projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge tag 'xfs-6.9-merge-9' of git://git.kernel.org/pub/scm/fs/xfs/xfs-linux
[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>
3721b9b6 18#include <net/tcp_states.h>
cf7da0d6
PK		 19#if IS_ENABLED(CONFIG_MPTCP_IPV6)
20#include <net/transp_v6.h>
21#endif
f870fa0b 22#include <net/mptcp.h>
e16163b6 23#include <net/xfrm.h>
644807e3 24#include <asm/ioctls.h>
f870fa0b 25#include "protocol.h"
fc518953 26#include "mib.h"
f870fa0b 27
e10a9892
GT		 28#define CREATE_TRACE_POINTS
29#include <trace/events/mptcp.h>
30
b0519de8
FW		 31#if IS_ENABLED(CONFIG_MPTCP_IPV6)
32struct mptcp6_sock {
33 struct mptcp_sock msk;
34 struct ipv6_pinfo np;
35};
36#endif
37
b7f653b2
FW		 38enum {
39 MPTCP_CMSG_TS = BIT(0),
2c9e7765 40 MPTCP_CMSG_INQ = BIT(1),
b7f653b2
FW		 41};
42
91b6d325 43static struct percpu_counter mptcp_sockets_allocated ____cacheline_aligned_in_smp;
d027236c 44
e16163b6 45static void __mptcp_destroy_sock(struct sock *sk);
81c1d029 46static void mptcp_check_send_data_fin(struct sock *sk);
e16163b6 47
b19bc294
PA		 48DEFINE_PER_CPU(struct mptcp_delegated_action, mptcp_delegated_actions);
49static struct net_device mptcp_napi_dev;
50
6f8a612a
FW		 51/* Returns end sequence number of the receiver's advertised window */
52static u64 mptcp_wnd_end(const struct mptcp_sock *msk)
53{
7439d687 54 return READ_ONCE(msk->wnd_end);
6f8a612a
FW		 55}
56
8e2b8a9f 57static const struct proto_ops *mptcp_fallback_tcp_ops(const struct sock *sk)
0b4f33de 58{
0b4f33de 59#if IS_ENABLED(CONFIG_MPTCP_IPV6)
8e2b8a9f
DC		 60 if (sk->sk_prot == &tcpv6_prot)
61 return &inet6_stream_ops;
0b4f33de 62#endif
8e2b8a9f
DC		 63 WARN_ON_ONCE(sk->sk_prot != &tcp_prot);
64 return &inet_stream_ops;
0b4f33de
FW		 65}
66
fa68018d 67static int __mptcp_socket_create(struct mptcp_sock *msk)
2303f994
PK		 68{
69 struct mptcp_subflow_context *subflow;
70 struct sock *sk = (struct sock *)msk;
71 struct socket *ssock;
72 int err;
73
6bc1fe7d 74 err = mptcp_subflow_create_socket(sk, sk->sk_family, &ssock);
2303f994 75 if (err)
fa68018d 76 return err;
2303f994 77
b8dc6d6c 78 msk->scaling_ratio = tcp_sk(ssock->sk)->scaling_ratio;
1b1b43ee 79 WRITE_ONCE(msk->first, ssock->sk);
2303f994 80 subflow = mptcp_subflow_ctx(ssock->sk);
cec37a6e 81 list_add(&subflow->node, &msk->conn_list);
e16163b6 82 sock_hold(ssock->sk);
2303f994 83 subflow->request_mptcp = 1;
6f06b4d4 84 subflow->subflow_id = msk->subflow_id++;
4cf86ae8
PA		 85
86 /* This is the first subflow, always with id 0 */
a7cfe776 87 WRITE_ONCE(subflow->local_id, 0);
866f26f2 88 mptcp_sock_graft(msk->first, sk->sk_socket);
39880bd8 89 iput(SOCK_INODE(ssock));
e1ff9e82 90
fa68018d 91 return 0;
2303f994
PK		 92}
93
ddb1a072
PA		 94/* If the MPC handshake is not started, returns the first subflow,
95 * eventually allocating it.
96 */
3f326a82 97struct sock *__mptcp_nmpc_sk(struct mptcp_sock *msk)
ddb1a072
PA		 98{
99 struct sock *sk = (struct sock *)msk;
100 int ret;
101
102 if (!((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)))
103 return ERR_PTR(-EINVAL);
104
3f326a82 105 if (!msk->first) {
ddb1a072
PA		 106 ret = __mptcp_socket_create(msk);
107 if (ret)
108 return ERR_PTR(ret);
ddb1a072
PA		 109 }
110
3f326a82 111 return msk->first;
ddb1a072
PA		 112}
113
ab174ad8
PA		 114static void mptcp_drop(struct sock *sk, struct sk_buff *skb)
115{
116 sk_drops_add(sk, skb);
117 __kfree_skb(skb);
118}
119
9531e4a8
ED		 120static void mptcp_rmem_fwd_alloc_add(struct sock *sk, int size)
121{
122 WRITE_ONCE(mptcp_sk(sk)->rmem_fwd_alloc,
123 mptcp_sk(sk)->rmem_fwd_alloc + size);
124}
125
6511882c
PA		 126static void mptcp_rmem_charge(struct sock *sk, int size)
127{
9531e4a8 128 mptcp_rmem_fwd_alloc_add(sk, -size);
6511882c
PA		 129}
130
8268ed4c
PA		 131static bool mptcp_try_coalesce(struct sock *sk, struct sk_buff *to,
132 struct sk_buff *from)
133{
134 bool fragstolen;
135 int delta;
136
137 if (MPTCP_SKB_CB(from)->offset ||
138 !skb_try_coalesce(to, from, &fragstolen, &delta))
139 return false;
140
06242e44
PA		 141 pr_debug("colesced seq %llx into %llx new len %d new end seq %llx",
142 MPTCP_SKB_CB(from)->map_seq, MPTCP_SKB_CB(to)->map_seq,
143 to->len, MPTCP_SKB_CB(from)->end_seq);
ab174ad8 144 MPTCP_SKB_CB(to)->end_seq = MPTCP_SKB_CB(from)->end_seq;
7288ff6e
PA		 145
146 /* note the fwd memory can reach a negative value after accounting
147 * for the delta, but the later skb free will restore a non
148 * negative one
149 */
8268ed4c 150 atomic_add(delta, &sk->sk_rmem_alloc);
6511882c 151 mptcp_rmem_charge(sk, delta);
7288ff6e
PA		 152 kfree_skb_partial(from, fragstolen);
153
8268ed4c
PA		 154 return true;
155}
156
ab174ad8
PA		 157static bool mptcp_ooo_try_coalesce(struct mptcp_sock *msk, struct sk_buff *to,
158 struct sk_buff *from)
159{
160 if (MPTCP_SKB_CB(from)->map_seq != MPTCP_SKB_CB(to)->end_seq)
161 return false;
162
163 return mptcp_try_coalesce((struct sock *)msk, to, from);
164}
165
6511882c
PA		 166static void __mptcp_rmem_reclaim(struct sock *sk, int amount)
167{
100fdd1f 168 amount >>= PAGE_SHIFT;
9531e4a8 169 mptcp_rmem_charge(sk, amount << PAGE_SHIFT);
6511882c
PA		 170 __sk_mem_reduce_allocated(sk, amount);
171}
172
173static void mptcp_rmem_uncharge(struct sock *sk, int size)
174{
175 struct mptcp_sock *msk = mptcp_sk(sk);
176 int reclaimable;
177
9531e4a8 178 mptcp_rmem_fwd_alloc_add(sk, size);
6511882c
PA		 179 reclaimable = msk->rmem_fwd_alloc - sk_unused_reserved_mem(sk);
180
181 /* see sk_mem_uncharge() for the rationale behind the following schema */
d24141fe
PA		 182 if (unlikely(reclaimable >= PAGE_SIZE))
183 __mptcp_rmem_reclaim(sk, reclaimable);
6511882c
PA		 184}
185
186static void mptcp_rfree(struct sk_buff *skb)
187{
188 unsigned int len = skb->truesize;
189 struct sock *sk = skb->sk;
190
191 atomic_sub(len, &sk->sk_rmem_alloc);
192 mptcp_rmem_uncharge(sk, len);
193}
194
36b122ba 195void mptcp_set_owner_r(struct sk_buff *skb, struct sock *sk)
6511882c
PA		 196{
197 skb_orphan(skb);
198 skb->sk = sk;
199 skb->destructor = mptcp_rfree;
200 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
201 mptcp_rmem_charge(sk, skb->truesize);
202}
203
ab174ad8
PA		 204/* "inspired" by tcp_data_queue_ofo(), main differences:
205 * - use mptcp seqs
206 * - don't cope with sacks
207 */
208static void mptcp_data_queue_ofo(struct mptcp_sock *msk, struct sk_buff *skb)
209{
210 struct sock *sk = (struct sock *)msk;
211 struct rb_node **p, *parent;
212 u64 seq, end_seq, max_seq;
213 struct sk_buff *skb1;
214
215 seq = MPTCP_SKB_CB(skb)->map_seq;
216 end_seq = MPTCP_SKB_CB(skb)->end_seq;
f3589be0 217 max_seq = atomic64_read(&msk->rcv_wnd_sent);
ab174ad8 218
06242e44
PA		 219 pr_debug("msk=%p seq=%llx limit=%llx empty=%d", msk, seq, max_seq,
220 RB_EMPTY_ROOT(&msk->out_of_order_queue));
fa3fe2b1 221 if (after64(end_seq, max_seq)) {
ab174ad8
PA		 222 /* out of window */
223 mptcp_drop(sk, skb);
fa3fe2b1
FW		 224 pr_debug("oow by %lld, rcv_wnd_sent %llu\n",
225 (unsigned long long)end_seq - (unsigned long)max_seq,
f3589be0 226 (unsigned long long)atomic64_read(&msk->rcv_wnd_sent));
06242e44 227 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_NODSSWINDOW);
ab174ad8
PA		 228 return;
229 }
230
231 p = &msk->out_of_order_queue.rb_node;
06242e44 232 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOQUEUE);
ab174ad8
PA		 233 if (RB_EMPTY_ROOT(&msk->out_of_order_queue)) {
234 rb_link_node(&skb->rbnode, NULL, p);
235 rb_insert_color(&skb->rbnode, &msk->out_of_order_queue);
236 msk->ooo_last_skb = skb;
237 goto end;
238 }
239
240 /* with 2 subflows, adding at end of ooo queue is quite likely
241 * Use of ooo_last_skb avoids the O(Log(N)) rbtree lookup.
242 */
06242e44
PA		 243 if (mptcp_ooo_try_coalesce(msk, msk->ooo_last_skb, skb)) {
244 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOMERGE);
245 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOQUEUETAIL);
ab174ad8 246 return;
06242e44 247 }
ab174ad8
PA		 248
249 /* Can avoid an rbtree lookup if we are adding skb after ooo_last_skb */
250 if (!before64(seq, MPTCP_SKB_CB(msk->ooo_last_skb)->end_seq)) {
06242e44 251 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOQUEUETAIL);
ab174ad8
PA		 252 parent = &msk->ooo_last_skb->rbnode;
253 p = &parent->rb_right;
254 goto insert;
255 }
256
257 /* Find place to insert this segment. Handle overlaps on the way. */
258 parent = NULL;
259 while (*p) {
260 parent = *p;
261 skb1 = rb_to_skb(parent);
262 if (before64(seq, MPTCP_SKB_CB(skb1)->map_seq)) {
263 p = &parent->rb_left;
264 continue;
265 }
266 if (before64(seq, MPTCP_SKB_CB(skb1)->end_seq)) {
267 if (!after64(end_seq, MPTCP_SKB_CB(skb1)->end_seq)) {
268 /* All the bits are present. Drop. */
269 mptcp_drop(sk, skb);
06242e44 270 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
ab174ad8
PA		 271 return;
272 }
273 if (after64(seq, MPTCP_SKB_CB(skb1)->map_seq)) {
274 /* partial overlap:
275 * | skb |
276 * | skb1 |
277 * continue traversing
278 */
279 } else {
280 /* skb's seq == skb1's seq and skb covers skb1.
281 * Replace skb1 with skb.
282 */
283 rb_replace_node(&skb1->rbnode, &skb->rbnode,
284 &msk->out_of_order_queue);
285 mptcp_drop(sk, skb1);
06242e44 286 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
ab174ad8
PA		 287 goto merge_right;
288 }
289 } else if (mptcp_ooo_try_coalesce(msk, skb1, skb)) {
06242e44 290 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOMERGE);
ab174ad8
PA		 291 return;
292 }
293 p = &parent->rb_right;
294 }
06242e44 295
ab174ad8
PA		 296insert:
297 /* Insert segment into RB tree. */
298 rb_link_node(&skb->rbnode, parent, p);
299 rb_insert_color(&skb->rbnode, &msk->out_of_order_queue);
300
301merge_right:
302 /* Remove other segments covered by skb. */
303 while ((skb1 = skb_rb_next(skb)) != NULL) {
304 if (before64(end_seq, MPTCP_SKB_CB(skb1)->end_seq))
305 break;
306 rb_erase(&skb1->rbnode, &msk->out_of_order_queue);
307 mptcp_drop(sk, skb1);
06242e44 308 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
ab174ad8
PA		 309 }
310 /* If there is no skb after us, we are the last_skb ! */
311 if (!skb1)
312 msk->ooo_last_skb = skb;
313
314end:
315 skb_condense(skb);
6511882c
PA		 316 mptcp_set_owner_r(skb, sk);
317}
318
319static bool mptcp_rmem_schedule(struct sock *sk, struct sock *ssk, int size)
320{
321 struct mptcp_sock *msk = mptcp_sk(sk);
322 int amt, amount;
323
69d93dae 324 if (size <= msk->rmem_fwd_alloc)
6511882c
PA		 325 return true;
326
69d93dae 327 size -= msk->rmem_fwd_alloc;
6511882c 328 amt = sk_mem_pages(size);
100fdd1f 329 amount = amt << PAGE_SHIFT;
4aaa1685
PA		 330 if (!__sk_mem_raise_allocated(sk, size, amt, SK_MEM_RECV))
331 return false;
6511882c 332
9531e4a8 333 mptcp_rmem_fwd_alloc_add(sk, amount);
6511882c 334 return true;
ab174ad8
PA		 335}
336
337static bool __mptcp_move_skb(struct mptcp_sock *msk, struct sock *ssk,
338 struct sk_buff *skb, unsigned int offset,
339 size_t copy_len)
6771bfd9 340{
ab174ad8 341 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
6771bfd9 342 struct sock *sk = (struct sock *)msk;
4e637c70 343 struct sk_buff *tail;
b7f653b2 344 bool has_rxtstamp;
6771bfd9
FW		 345
346 __skb_unlink(skb, &ssk->sk_receive_queue);
6771bfd9 347
4e637c70
FW		 348 skb_ext_reset(skb);
349 skb_orphan(skb);
ab174ad8 350
9c3f94e1 351 /* try to fetch required memory from subflow */
6511882c
PA		 352 if (!mptcp_rmem_schedule(sk, ssk, skb->truesize))
353 goto drop;
9c3f94e1 354
b7f653b2
FW		 355 has_rxtstamp = TCP_SKB_CB(skb)->has_rxtstamp;
356
ab174ad8
PA		 357 /* the skb map_seq accounts for the skb offset:
358 * mptcp_subflow_get_mapped_dsn() is based on the current tp->copied_seq
359 * value
360 */
361 MPTCP_SKB_CB(skb)->map_seq = mptcp_subflow_get_mapped_dsn(subflow);
362 MPTCP_SKB_CB(skb)->end_seq = MPTCP_SKB_CB(skb)->map_seq + copy_len;
8268ed4c 363 MPTCP_SKB_CB(skb)->offset = offset;
b7f653b2 364 MPTCP_SKB_CB(skb)->has_rxtstamp = has_rxtstamp;
4e637c70 365
ab174ad8
PA		 366 if (MPTCP_SKB_CB(skb)->map_seq == msk->ack_seq) {
367 /* in sequence */
38967f42 368 msk->bytes_received += copy_len;
8b0308fe 369 WRITE_ONCE(msk->ack_seq, msk->ack_seq + copy_len);
ab174ad8
PA		 370 tail = skb_peek_tail(&sk->sk_receive_queue);
371 if (tail && mptcp_try_coalesce(sk, tail, skb))
372 return true;
4e637c70 373
6511882c 374 mptcp_set_owner_r(skb, sk);
ab174ad8
PA		 375 __skb_queue_tail(&sk->sk_receive_queue, skb);
376 return true;
377 } else if (after64(MPTCP_SKB_CB(skb)->map_seq, msk->ack_seq)) {
378 mptcp_data_queue_ofo(msk, skb);
379 return false;
380 }
381
382 /* old data, keep it simple and drop the whole pkt, sender
383 * will retransmit as needed, if needed.
384 */
06242e44 385 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
9c3f94e1 386drop:
ab174ad8
PA		 387 mptcp_drop(sk, skb);
388 return false;
6771bfd9
FW		 389}
390
f6909dc1 391static void mptcp_stop_rtx_timer(struct sock *sk)
16a9a9da
MM		 392{
393 struct inet_connection_sock *icsk = inet_csk(sk);
394
395 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
396 mptcp_sk(sk)->timer_ival = 0;
397}
398
e16163b6
PA		 399static void mptcp_close_wake_up(struct sock *sk)
400{
401 if (sock_flag(sk, SOCK_DEAD))
402 return;
403
404 sk->sk_state_change(sk);
405 if (sk->sk_shutdown == SHUTDOWN_MASK ||
406 sk->sk_state == TCP_CLOSE)
407 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_HUP);
408 else
409 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
410}
411
9426ce47 412/* called under the msk socket lock */
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:
c693a851 432 mptcp_set_state(sk, TCP_FIN_WAIT2);
16a9a9da
MM		 433 break;
434 case TCP_CLOSING:
16a9a9da 435 case TCP_LAST_ACK:
c693a851 436 mptcp_set_state(sk, TCP_CLOSE);
16a9a9da
MM		 437 break;
438 }
439
e16163b6 440 mptcp_close_wake_up(sk);
16a9a9da
MM		 441 }
442}
443
9426ce47 444/* can be called with no lock acquired */
3721b9b6
MM		 445static bool mptcp_pending_data_fin(struct sock *sk, u64 *seq)
446{
447 struct mptcp_sock *msk = mptcp_sk(sk);
448
449 if (READ_ONCE(msk->rcv_data_fin) &&
9426ce47 450 ((1 << inet_sk_state_load(sk)) &
3721b9b6
MM		 451 (TCPF_ESTABLISHED | TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2))) {
452 u64 rcv_data_fin_seq = READ_ONCE(msk->rcv_data_fin_seq);
453
9426ce47 454 if (READ_ONCE(msk->ack_seq) == rcv_data_fin_seq) {
3721b9b6
MM		 455 if (seq)
456 *seq = rcv_data_fin_seq;
457
458 return true;
459 }
460 }
461
462 return false;
463}
464
403a40f2 465static void mptcp_set_datafin_timeout(struct sock *sk)
6477dd39
MM		 466{
467 struct inet_connection_sock *icsk = inet_csk(sk);
877d11f0 468 u32 retransmits;
6477dd39 469
877d11f0
MM		 470 retransmits = min_t(u32, icsk->icsk_retransmits,
471 ilog2(TCP_RTO_MAX / TCP_RTO_MIN));
472
473 mptcp_sk(sk)->timer_ival = TCP_RTO_MIN << retransmits;
6477dd39
MM		 474}
475
33d41c9c 476static void __mptcp_set_timeout(struct sock *sk, long tout)
3721b9b6 477{
3721b9b6
MM		 478 mptcp_sk(sk)->timer_ival = tout > 0 ? tout : TCP_RTO_MIN;
479}
480
33d41c9c
PA		 481static long mptcp_timeout_from_subflow(const struct mptcp_subflow_context *subflow)
482{
483 const struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
484
71b7dec2
PA		 485 return inet_csk(ssk)->icsk_pending && !subflow->stale_count ?
