 * subflow_syn_recv_sock()
3832 */
3833 if (WARN_ON_ONCE(!new_mptcp_sock)) {
3834 tcp_sk(newsk)->is_mptcp = 0;
3835 goto tcpfallback;
3836 }
3837
3838 newsk = new_mptcp_sock;
3839 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_MPCAPABLEPASSIVEACK);
3840
3841 newsk->sk_kern_sock = kern;
3842 lock_sock(newsk);
3843 __inet_accept(sock, newsock, newsk);
0397c6d8 3844
a5ef058d 3845 set_bit(SOCK_CUSTOM_SOCKOPT, &newsock->flags);
8e2b8a9f 3846 msk = mptcp_sk(newsk);
b6985b9b 3847 msk->in_accept_queue = 0;
a5ef058d 3848
cf7da0d6
PK		 3849 /* set ssk->sk_socket of accept()ed flows to mptcp socket.
3850 * This is needed so NOSPACE flag can be set from tcp stack.
3851 */
190f8b06 3852 mptcp_for_each_subflow(msk, subflow) {
cf7da0d6
PK		 3853 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
3854
3855 if (!ssk->sk_socket)
3856 mptcp_sock_graft(ssk, newsock);
3857 }
63740448
PA		 3858
3859 /* Do late cleanup for the first subflow as necessary. Also
3860 * deal with bad peers not doing a complete shutdown.
3861 */
39880bd8 3862 if (unlikely(inet_sk_state_load(msk->first) == TCP_CLOSE)) {
63740448
PA		 3863 __mptcp_close_ssk(newsk, msk->first,
3864 mptcp_subflow_ctx(msk->first), 0);
39880bd8 3865 if (unlikely(list_is_singular(&msk->conn_list)))
63740448
PA		 3866 inet_sk_state_store(newsk, TCP_CLOSE);
3867 }
8e2b8a9f
DC		 3868 } else {
3869 MPTCP_INC_STATS(sock_net(ssk),
3870 MPTCP_MIB_MPCAPABLEPASSIVEFALLBACK);
3871tcpfallback:
3872 newsk->sk_kern_sock = kern;
3873 lock_sock(newsk);
3874 __inet_accept(sock, newsock, newsk);
3875 /* we are being invoked after accepting a non-mp-capable
3876 * flow: sk is a tcp_sk, not an mptcp one.
3877 *
3878 * Hand the socket over to tcp so all further socket ops
3879 * bypass mptcp.
3880 */
3881 WRITE_ONCE(newsock->sk->sk_socket->ops,
3882 mptcp_fallback_tcp_ops(newsock->sk));
cf7da0d6 3883 }
e76c8ef5 3884 release_sock(newsk);
cf7da0d6 3885
e76c8ef5 3886 return 0;
cf7da0d6
PK		 3887}
3888
8edf0864
FW		 3889static __poll_t mptcp_check_writeable(struct mptcp_sock *msk)
3890{
3891 struct sock *sk = (struct sock *)msk;
8edf0864 3892
8edf0864
FW		 3893 if (sk_stream_is_writeable(sk))
3894 return EPOLLOUT | EPOLLWRNORM;
3895
5cf92bba 3896 mptcp_set_nospace(sk);
6e628cd3
PA		 3897 smp_mb__after_atomic(); /* msk->flags is changed by write_space cb */
3898 if (sk_stream_is_writeable(sk))
3899 return EPOLLOUT | EPOLLWRNORM;
8edf0864 3900
6e628cd3 3901 return 0;
8edf0864
FW		 3902}
3903
2303f994
PK		 3904static __poll_t mptcp_poll(struct file *file, struct socket *sock,
3905 struct poll_table_struct *wait)
3906{
1891c4a0 3907 struct sock *sk = sock->sk;
8ab183de 3908 struct mptcp_sock *msk;
2303f994 3909 __poll_t mask = 0;
6b9831bf 3910 u8 shutdown;
8a05661b 3911 int state;
2303f994 3912
1891c4a0 3913 msk = mptcp_sk(sk);
1891c4a0 3914 sock_poll_wait(file, sock, wait);
1891c4a0 3915
8a05661b 3916 state = inet_sk_state_load(sk);
6719331c 3917 pr_debug("msk=%p state=%d flags=%lx", msk, state, msk->flags);
71ba088c 3918 if (state == TCP_LISTEN) {