486 inet_csk(ssk)->icsk_timeout - jiffies : 0;
33d41c9c
PA		 487}
488
489static void mptcp_set_timeout(struct sock *sk)
490{
491 struct mptcp_subflow_context *subflow;
492 long tout = 0;
493
494 mptcp_for_each_subflow(mptcp_sk(sk), subflow)
495 tout = max(tout, mptcp_timeout_from_subflow(subflow));
496 __mptcp_set_timeout(sk, tout);
497}
498
76a13b31 499static inline bool tcp_can_send_ack(const struct sock *ssk)
fd897679
PA		 500{
501 return !((1 << inet_sk_state_load(ssk)) &
20bc80b6 502 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_TIME_WAIT | TCPF_CLOSE | TCPF_LISTEN));
fd897679
PA		 503}
504
a6574302
MM		 505void __mptcp_subflow_send_ack(struct sock *ssk)
506{
507 if (tcp_can_send_ack(ssk))
508 tcp_send_ack(ssk);
509}
510
f5360e9b 511static void mptcp_subflow_send_ack(struct sock *ssk)
340fa666
MM		 512{
513 bool slow;
514
515 slow = lock_sock_fast(ssk);
a6574302 516 __mptcp_subflow_send_ack(ssk);
340fa666
MM		 517 unlock_sock_fast(ssk, slow);
518}
519
fd897679 520static void mptcp_send_ack(struct mptcp_sock *msk)
7ed90803
PA		 521{
522 struct mptcp_subflow_context *subflow;
523
340fa666
MM		 524 mptcp_for_each_subflow(msk, subflow)
525 mptcp_subflow_send_ack(mptcp_subflow_tcp_sock(subflow));
fd897679
PA		 526}
527
fde56eea 528static void mptcp_subflow_cleanup_rbuf(struct sock *ssk)
fd897679 529{
75e908c3 530 bool slow;
fd897679 531
75e908c3 532 slow = lock_sock_fast(ssk);
fde56eea 533 if (tcp_can_send_ack(ssk))
fd897679 534 tcp_cleanup_rbuf(ssk, 1);
75e908c3 535 unlock_sock_fast(ssk, slow);
fde56eea
PA		 536}
537
538static bool mptcp_subflow_could_cleanup(const struct sock *ssk, bool rx_empty)
539{
540 const struct inet_connection_sock *icsk = inet_csk(ssk);
c4512c63 541 u8 ack_pending = READ_ONCE(icsk->icsk_ack.pending);
fde56eea
PA		 542 const struct tcp_sock *tp = tcp_sk(ssk);
543
544 return (ack_pending & ICSK_ACK_SCHED) &&
545 ((READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->rcv_wup) >
546 READ_ONCE(icsk->icsk_ack.rcv_mss)) ||
547 (rx_empty && ack_pending &
548 (ICSK_ACK_PUSHED2 | ICSK_ACK_PUSHED)));
fd897679
PA		 549}
550
551static void mptcp_cleanup_rbuf(struct mptcp_sock *msk)
552{
e3859603 553 int old_space = READ_ONCE(msk->old_wspace);
fd897679 554 struct mptcp_subflow_context *subflow;
e3859603 555 struct sock *sk = (struct sock *)msk;
fde56eea
PA		 556 int space = __mptcp_space(sk);
557 bool cleanup, rx_empty;
e3859603 558
fde56eea 559 cleanup = (space > 0) && (space >= (old_space << 1));
ce599c51 560 rx_empty = !__mptcp_rmem(sk);
fd897679 561
fde56eea
PA		 562 mptcp_for_each_subflow(msk, subflow) {
563 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
fd897679 564
fde56eea
PA		 565 if (cleanup || mptcp_subflow_could_cleanup(ssk, rx_empty))
566 mptcp_subflow_cleanup_rbuf(ssk);
7ed90803
PA		 567 }
568}
569
570static bool mptcp_check_data_fin(struct sock *sk)
3721b9b6
MM		 571{
572 struct mptcp_sock *msk = mptcp_sk(sk);
573 u64 rcv_data_fin_seq;
7ed90803 574 bool ret = false;
3721b9b6 575
3721b9b6
MM		 576 /* Need to ack a DATA_FIN received from a peer while this side
577 * of the connection is in ESTABLISHED, FIN_WAIT1, or FIN_WAIT2.
578 * msk->rcv_data_fin was set when parsing the incoming options
579 * at the subflow level and the msk lock was not held, so this
580 * is the first opportunity to act on the DATA_FIN and change
581 * the msk state.
582 *
583 * If we are caught up to the sequence number of the incoming
584 * DATA_FIN, send the DATA_ACK now and do state transition. If
585 * not caught up, do nothing and let the recv code send DATA_ACK
586 * when catching up.
587 */
588
589 if (mptcp_pending_data_fin(sk, &rcv_data_fin_seq)) {
917944da 590 WRITE_ONCE(msk->ack_seq, msk->ack_seq + 1);
3721b9b6
MM		 591 WRITE_ONCE(msk->rcv_data_fin, 0);
592
6b9831bf 593 WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | RCV_SHUTDOWN);
16a9a9da 594 smp_mb__before_atomic(); /* SHUTDOWN must be visible first */
3721b9b6
MM		 595
596 switch (sk->sk_state) {
597 case TCP_ESTABLISHED:
c693a851 598 mptcp_set_state(sk, TCP_CLOSE_WAIT);
3721b9b6
MM		 599 break;
600 case TCP_FIN_WAIT1:
c693a851 601 mptcp_set_state(sk, TCP_CLOSING);
3721b9b6
MM		 602 break;
603 case TCP_FIN_WAIT2:
c693a851 604 mptcp_set_state(sk, TCP_CLOSE);
3721b9b6
MM		 605 break;
606 default:
607 /* Other states not expected */
608 WARN_ON_ONCE(1);
609 break;
610 }
611
7ed90803 612 ret = true;
81c1d029
PA		 613 if (!__mptcp_check_fallback(msk))
614 mptcp_send_ack(msk);
e16163b6 615 mptcp_close_wake_up(sk);
3721b9b6 616 }
7ed90803 617 return ret;
3721b9b6
MM		 618}
619
6771bfd9
FW		 620static bool __mptcp_move_skbs_from_subflow(struct mptcp_sock *msk,
621 struct sock *ssk,
622 unsigned int *bytes)
623{
624 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
600911ff 625 struct sock *sk = (struct sock *)msk;
6771bfd9
FW		 626 unsigned int moved = 0;
627 bool more_data_avail;
628 struct tcp_sock *tp;
629 bool done = false;
13c7ba0c
FW		 630 int sk_rbuf;
631
632 sk_rbuf = READ_ONCE(sk->sk_rcvbuf);
633
634 if (!(sk->sk_userlocks & SOCK_RCVBUF_LOCK)) {
635 int ssk_rbuf = READ_ONCE(ssk->sk_rcvbuf);
636
637 if (unlikely(ssk_rbuf > sk_rbuf)) {
638 WRITE_ONCE(sk->sk_rcvbuf, ssk_rbuf);
639 sk_rbuf = ssk_rbuf;
640 }
641 }
600911ff 642
ab174ad8 643 pr_debug("msk=%p ssk=%p", msk, ssk);
6771bfd9
FW		 644 tp = tcp_sk(ssk);
645 do {
646 u32 map_remaining, offset;
647 u32 seq = tp->copied_seq;
648 struct sk_buff *skb;
649 bool fin;
650
651 /* try to move as much data as available */
652 map_remaining = subflow->map_data_len -
653 mptcp_subflow_get_map_offset(subflow);
654
655 skb = skb_peek(&ssk->sk_receive_queue);
d9fb8c50 656 if (!skb) {
d89e3ed7
PA		 657 /* With racing move_skbs_to_msk() and __mptcp_move_skbs(),
658 * a different CPU can have already processed the pending
659 * data, stop here or we can enter an infinite loop
d9fb8c50
PA		 660 */
661 if (!moved)
662 done = true;
6771bfd9 663 break;
d9fb8c50 664 }
6771bfd9 665
e1ff9e82 666 if (__mptcp_check_fallback(msk)) {
d89e3ed7
PA		 667 /* Under fallback skbs have no MPTCP extension and TCP could
668 * collapse them between the dummy map creation and the
669 * current dequeue. Be sure to adjust the map size.
e1ff9e82
DC		 670 */
671 map_remaining = skb->len;
672 subflow->map_data_len = skb->len;
673 }
674
6771bfd9
FW		 675 offset = seq - TCP_SKB_CB(skb)->seq;
676 fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
677 if (fin) {
678 done = true;
679 seq++;
680 }
681
682 if (offset < skb->len) {
683 size_t len = skb->len - offset;
684
685 if (tp->urg_data)
686 done = true;
687
ab174ad8
PA		 688 if (__mptcp_move_skb(msk, ssk, skb, offset, len))
689 moved += len;
6771bfd9 690 seq += len;
6771bfd9
FW		 691
692 if (WARN_ON_ONCE(map_remaining < len))
693 break;
694 } else {
695 WARN_ON_ONCE(!fin);
696 sk_eat_skb(ssk, skb);
697 done = true;
698 }
699
700 WRITE_ONCE(tp->copied_seq, seq);
701 more_data_avail = mptcp_subflow_data_available(ssk);
600911ff 702
13c7ba0c 703 if (atomic_read(&sk->sk_rmem_alloc) > sk_rbuf) {
600911ff
FW		 704 done = true;
705 break;
706 }
6771bfd9
FW		 707 } while (more_data_avail);
708
6719331c 709 *bytes += moved;
6771bfd9
FW		 710 return done;
711}
712
87952603 713static bool __mptcp_ofo_queue(struct mptcp_sock *msk)
ab174ad8
PA		 714{
715 struct sock *sk = (struct sock *)msk;
716 struct sk_buff *skb, *tail;
717 bool moved = false;
718 struct rb_node *p;
719 u64 end_seq;
720
721 p = rb_first(&msk->out_of_order_queue);
06242e44 722 pr_debug("msk=%p empty=%d", msk, RB_EMPTY_ROOT(&msk->out_of_order_queue));
ab174ad8
PA		 723 while (p) {
724 skb = rb_to_skb(p);
725 if (after64(MPTCP_SKB_CB(skb)->map_seq, msk->ack_seq))
726 break;
727
728 p = rb_next(p);
729 rb_erase(&skb->rbnode, &msk->out_of_order_queue);
730
731 if (unlikely(!after64(MPTCP_SKB_CB(skb)->end_seq,
732 msk->ack_seq))) {
733 mptcp_drop(sk, skb);
06242e44 734 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
ab174ad8
PA		 735 continue;
736 }
737
738 end_seq = MPTCP_SKB_CB(skb)->end_seq;
739 tail = skb_peek_tail(&sk->sk_receive_queue);
740 if (!tail || !mptcp_ooo_try_coalesce(msk, tail, skb)) {
741 int delta = msk->ack_seq - MPTCP_SKB_CB(skb)->map_seq;
742
743 /* skip overlapping data, if any */
06242e44
PA		 744 pr_debug("uncoalesced seq=%llx ack seq=%llx delta=%d",
745 MPTCP_SKB_CB(skb)->map_seq, msk->ack_seq,
746 delta);
ab174ad8 747 MPTCP_SKB_CB(skb)->offset += delta;
2c9e7765 748 MPTCP_SKB_CB(skb)->map_seq += delta;
ab174ad8
PA		 749 __skb_queue_tail(&sk->sk_receive_queue, skb);
750 }
38967f42 751 msk->bytes_received += end_seq - msk->ack_seq;
9426ce47 752 WRITE_ONCE(msk->ack_seq, end_seq);
ab174ad8
PA		 753 moved = true;
754 }
755 return moved;
756}
757
9f1a9881 758static bool __mptcp_subflow_error_report(struct sock *sk, struct sock *ssk)
d5fbeff1 759{
9f1a9881
PA		 760 int err = sock_error(ssk);
761 int ssk_state;
d5fbeff1 762
9f1a9881
PA		 763 if (!err)
764 return false;
d5fbeff1 765
9f1a9881
PA		 766 /* only propagate errors on fallen-back sockets or
767 * on MPC connect
768 */
769 if (sk->sk_state != TCP_SYN_SENT && !__mptcp_check_fallback(mptcp_sk(sk)))
770 return false;
d5fbeff1 771
9f1a9881
PA		 772 /* We need to propagate only transition to CLOSE state.
773 * Orphaned socket will see such state change via
774 * subflow_sched_work_if_closed() and that path will properly
775 * destroy the msk as needed.
776 */
777 ssk_state = inet_sk_state_load(ssk);
778 if (ssk_state == TCP_CLOSE && !sock_flag(sk, SOCK_DEAD))
c693a851 779 mptcp_set_state(sk, ssk_state);
9f1a9881
PA		 780 WRITE_ONCE(sk->sk_err, -err);
781
782 /* This barrier is coupled with smp_rmb() in mptcp_poll() */
783 smp_wmb();
784 sk_error_report(sk);
785 return true;
786}
d5fbeff1 787
9f1a9881
PA		 788void __mptcp_error_report(struct sock *sk)
789{
790 struct mptcp_subflow_context *subflow;
791 struct mptcp_sock *msk = mptcp_sk(sk);
792
793 mptcp_for_each_subflow(msk, subflow)
794 if (__mptcp_subflow_error_report(sk, mptcp_subflow_tcp_sock(subflow)))
795 break;
d5fbeff1
PA		 796}
797
2e52213c
FW		 798/* In most cases we will be able to lock the mptcp socket. If its already
799 * owned, we need to defer to the work queue to avoid ABBA deadlock.
800 */
99d1055c 801static bool move_skbs_to_msk(struct mptcp_sock *msk, struct sock *ssk)
2e52213c
FW		 802{
803 struct sock *sk = (struct sock *)msk;
804 unsigned int moved = 0;
805
87952603
PA		 806 __mptcp_move_skbs_from_subflow(msk, ssk, &moved);
807 __mptcp_ofo_queue(msk);
499ada50
PA		 808 if (unlikely(ssk->sk_err)) {
809 if (!sock_owned_by_user(sk))
810 __mptcp_error_report(sk);
811 else
e9d09bac 812 __set_bit(MPTCP_ERROR_REPORT, &msk->cb_flags);
499ada50 813 }
2e52213c 814
87952603
PA		 815 /* If the moves have caught up with the DATA_FIN sequence number
816 * it's time to ack the DATA_FIN and change socket state, but
817 * this is not a good place to change state. Let the workqueue
818 * do it.
819 */
820 if (mptcp_pending_data_fin(sk, NULL))
821 mptcp_schedule_work(sk);
99d1055c 822 return moved > 0;
2e52213c
FW		 823}
824
825void mptcp_data_ready(struct sock *sk, struct sock *ssk)
101f6f85 826{
6719331c 827 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
101f6f85 828 struct mptcp_sock *msk = mptcp_sk(sk);
13c7ba0c 829 int sk_rbuf, ssk_rbuf;
101f6f85 830
d7b1bfd0
PA		 831 /* The peer can send data while we are shutting down this
832 * subflow at msk destruction time, but we must avoid enqueuing
833 * more data to the msk receive queue
834 */
835 if (unlikely(subflow->disposable))
836 return;
837
13c7ba0c
FW		 838 ssk_rbuf = READ_ONCE(ssk->sk_rcvbuf);
839 sk_rbuf = READ_ONCE(sk->sk_rcvbuf);
840 if (unlikely(ssk_rbuf > sk_rbuf))
841 sk_rbuf = ssk_rbuf;
842
99d1055c 843 /* over limit? can't append more skbs to msk, Also, no need to wake-up*/
ce599c51
PA		 844 if (__mptcp_rmem(sk) > sk_rbuf) {
845 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_RCVPRUNED);
99d1055c 846 return;
ce599c51 847 }
600911ff 848
99d1055c
PA		 849 /* Wake-up the reader only for in-sequence data */
850 mptcp_data_lock(sk);
5684ab1a 851 if (move_skbs_to_msk(msk, ssk) && mptcp_epollin_ready(sk))
6719331c 852 sk->sk_data_ready(sk);
99d1055c 853 mptcp_data_unlock(sk);
101f6f85
FW		 854}
855
55b47ca7
PA		 856static void mptcp_subflow_joined(struct mptcp_sock *msk, struct sock *ssk)
857{
858 mptcp_subflow_ctx(ssk)->map_seq = READ_ONCE(msk->ack_seq);
859 WRITE_ONCE(msk->allow_infinite_fallback, false);
860 mptcp_event(MPTCP_EVENT_SUB_ESTABLISHED, msk, ssk, GFP_ATOMIC);
861}
862
3e501490 863static bool __mptcp_finish_join(struct mptcp_sock *msk, struct sock *ssk)
ec3edaa7 864{
3e501490 865 struct sock *sk = (struct sock *)msk;
5cf92bba 866
3e501490 867 if (sk->sk_state != TCP_ESTABLISHED)
78962489 868 return false;
ec3edaa7 869
3e501490
PA		 870 /* attach to msk socket only after we are sure we will deal with it
871 * at close time
872 */
873 if (sk->sk_socket && !ssk->sk_socket)
874 mptcp_sock_graft(ssk, sk->sk_socket);
78962489 875
6f06b4d4 876 mptcp_subflow_ctx(ssk)->subflow_id = msk->subflow_id++;
3e501490 877 mptcp_sockopt_sync_locked(msk, ssk);
55b47ca7 878 mptcp_subflow_joined(msk, ssk);
27e5ccc2 879 mptcp_stop_tout_timer(sk);
8005184f 880 __mptcp_propagate_sndbuf(sk, ssk);
3e501490 881 return true;
78962489
FW		 882}
883
56a666c4 884static void __mptcp_flush_join_list(struct sock *sk, struct list_head *join_list)
78962489 885{
3e501490
PA		 886 struct mptcp_subflow_context *tmp, *subflow;
887 struct mptcp_sock *msk = mptcp_sk(sk);
78962489 888
56a666c4 889 list_for_each_entry_safe(subflow, tmp, join_list, node) {
3e501490
PA		 890 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
891 bool slow = lock_sock_fast(ssk);
78962489 892
3e501490
PA		 893 list_move_tail(&subflow->node, &msk->conn_list);
894 if (!__mptcp_finish_join(msk, ssk))
895 mptcp_subflow_reset(ssk);
896 unlock_sock_fast(ssk, slow);
897 }
ec3edaa7
PK		 898}
899
f6909dc1 900static bool mptcp_rtx_timer_pending(struct sock *sk)
b51f9b80
PA		 901{
902 return timer_pending(&inet_csk(sk)->icsk_retransmit_timer);
903}
904
f6909dc1 905static void mptcp_reset_rtx_timer(struct sock *sk)
b51f9b80
PA		 906{
907 struct inet_connection_sock *icsk = inet_csk(sk);
908 unsigned long tout;
909
e16163b6
PA		 910 /* prevent rescheduling on close */
911 if (unlikely(inet_sk_state_load(sk) == TCP_CLOSE))
912 return;
913
33d41c9c 914 tout = mptcp_sk(sk)->timer_ival;
b51f9b80
PA		 915 sk_reset_timer(sk, &icsk->icsk_retransmit_timer, jiffies + tout);
916}
917
ba8f48f7
PA		 918bool mptcp_schedule_work(struct sock *sk)
919{
920 if (inet_sk_state_load(sk) != TCP_CLOSE &&
921 schedule_work(&mptcp_sk(sk)->work)) {
922 /* each subflow already holds a reference to the sk, and the
923 * workqueue is invoked by a subflow, so sk can't go away here.
924 */
925 sock_hold(sk);
926 return true;
927 }
928 return false;
929}
930
7a6a6cbc
PA		 931static struct sock *mptcp_subflow_recv_lookup(const struct mptcp_sock *msk)
932{
933 struct mptcp_subflow_context *subflow;
7a6a6cbc 934
109cdeb8 935 msk_owned_by_me(msk);
7a6a6cbc
PA		 936
937 mptcp_for_each_subflow(msk, subflow) {
99d1055c 938 if (READ_ONCE(subflow->data_avail))
7a6a6cbc
PA		 939 return mptcp_subflow_tcp_sock(subflow);
940 }
941
942 return NULL;
943}
944
3f8e0aae
PA		 945static bool mptcp_skb_can_collapse_to(u64 write_seq,
946 const struct sk_buff *skb,
947 const struct mptcp_ext *mpext)
57040755
PA		 948{
949 if (!tcp_skb_can_collapse_to(skb))
950 return false;
951
5a369ca6
PA		 952 /* can collapse only if MPTCP level sequence is in order and this
953 * mapping has not been xmitted yet
954 */
955 return mpext && mpext->data_seq + mpext->data_len == write_seq &&
956 !mpext->frozen;
57040755
PA		 957}
958
29249eac
PA		 959/* we can append data to the given data frag if:
960 * - there is space available in the backing page_frag
961 * - the data frag tail matches the current page_frag free offset
962 * - the data frag end sequence number matches the current write seq
963 */
18b683bf
PA		 964static bool mptcp_frag_can_collapse_to(const struct mptcp_sock *msk,
965 const struct page_frag *pfrag,
966 const struct mptcp_data_frag *df)
967{
968 return df && pfrag->page == df->page &&
d9ca1de8 969 pfrag->size - pfrag->offset > 0 &&
29249eac 970 pfrag->offset == (df->offset + df->data_len) &&
18b683bf
PA		 971 df->data_seq + df->data_len == msk->write_seq;
972}
973
d027236c
PA		 974static void dfrag_uncharge(struct sock *sk, int len)
975{
976 sk_mem_uncharge(sk, len);
7948f6cc 977 sk_wmem_queued_add(sk, -len);
d027236c
PA		 978}
979
980static void dfrag_clear(struct sock *sk, struct mptcp_data_frag *dfrag)
18b683bf 981{
d027236c
PA		 982 int len = dfrag->data_len + dfrag->overhead;
983
18b683bf 984 list_del(&dfrag->list);
d027236c 985 dfrag_uncharge(sk, len);
18b683bf
PA		 986 put_page(dfrag->page);
987}
988
9426ce47 989/* called under both the msk socket lock and the data lock */
6e628cd3 990static void __mptcp_clean_una(struct sock *sk)
18b683bf
PA		 991{
992 struct mptcp_sock *msk = mptcp_sk(sk);
993 struct mptcp_data_frag *dtmp, *dfrag;
e1ff9e82
DC		 994 u64 snd_una;
995
7439d687 996 snd_una = msk->snd_una;
18b683bf
PA		 997 list_for_each_entry_safe(dfrag, dtmp, &msk->rtx_queue, list) {
998 if (after64(dfrag->data_seq + dfrag->data_len, snd_una))
999 break;
1000
1e1d9d6f
PA		 1001 if (unlikely(dfrag == msk->first_pending)) {
1002 /* in recovery mode can see ack after the current snd head */
1003 if (WARN_ON_ONCE(!msk->recovery))
1004 break;
1005
1006 WRITE_ONCE(msk->first_pending, mptcp_send_next(sk));
1007 }
1008
d027236c 1009 dfrag_clear(sk, dfrag);
d027236c
PA		 1010 }
1011
7948f6cc
FW		 1012 dfrag = mptcp_rtx_head(sk);
1013 if (dfrag && after64(snd_una, dfrag->data_seq)) {
53eb4c38
PA		 1014 u64 delta = snd_una - dfrag->data_seq;
1015
1e1d9d6f
PA		 1016 /* prevent wrap around in recovery mode */
1017 if (unlikely(delta > dfrag->already_sent)) {
1018 if (WARN_ON_ONCE(!msk->recovery))
1019 goto out;
1020 if (WARN_ON_ONCE(delta > dfrag->data_len))
1021 goto out;
1022 dfrag->already_sent += delta - dfrag->already_sent;
1023 }
7948f6cc
FW		 1024
1025 dfrag->data_seq += delta;
53eb4c38 1026 dfrag->offset += delta;
7948f6cc 1027 dfrag->data_len -= delta;
d9ca1de8 1028 dfrag->already_sent -= delta;
7948f6cc
FW		 1029
1030 dfrag_uncharge(sk, delta);
7948f6cc
FW		 1031 }
1032
1e1d9d6f
PA		 1033 /* all retransmitted data acked, recovery completed */
1034 if (unlikely(msk->recovery) && after64(msk->snd_una, msk->recovery_snd_nxt))
1035 msk->recovery = false;
1036
53eb4c38 1037out:
d440a4e2 1038 if (snd_una == msk->snd_nxt && snd_una == msk->write_seq) {
f6909dc1
PA		 1039 if (mptcp_rtx_timer_pending(sk) && !mptcp_data_fin_enabled(msk))
1040 mptcp_stop_rtx_timer(sk);
6e628cd3 1041 } else {
f6909dc1 1042 mptcp_reset_rtx_timer(sk);
18b683bf 1043 }
d440a4e2
PA		 1044
1045 if (mptcp_pending_data_fin_ack(sk))
1046 mptcp_schedule_work(sk);
18b683bf
PA		 1047}
1048
417789df
PA		 1049static void __mptcp_clean_una_wakeup(struct sock *sk)
1050{
765ff425
PA		 1051 lockdep_assert_held_once(&sk->sk_lock.slock);
1052
417789df
PA		 1053 __mptcp_clean_una(sk);
1054 mptcp_write_space(sk);
1055}
1056
b5941f06
PA		 1057static void mptcp_clean_una_wakeup(struct sock *sk)
1058{
1059 mptcp_data_lock(sk);
1060 __mptcp_clean_una_wakeup(sk);
1061 mptcp_data_unlock(sk);
1062}
1063
724cfd2e 1064static void mptcp_enter_memory_pressure(struct sock *sk)
18b683bf 1065{
d9ca1de8
PA		 1066 struct mptcp_subflow_context *subflow;
1067 struct mptcp_sock *msk = mptcp_sk(sk);
1068 bool first = true;
1069
d9ca1de8
PA		 1070 mptcp_for_each_subflow(msk, subflow) {
1071 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1072
1073 if (first)
1074 tcp_enter_memory_pressure(ssk);
1075 sk_stream_moderate_sndbuf(ssk);
8005184f 1076
d9ca1de8
PA		 1077 first = false;
1078 }
8005184f 1079 __mptcp_sync_sndbuf(sk);
724cfd2e
PA		 1080}
1081
1082/* ensure we get enough memory for the frag hdr, beyond some minimal amount of
1083 * data
1084 */
1085static bool mptcp_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1086{
1087 if (likely(skb_page_frag_refill(32U + sizeof(struct mptcp_data_frag),
1088 pfrag, sk->sk_allocation)))
1089 return true;
1090
1091 mptcp_enter_memory_pressure(sk);
18b683bf
PA		 1092 return false;
1093}
1094
1095static struct mptcp_data_frag *
1096mptcp_carve_data_frag(const struct mptcp_sock *msk, struct page_frag *pfrag,
1097 int orig_offset)
1098{
1099 int offset = ALIGN(orig_offset, sizeof(long));
1100 struct mptcp_data_frag *dfrag;
1101
1102 dfrag = (struct mptcp_data_frag *)(page_to_virt(pfrag->page) + offset);
1103 dfrag->data_len = 0;
1104 dfrag->data_seq = msk->write_seq;
1105 dfrag->overhead = offset - orig_offset + sizeof(struct mptcp_data_frag);
1106 dfrag->offset = offset + sizeof(struct mptcp_data_frag);
d9ca1de8 1107 dfrag->already_sent = 0;
18b683bf
PA		 1108 dfrag->page = pfrag->page;
1109
1110 return dfrag;
1111}
1112
caf971df
PA		 1113struct mptcp_sendmsg_info {
1114 int mss_now;
1115 int size_goal;
d9ca1de8
PA		 1116 u16 limit;
1117 u16 sent;
1118 unsigned int flags;
1094c6fe 1119 bool data_lock_held;
caf971df
PA		 1120};
1121
b713d006
PA		 1122static int mptcp_check_allowed_size(const struct mptcp_sock *msk, struct sock *ssk,
1123 u64 data_seq, int avail_size)
6f8a612a
FW		 1124{
1125 u64 window_end = mptcp_wnd_end(msk);
b713d006 1126 u64 mptcp_snd_wnd;
6f8a612a
FW		 1127
1128 if (__mptcp_check_fallback(msk))
1129 return avail_size;
1130
b713d006
PA		 1131 mptcp_snd_wnd = window_end - data_seq;
1132 avail_size = min_t(unsigned int, mptcp_snd_wnd, avail_size);
6f8a612a 1133
92be2f52 1134 if (unlikely(tcp_sk(ssk)->snd_wnd < mptcp_snd_wnd)) {
b713d006 1135 tcp_sk(ssk)->snd_wnd = min_t(u64, U32_MAX, mptcp_snd_wnd);
92be2f52
PA		 1136 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_SNDWNDSHARED);
1137 }
6f8a612a
FW		 1138
1139 return avail_size;
1140}
1141
724cfd2e
PA		 1142static bool __mptcp_add_ext(struct sk_buff *skb, gfp_t gfp)
1143{
1144 struct skb_ext *mpext = __skb_ext_alloc(gfp);
1145
1146 if (!mpext)
1147 return false;
1148 __skb_ext_set(skb, SKB_EXT_MPTCP, mpext);
1149 return true;
1150}
1151
6e628cd3 1152static struct sk_buff *__mptcp_do_alloc_tx_skb(struct sock *sk, gfp_t gfp)
724cfd2e
PA		 1153{
1154 struct sk_buff *skb;
1155
6e628cd3 1156 skb = alloc_skb_fclone(MAX_TCP_HEADER, gfp);
724cfd2e 1157 if (likely(skb)) {
6e628cd3 1158 if (likely(__mptcp_add_ext(skb, gfp))) {
724cfd2e 1159 skb_reserve(skb, MAX_TCP_HEADER);
a52fe46e 1160 skb->ip_summed = CHECKSUM_PARTIAL;
f70cad10 1161 INIT_LIST_HEAD(&skb->tcp_tsorted_anchor);
724cfd2e
PA		 1162 return skb;
1163 }
1164 __kfree_skb(skb);
1165 } else {
1166 mptcp_enter_memory_pressure(sk);
1167 }
1168 return NULL;
1169}
1170
f70cad10 1171static struct sk_buff *__mptcp_alloc_tx_skb(struct sock *sk, struct sock *ssk, gfp_t gfp)
724cfd2e 1172{
724cfd2e
PA		 1173 struct sk_buff *skb;
1174
6e628cd3 1175 skb = __mptcp_do_alloc_tx_skb(sk, gfp);
724cfd2e 1176 if (!skb)
f70cad10 1177 return NULL;
724cfd2e
PA		 1178
1179 if (likely(sk_wmem_schedule(ssk, skb->truesize))) {
f70cad10
PA		 1180 tcp_skb_entail(ssk, skb);
1181 return skb;
724cfd2e 1182 }
3ef3905a 1183 tcp_skb_tsorted_anchor_cleanup(skb);
724cfd2e 1184 kfree_skb(skb);
f70cad10 1185 return NULL;
724cfd2e
PA		 1186}
1187
f70cad10 1188static struct sk_buff *mptcp_alloc_tx_skb(struct sock *sk, struct sock *ssk, bool data_lock_held)
724cfd2e 1189{
1094c6fe 1190 gfp_t gfp = data_lock_held ? GFP_ATOMIC : sk->sk_allocation;
724cfd2e 1191
1094c6fe 1192 return __mptcp_alloc_tx_skb(sk, ssk, gfp);
724cfd2e
PA		 1193}
1194
d0cc2987
GT		 1195/* note: this always recompute the csum on the whole skb, even
1196 * if we just appended a single frag. More status info needed
1197 */
1198static void mptcp_update_data_checksum(struct sk_buff *skb, int added)
1199{
1200 struct mptcp_ext *mpext = mptcp_get_ext(skb);
1201 __wsum csum = ~csum_unfold(mpext->csum);
1202 int offset = skb->len - added;
1203
1204 mpext->csum = csum_fold(csum_block_add(csum, skb_checksum(skb, offset, added, 0), offset));
1205}
1206
1e39e5a3
GT		 1207static void mptcp_update_infinite_map(struct mptcp_sock *msk,
1208 struct sock *ssk,
1209 struct mptcp_ext *mpext)
1210{
1211 if (!mpext)
1212 return;
1213
1214 mpext->infinite_map = 1;
1215 mpext->data_len = 0;
1216
104125b8 1217 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPTX);
1e39e5a3
GT		 1218 mptcp_subflow_ctx(ssk)->send_infinite_map = 0;
1219 pr_fallback(msk);
d51991e2 1220 mptcp_do_fallback(ssk);
1e39e5a3
GT		 1221}
1222
9fce92f0
PA		 1223#define MPTCP_MAX_GSO_SIZE (GSO_LEGACY_MAX_SIZE - (MAX_TCP_HEADER + 1))
1224
6d0060f6 1225static int mptcp_sendmsg_frag(struct sock *sk, struct sock *ssk,
d9ca1de8 1226 struct mptcp_data_frag *dfrag,
caf971df 1227 struct mptcp_sendmsg_info *info)
6d0060f6 1228{
d9ca1de8 1229 u64 data_seq = dfrag->data_seq + info->sent;
f70cad10 1230 int offset = dfrag->offset + info->sent;
6d0060f6 1231 struct mptcp_sock *msk = mptcp_sk(sk);
6f8a612a 1232 bool zero_window_probe = false;
6d0060f6 1233 struct mptcp_ext *mpext = NULL;
f70cad10
PA		 1234 bool can_coalesce = false;
1235 bool reuse_skb = true;
1236 struct sk_buff *skb;
1237 size_t copy;
1238 int i;
6d0060f6 1239
e4b61351 1240 pr_debug("msk=%p ssk=%p sending dfrag at seq=%llu len=%u already sent=%u",
d9ca1de8
PA		 1241 msk, ssk, dfrag->data_seq, dfrag->data_len, info->sent);
1242
f70cad10
PA		 1243 if (WARN_ON_ONCE(info->sent > info->limit ||
1244 info->limit > dfrag->data_len))
1245 return 0;
1246
c886d702
PA		 1247 if (unlikely(!__tcp_can_send(ssk)))
1248 return -EAGAIN;
1249
d9ca1de8 1250 /* compute send limit */
9fce92f0
PA		 1251 if (unlikely(ssk->sk_gso_max_size > MPTCP_MAX_GSO_SIZE))
1252 ssk->sk_gso_max_size = MPTCP_MAX_GSO_SIZE;
d9ca1de8 1253 info->mss_now = tcp_send_mss(ssk, &info->size_goal, info->flags);
f70cad10
PA		 1254 copy = info->size_goal;
1255
57040755 1256 skb = tcp_write_queue_tail(ssk);
f70cad10 1257 if (skb && copy > skb->len) {
57040755
PA		 1258 /* Limit the write to the size available in the
1259 * current skb, if any, so that we create at most a new skb.