5426a4ef 3919 struct sock *ssk = READ_ONCE(msk->first);
5b825727 3920
5426a4ef 3921 if (WARN_ON_ONCE(!ssk))
71ba088c
PA		 3922 return 0;
3923
5426a4ef 3924 return inet_csk_listen_poll(ssk);
71ba088c 3925 }
8a05661b 3926
6b9831bf
PA		 3927 shutdown = READ_ONCE(sk->sk_shutdown);
3928 if (shutdown == SHUTDOWN_MASK || state == TCP_CLOSE)
3929 mask |= EPOLLHUP;
3930 if (shutdown & RCV_SHUTDOWN)
3931 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
3932
8a05661b 3933 if (state != TCP_SYN_SENT && state != TCP_SYN_RECV) {
5684ab1a 3934 mask |= mptcp_check_readable(sk);
6b9831bf
PA		 3935 if (shutdown & SEND_SHUTDOWN)
3936 mask |= EPOLLOUT | EPOLLWRNORM;
3937 else
3938 mask |= mptcp_check_writeable(msk);
08e39c0d
ED		 3939 } else if (state == TCP_SYN_SENT &&
3940 inet_test_bit(DEFER_CONNECT, sk)) {
a42cf9d1
BH		 3941 /* cf tcp_poll() note about TFO */
3942 mask |= EPOLLOUT | EPOLLWRNORM;
8a05661b 3943 }
1891c4a0 3944
d89e3ed7 3945 /* This barrier is coupled with smp_wmb() in __mptcp_error_report() */
15cc1045 3946 smp_rmb();
9ae8e5ad 3947 if (READ_ONCE(sk->sk_err))
15cc1045
PA		 3948 mask |= EPOLLERR;
3949
2303f994
PK		 3950 return mask;
3951}
3952
e42f1ac6
FW		 3953static const struct proto_ops mptcp_stream_ops = {
3954 .family = PF_INET,
3955 .owner = THIS_MODULE,
0a3cc579 3956 .release = inet_release,
e42f1ac6 3957 .bind = mptcp_bind,
786fc124 3958 .connect = inet_stream_connect,
e42f1ac6
FW		 3959 .socketpair = sock_no_socketpair,
3960 .accept = mptcp_stream_accept,
d2f77c53 3961 .getname = inet_getname,
e42f1ac6
FW		 3962 .poll = mptcp_poll,
3963 .ioctl = inet_ioctl,
3964 .gettstamp = sock_gettstamp,
3965 .listen = mptcp_listen,
76e2a55d 3966 .shutdown = inet_shutdown,
e42f1ac6
FW		 3967 .setsockopt = sock_common_setsockopt,
3968 .getsockopt = sock_common_getsockopt,
3969 .sendmsg = inet_sendmsg,
3970 .recvmsg = inet_recvmsg,
3971 .mmap = sock_no_mmap,
5684ab1a 3972 .set_rcvlowat = mptcp_set_rcvlowat,
e42f1ac6 3973};
2303f994 3974
f870fa0b
MM		 3975static struct inet_protosw mptcp_protosw = {
3976 .type = SOCK_STREAM,
3977 .protocol = IPPROTO_MPTCP,
3978 .prot = &mptcp_prot,
2303f994
PK		 3979 .ops = &mptcp_stream_ops,
3980 .flags = INET_PROTOSW_ICSK,
f870fa0b
MM		 3981};
3982
b19bc294
PA		 3983static int mptcp_napi_poll(struct napi_struct *napi, int budget)
3984{
3985 struct mptcp_delegated_action *delegated;
3986 struct mptcp_subflow_context *subflow;
3987 int work_done = 0;
3988
3989 delegated = container_of(napi, struct mptcp_delegated_action, napi);
3990 while ((subflow = mptcp_subflow_delegated_next(delegated)) != NULL) {
3991 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
3992
3993 bh_lock_sock_nested(ssk);
a5efdbce
PA		 3994 if (!sock_owned_by_user(ssk)) {
3995 mptcp_subflow_process_delegated(ssk, xchg(&subflow->delegated_status, 0));
3996 } else {
3997 /* tcp_release_cb_override already processed
3998 * the action or will do at next release_sock().
3999 * In both case must dequeue the subflow here - on the same
4000 * CPU that scheduled it.