1260 * Explicitly tells TCP internals to avoid collapsing on later
1261 * queue management operation, to avoid breaking the ext <->
1262 * SSN association set here
1263 */
a16c054b 1264 mpext = mptcp_get_ext(skb);
1094c6fe 1265 if (!mptcp_skb_can_collapse_to(data_seq, skb, mpext)) {
57040755 1266 TCP_SKB_CB(skb)->eor = 1;
b9cd26f6 1267 tcp_mark_push(tcp_sk(ssk), skb);
1094c6fe
PA		 1268 goto alloc_skb;
1269 }
1270
f70cad10
PA		 1271 i = skb_shinfo(skb)->nr_frags;
1272 can_coalesce = skb_can_coalesce(skb, i, dfrag->page, offset);
657b991a 1273 if (!can_coalesce && i >= READ_ONCE(sysctl_max_skb_frags)) {
f70cad10
PA		 1274 tcp_mark_push(tcp_sk(ssk), skb);
1275 goto alloc_skb;
15e6ca97 1276 }
18b683bf 1277
f70cad10
PA		 1278 copy -= skb->len;
1279 } else {
1094c6fe 1280alloc_skb:
f70cad10
PA		 1281 skb = mptcp_alloc_tx_skb(sk, ssk, info->data_lock_held);
1282 if (!skb)
1283 return -ENOMEM;
1284
1285 i = skb_shinfo(skb)->nr_frags;
1286 reuse_skb = false;
a16c054b 1287 mpext = mptcp_get_ext(skb);
f70cad10 1288 }
1094c6fe 1289
6f8a612a 1290 /* Zero window and all data acked? Probe. */
b713d006 1291 copy = mptcp_check_allowed_size(msk, ssk, data_seq, copy);
f70cad10 1292 if (copy == 0) {
7439d687
PA		 1293 u64 snd_una = READ_ONCE(msk->snd_una);
1294
72377ab2 1295 if (snd_una != msk->snd_nxt || tcp_write_queue_tail(ssk)) {
27728ba8 1296 tcp_remove_empty_skb(ssk);
6f8a612a 1297 return 0;
f70cad10
PA		 1298 }
1299
6f8a612a 1300 zero_window_probe = true;
7439d687 1301 data_seq = snd_una - 1;
f70cad10 1302 copy = 1;
f70cad10 1303 }
d027236c 1304
f70cad10
PA		 1305 copy = min_t(size_t, copy, info->limit - info->sent);
1306 if (!sk_wmem_schedule(ssk, copy)) {
27728ba8 1307 tcp_remove_empty_skb(ssk);
e2223995 1308 return -ENOMEM;
35759383 1309 }
18b683bf 1310
f70cad10
PA		 1311 if (can_coalesce) {
1312 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1313 } else {
1314 get_page(dfrag->page);
1315 skb_fill_page_desc(skb, i, dfrag->page, offset, copy);
57040755
PA		 1316 }
1317
f70cad10
PA		 1318 skb->len += copy;
1319 skb->data_len += copy;
1320 skb->truesize += copy;
1321 sk_wmem_queued_add(ssk, copy);
1322 sk_mem_charge(ssk, copy);
f70cad10
PA		 1323 WRITE_ONCE(tcp_sk(ssk)->write_seq, tcp_sk(ssk)->write_seq + copy);
1324 TCP_SKB_CB(skb)->end_seq += copy;
1325 tcp_skb_pcount_set(skb, 0);
1326
1327 /* on skb reuse we just need to update the DSS len */
1328 if (reuse_skb) {
1329 TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH;
1330 mpext->data_len += copy;
f70cad10 1331 goto out;
724cfd2e 1332 }
6d0060f6
MM		 1333
1334 memset(mpext, 0, sizeof(*mpext));
d9ca1de8 1335 mpext->data_seq = data_seq;
6d0060f6 1336 mpext->subflow_seq = mptcp_subflow_ctx(ssk)->rel_write_seq;
f70cad10 1337 mpext->data_len = copy;
6d0060f6
MM		 1338 mpext->use_map = 1;
1339 mpext->dsn64 = 1;
1340
1341 pr_debug("data_seq=%llu subflow_seq=%u data_len=%u dsn64=%d",
1342 mpext->data_seq, mpext->subflow_seq, mpext->data_len,
1343 mpext->dsn64);
1344
6f8a612a 1345 if (zero_window_probe) {
f70cad10 1346 mptcp_subflow_ctx(ssk)->rel_write_seq += copy;
6f8a612a 1347 mpext->frozen = 1;
d0cc2987 1348 if (READ_ONCE(msk->csum_enabled))
f70cad10 1349 mptcp_update_data_checksum(skb, copy);
6f8a612a 1350 tcp_push_pending_frames(ssk);
d0cc2987 1351 return 0;
6f8a612a 1352 }
57040755 1353out:
d0cc2987 1354 if (READ_ONCE(msk->csum_enabled))
f70cad10 1355 mptcp_update_data_checksum(skb, copy);
1e39e5a3
GT		 1356 if (mptcp_subflow_ctx(ssk)->send_infinite_map)
1357 mptcp_update_infinite_map(msk, ssk, mpext);
0eb4e7ee 1358 trace_mptcp_sendmsg_frag(mpext);
f70cad10
PA		 1359 mptcp_subflow_ctx(ssk)->rel_write_seq += copy;
1360 return copy;
6d0060f6
MM		 1361}
1362
d5f49190
PA		 1363#define MPTCP_SEND_BURST_SIZE ((1 << 16) - \
1364 sizeof(struct tcphdr) - \
1365 MAX_TCP_OPTION_SPACE - \
1366 sizeof(struct ipv6hdr) - \
1367 sizeof(struct frag_hdr))
1368
1369struct subflow_send_info {
1370 struct sock *ssk;
3ce0852c 1371 u64 linger_time;
d5f49190
PA		 1372};
1373
ff5a0b42
PA		 1374void mptcp_subflow_set_active(struct mptcp_subflow_context *subflow)
1375{
1376 if (!subflow->stale)
1377 return;
1378
1379 subflow->stale = 0;
fc1b4e3b 1380 MPTCP_INC_STATS(sock_net(mptcp_subflow_tcp_sock(subflow)), MPTCP_MIB_SUBFLOWRECOVER);
ff5a0b42
PA		 1381}
1382
1383bool mptcp_subflow_active(struct mptcp_subflow_context *subflow)
1384{
1385 if (unlikely(subflow->stale)) {
1386 u32 rcv_tstamp = READ_ONCE(tcp_sk(mptcp_subflow_tcp_sock(subflow))->rcv_tstamp);
1387
1388 if (subflow->stale_rcv_tstamp == rcv_tstamp)
1389 return false;
1390
1391 mptcp_subflow_set_active(subflow);
1392 }
1393 return __mptcp_subflow_active(subflow);
1394}
1395
3ce0852c
PA		 1396#define SSK_MODE_ACTIVE 0
1397#define SSK_MODE_BACKUP 1
1398#define SSK_MODE_MAX 2
1399
33d41c9c
PA		 1400/* implement the mptcp packet scheduler;
1401 * returns the subflow that will transmit the next DSS
1402 * additionally updates the rtx timeout
1403 */
07336a87 1404struct sock *mptcp_subflow_get_send(struct mptcp_sock *msk)
f296234c 1405{
3ce0852c 1406 struct subflow_send_info send_info[SSK_MODE_MAX];
f296234c 1407 struct mptcp_subflow_context *subflow;
33d41c9c 1408 struct sock *sk = (struct sock *)msk;
3ce0852c 1409 u32 pace, burst, wmem;
d5f49190
PA		 1410 int i, nr_active = 0;
1411 struct sock *ssk;
3ce0852c 1412 u64 linger_time;
33d41c9c 1413 long tout = 0;
f296234c 1414
d5f49190 1415 /* pick the subflow with the lower wmem/wspace ratio */
3ce0852c 1416 for (i = 0; i < SSK_MODE_MAX; ++i) {
d5f49190 1417 send_info[i].ssk = NULL;
3ce0852c 1418 send_info[i].linger_time = -1;
d5f49190 1419 }
3ce0852c 1420
d5f49190 1421 mptcp_for_each_subflow(msk, subflow) {
e10a9892 1422 trace_mptcp_subflow_get_send(subflow);
d5f49190
PA		 1423 ssk = mptcp_subflow_tcp_sock(subflow);
1424 if (!mptcp_subflow_active(subflow))
1425 continue;
1426
33d41c9c 1427 tout = max(tout, mptcp_timeout_from_subflow(subflow));
d5f49190 1428 nr_active += !subflow->backup;
3ce0852c
PA		 1429 pace = subflow->avg_pacing_rate;
1430 if (unlikely(!pace)) {
1431 /* init pacing rate from socket */
1432 subflow->avg_pacing_rate = READ_ONCE(ssk->sk_pacing_rate);
1433 pace = subflow->avg_pacing_rate;
1434 if (!pace)
1435 continue;
1436 }
f296234c 1437
3ce0852c
PA		 1438 linger_time = div_u64((u64)READ_ONCE(ssk->sk_wmem_queued) << 32, pace);
1439 if (linger_time < send_info[subflow->backup].linger_time) {
d5f49190 1440 send_info[subflow->backup].ssk = ssk;
3ce0852c 1441 send_info[subflow->backup].linger_time = linger_time;
d5f49190 1442 }
f296234c 1443 }
33d41c9c 1444 __mptcp_set_timeout(sk, tout);
f296234c 1445
d5f49190
PA		 1446 /* pick the best backup if no other subflow is active */
1447 if (!nr_active)
3ce0852c
PA		 1448 send_info[SSK_MODE_ACTIVE].ssk = send_info[SSK_MODE_BACKUP].ssk;
1449
1450 /* According to the blest algorithm, to avoid HoL blocking for the
1451 * faster flow, we need to:
1452 * - estimate the faster flow linger time
1453 * - use the above to estimate the amount of byte transferred
1454 * by the faster flow
1455 * - check that the amount of queued data is greter than the above,
1456 * otherwise do not use the picked, slower, subflow
1457 * We select the subflow with the shorter estimated time to flush
1458 * the queued mem, which basically ensure the above. We just need
1459 * to check that subflow has a non empty cwin.
1460 */
1461 ssk = send_info[SSK_MODE_ACTIVE].ssk;
b713d006 1462 if (!ssk || !sk_stream_memory_free(ssk))
3ce0852c 1463 return NULL;
5cf92bba 1464
b713d006 1465 burst = min_t(int, MPTCP_SEND_BURST_SIZE, mptcp_wnd_end(msk) - msk->snd_nxt);
3ce0852c 1466 wmem = READ_ONCE(ssk->sk_wmem_queued);
ebc1e08f 1467 if (!burst)
b713d006 1468 return ssk;
b713d006 1469
3ce0852c
PA		 1470 subflow = mptcp_subflow_ctx(ssk);
1471 subflow->avg_pacing_rate = div_u64((u64)subflow->avg_pacing_rate * wmem +
1472 READ_ONCE(ssk->sk_pacing_rate) * burst,
1473 burst + wmem);
3ce0852c
PA		 1474 msk->snd_burst = burst;
1475 return ssk;
f296234c
PK		 1476}
1477
b8e0def3 1478static void mptcp_push_release(struct sock *ssk, struct mptcp_sendmsg_info *info)
d9ca1de8 1479{
d9ca1de8
PA		 1480 tcp_push(ssk, 0, info->mss_now, tcp_sk(ssk)->nonagle, info->size_goal);
1481 release_sock(ssk);
1482}
1483
0d199e43
FW		 1484static void mptcp_update_post_push(struct mptcp_sock *msk,
1485 struct mptcp_data_frag *dfrag,
1486 u32 sent)
1487{
1488 u64 snd_nxt_new = dfrag->data_seq;
1489
1490 dfrag->already_sent += sent;
1491
1492 msk->snd_burst -= sent;
0d199e43
FW		 1493
1494 snd_nxt_new += dfrag->already_sent;
1495
1496 /* snd_nxt_new can be smaller than snd_nxt in case mptcp
1497 * is recovering after a failover. In that event, this re-sends
1498 * old segments.
1499 *
1500 * Thus compute snd_nxt_new candidate based on
1501 * the dfrag->data_seq that was sent and the data
1502 * that has been handed to the subflow for transmission
1503 * and skip update in case it was old dfrag.
1504 */
38967f42
PA		 1505 if (likely(after64(snd_nxt_new, msk->snd_nxt))) {
1506 msk->bytes_sent += snd_nxt_new - msk->snd_nxt;
d440a4e2 1507 WRITE_ONCE(msk->snd_nxt, snd_nxt_new);
38967f42 1508 }
0d199e43
FW		 1509}
1510
8b38217a 1511void mptcp_check_and_set_pending(struct sock *sk)
3241a9c0 1512{
bdd70eb6
PA		 1513 if (mptcp_send_head(sk)) {
1514 mptcp_data_lock(sk);
1515 mptcp_sk(sk)->cb_flags |= BIT(MPTCP_PUSH_PENDING);
1516 mptcp_data_unlock(sk);
1517 }
3241a9c0
FW		 1518}
1519
c5b4297d
GT		 1520static int __subflow_push_pending(struct sock *sk, struct sock *ssk,
1521 struct mptcp_sendmsg_info *info)
f870fa0b
MM		 1522{
1523 struct mptcp_sock *msk = mptcp_sk(sk);
d9ca1de8 1524 struct mptcp_data_frag *dfrag;
c5b4297d 1525 int len, copied = 0, err = 0;
d9ca1de8
PA		 1526
1527 while ((dfrag = mptcp_send_head(sk))) {
c5b4297d
GT		 1528 info->sent = dfrag->already_sent;
1529 info->limit = dfrag->data_len;
d9ca1de8
PA		 1530 len = dfrag->data_len - dfrag->already_sent;
1531 while (len > 0) {
1532 int ret = 0;
1533
c5b4297d 1534 ret = mptcp_sendmsg_frag(sk, ssk, dfrag, info);
d9ca1de8 1535 if (ret <= 0) {
c5b4297d 1536 err = copied ? : ret;
d9ca1de8
PA		 1537 goto out;
1538 }
1539
c5b4297d
GT		 1540 info->sent += ret;
1541 copied += ret;
d9ca1de8 1542 len -= ret;
0d199e43
FW		 1543
1544 mptcp_update_post_push(msk, dfrag, ret);
d9ca1de8
PA		 1545 }
1546 WRITE_ONCE(msk->first_pending, mptcp_send_next(sk));
c5b4297d
GT		 1547
1548 if (msk->snd_burst <= 0 ||
1549 !sk_stream_memory_free(ssk) ||
1550 !mptcp_subflow_active(mptcp_subflow_ctx(ssk))) {
1551 err = copied;
1552 goto out;
1553 }
1554 mptcp_set_timeout(sk);
1555 }
1556 err = copied;
1557
1558out:
1559 return err;
1560}
1561
1562void __mptcp_push_pending(struct sock *sk, unsigned int flags)
1563{
1564 struct sock *prev_ssk = NULL, *ssk = NULL;
1565 struct mptcp_sock *msk = mptcp_sk(sk);
1566 struct mptcp_sendmsg_info info = {
1567 .flags = flags,
1568 };
1569 bool do_check_data_fin = false;
0fa1b378 1570 int push_count = 1;
c5b4297d 1571
0fa1b378
GT		 1572 while (mptcp_send_head(sk) && (push_count > 0)) {
1573 struct mptcp_subflow_context *subflow;
c5b4297d
GT		 1574 int ret = 0;
1575
0fa1b378
GT		 1576 if (mptcp_sched_get_send(msk))
1577 break;
c5b4297d 1578
0fa1b378
GT		 1579 push_count = 0;
1580
1581 mptcp_for_each_subflow(msk, subflow) {
1582 if (READ_ONCE(subflow->scheduled)) {
1583 mptcp_subflow_set_scheduled(subflow, false);
1584
1585 prev_ssk = ssk;
1586 ssk = mptcp_subflow_tcp_sock(subflow);
1587 if (ssk != prev_ssk) {
1588 /* First check. If the ssk has changed since
1589 * the last round, release prev_ssk
1590 */
1591 if (prev_ssk)
1592 mptcp_push_release(prev_ssk, &info);
1593
1594 /* Need to lock the new subflow only if different
1595 * from the previous one, otherwise we are still
1596 * helding the relevant lock
1597 */
1598 lock_sock(ssk);
1599 }
1600
1601 push_count++;
1602
1603 ret = __subflow_push_pending(sk, ssk, &info);
1604 if (ret <= 0) {
1605 if (ret != -EAGAIN ||
1606 (1 << ssk->sk_state) &
1607 (TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2 | TCPF_CLOSE))
1608 push_count--;
1609 continue;
1610 }
1611 do_check_data_fin = true;
1612 }
c5b4297d 1613 }
d9ca1de8
PA		 1614 }
1615
1616 /* at this point we held the socket lock for the last subflow we used */
1617 if (ssk)
b8e0def3 1618 mptcp_push_release(ssk, &info);
d9ca1de8 1619
33d41c9c 1620 /* ensure the rtx timer is running */
f6909dc1
PA		 1621 if (!mptcp_rtx_timer_pending(sk))
1622 mptcp_reset_rtx_timer(sk);
0522b424 1623 if (do_check_data_fin)
81c1d029 1624 mptcp_check_send_data_fin(sk);
d9ca1de8
PA		 1625}
1626
73a0052a 1627static void __mptcp_subflow_push_pending(struct sock *sk, struct sock *ssk, bool first)
6e628cd3
PA		 1628{
1629 struct mptcp_sock *msk = mptcp_sk(sk);
1094c6fe
PA		 1630 struct mptcp_sendmsg_info info = {
1631 .data_lock_held = true,
1632 };
0fa1b378 1633 bool keep_pushing = true;
b19bc294 1634 struct sock *xmit_ssk;
c5b4297d 1635 int copied = 0;
6e628cd3
PA		 1636
1637 info.flags = 0;
0fa1b378
GT		 1638 while (mptcp_send_head(sk) && keep_pushing) {
1639 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
c5b4297d 1640 int ret = 0;
b19bc294 1641
c5b4297d
GT		 1642 /* check for a different subflow usage only after
1643 * spooling the first chunk of data
1644 */
0fa1b378
GT		 1645 if (first) {
1646 mptcp_subflow_set_scheduled(subflow, false);
1647 ret = __subflow_push_pending(sk, ssk, &info);
1648 first = false;
1649 if (ret <= 0)
1650 break;
1651 copied += ret;
1652 continue;
1653 }
1654
1655 if (mptcp_sched_get_send(msk))
c5b4297d 1656 goto out;
0fa1b378
GT		 1657
1658 if (READ_ONCE(subflow->scheduled)) {
1659 mptcp_subflow_set_scheduled(subflow, false);
1660 ret = __subflow_push_pending(sk, ssk, &info);
1661 if (ret <= 0)
1662 keep_pushing = false;
1663 copied += ret;
6e628cd3 1664 }
c5b4297d 1665
0fa1b378
GT		 1666 mptcp_for_each_subflow(msk, subflow) {
1667 if (READ_ONCE(subflow->scheduled)) {
1668 xmit_ssk = mptcp_subflow_tcp_sock(subflow);
1669 if (xmit_ssk != ssk) {
1670 mptcp_subflow_delegate(subflow,
1671 MPTCP_DELEGATE_SEND);
1672 keep_pushing = false;
1673 }
1674 }
1675 }
6e628cd3
PA		 1676 }
1677
1678out:
1679 /* __mptcp_alloc_tx_skb could have released some wmem and we are
1680 * not going to flush it via release_sock()
1681 */
6e628cd3 1682 if (copied) {
6e628cd3
PA		 1683 tcp_push(ssk, 0, info.mss_now, tcp_sk(ssk)->nonagle,
1684 info.size_goal);
f6909dc1
PA		 1685 if (!mptcp_rtx_timer_pending(sk))
1686 mptcp_reset_rtx_timer(sk);
d09d818e 1687
6e628cd3
PA		 1688 if (msk->snd_data_fin_enable &&
1689 msk->snd_nxt + 1 == msk->write_seq)
1690 mptcp_schedule_work(sk);
1691 }
1692}
1693
7d803344
PA		 1694static int mptcp_disconnect(struct sock *sk, int flags);
1695
a2702a07 1696static int mptcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg,
fa9e5746
PA		 1697 size_t len, int *copied_syn)
1698{
1699 unsigned int saved_flags = msg->msg_flags;
1700 struct mptcp_sock *msk = mptcp_sk(sk);
a2702a07 1701 struct sock *ssk;
fa9e5746
PA		 1702 int ret;
1703
a2702a07
PA		 1704 /* on flags based fastopen the mptcp is supposed to create the
1705 * first subflow right now. Otherwise we are in the defer_connect
1706 * path, and the first subflow must be already present.
1707 * Since the defer_connect flag is cleared after the first succsful
1708 * fastopen attempt, no need to check for additional subflow status.
1709 */
ddb1a072 1710 if (msg->msg_flags & MSG_FASTOPEN) {
3f326a82
PA		 1711 ssk = __mptcp_nmpc_sk(msk);
1712 if (IS_ERR(ssk))
1713 return PTR_ERR(ssk);
ddb1a072 1714 }
a2702a07
PA		 1715 if (!msk->first)
1716 return -EINVAL;
1717
1718 ssk = msk->first;
1719
fa9e5746
PA		 1720 lock_sock(ssk);
1721 msg->msg_flags |= MSG_DONTWAIT;
7d803344 1722 msk->fastopening = 1;
fa9e5746 1723 ret = tcp_sendmsg_fastopen(ssk, msg, copied_syn, len, NULL);
7d803344 1724 msk->fastopening = 0;
fa9e5746
PA		 1725 msg->msg_flags = saved_flags;
1726 release_sock(ssk);
1727
1728 /* do the blocking bits of inet_stream_connect outside the ssk socket lock */
1729 if (ret == -EINPROGRESS && !(msg->msg_flags & MSG_DONTWAIT)) {
1730 ret = __inet_stream_connect(sk->sk_socket, msg->msg_name,
1731 msg->msg_namelen, msg->msg_flags, 1);
1732
1733 /* Keep the same behaviour of plain TCP: zero the copied bytes in
1734 * case of any error, except timeout or signal
1735 */
1736 if (ret && ret != -EINPROGRESS && ret != -ERESTARTSYS && ret != -EINTR)
1737 *copied_syn = 0;
7d803344 1738 } else if (ret && ret != -EINPROGRESS) {
c2b2ae39
PA		 1739 /* The disconnect() op called by tcp_sendmsg_fastopen()/
1740 * __inet_stream_connect() can fail, due to looking check,
1741 * see mptcp_disconnect().
1742 * Attempt it again outside the problematic scope.
1743 */
1744 if (!mptcp_disconnect(sk, 0))
1745 sk->sk_socket->state = SS_UNCONNECTED;
fa9e5746 1746 }
08e39c0d 1747 inet_clear_bit(DEFER_CONNECT, sk);
fa9e5746
PA		 1748
1749 return ret;
1750}
1751
0ffe8e74
PA		 1752static int do_copy_data_nocache(struct sock *sk, int copy,
1753 struct iov_iter *from, char *to)
1754{
1755 if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1756 if (!copy_from_iter_full_nocache(to, copy, from))
1757 return -EFAULT;
1758 } else if (!copy_from_iter_full(to, copy, from)) {
1759 return -EFAULT;
1760 }
1761 return 0;
1762}
1763
29b5e5ef
PA		 1764/* open-code sk_stream_memory_free() plus sent limit computation to
1765 * avoid indirect calls in fast-path.
1766 * Called under the msk socket lock, so we can avoid a bunch of ONCE
1767 * annotations.
1768 */
1769static u32 mptcp_send_limit(const struct sock *sk)
1770{
1771 const struct mptcp_sock *msk = mptcp_sk(sk);
1772 u32 limit, not_sent;
1773
1774 if (sk->sk_wmem_queued >= READ_ONCE(sk->sk_sndbuf))
1775 return 0;
1776
1777 limit = mptcp_notsent_lowat(sk);
1778 if (limit == UINT_MAX)
1779 return UINT_MAX;
1780
1781 not_sent = msk->write_seq - msk->snd_nxt;
1782 if (not_sent >= limit)
1783 return 0;
1784
1785 return limit - not_sent;
1786}
1787
d9ca1de8
PA		 1788static int mptcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1789{
1790 struct mptcp_sock *msk = mptcp_sk(sk);
17091708 1791 struct page_frag *pfrag;
6d0060f6 1792 size_t copied = 0;
caf971df 1793 int ret = 0;
6d0060f6 1794 long timeo;
f870fa0b 1795
987858e5 1796 /* silently ignore everything else */
1e777f39 1797 msg->msg_flags &= MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL | MSG_FASTOPEN;
987858e5 1798
6511882c 1799 lock_sock(sk);
1954b860 1800
08e39c0d
ED		 1801 if (unlikely(inet_test_bit(DEFER_CONNECT, sk) ||
1802 msg->msg_flags & MSG_FASTOPEN)) {
d98a82a6
DS		 1803 int copied_syn = 0;
1804
a2702a07 1805 ret = mptcp_sendmsg_fastopen(sk, msg, len, &copied_syn);
d98a82a6 1806 copied += copied_syn;
fa9e5746 1807 if (ret == -EINPROGRESS && copied_syn > 0)
d98a82a6 1808 goto out;
fa9e5746 1809 else if (ret)
69800e51 1810 goto do_error;
d98a82a6
DS		 1811 }
1812
1954b860
MM		 1813 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1814
1815 if ((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) {
1816 ret = sk_stream_wait_connect(sk, &timeo);
1817 if (ret)
69800e51 1818 goto do_error;
1954b860
MM		 1819 }
1820
69800e51
PA		 1821 ret = -EPIPE;
1822 if (unlikely(sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN)))
1823 goto do_error;
1824
17091708 1825 pfrag = sk_page_frag(sk);
18b683bf 1826
d9ca1de8 1827 while (msg_data_left(msg)) {
724cfd2e 1828 int total_ts, frag_truesize = 0;
d9ca1de8 1829 struct mptcp_data_frag *dfrag;
d9ca1de8
PA		 1830 bool dfrag_collapsed;
1831 size_t psize, offset;
29b5e5ef
PA		 1832 u32 copy_limit;
1833
1834 /* ensure fitting the notsent_lowat() constraint */
1835 copy_limit = mptcp_send_limit(sk);
1836 if (!copy_limit)
1837 goto wait_for_memory;
18b683bf 1838
d9ca1de8
PA		 1839 /* reuse tail pfrag, if possible, or carve a new one from the
1840 * page allocator
1841 */
1842 dfrag = mptcp_pending_tail(sk);
1843 dfrag_collapsed = mptcp_frag_can_collapse_to(msk, pfrag, dfrag);
1844 if (!dfrag_collapsed) {
d9ca1de8
PA		 1845 if (!mptcp_page_frag_refill(sk, pfrag))
1846 goto wait_for_memory;
1847
1848 dfrag = mptcp_carve_data_frag(msk, pfrag, pfrag->offset);
1849 frag_truesize = dfrag->overhead;
72511aab 1850 }
6d0060f6 1851
d9ca1de8
PA		 1852 /* we do not bound vs wspace, to allow a single packet.
1853 * memory accounting will prevent execessive memory usage
1854 * anyway
d5f49190 1855 */
d9ca1de8
PA		 1856 offset = dfrag->offset + dfrag->data_len;
1857 psize = pfrag->size - offset;
1858 psize = min_t(size_t, psize, msg_data_left(msg));
29b5e5ef 1859 psize = min_t(size_t, psize, copy_limit);
724cfd2e 1860 total_ts = psize + frag_truesize;
d9ca1de8 1861
6511882c 1862 if (!sk_wmem_schedule(sk, total_ts))
724cfd2e 1863 goto wait_for_memory;
724cfd2e 1864
0ffe8e74
PA		 1865 ret = do_copy_data_nocache(sk, psize, &msg->msg_iter,
1866 page_address(dfrag->page) + offset);
1867 if (ret)
69800e51 1868 goto do_error;
72511aab 1869
d9ca1de8 1870 /* data successfully copied into the write queue */
5e6300e7 1871 sk_forward_alloc_add(sk, -total_ts);
d9ca1de8
PA		 1872 copied += psize;
1873 dfrag->data_len += psize;
1874 frag_truesize += psize;
1875 pfrag->offset += frag_truesize;
1876 WRITE_ONCE(msk->write_seq, msk->write_seq + psize);
1877
1878 /* charge data on mptcp pending queue to the msk socket
1879 * Note: we charge such data both to sk and ssk
fb529e62 1880 */
d9ca1de8 1881 sk_wmem_queued_add(sk, frag_truesize);
d9ca1de8
PA		 1882 if (!dfrag_collapsed) {
1883 get_page(dfrag->page);
1884 list_add_tail(&dfrag->list, &msk->rtx_queue);
1885 if (!msk->first_pending)
1886 WRITE_ONCE(msk->first_pending, dfrag);
fb529e62 1887 }
e4b61351 1888 pr_debug("msk=%p dfrag at seq=%llu len=%u sent=%u new=%d", msk,
d9ca1de8
PA		 1889 dfrag->data_seq, dfrag->data_len, dfrag->already_sent,
1890 !dfrag_collapsed);
6d0060f6 1891
d9ca1de8 1892 continue;
b51f9b80 1893
d9ca1de8 1894wait_for_memory:
037db6ea 1895 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
c2e6048f 1896 __mptcp_push_pending(sk, msg->msg_flags);
d9ca1de8
PA		 1897 ret = sk_stream_wait_memory(sk, &timeo);
1898 if (ret)
69800e51 1899 goto do_error;
57040755 1900 }
6d0060f6 1901
13e16037 1902 if (copied)
c2e6048f 1903 __mptcp_push_pending(sk, msg->msg_flags);
d9ca1de8 1904
1954b860 1905out:
cec37a6e 1906 release_sock(sk);
69800e51
PA		 1907 return copied;
1908
1909do_error:
1910 if (copied)
1911 goto out;
1912
1913 copied = sk_stream_error(sk, msg->msg_flags, ret);
1914 goto out;
f870fa0b
MM		 1915}
1916
6771bfd9
FW		 1917static int __mptcp_recvmsg_mskq(struct mptcp_sock *msk,
1918 struct msghdr *msg,
b7f653b2
FW		 1919 size_t len, int flags,
1920 struct scm_timestamping_internal *tss,
1921 int *cmsg_flags)
6771bfd9 1922{
ca4fb892 1923 struct sk_buff *skb, *tmp;
6771bfd9
FW		 1924 int copied = 0;
1925
ca4fb892 1926 skb_queue_walk_safe(&msk->receive_queue, skb, tmp) {
6771bfd9
FW		 1927 u32 offset = MPTCP_SKB_CB(skb)->offset;
1928 u32 data_len = skb->len - offset;
1929 u32 count = min_t(size_t, len - copied, data_len);
1930 int err;
1931
d976092c
PA		 1932 if (!(flags & MSG_TRUNC)) {
1933 err = skb_copy_datagram_msg(skb, offset, msg, count);
1934 if (unlikely(err < 0)) {
1935 if (!copied)
1936 return err;
1937 break;
1938 }
6771bfd9
FW		 1939 }
1940
b7f653b2
FW		 1941 if (MPTCP_SKB_CB(skb)->has_rxtstamp) {
1942 tcp_update_recv_tstamps(skb, tss);
1943 *cmsg_flags |= MPTCP_CMSG_TS;
1944 }
1945
6771bfd9
FW		 1946 copied += count;
1947
1948 if (count < data_len) {
2c9e7765 1949 if (!(flags & MSG_PEEK)) {
ca4fb892 1950 MPTCP_SKB_CB(skb)->offset += count;
2c9e7765 1951 MPTCP_SKB_CB(skb)->map_seq += count;
5684ab1a 1952 msk->bytes_consumed += count;
2c9e7765 1953 }
6771bfd9
FW		 1954 break;
1955 }
1956
ca4fb892
YL		 1957 if (!(flags & MSG_PEEK)) {
1958 /* we will bulk release the skb memory later */
1959 skb->destructor = NULL;
ce599c51 1960 WRITE_ONCE(msk->rmem_released, msk->rmem_released + skb->truesize);
ca4fb892
YL		 1961 __skb_unlink(skb, &msk->receive_queue);
1962 __kfree_skb(skb);
5684ab1a 1963 msk->bytes_consumed += count;
ca4fb892 1964 }
6771bfd9
FW		 1965
1966 if (copied >= len)
1967 break;
1968 }
1969
1970 return copied;
1971}
1972
a6b118fe
FW		 1973/* receive buffer autotuning. See tcp_rcv_space_adjust for more information.
1974 *
1975 * Only difference: Use highest rtt estimate of the subflows in use.
1976 */
1977static void mptcp_rcv_space_adjust(struct mptcp_sock *msk, int copied)
1978{
1979 struct mptcp_subflow_context *subflow;
1980 struct sock *sk = (struct sock *)msk;
b8dc6d6c 1981 u8 scaling_ratio = U8_MAX;
a6b118fe
FW		 1982 u32 time, advmss = 1;
1983 u64 rtt_us, mstamp;
1984
109cdeb8 1985 msk_owned_by_me(msk);
a6b118fe
FW		 1986
1987 if (copied <= 0)
1988 return;
1989
013e3179
PA		 1990 if (!msk->rcvspace_init)
1991 mptcp_rcv_space_init(msk, msk->first);
1992
a6b118fe
FW		 1993 msk->rcvq_space.copied += copied;
1994
1995 mstamp = div_u64(tcp_clock_ns(), NSEC_PER_USEC);
1996 time = tcp_stamp_us_delta(mstamp, msk->rcvq_space.time);
1997
1998 rtt_us = msk->rcvq_space.rtt_us;
1999 if (rtt_us && time < (rtt_us >> 3))
2000 return;
2001
2002 rtt_us = 0;
2003 mptcp_for_each_subflow(msk, subflow) {
2004 const struct tcp_sock *tp;
2005 u64 sf_rtt_us;
2006 u32 sf_advmss;
2007
2008 tp = tcp_sk(mptcp_subflow_tcp_sock(subflow));
2009
2010 sf_rtt_us = READ_ONCE(tp->rcv_rtt_est.rtt_us);
2011 sf_advmss = READ_ONCE(tp->advmss);
2012
2013 rtt_us = max(sf_rtt_us, rtt_us);
2014 advmss = max(sf_advmss, advmss);
b8dc6d6c 2015 scaling_ratio = min(tp->scaling_ratio, scaling_ratio);
a6b118fe
FW		 2016 }
2017
2018 msk->rcvq_space.rtt_us = rtt_us;
b8dc6d6c 2019 msk->scaling_ratio = scaling_ratio;
a6b118fe
FW		 2020 if (time < (rtt_us >> 3) || rtt_us == 0)
2021 return;
2022
2023 if (msk->rcvq_space.copied <= msk->rcvq_space.space)
2024 goto new_measure;
2025
78047648 2026 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_moderate_rcvbuf) &&
a6b118fe 2027 !(sk->sk_userlocks & SOCK_RCVBUF_LOCK)) {
a6b118fe 2028 u64 rcvwin, grow;
b8dc6d6c 2029 int rcvbuf;
a6b118fe
FW		 2030
2031 rcvwin = ((u64)msk->rcvq_space.copied << 1) + 16 * advmss;
2032
2033 grow = rcvwin * (msk->rcvq_space.copied - msk->rcvq_space.space);
2034
2035 do_div(grow, msk->rcvq_space.space);
2036 rcvwin += (grow << 1);
2037
b8dc6d6c 2038 rcvbuf = min_t(u64, __tcp_space_from_win(scaling_ratio, rcvwin),
02739545 2039 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_rmem[2]));
a6b118fe
FW		 2040
2041 if (rcvbuf > sk->sk_rcvbuf) {
2042 u32 window_clamp;
2043
b8dc6d6c 2044 window_clamp = __tcp_win_from_space(scaling_ratio, rcvbuf);
a6b118fe
FW		 2045 WRITE_ONCE(sk->sk_rcvbuf, rcvbuf);
2046
2047 /* Make subflows follow along. If we do not do this, we
2048 * get drops at subflow level if skbs can't be moved to
2049 * the mptcp rx queue fast enough (announced rcv_win can
2050 * exceed ssk->sk_rcvbuf).
2051 */
2052 mptcp_for_each_subflow(msk, subflow) {
2053 struct sock *ssk;
c76c6956 2054 bool slow;
a6b118fe
FW		 2055
2056 ssk = mptcp_subflow_tcp_sock(subflow);
c76c6956 2057 slow = lock_sock_fast(ssk);
a6b118fe
FW		 2058 WRITE_ONCE(ssk->sk_rcvbuf, rcvbuf);
2059 tcp_sk(ssk)->window_clamp = window_clamp;
c76c6956
PA		 2060 tcp_cleanup_rbuf(ssk, 1);
2061 unlock_sock_fast(ssk, slow);
a6b118fe
FW		 2062 }
2063 }
2064 }
2065
2066 msk->rcvq_space.space = msk->rcvq_space.copied;
2067new_measure:
2068 msk->rcvq_space.copied = 0;
2069 msk->rcvq_space.time = mstamp;
2070}
2071
87952603
PA		 2072static void __mptcp_update_rmem(struct sock *sk)
2073{
2074 struct mptcp_sock *msk = mptcp_sk(sk);
2075
2076 if (!msk->rmem_released)
2077 return;
2078
2079 atomic_sub(msk->rmem_released, &sk->sk_rmem_alloc);
6511882c 2080 mptcp_rmem_uncharge(sk, msk->rmem_released);
ce599c51 2081 WRITE_ONCE(msk->rmem_released, 0);
87952603
PA		 2082}
2083
2084static void __mptcp_splice_receive_queue(struct sock *sk)
2085{
2086 struct mptcp_sock *msk = mptcp_sk(sk);
2087
2088 skb_queue_splice_tail_init(&sk->sk_receive_queue, &msk->receive_queue);
2089}
2090
e3859603 2091static bool __mptcp_move_skbs(struct mptcp_sock *msk)
6771bfd9 2092{
87952603 2093 struct sock *sk = (struct sock *)msk;
6771bfd9 2094 unsigned int moved = 0;
87952603 2095 bool ret, done;
d5f49190 2096
6771bfd9
FW		 2097 do {
2098 struct sock *ssk = mptcp_subflow_recv_lookup(msk);
65f49fe7 2099 bool slowpath;
6771bfd9 2100
87952603
PA		 2101 /* we can have data pending in the subflows only if the msk
2102 * receive buffer was full at subflow_data_ready() time,
2103 * that is an unlikely slow path.
2104 */
2105 if (likely(!ssk))
6771bfd9
FW		 2106 break;
2107
65f49fe7 2108 slowpath = lock_sock_fast(ssk);
87952603 2109 mptcp_data_lock(sk);
e3859603 2110 __mptcp_update_rmem(sk);
6771bfd9 2111 done = __mptcp_move_skbs_from_subflow(msk, ssk, &moved);
87952603 2112 mptcp_data_unlock(sk);
499ada50
PA		 2113
2114 if (unlikely(ssk->sk_err))
2115 __mptcp_error_report(sk);
65f49fe7 2116 unlock_sock_fast(ssk, slowpath);
6771bfd9
FW		 2117 } while (!done);
2118
87952603
PA		 2119 /* acquire the data lock only if some input data is pending */
2120 ret = moved > 0;
2121 if (!RB_EMPTY_ROOT(&msk->out_of_order_queue) ||
2122 !skb_queue_empty_lockless(&sk->sk_receive_queue)) {
2123 mptcp_data_lock(sk);
2124 __mptcp_update_rmem(sk);
2125 ret |= __mptcp_ofo_queue(msk);
2126 __mptcp_splice_receive_queue(sk);
2127 mptcp_data_unlock(sk);
ab174ad8 2128 }
87952603
PA		 2129 if (ret)
2130 mptcp_check_data_fin((struct sock *)msk);
2131 return !skb_queue_empty(&msk->receive_queue);
6771bfd9
FW		 2132}
2133
2c9e7765
FW		 2134static unsigned int mptcp_inq_hint(const struct sock *sk)
2135{
2136 const struct mptcp_sock *msk = mptcp_sk(sk);
2137 const struct sk_buff *skb;
2138
2139 skb = skb_peek(&msk->receive_queue);
2140 if (skb) {
9426ce47 2141 u64 hint_val = READ_ONCE(msk->ack_seq) - MPTCP_SKB_CB(skb)->map_seq;
2c9e7765
FW		 2142
2143 if (hint_val >= INT_MAX)
2144 return INT_MAX;
2145
2146 return (unsigned int)hint_val;
2147 }
2148
2149 if (sk->sk_state == TCP_CLOSE || (sk->sk_shutdown & RCV_SHUTDOWN))
2150 return 1;
2151
2152 return 0;
2153}
2154
f870fa0b 2155static int mptcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
ec095263 2156 int flags, int *addr_len)
f870fa0b
MM		 2157{
2158 struct mptcp_sock *msk = mptcp_sk(sk);
b7f653b2
FW		 2159 struct scm_timestamping_internal tss;
2160 int copied = 0, cmsg_flags = 0;
7a6a6cbc
PA		 2161 int target;
2162 long timeo;
f870fa0b 2163
cb9d80f4
PA		 2164 /* MSG_ERRQUEUE is really a no-op till we support IP_RECVERR */
2165 if (unlikely(flags & MSG_ERRQUEUE))
2166 return inet_recv_error(sk, msg, len, addr_len);
2167
6511882c 2168 lock_sock(sk);
fd897679
PA		 2169 if (unlikely(sk->sk_state == TCP_LISTEN)) {
2170 copied = -ENOTCONN;
2171 goto out_err;
2172 }
2173
ec095263 2174 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
7a6a6cbc
PA		 2175
2176 len = min_t(size_t, len, INT_MAX);
2177 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
2178
2c9e7765
FW		 2179 if (unlikely(msk->recvmsg_inq))
2180 cmsg_flags = MPTCP_CMSG_INQ;
2181
05e3ecea 2182 while (copied < len) {
e3859603 2183 int bytes_read;
7a6a6cbc 2184
b7f653b2 2185 bytes_read = __mptcp_recvmsg_mskq(msk, msg, len - copied, flags, &tss, &cmsg_flags);
6771bfd9
FW		 2186 if (unlikely(bytes_read < 0)) {
2187 if (!copied)
2188 copied = bytes_read;
2189 goto out_err;
2190 }
7a6a6cbc 2191
6771bfd9 2192 copied += bytes_read;
7a6a6cbc 2193
ea4ca586 2194 /* be sure to advertise window change */
e3859603
PA		 2195 mptcp_cleanup_rbuf(msk);
2196
2197 if (skb_queue_empty(&msk->receive_queue) && __mptcp_move_skbs(msk))
2198 continue;
ea4ca586 2199
7a6a6cbc
PA		 2200 /* only the master socket status is relevant here. The exit
2201 * conditions mirror closely tcp_recvmsg()
2202 */
2203 if (copied >= target)
2204 break;
2205
2206 if (copied) {
2207 if (sk->sk_err ||
2208 sk->sk_state == TCP_CLOSE ||
2209 (sk->sk_shutdown & RCV_SHUTDOWN) ||
2210 !timeo ||
2211 signal_pending(current))
2212 break;
2213 } else {
2214 if (sk->sk_err) {
2215 copied = sock_error(sk);
2216 break;
2217 }
2218
87952603
PA		 2219 if (sk->sk_shutdown & RCV_SHUTDOWN) {
2220 /* race breaker: the shutdown could be after the
2221 * previous receive queue check
2222 */
e3859603 2223 if (__mptcp_move_skbs(msk))
87952603 2224 continue;
7a6a6cbc 2225 break;
87952603 2226 }
7a6a6cbc
PA		 2227
2228 if (sk->sk_state == TCP_CLOSE) {
2229 copied = -ENOTCONN;
2230 break;
2231 }
2232
2233 if (!timeo) {
2234 copied = -EAGAIN;
2235 break;
2236 }
2237
2238 if (signal_pending(current)) {
2239 copied = sock_intr_errno(timeo);
2240 break;
2241 }
2242 }
2243
2244 pr_debug("block timeout %ld", timeo);
612f71d7 2245 sk_wait_data(sk, &timeo, NULL);
7a6a6cbc 2246 }
3c90e377 2247
6771bfd9 2248out_err:
b7f653b2
FW		 2249 if (cmsg_flags && copied >= 0) {
2250 if (cmsg_flags & MPTCP_CMSG_TS)
2251 tcp_recv_timestamp(msg, sk, &tss);
2c9e7765
FW		 2252
2253 if (cmsg_flags & MPTCP_CMSG_INQ) {
2254 unsigned int inq = mptcp_inq_hint(sk);
2255
2256 put_cmsg(msg, SOL_TCP, TCP_CM_INQ, sizeof(inq), &inq);
2257 }
b7f653b2
FW		 2258 }
2259
612f71d7
PA		 2260 pr_debug("msk=%p rx queue empty=%d:%d copied=%d",
2261 msk, skb_queue_empty_lockless(&sk->sk_receive_queue),
2262 skb_queue_empty(&msk->receive_queue), copied);
ca4fb892
YL		 2263 if (!(flags & MSG_PEEK))
2264 mptcp_rcv_space_adjust(msk, copied);
a6b118fe 2265
7a6a6cbc 2266 release_sock(sk);
cec37a6e
PK		 2267 return copied;
2268}
2269
b51f9b80
PA		 2270static void mptcp_retransmit_timer(struct timer_list *t)
2271{
2272 struct inet_connection_sock *icsk = from_timer(icsk, t,
2273 icsk_retransmit_timer);
2274 struct sock *sk = &icsk->icsk_inet.sk;
2d6f5a2b 2275 struct mptcp_sock *msk = mptcp_sk(sk);
b51f9b80
PA		 2276
2277 bh_lock_sock(sk);
2278 if (!sock_owned_by_user(sk)) {
2d6f5a2b
PA		 2279 /* we need a process context to retransmit */
2280 if (!test_and_set_bit(MPTCP_WORK_RTX, &msk->flags))
2281 mptcp_schedule_work(sk);
b51f9b80
PA		 2282 } else {
2283 /* delegate our work to tcp_release_cb() */
e9d09bac 2284 __set_bit(MPTCP_RETRANSMIT, &msk->cb_flags);
b51f9b80
PA		 2285 }
2286 bh_unlock_sock(sk);
2287 sock_put(sk);
2288}
2289
f6909dc1 2290static void mptcp_tout_timer(struct timer_list *t)
e16163b6
PA		 2291{
2292 struct sock *sk = from_timer(sk, t, sk_timer);
2293
2294 mptcp_schedule_work(sk);
b6d69fc8 2295 sock_put(sk);
e16163b6
PA		 2296}
2297
3b1d6210
PA		 2298/* Find an idle subflow. Return NULL if there is unacked data at tcp
2299 * level.