4001 */
4002 smp_wmb();
4003 clear_bit(MPTCP_DELEGATE_SCHEDULED, &subflow->delegated_status);
4004 }
b19bc294
PA		 4005 bh_unlock_sock(ssk);
4006 sock_put(ssk);
4007
4008 if (++work_done == budget)
4009 return budget;
4010 }
4011
4012 /* always provide a 0 'work_done' argument, so that napi_complete_done
4013 * will not try accessing the NULL napi->dev ptr
4014 */
4015 napi_complete_done(napi, 0);
4016 return work_done;
4017}
4018
d39dceca 4019void __init mptcp_proto_init(void)
f870fa0b 4020{
b19bc294
PA		 4021 struct mptcp_delegated_action *delegated;
4022 int cpu;
4023
2303f994 4024 mptcp_prot.h.hashinfo = tcp_prot.h.hashinfo;
2303f994 4025
d027236c
PA		 4026 if (percpu_counter_init(&mptcp_sockets_allocated, 0, GFP_KERNEL))
4027 panic("Failed to allocate MPTCP pcpu counter\n");
4028
b19bc294
PA		 4029 init_dummy_netdev(&mptcp_napi_dev);
4030 for_each_possible_cpu(cpu) {
4031 delegated = per_cpu_ptr(&mptcp_delegated_actions, cpu);
4032 INIT_LIST_HEAD(&delegated->head);
16d083e2
JK		 4033 netif_napi_add_tx(&mptcp_napi_dev, &delegated->napi,
4034 mptcp_napi_poll);
b19bc294
PA		 4035 napi_enable(&delegated->napi);
4036 }
4037
2303f994 4038 mptcp_subflow_init();
1b1c7a0e 4039 mptcp_pm_init();
ed1ad86b 4040 mptcp_sched_init();
2c5ebd00 4041 mptcp_token_init();
2303f994 4042
f870fa0b
MM		 4043 if (proto_register(&mptcp_prot, 1) != 0)
4044 panic("Failed to register MPTCP proto.\n");
4045
4046 inet_register_protosw(&mptcp_protosw);
6771bfd9
FW		 4047
4048 BUILD_BUG_ON(sizeof(struct mptcp_skb_cb) > sizeof_field(struct sk_buff, cb));
f870fa0b
MM		 4049}
4050
4051#if IS_ENABLED(CONFIG_MPTCP_IPV6)
e42f1ac6
FW		 4052static const struct proto_ops mptcp_v6_stream_ops = {
4053 .family = PF_INET6,
4054 .owner = THIS_MODULE,
0a3cc579 4055 .release = inet6_release,
e42f1ac6 4056 .bind = mptcp_bind,
786fc124 4057 .connect = inet_stream_connect,
e42f1ac6
FW		 4058 .socketpair = sock_no_socketpair,
4059 .accept = mptcp_stream_accept,
d2f77c53 4060 .getname = inet6_getname,
e42f1ac6
FW		 4061 .poll = mptcp_poll,
4062 .ioctl = inet6_ioctl,
4063 .gettstamp = sock_gettstamp,
4064 .listen = mptcp_listen,
76e2a55d 4065 .shutdown = inet_shutdown,
e42f1ac6
FW		 4066 .setsockopt = sock_common_setsockopt,
4067 .getsockopt = sock_common_getsockopt,
4068 .sendmsg = inet6_sendmsg,
4069 .recvmsg = inet6_recvmsg,
4070 .mmap = sock_no_mmap,
e42f1ac6 4071#ifdef CONFIG_COMPAT
3986912f 4072 .compat_ioctl = inet6_compat_ioctl,
e42f1ac6 4073#endif
5684ab1a 4074 .set_rcvlowat = mptcp_set_rcvlowat,
e42f1ac6
FW		 4075};
4076
f870fa0b
MM		 4077static struct proto mptcp_v6_prot;
4078
4079static struct inet_protosw mptcp_v6_protosw = {
4080 .type = SOCK_STREAM,
4081 .protocol = IPPROTO_MPTCP,
4082 .prot = &mptcp_v6_prot,
2303f994 4083 .ops = &mptcp_v6_stream_ops,
f870fa0b
MM		 4084 .flags = INET_PROTOSW_ICSK,
4085};
4086
d39dceca 4087int __init mptcp_proto_v6_init(void)
f870fa0b
MM		 4088{
4089 int err;
4090
4091 mptcp_v6_prot = mptcp_prot;
4092 strcpy(mptcp_v6_prot.name, "MPTCPv6");
4093 mptcp_v6_prot.slab = NULL;
b0519de8 4094 mptcp_v6_prot.obj_size = sizeof(struct mptcp6_sock);
f5f80e32 4095 mptcp_v6_prot.ipv6_pinfo_offset = offsetof(struct mptcp6_sock, np);
f870fa0b
MM		 4096
4097 err = proto_register(&mptcp_v6_prot, 1);
4098 if (err)
4099 return err;
4100
4101 err = inet6_register_protosw(&mptcp_v6_protosw);
4102 if (err)
4103 proto_unregister(&mptcp_v6_prot);
4104
4105 return err;
4106}
4107#endif
Linux 6.x block layer and io_uring tree(s)