2300 *
2301 * A backup subflow is returned only if that is the only kind available.
2302 */
07336a87 2303struct sock *mptcp_subflow_get_retrans(struct mptcp_sock *msk)
3b1d6210 2304{
71b7dec2 2305 struct sock *backup = NULL, *pick = NULL;
3b1d6210 2306 struct mptcp_subflow_context *subflow;
71b7dec2 2307 int min_stale_count = INT_MAX;
3b1d6210 2308
3b1d6210
PA		 2309 mptcp_for_each_subflow(msk, subflow) {
2310 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2311
ff5a0b42 2312 if (!__mptcp_subflow_active(subflow))
d5f49190
PA		 2313 continue;
2314
71b7dec2
PA		 2315 /* still data outstanding at TCP level? skip this */
2316 if (!tcp_rtx_and_write_queues_empty(ssk)) {
2317 mptcp_pm_subflow_chk_stale(msk, ssk);
2318 min_stale_count = min_t(int, min_stale_count, subflow->stale_count);
2319 continue;
860975c6 2320 }
3b1d6210
PA		 2321
2322 if (subflow->backup) {
2323 if (!backup)
2324 backup = ssk;
2325 continue;
2326 }
2327
71b7dec2
PA		 2328 if (!pick)
2329 pick = ssk;
3b1d6210
PA		 2330 }
2331
71b7dec2
PA		 2332 if (pick)
2333 return pick;
2334
2335 /* use backup only if there are no progresses anywhere */
2336 return min_stale_count > 1 ? backup : NULL;
3b1d6210
PA		 2337}
2338
1e1d9d6f
PA		 2339bool __mptcp_retransmit_pending_data(struct sock *sk)
2340{
2341 struct mptcp_data_frag *cur, *rtx_head;
2342 struct mptcp_sock *msk = mptcp_sk(sk);
2343
00df24f1 2344 if (__mptcp_check_fallback(msk))
1e1d9d6f
PA		 2345 return false;
2346
1e1d9d6f
PA		 2347 /* the closing socket has some data untransmitted and/or unacked:
2348 * some data in the mptcp rtx queue has not really xmitted yet.
2349 * keep it simple and re-inject the whole mptcp level rtx queue
2350 */
2351 mptcp_data_lock(sk);
2352 __mptcp_clean_una_wakeup(sk);
2353 rtx_head = mptcp_rtx_head(sk);
2354 if (!rtx_head) {
2355 mptcp_data_unlock(sk);
2356 return false;
2357 }
2358
0d199e43 2359 msk->recovery_snd_nxt = msk->snd_nxt;
1e1d9d6f
PA		 2360 msk->recovery = true;
2361 mptcp_data_unlock(sk);
2362
2363 msk->first_pending = rtx_head;
1e1d9d6f
PA		 2364 msk->snd_burst = 0;
2365
2366 /* be sure to clear the "sent status" on all re-injected fragments */
2367 list_for_each_entry(cur, &msk->rtx_queue, list) {
2368 if (!cur->already_sent)
2369 break;
2370 cur->already_sent = 0;
2371 }
2372
2373 return true;
2374}
2375
b29fcfb5
PA		 2376/* flags for __mptcp_close_ssk() */
2377#define MPTCP_CF_PUSH BIT(1)
2378#define MPTCP_CF_FASTCLOSE BIT(2)
2379
14c56686
GT		 2380/* be sure to send a reset only if the caller asked for it, also
2381 * clean completely the subflow status when the subflow reaches
2382 * TCP_CLOSE state
2383 */
2384static void __mptcp_subflow_disconnect(struct sock *ssk,
2385 struct mptcp_subflow_context *subflow,
2386 unsigned int flags)
2387{
2388 if (((1 << ssk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)) ||
2389 (flags & MPTCP_CF_FASTCLOSE)) {
2390 /* The MPTCP code never wait on the subflow sockets, TCP-level
2391 * disconnect should never fail
2392 */
2393 WARN_ON_ONCE(tcp_disconnect(ssk, 0));
2394 mptcp_subflow_ctx_reset(subflow);
2395 } else {
2396 tcp_shutdown(ssk, SEND_SHUTDOWN);
2397 }
2398}
2399
cec37a6e
PK		 2400/* subflow sockets can be either outgoing (connect) or incoming
2401 * (accept).
2402 *
2403 * Outgoing subflows use in-kernel sockets.
2404 * Incoming subflows do not have their own 'struct socket' allocated,
2405 * so we need to use tcp_close() after detaching them from the mptcp
2406 * parent socket.
2407 */
a141e02e 2408static void __mptcp_close_ssk(struct sock *sk, struct sock *ssk,
b29fcfb5
PA		 2409 struct mptcp_subflow_context *subflow,
2410 unsigned int flags)
cec37a6e 2411{
e0be4931 2412 struct mptcp_sock *msk = mptcp_sk(sk);
63740448
PA		 2413 bool dispose_it, need_push = false;
2414
2415 /* If the first subflow moved to a close state before accept, e.g. due
27e5ccc2
PA		 2416 * to an incoming reset or listener shutdown, the subflow socket is
2417 * already deleted by inet_child_forget() and the mptcp socket can't
2418 * survive too.
63740448 2419 */
27e5ccc2
PA		 2420 if (msk->in_accept_queue && msk->first == ssk &&
2421 (sock_flag(sk, SOCK_DEAD) || sock_flag(ssk, SOCK_DEAD))) {
63740448
PA		 2422 /* ensure later check in mptcp_worker() will dispose the msk */
2423 sock_set_flag(sk, SOCK_DEAD);
d866ae9a 2424 mptcp_set_close_tout(sk, tcp_jiffies32 - (mptcp_close_timeout(sk) + 1));
63740448
PA		 2425 lock_sock_nested(ssk, SINGLE_DEPTH_NESTING);
2426 mptcp_subflow_drop_ctx(ssk);
2427 goto out_release;
2428 }
e0be4931 2429
39880bd8 2430 dispose_it = msk->free_first || ssk != msk->first;
b29fcfb5
PA		 2431 if (dispose_it)
2432 list_del(&subflow->node);
cec37a6e 2433
3f8b2667 2434 lock_sock_nested(ssk, SINGLE_DEPTH_NESTING);
e16163b6 2435
ff18f9ef 2436 if ((flags & MPTCP_CF_FASTCLOSE) && !__mptcp_check_fallback(msk)) {
14c56686 2437 /* be sure to force the tcp_close path
d21f8348
PA		 2438 * to generate the egress reset
2439 */
2440 ssk->sk_lingertime = 0;
2441 sock_set_flag(ssk, SOCK_LINGER);
b29fcfb5 2442 subflow->send_fastclose = 1;
d21f8348 2443 }
b29fcfb5
PA		 2444
2445 need_push = (flags & MPTCP_CF_PUSH) && __mptcp_retransmit_pending_data(sk);
2446 if (!dispose_it) {
14c56686 2447 __mptcp_subflow_disconnect(ssk, subflow, flags);
b29fcfb5
PA		 2448 release_sock(ssk);
2449
2450 goto out;
2451 }
2452
d7b1bfd0
PA		 2453 subflow->disposable = 1;
2454
e16163b6
PA		 2455 /* if ssk hit tcp_done(), tcp_cleanup_ulp() cleared the related ops
2456 * the ssk has been already destroyed, we just need to release the
2457 * reference owned by msk;
2458 */
2459 if (!inet_csk(ssk)->icsk_ulp_ops) {
b6985b9b 2460 WARN_ON_ONCE(!sock_flag(ssk, SOCK_DEAD));
e16163b6 2461 kfree_rcu(subflow, rcu);
cec37a6e 2462 } else {
d7b1bfd0 2463 /* otherwise tcp will dispose of the ssk and subflow ctx */
e16163b6
PA		 2464 __tcp_close(ssk, 0);
2465
2466 /* close acquired an extra ref */
2467 __sock_put(ssk);
cec37a6e 2468 }
63740448
PA		 2469
2470out_release:
9f1a9881 2471 __mptcp_subflow_error_report(sk, ssk);
e16163b6 2472 release_sock(ssk);
e16163b6
PA		 2473
2474 sock_put(ssk);
e0be4931 2475
c8fe62f0 2476 if (ssk == msk->first)
1b1b43ee 2477 WRITE_ONCE(msk->first, NULL);
c8fe62f0 2478
b29fcfb5 2479out:
8005184f 2480 __mptcp_sync_sndbuf(sk);
1e1d9d6f
PA		 2481 if (need_push)
2482 __mptcp_push_pending(sk, 0);
27e5ccc2
PA		 2483
2484 /* Catch every 'all subflows closed' scenario, including peers silently
2485 * closing them, e.g. due to timeout.
2486 * For established sockets, allow an additional timeout before closing,
2487 * as the protocol can still create more subflows.
2488 */
2489 if (list_is_singular(&msk->conn_list) && msk->first &&
2490 inet_sk_state_load(msk->first) == TCP_CLOSE) {
2491 if (sk->sk_state != TCP_ESTABLISHED ||
2492 msk->in_accept_queue || sock_flag(sk, SOCK_DEAD)) {
c693a851 2493 mptcp_set_state(sk, TCP_CLOSE);
27e5ccc2
PA		 2494 mptcp_close_wake_up(sk);
2495 } else {
2496 mptcp_start_tout_timer(sk);
2497 }
2498 }
f870fa0b
MM		 2499}
2500
a141e02e
FW		 2501void mptcp_close_ssk(struct sock *sk, struct sock *ssk,
2502 struct mptcp_subflow_context *subflow)
2503{
b911c97c
FW		 2504 if (sk->sk_state == TCP_ESTABLISHED)
2505 mptcp_event(MPTCP_EVENT_SUB_CLOSED, mptcp_sk(sk), ssk, GFP_KERNEL);
a88c9e49
PA		 2506
2507 /* subflow aborted before reaching the fully_established status
2508 * attempt the creation of the next subflow
2509 */
74cbb0c6 2510 mptcp_pm_subflow_check_next(mptcp_sk(sk), subflow);
a88c9e49 2511
b29fcfb5 2512 __mptcp_close_ssk(sk, ssk, subflow, MPTCP_CF_PUSH);
a141e02e
FW		 2513}
2514
dc24f8b4
PA		 2515static unsigned int mptcp_sync_mss(struct sock *sk, u32 pmtu)
2516{
2517 return 0;
2518}
2519
b6985b9b 2520static void __mptcp_close_subflow(struct sock *sk)
0e4f35d7
PA		 2521{
2522 struct mptcp_subflow_context *subflow, *tmp;
b6985b9b 2523 struct mptcp_sock *msk = mptcp_sk(sk);
0e4f35d7 2524
3abc05d9
FW		 2525 might_sleep();
2526
5efbf6f7 2527 mptcp_for_each_subflow_safe(msk, subflow, tmp) {
0e4f35d7
PA		 2528 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2529
2530 if (inet_sk_state_load(ssk) != TCP_CLOSE)
2531 continue;
2532
40947e13
FW		 2533 /* 'subflow_data_ready' will re-sched once rx queue is empty */
2534 if (!skb_queue_empty_lockless(&ssk->sk_receive_queue))
2535 continue;
2536
b6985b9b
PA		 2537 mptcp_close_ssk(sk, ssk, subflow);
2538 }
2539
0e4f35d7
PA		 2540}
2541
27e5ccc2 2542static bool mptcp_close_tout_expired(const struct sock *sk)
e16163b6 2543{
27e5ccc2
PA		 2544 if (!inet_csk(sk)->icsk_mtup.probe_timestamp ||
2545 sk->sk_state == TCP_CLOSE)
2546 return false;
e16163b6 2547
27e5ccc2 2548 return time_after32(tcp_jiffies32,
d866ae9a 2549 inet_csk(sk)->icsk_mtup.probe_timestamp + mptcp_close_timeout(sk));
e16163b6
PA		 2550}
2551
50c504a2
FW		 2552static void mptcp_check_fastclose(struct mptcp_sock *msk)
2553{
2554 struct mptcp_subflow_context *subflow, *tmp;
80638684 2555 struct sock *sk = (struct sock *)msk;
50c504a2
FW		 2556
2557 if (likely(!READ_ONCE(msk->rcv_fastclose)))
2558 return;
2559
2560 mptcp_token_destroy(msk);
2561
5efbf6f7 2562 mptcp_for_each_subflow_safe(msk, subflow, tmp) {
50c504a2 2563 struct sock *tcp_sk = mptcp_subflow_tcp_sock(subflow);
75e908c3 2564 bool slow;
50c504a2 2565
75e908c3 2566 slow = lock_sock_fast(tcp_sk);
50c504a2
FW		 2567 if (tcp_sk->sk_state != TCP_CLOSE) {
2568 tcp_send_active_reset(tcp_sk, GFP_ATOMIC);
2569 tcp_set_state(tcp_sk, TCP_CLOSE);
2570 }
75e908c3 2571 unlock_sock_fast(tcp_sk, slow);
50c504a2
FW		 2572 }
2573
69800e51
PA		 2574 /* Mirror the tcp_reset() error propagation */
2575 switch (sk->sk_state) {
2576 case TCP_SYN_SENT:
9ae8e5ad 2577 WRITE_ONCE(sk->sk_err, ECONNREFUSED);
69800e51
PA		 2578 break;
2579 case TCP_CLOSE_WAIT:
9ae8e5ad 2580 WRITE_ONCE(sk->sk_err, EPIPE);
69800e51
PA		 2581 break;
2582 case TCP_CLOSE:
2583 return;
2584 default:
9ae8e5ad 2585 WRITE_ONCE(sk->sk_err, ECONNRESET);
69800e51
PA		 2586 }
2587
c693a851 2588 mptcp_set_state(sk, TCP_CLOSE);
6b9831bf 2589 WRITE_ONCE(sk->sk_shutdown, SHUTDOWN_MASK);
50c504a2 2590 smp_mb__before_atomic(); /* SHUTDOWN must be visible first */
50c504a2
FW		 2591 set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags);
2592
69800e51
PA		 2593 /* the calling mptcp_worker will properly destroy the socket */
2594 if (sock_flag(sk, SOCK_DEAD))
2595 return;
2596
2597 sk->sk_state_change(sk);
2598 sk_error_report(sk);
50c504a2
FW		 2599}
2600
2948d0a1 2601static void __mptcp_retrans(struct sock *sk)
80992017 2602{
2948d0a1 2603 struct mptcp_sock *msk = mptcp_sk(sk);
ee2708ae 2604 struct mptcp_subflow_context *subflow;
caf971df 2605 struct mptcp_sendmsg_info info = {};
3b1d6210 2606 struct mptcp_data_frag *dfrag;
2948d0a1 2607 struct sock *ssk;
ee2708ae
GT		 2608 int ret, err;
2609 u16 len = 0;
3b1d6210 2610
b5941f06 2611 mptcp_clean_una_wakeup(sk);
3241a9c0
FW		 2612
2613 /* first check ssk: need to kick "stale" logic */
ee2708ae 2614 err = mptcp_sched_get_retrans(msk);
3b1d6210 2615 dfrag = mptcp_rtx_head(sk);
6477dd39
MM		 2616 if (!dfrag) {
2617 if (mptcp_data_fin_enabled(msk)) {
2618 struct inet_connection_sock *icsk = inet_csk(sk);
2619
2620 icsk->icsk_retransmits++;
2621 mptcp_set_datafin_timeout(sk);
2622 mptcp_send_ack(msk);
2623
2624 goto reset_timer;
2625 }
2626
3241a9c0
FW		 2627 if (!mptcp_send_head(sk))
2628 return;
2629
2630 goto reset_timer;
6477dd39 2631 }
3b1d6210 2632
ee2708ae 2633 if (err)
2948d0a1 2634 goto reset_timer;
3b1d6210 2635
ee2708ae
GT		 2636 mptcp_for_each_subflow(msk, subflow) {
2637 if (READ_ONCE(subflow->scheduled)) {
2638 u16 copied = 0;
3b1d6210 2639
ee2708ae 2640 mptcp_subflow_set_scheduled(subflow, false);
3b1d6210 2641
ee2708ae
GT		 2642 ssk = mptcp_subflow_tcp_sock(subflow);
2643
2644 lock_sock(ssk);
2645
2646 /* limit retransmission to the bytes already sent on some subflows */
2647 info.sent = 0;
2648 info.limit = READ_ONCE(msk->csum_enabled) ? dfrag->data_len :
2649 dfrag->already_sent;
2650 while (info.sent < info.limit) {
2651 ret = mptcp_sendmsg_frag(sk, ssk, dfrag, &info);
2652 if (ret <= 0)
2653 break;
2654
2655 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_RETRANSSEGS);
2656 copied += ret;
2657 info.sent += ret;
2658 }
2659 if (copied) {
2660 len = max(copied, len);
2661 tcp_push(ssk, 0, info.mss_now, tcp_sk(ssk)->nonagle,
2662 info.size_goal);
2663 WRITE_ONCE(msk->allow_infinite_fallback, false);
2664 }
2665
2666 release_sock(ssk);
2667 }
4e14867d 2668 }
3b1d6210 2669
ee2708ae
GT		 2670 msk->bytes_retrans += len;
2671 dfrag->already_sent = max(dfrag->already_sent, len);
3b1d6210 2672
2948d0a1 2673reset_timer:
3241a9c0
FW		 2674 mptcp_check_and_set_pending(sk);
2675
f6909dc1
PA		 2676 if (!mptcp_rtx_timer_pending(sk))
2677 mptcp_reset_rtx_timer(sk);
2948d0a1
PA		 2678}
2679
76a13b31
GT		 2680/* schedule the timeout timer for the relevant event: either close timeout
2681 * or mp_fail timeout. The close timeout takes precedence on the mp_fail one
2682 */
f6909dc1 2683void mptcp_reset_tout_timer(struct mptcp_sock *msk, unsigned long fail_tout)
76a13b31
GT		 2684{
2685 struct sock *sk = (struct sock *)msk;
2686 unsigned long timeout, close_timeout;
2687
27e5ccc2 2688 if (!fail_tout && !inet_csk(sk)->icsk_mtup.probe_timestamp)
76a13b31
GT		 2689 return;
2690
27e5ccc2 2691 close_timeout = inet_csk(sk)->icsk_mtup.probe_timestamp - tcp_jiffies32 + jiffies +
d866ae9a 2692 mptcp_close_timeout(sk);
76a13b31
GT		 2693
2694 /* the close timeout takes precedence on the fail one, and here at least one of
2695 * them is active
2696 */
27e5ccc2 2697 timeout = inet_csk(sk)->icsk_mtup.probe_timestamp ? close_timeout : fail_tout;
76a13b31
GT		 2698
2699 sk_reset_timer(sk, &sk->sk_timer, timeout);
2700}
2701
49fa1919
GT		 2702static void mptcp_mp_fail_no_response(struct mptcp_sock *msk)
2703{
76a13b31 2704 struct sock *ssk = msk->first;
49fa1919
GT		 2705 bool slow;
2706
76a13b31
GT		 2707 if (!ssk)
2708 return;
49fa1919 2709
76a13b31
GT		 2710 pr_debug("MP_FAIL doesn't respond, reset the subflow");
2711
2712 slow = lock_sock_fast(ssk);
2713 mptcp_subflow_reset(ssk);
2714 WRITE_ONCE(mptcp_subflow_ctx(ssk)->fail_tout, 0);
2715 unlock_sock_fast(ssk, slow);
49fa1919
GT		 2716}
2717
d21f8348
PA		 2718static void mptcp_do_fastclose(struct sock *sk)
2719{
2720 struct mptcp_subflow_context *subflow, *tmp;
2721 struct mptcp_sock *msk = mptcp_sk(sk);
2722
c693a851 2723 mptcp_set_state(sk, TCP_CLOSE);
d21f8348
PA		 2724 mptcp_for_each_subflow_safe(msk, subflow, tmp)
2725 __mptcp_close_ssk(sk, mptcp_subflow_tcp_sock(subflow),
2726 subflow, MPTCP_CF_FASTCLOSE);
2727}
2728
2948d0a1
PA		 2729static void mptcp_worker(struct work_struct *work)
2730{
2731 struct mptcp_sock *msk = container_of(work, struct mptcp_sock, work);
80638684 2732 struct sock *sk = (struct sock *)msk;
76a13b31 2733 unsigned long fail_tout;
2948d0a1
PA		 2734 int state;
2735
2736 lock_sock(sk);
2737 state = sk->sk_state;
d6a04437 2738 if (unlikely((1 << state) & (TCPF_CLOSE | TCPF_LISTEN)))
2948d0a1
PA		 2739 goto unlock;
2740
2948d0a1
PA		 2741 mptcp_check_fastclose(msk);
2742
86e39e04 2743 mptcp_pm_nl_work(msk);
2948d0a1 2744
81c1d029
PA		 2745 mptcp_check_send_data_fin(sk);
2746 mptcp_check_data_fin_ack(sk);
2948d0a1
PA		 2747 mptcp_check_data_fin(sk);
2748
b6985b9b
PA		 2749 if (test_and_clear_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags))
2750 __mptcp_close_subflow(sk);
2751
27e5ccc2
PA		 2752 if (mptcp_close_tout_expired(sk)) {
2753 mptcp_do_fastclose(sk);
2754 mptcp_close_wake_up(sk);
2755 }
bbd49d11 2756
27e5ccc2
PA		 2757 if (sock_flag(sk, SOCK_DEAD) && sk->sk_state == TCP_CLOSE) {
2758 __mptcp_destroy_sock(sk);
2759 goto unlock;
2948d0a1
PA		 2760 }
2761
2948d0a1
PA		 2762 if (test_and_clear_bit(MPTCP_WORK_RTX, &msk->flags))
2763 __mptcp_retrans(sk);
3b1d6210 2764
76a13b31
GT		 2765 fail_tout = msk->first ? READ_ONCE(mptcp_subflow_ctx(msk->first)->fail_tout) : 0;
2766 if (fail_tout && time_after(jiffies, fail_tout))
2767 mptcp_mp_fail_no_response(msk);
49fa1919 2768
3b1d6210 2769unlock:
80992017
PA		 2770 release_sock(sk);
2771 sock_put(sk);
2772}
2773
e2636917 2774static void __mptcp_init_sock(struct sock *sk)
f870fa0b 2775{
cec37a6e
PK		 2776 struct mptcp_sock *msk = mptcp_sk(sk);
2777
2778 INIT_LIST_HEAD(&msk->conn_list);
ec3edaa7 2779 INIT_LIST_HEAD(&msk->join_list);
18b683bf 2780 INIT_LIST_HEAD(&msk->rtx_queue);
80992017 2781 INIT_WORK(&msk->work, mptcp_worker);
87952603 2782 __skb_queue_head_init(&msk->receive_queue);
ab174ad8 2783 msk->out_of_order_queue = RB_ROOT;
f0e6a4cf 2784 msk->first_pending = NULL;
9426ce47 2785 WRITE_ONCE(msk->rmem_fwd_alloc, 0);
ce599c51 2786 WRITE_ONCE(msk->rmem_released, 0);
33d41c9c 2787 msk->timer_ival = TCP_RTO_MIN;
5684ab1a 2788 msk->scaling_ratio = TCP_DEFAULT_SCALING_RATIO;
cec37a6e 2789
1b1b43ee 2790 WRITE_ONCE(msk->first, NULL);
dc24f8b4 2791 inet_csk(sk)->icsk_sync_mss = mptcp_sync_mss;
752e9067 2792 WRITE_ONCE(msk->csum_enabled, mptcp_is_checksum_enabled(sock_net(sk)));
0530020a 2793 WRITE_ONCE(msk->allow_infinite_fallback, true);
1e1d9d6f 2794 msk->recovery = false;
6f06b4d4 2795 msk->subflow_id = 1;
8ab183de 2796
1b1c7a0e
PK		 2797 mptcp_pm_data_init(msk);
2798
b51f9b80
PA		 2799 /* re-use the csk retrans timer for MPTCP-level retrans */
2800 timer_setup(&msk->sk.icsk_retransmit_timer, mptcp_retransmit_timer, 0);
f6909dc1 2801 timer_setup(&sk->sk_timer, mptcp_tout_timer, 0);
f870fa0b
MM		 2802}
2803
b29fcfb5 2804static void mptcp_ca_reset(struct sock *sk)
784325e9 2805{
20b5759f 2806 struct inet_connection_sock *icsk = inet_csk(sk);
b29fcfb5
PA		 2807
2808 tcp_assign_congestion_control(sk);
2809 strcpy(mptcp_sk(sk)->ca_name, icsk->icsk_ca_ops->name);
2810
2811 /* no need to keep a reference to the ops, the name will suffice */
2812 tcp_cleanup_congestion_control(sk);
2813 icsk->icsk_ca_ops = NULL;
2814}
2815
2816static int mptcp_init_sock(struct sock *sk)
2817{
fc518953 2818 struct net *net = sock_net(sk);
1730b2b2 2819 int ret;
18b683bf 2820
e2636917 2821 __mptcp_init_sock(sk);
b6c08380 2822
fc518953
FW		 2823 if (!mptcp_is_enabled(net))
2824 return -ENOPROTOOPT;
2825
2826 if (unlikely(!net->mib.mptcp_statistics) && !mptcp_mib_alloc(net))
2827 return -ENOMEM;
2828
1730b2b2
GT		 2829 ret = mptcp_init_sched(mptcp_sk(sk),
2830 mptcp_sched_find(mptcp_get_scheduler(net)));
2831 if (ret)
2832 return ret;
2833
a5ef058d
PA		 2834 set_bit(SOCK_CUSTOM_SOCKOPT, &sk->sk_socket->flags);
2835
20b5759f
PA		 2836 /* fetch the ca name; do it outside __mptcp_init_sock(), so that clone will
2837 * propagate the correct value
2838 */
b29fcfb5 2839 mptcp_ca_reset(sk);
20b5759f 2840
d027236c 2841 sk_sockets_allocated_inc(sk);
a963853f
GT		 2842 sk->sk_rcvbuf = READ_ONCE(net->ipv4.sysctl_tcp_rmem[1]);
2843 sk->sk_sndbuf = READ_ONCE(net->ipv4.sysctl_tcp_wmem[1]);
d027236c 2844
18b683bf
PA		 2845 return 0;
2846}
2847
2848static void __mptcp_clear_xmit(struct sock *sk)
2849{
2850 struct mptcp_sock *msk = mptcp_sk(sk);
2851 struct mptcp_data_frag *dtmp, *dfrag;
2852
d9ca1de8 2853 WRITE_ONCE(msk->first_pending, NULL);
18b683bf 2854 list_for_each_entry_safe(dfrag, dtmp, &msk->rtx_queue, list)
d027236c 2855 dfrag_clear(sk, dfrag);
784325e9
MB		 2856}
2857
30e51b92 2858void mptcp_cancel_work(struct sock *sk)
80992017
PA		 2859{
2860 struct mptcp_sock *msk = mptcp_sk(sk);
2861
b2771d24 2862 if (cancel_work_sync(&msk->work))
e16163b6 2863 __sock_put(sk);
80992017
PA		 2864}
2865
d0876b22 2866void mptcp_subflow_shutdown(struct sock *sk, struct sock *ssk, int how)
21498490
PK		 2867{
2868 lock_sock(ssk);
2869
2870 switch (ssk->sk_state) {
2871 case TCP_LISTEN:
2872 if (!(how & RCV_SHUTDOWN))
2873 break;
df561f66 2874 fallthrough;
21498490 2875 case TCP_SYN_SENT:
c2b2ae39 2876 WARN_ON_ONCE(tcp_disconnect(ssk, O_NONBLOCK));
21498490
PK		 2877 break;
2878 default:
43b54c6e
MM		 2879 if (__mptcp_check_fallback(mptcp_sk(sk))) {
2880 pr_debug("Fallback");
2881 ssk->sk_shutdown |= how;
2882 tcp_shutdown(ssk, how);
81c1d029
PA		 2883
2884 /* simulate the data_fin ack reception to let the state
2885 * machine move forward
2886 */
2887 WRITE_ONCE(mptcp_sk(sk)->snd_una, mptcp_sk(sk)->snd_nxt);
2888 mptcp_schedule_work(sk);
43b54c6e
MM		 2889 } else {
2890 pr_debug("Sending DATA_FIN on subflow %p", ssk);
43b54c6e 2891 tcp_send_ack(ssk);
f6909dc1
PA		 2892 if (!mptcp_rtx_timer_pending(sk))
2893 mptcp_reset_rtx_timer(sk);
43b54c6e 2894 }
21498490
PK		 2895 break;
2896 }
2897
21498490
PK		 2898 release_sock(ssk);
2899}
2900
d9cd27b8
GT		 2901void mptcp_set_state(struct sock *sk, int state)
2902{
2903 int oldstate = sk->sk_state;
2904
2905 switch (state) {
2906 case TCP_ESTABLISHED:
2907 if (oldstate != TCP_ESTABLISHED)
2908 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_CURRESTAB);
2909 break;
2910
2911 default:
2912 if (oldstate == TCP_ESTABLISHED)
2913 MPTCP_DEC_STATS(sock_net(sk), MPTCP_MIB_CURRESTAB);
2914 }
2915
2916 inet_sk_state_store(sk, state);
2917}
2918
6920b851
MM		 2919static const unsigned char new_state[16] = {
2920 /* current state: new state: action: */
2921 [0 /* (Invalid) */] = TCP_CLOSE,
2922 [TCP_ESTABLISHED] = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2923 [TCP_SYN_SENT] = TCP_CLOSE,
2924 [TCP_SYN_RECV] = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2925 [TCP_FIN_WAIT1] = TCP_FIN_WAIT1,
2926 [TCP_FIN_WAIT2] = TCP_FIN_WAIT2,
2927 [TCP_TIME_WAIT] = TCP_CLOSE, /* should not happen ! */
2928 [TCP_CLOSE] = TCP_CLOSE,
2929 [TCP_CLOSE_WAIT] = TCP_LAST_ACK | TCP_ACTION_FIN,
2930 [TCP_LAST_ACK] = TCP_LAST_ACK,
2931 [TCP_LISTEN] = TCP_CLOSE,
2932 [TCP_CLOSING] = TCP_CLOSING,
2933 [TCP_NEW_SYN_RECV] = TCP_CLOSE, /* should not happen ! */
2934};
2935
2936static int mptcp_close_state(struct sock *sk)
2937{
2938 int next = (int)new_state[sk->sk_state];
2939 int ns = next & TCP_STATE_MASK;
2940
c693a851 2941 mptcp_set_state(sk, ns);
6920b851
MM		 2942
2943 return next & TCP_ACTION_FIN;
2944}
2945
81c1d029 2946static void mptcp_check_send_data_fin(struct sock *sk)
f870fa0b 2947{
e16163b6 2948 struct mptcp_subflow_context *subflow;
f870fa0b
MM		 2949 struct mptcp_sock *msk = mptcp_sk(sk);
2950
e16163b6
PA		 2951 pr_debug("msk=%p snd_data_fin_enable=%d pending=%d snd_nxt=%llu write_seq=%llu",
2952 msk, msk->snd_data_fin_enable, !!mptcp_send_head(sk),
2953 msk->snd_nxt, msk->write_seq);
43b54c6e 2954
e16163b6
PA		 2955 /* we still need to enqueue subflows or not really shutting down,
2956 * skip this
2957 */
2958 if (!msk->snd_data_fin_enable || msk->snd_nxt + 1 != msk->write_seq ||
2959 mptcp_send_head(sk))
2960 return;
2961
2962 WRITE_ONCE(msk->snd_nxt, msk->write_seq);
2963
e16163b6
PA		 2964 mptcp_for_each_subflow(msk, subflow) {
2965 struct sock *tcp_sk = mptcp_subflow_tcp_sock(subflow);
43b54c6e 2966
e16163b6 2967 mptcp_subflow_shutdown(sk, tcp_sk, SEND_SHUTDOWN);
43b54c6e 2968 }
e16163b6 2969}
2c22c06c 2970
e16163b6
PA		 2971static void __mptcp_wr_shutdown(struct sock *sk)
2972{
2973 struct mptcp_sock *msk = mptcp_sk(sk);
43b54c6e 2974
e16163b6
PA		 2975 pr_debug("msk=%p snd_data_fin_enable=%d shutdown=%x state=%d pending=%d",
2976 msk, msk->snd_data_fin_enable, sk->sk_shutdown, sk->sk_state,
2977 !!mptcp_send_head(sk));
2978
2979 /* will be ignored by fallback sockets */
2980 WRITE_ONCE(msk->write_seq, msk->write_seq + 1);
2981 WRITE_ONCE(msk->snd_data_fin_enable, 1);
2982
81c1d029 2983 mptcp_check_send_data_fin(sk);
e16163b6
PA		 2984}
2985
2986static void __mptcp_destroy_sock(struct sock *sk)
2987{
e16163b6 2988 struct mptcp_sock *msk = mptcp_sk(sk);
e16163b6
PA		 2989
2990 pr_debug("msk=%p", msk);
f870fa0b 2991
3abc05d9
FW		 2992 might_sleep();
2993
f6909dc1 2994 mptcp_stop_rtx_timer(sk);
e16163b6
PA		 2995 sk_stop_timer(sk, &sk->sk_timer);
2996 msk->pm.status = 0;
1730b2b2 2997 mptcp_release_sched(msk);
b2c5b614 2998
e16163b6 2999 sk->sk_prot->destroy(sk);
80992017 3000
9426ce47 3001 WARN_ON_ONCE(READ_ONCE(msk->rmem_fwd_alloc));
87952603 3002 WARN_ON_ONCE(msk->rmem_released);
e16163b6
PA		 3003 sk_stream_kill_queues(sk);
3004 xfrm_sk_free_policy(sk);
aa1fbd94 3005
e16163b6
PA		 3006 sock_put(sk);
3007}
3008
63740448
PA		 3009void __mptcp_unaccepted_force_close(struct sock *sk)
3010{
3011 sock_set_flag(sk, SOCK_DEAD);
63740448
PA		 3012 mptcp_do_fastclose(sk);
3013 __mptcp_destroy_sock(sk);
3014}
3015
5684ab1a 3016static __poll_t mptcp_check_readable(struct sock *sk)
d21f8348 3017{
5684ab1a 3018 return mptcp_epollin_ready(sk) ? EPOLLIN | EPOLLRDNORM : 0;
d21f8348
PA		 3019}
3020
57fc0f1c 3021static void mptcp_check_listen_stop(struct sock *sk)
c558246e 3022{
57fc0f1c
PA		 3023 struct sock *ssk;
3024
3025 if (inet_sk_state_load(sk) != TCP_LISTEN)
3026 return;
3027
3028 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
3029 ssk = mptcp_sk(sk)->first;
3030 if (WARN_ON_ONCE(!ssk || inet_sk_state_load(ssk) != TCP_LISTEN))
3031 return;
3032
3033 lock_sock_nested(ssk, SINGLE_DEPTH_NESTING);
3fffa15b 3034 tcp_set_state(ssk, TCP_CLOSE);
57fc0f1c
PA		 3035 mptcp_subflow_queue_clean(sk, ssk);
3036 inet_csk_listen_stop(ssk);
3037 mptcp_event_pm_listener(ssk, MPTCP_EVENT_LISTENER_CLOSED);
57fc0f1c 3038 release_sock(ssk);
c558246e
MD		 3039}
3040
26d3e21c 3041bool __mptcp_close(struct sock *sk, long timeout)
e16163b6
PA		 3042{
3043 struct mptcp_subflow_context *subflow;
76a13b31 3044 struct mptcp_sock *msk = mptcp_sk(sk);
e16163b6 3045 bool do_cancel_work = false;
d4e85922 3046 int subflows_alive = 0;
e16163b6 3047
6b9831bf 3048 WRITE_ONCE(sk->sk_shutdown, SHUTDOWN_MASK);
e16163b6
PA		 3049
3050 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE)) {
57fc0f1c 3051 mptcp_check_listen_stop(sk);
c693a851 3052 mptcp_set_state(sk, TCP_CLOSE);
e16163b6
PA		 3053 goto cleanup;
3054 }
6771bfd9 3055
5684ab1a 3056 if (mptcp_data_avail(msk) || timeout < 0) {
8d547809
PA		 3057 /* If the msk has read data, or the caller explicitly ask it,
3058 * do the MPTCP equivalent of TCP reset, aka MPTCP fastclose
3059 */
d21f8348 3060 mptcp_do_fastclose(sk);
8d547809 3061 timeout = 0;
d21f8348 3062 } else if (mptcp_close_state(sk)) {
e16163b6 3063 __mptcp_wr_shutdown(sk);
d21f8348 3064 }
e16163b6
PA		 3065
3066 sk_stream_wait_close(sk, timeout);
3067
3068cleanup:
3069 /* orphan all the subflows */
76a13b31 3070 mptcp_for_each_subflow(msk, subflow) {
e16163b6 3071 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
49054556 3072 bool slow = lock_sock_fast_nested(ssk);
e16163b6 3073
d4e85922
PA		 3074 subflows_alive += ssk->sk_state != TCP_CLOSE;
3075
76a13b31
GT		 3076 /* since the close timeout takes precedence on the fail one,
3077 * cancel the latter
3078 */
3079 if (ssk == msk->first)
3080 subflow->fail_tout = 0;
3081
fe948001
MD		 3082 /* detach from the parent socket, but allow data_ready to
3083 * push incoming data into the mptcp stack, to properly ack it
3084 */
3085 ssk->sk_socket = NULL;
3086 ssk->sk_wq = NULL;
e16163b6 3087 unlock_sock_fast(ssk, slow);
e16163b6
PA		 3088 }
3089 sock_orphan(sk);
3090
d4e85922
PA		 3091 /* all the subflows are closed, only timeout can change the msk
3092 * state, let's not keep resources busy for no reasons
3093 */
3094 if (subflows_alive == 0)
c693a851 3095 mptcp_set_state(sk, TCP_CLOSE);
d4e85922 3096
e16163b6
PA		 3097 sock_hold(sk);
3098 pr_debug("msk=%p state=%d", sk, sk->sk_state);
00df24f1 3099 if (msk->token)
76a13b31 3100 mptcp_event(MPTCP_EVENT_CLOSED, msk, NULL, GFP_KERNEL);
b29fcfb5 3101
e16163b6
PA		 3102 if (sk->sk_state == TCP_CLOSE) {
3103 __mptcp_destroy_sock(sk);
3104 do_cancel_work = true;
3105 } else {
27e5ccc2 3106 mptcp_start_tout_timer(sk);
e16163b6 3107 }
26d3e21c
MD		 3108
3109 return do_cancel_work;
3110}
3111
3112static void mptcp_close(struct sock *sk, long timeout)
3113{
3114 bool do_cancel_work;
3115
3116 lock_sock(sk);
3117
3118 do_cancel_work = __mptcp_close(sk, timeout);
e16163b6
PA		 3119 release_sock(sk);
3120 if (do_cancel_work)
3121 mptcp_cancel_work(sk);
b911c97c 3122
e16163b6 3123 sock_put(sk);
f870fa0b
MM		 3124}
3125
7e8b88ec 3126static void mptcp_copy_inaddrs(struct sock *msk, const struct sock *ssk)
cf7da0d6
PK		 3127{
3128#if IS_ENABLED(CONFIG_MPTCP_IPV6)
3129 const struct ipv6_pinfo *ssk6 = inet6_sk(ssk);
3130 struct ipv6_pinfo *msk6 = inet6_sk(msk);
3131
3132 msk->sk_v6_daddr = ssk->sk_v6_daddr;
3133 msk->sk_v6_rcv_saddr = ssk->sk_v6_rcv_saddr;
3134
3135 if (msk6 && ssk6) {
3136 msk6->saddr = ssk6->saddr;
3137 msk6->flow_label = ssk6->flow_label;
3138 }
3139#endif
3140
3141 inet_sk(msk)->inet_num = inet_sk(ssk)->inet_num;
3142 inet_sk(msk)->inet_dport = inet_sk(ssk)->inet_dport;
3143 inet_sk(msk)->inet_sport = inet_sk(ssk)->inet_sport;
3144 inet_sk(msk)->inet_daddr = inet_sk(ssk)->inet_daddr;
3145 inet_sk(msk)->inet_saddr = inet_sk(ssk)->inet_saddr;
3146 inet_sk(msk)->inet_rcv_saddr = inet_sk(ssk)->inet_rcv_saddr;
3147}
3148
18b683bf
PA		 3149static int mptcp_disconnect(struct sock *sk, int flags)
3150{
76e2a55d
PA		 3151 struct mptcp_sock *msk = mptcp_sk(sk);
3152
7d803344
PA		 3153 /* We are on the fastopen error path. We can't call straight into the
3154 * subflows cleanup code due to lock nesting (we are already under
c2b2ae39 3155 * msk->firstsocket lock).
7d803344
PA		 3156 */
3157 if (msk->fastopening)
c2b2ae39 3158 return -EBUSY;
7d803344 3159
57fc0f1c 3160 mptcp_check_listen_stop(sk);
c693a851 3161 mptcp_set_state(sk, TCP_CLOSE);
b29fcfb5 3162
f6909dc1 3163 mptcp_stop_rtx_timer(sk);
27e5ccc2 3164 mptcp_stop_tout_timer(sk);
b29fcfb5 3165
00df24f1
GT		 3166 if (msk->token)
3167 mptcp_event(MPTCP_EVENT_CLOSED, msk, NULL, GFP_KERNEL);
b29fcfb5 3168
c0bf3c6a
PA		 3169 /* msk->subflow is still intact, the following will not free the first
3170 * subflow
3171 */
3172 mptcp_destroy_common(msk, MPTCP_CF_FASTCLOSE);
e9d09bac
PA		 3173 WRITE_ONCE(msk->flags, 0);
3174 msk->cb_flags = 0;
b29fcfb5 3175 msk->recovery = false;
28e5c138
PA		 3176 WRITE_ONCE(msk->can_ack, false);
3177 WRITE_ONCE(msk->fully_established, false);
3178 WRITE_ONCE(msk->rcv_data_fin, false);
3179 WRITE_ONCE(msk->snd_data_fin_enable, false);
3180 WRITE_ONCE(msk->rcv_fastclose, false);
3181 WRITE_ONCE(msk->use_64bit_ack, false);
b29fcfb5
PA		 3182 WRITE_ONCE(msk->csum_enabled, mptcp_is_checksum_enabled(sock_net(sk)));
3183 mptcp_pm_data_reset(msk);
3184 mptcp_ca_reset(sk);
28e5c138 3185 msk->bytes_consumed = 0;
38967f42
PA		 3186 msk->bytes_acked = 0;
3187 msk->bytes_received = 0;
3188 msk->bytes_sent = 0;
3189 msk->bytes_retrans = 0;
013e3179 3190 msk->rcvspace_init = 0;
b29fcfb5 3191
6b9831bf 3192 WRITE_ONCE(sk->sk_shutdown, 0);
b29fcfb5 3193 sk_error_report(sk);
42c556fe 3194 return 0;
18b683bf
PA		 3195}
3196
b0519de8
FW		 3197#if IS_ENABLED(CONFIG_MPTCP_IPV6)
3198static struct ipv6_pinfo *mptcp_inet6_sk(const struct sock *sk)
3199{
3200 unsigned int offset = sizeof(struct mptcp6_sock) - sizeof(struct ipv6_pinfo);
3201
3202 return (struct ipv6_pinfo *)(((u8 *)sk) + offset);
3203}
10048689
DC		 3204
3205static void mptcp_copy_ip6_options(struct sock *newsk, const struct sock *sk)
3206{
3207 const struct ipv6_pinfo *np = inet6_sk(sk);
3208 struct ipv6_txoptions *opt;
3209 struct ipv6_pinfo *newnp;
3210
3211 newnp = inet6_sk(newsk);
3212
3213 rcu_read_lock();
3214 opt = rcu_dereference(np->opt);
3215 if (opt) {
3216 opt = ipv6_dup_options(newsk, opt);
3217 if (!opt)
3218 net_warn_ratelimited("%s: Failed to copy ip6 options\n", __func__);
3219 }
3220 RCU_INIT_POINTER(newnp->opt, opt);
3221 rcu_read_unlock();
3222}
b0519de8
FW		 3223#endif
3224
10048689
DC		 3225static void mptcp_copy_ip_options(struct sock *newsk, const struct sock *sk)
3226{
3227 struct ip_options_rcu *inet_opt, *newopt = NULL;
3228 const struct inet_sock *inet = inet_sk(sk);
3229 struct inet_sock *newinet;
3230
3231 newinet = inet_sk(newsk);
3232
3233 rcu_read_lock();
3234 inet_opt = rcu_dereference(inet->inet_opt);
3235 if (inet_opt) {
3236 newopt = sock_kmalloc(newsk, sizeof(*inet_opt) +
3237 inet_opt->opt.optlen, GFP_ATOMIC);
3238 if (newopt)
3239 memcpy(newopt, inet_opt, sizeof(*inet_opt) +
3240 inet_opt->opt.optlen);
3241 else
3242 net_warn_ratelimited("%s: Failed to copy ip options\n", __func__);
3243 }
3244 RCU_INIT_POINTER(newinet->inet_opt, newopt);
3245 rcu_read_unlock();
3246}
3247
7e8b88ec
PA		 3248struct sock *mptcp_sk_clone_init(const struct sock *sk,
3249 const struct mptcp_options_received *mp_opt,
3250 struct sock *ssk,
3251 struct request_sock *req)
b0519de8 3252{
58b09919 3253 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
b0519de8 3254 struct sock *nsk = sk_clone_lock(sk, GFP_ATOMIC);
e4a0fa47 3255 struct mptcp_subflow_context *subflow;
58b09919 3256 struct mptcp_sock *msk;
b0519de8
FW		 3257
3258 if (!nsk)
3259 return NULL;
3260
3261#if IS_ENABLED(CONFIG_MPTCP_IPV6)
3262 if (nsk->sk_family == AF_INET6)
3263 inet_sk(nsk)->pinet6 = mptcp_inet6_sk(nsk);
3264#endif
3265
58b09919
PA		 3266 __mptcp_init_sock(nsk);
3267
10048689
DC		 3268#if IS_ENABLED(CONFIG_MPTCP_IPV6)
3269 if (nsk->sk_family == AF_INET6)
3270 mptcp_copy_ip6_options(nsk, sk);
3271 else
3272#endif
3273 mptcp_copy_ip_options(nsk, sk);
3274
58b09919 3275 msk = mptcp_sk(nsk);
1c09d7cb 3276 WRITE_ONCE(msk->local_key, subflow_req->local_key);
b9f45543 3277 WRITE_ONCE(msk->token, subflow_req->token);
b6985b9b 3278 msk->in_accept_queue = 1;
b93df08c 3279 WRITE_ONCE(msk->fully_established, false);
74c7dfbe 3280 if (mp_opt->suboptions & OPTION_MPTCP_CSUMREQD)
06251181 3281 WRITE_ONCE(msk->csum_enabled, true);
58b09919 3282
d440a4e2
PA		 3283 WRITE_ONCE(msk->write_seq, subflow_req->idsn + 1);
3284 WRITE_ONCE(msk->snd_nxt, msk->write_seq);
9426ce47 3285 WRITE_ONCE(msk->snd_una, msk->write_seq);
65f5dd4f 3286 WRITE_ONCE(msk->wnd_end, msk->snd_nxt + tcp_sk(ssk)->snd_wnd);
df00b087 3287 msk->setsockopt_seq = mptcp_sk(sk)->setsockopt_seq;
1730b2b2 3288 mptcp_init_sched(msk, mptcp_sk(sk)->sched);
6f8a612a 3289
6f06b4d4
PA		 3290 /* passive msk is created after the first/MPC subflow */
3291 msk->subflow_id = 2;
3292
5e20087d 3293 sock_reset_flag(nsk, SOCK_RCU_FREE);
0c148460 3294 security_inet_csk_clone(nsk, req);
7e8b88ec
PA		 3295
3296 /* this can't race with mptcp_close(), as the msk is
3297 * not yet exposted to user-space
3298 */
c693a851 3299 mptcp_set_state(nsk, TCP_ESTABLISHED);
7e8b88ec
PA		 3300
3301 /* The msk maintain a ref to each subflow in the connections list */
3302 WRITE_ONCE(msk->first, ssk);
e4a0fa47
PA		 3303 subflow = mptcp_subflow_ctx(ssk);
3304 list_add(&subflow->node, &msk->conn_list);
7e8b88ec
PA		 3305 sock_hold(ssk);
3306
3307 /* new mpc subflow takes ownership of the newly
3308 * created mptcp socket
3309 */
3310 mptcp_token_accept(subflow_req, msk);
3311
3312 /* set msk addresses early to ensure mptcp_pm_get_local_id()
3313 * uses the correct data
3314 */
3315 mptcp_copy_inaddrs(nsk, ssk);
8005184f 3316 __mptcp_propagate_sndbuf(nsk, ssk);
7e8b88ec
PA		 3317
3318 mptcp_rcv_space_init(msk, ssk);
e4a0fa47
PA		 3319
3320 if (mp_opt->suboptions & OPTION_MPTCP_MPC_ACK)
3321 __mptcp_subflow_fully_established(msk, subflow, mp_opt);
58b09919
PA		 3322 bh_unlock_sock(nsk);
3323
b6985b9b 3324 /* note: the newly allocated socket refcount is 2 now */
b0519de8
FW		 3325 return nsk;
3326}
3327
a6b118fe
FW		 3328void mptcp_rcv_space_init(struct mptcp_sock *msk, const struct sock *ssk)
3329{
3330 const struct tcp_sock *tp = tcp_sk(ssk);
3331
013e3179 3332 msk->rcvspace_init = 1;
a6b118fe
FW		 3333 msk->rcvq_space.copied = 0;
3334 msk->rcvq_space.rtt_us = 0;
3335
3336 msk->rcvq_space.time = tp->tcp_mstamp;
3337
3338 /* initial rcv_space offering made to peer */
3339 msk->rcvq_space.space = min_t(u32, tp->rcv_wnd,
3340 TCP_INIT_CWND * tp->advmss);
3341 if (msk->rcvq_space.space == 0)
3342 msk->rcvq_space.space = TCP_INIT_CWND * TCP_MSS_DEFAULT;
3343}
3344
c0bf3c6a 3345void mptcp_destroy_common(struct mptcp_sock *msk, unsigned int flags)
5c8c1640 3346{
c0bf3c6a 3347 struct mptcp_subflow_context *subflow, *tmp;
87952603
PA		 3348 struct sock *sk = (struct sock *)msk;
3349
6e628cd3
PA		 3350 __mptcp_clear_xmit(sk);
3351
c0bf3c6a 3352 /* join list will be eventually flushed (with rst) at sock lock release time */
5efbf6f7 3353 mptcp_for_each_subflow_safe(msk, subflow, tmp)
c0bf3c6a
PA		 3354 __mptcp_close_ssk(sk, mptcp_subflow_tcp_sock(subflow), subflow, flags);
3355
87952603 3356 /* move to sk_receive_queue, sk_stream_kill_queues will purge it */
b29fcfb5 3357 mptcp_data_lock(sk);
87952603 3358 skb_queue_splice_tail_init(&msk->receive_queue, &sk->sk_receive_queue);
6511882c 3359 __skb_queue_purge(&sk->sk_receive_queue);
5c8c1640 3360 skb_rbtree_purge(&msk->out_of_order_queue);
b29fcfb5 3361 mptcp_data_unlock(sk);
6511882c
PA		 3362
3363 /* move all the rx fwd alloc into the sk_mem_reclaim_final in
3364 * inet_sock_destruct() will dispose it
3365 */
5e6300e7 3366 sk_forward_alloc_add(sk, msk->rmem_fwd_alloc);
9531e4a8 3367 WRITE_ONCE(msk->rmem_fwd_alloc, 0);
5c8c1640
GT		 3368 mptcp_token_destroy(msk);
3369 mptcp_pm_free_anno_list(msk);
4638de5a 3370 mptcp_free_local_addr_list(msk);
5c8c1640
GT		 3371}
3372
79c0949e
PK		 3373static void mptcp_destroy(struct sock *sk)
3374{
c9fd9c5f
FW		 3375 struct mptcp_sock *msk = mptcp_sk(sk);
3376
39880bd8
PA		 3377 /* allow the following to close even the initial subflow */
3378 msk->free_first = 1;
c0bf3c6a 3379 mptcp_destroy_common(msk, 0);
d027236c 3380 sk_sockets_allocated_dec(sk);
79c0949e
PK		 3381}
3382
6e628cd3
PA		 3383void __mptcp_data_acked(struct sock *sk)
3384{
3385 if (!sock_owned_by_user(sk))
3386 __mptcp_clean_una(sk);
3387 else
e9d09bac 3388 __set_bit(MPTCP_CLEAN_UNA, &mptcp_sk(sk)->cb_flags);
6e628cd3
PA		 3389}
3390
219d0499 3391void __mptcp_check_push(struct sock *sk, struct sock *ssk)
6e628cd3
PA		 3392{
3393 if (!mptcp_send_head(sk))
3394 return;
3395
73a0052a
GT		 3396 if (!sock_owned_by_user(sk))
3397 __mptcp_subflow_push_pending(sk, ssk, false);
3398 else
e9d09bac 3399 __set_bit(MPTCP_PUSH_PENDING, &mptcp_sk(sk)->cb_flags);
6e628cd3
PA		 3400}
3401
e9d09bac
PA		 3402#define MPTCP_FLAGS_PROCESS_CTX_NEED (BIT(MPTCP_PUSH_PENDING) | \
3403 BIT(MPTCP_RETRANSMIT) | \
3404 BIT(MPTCP_FLUSH_JOIN_LIST))
3405
e93da928 3406/* processes deferred events and flush wmem */
14c441b5 3407static void mptcp_release_cb(struct sock *sk)
e9d09bac 3408 __must_hold(&sk->sk_lock.slock)
14c441b5 3409{
e9d09bac
PA		 3410 struct mptcp_sock *msk = mptcp_sk(sk);
3411
c2e6048f 3412 for (;;) {
bdd70eb6 3413 unsigned long flags = (msk->cb_flags & MPTCP_FLAGS_PROCESS_CTX_NEED);
56a666c4
PA		 3414 struct list_head join_list;
3415
c2e6048f
PA		 3416 if (!flags)
3417 break;
3418
56a666c4
PA		 3419 INIT_LIST_HEAD(&join_list);
3420 list_splice_init(&msk->join_list, &join_list);
3421
c2e6048f 3422 /* the following actions acquire the subflow socket lock
6e628cd3
PA		 3423 *
3424 * 1) can't be invoked in atomic scope
3425 * 2) must avoid ABBA deadlock with msk socket spinlock: the RX
3426 * datapath acquires the msk socket spinlock while helding
3427 * the subflow socket lock
3428 */
e9d09bac 3429 msk->cb_flags &= ~flags;
6e628cd3 3430 spin_unlock_bh(&sk->sk_lock.slock);
56a666c4 3431
3e501490 3432 if (flags & BIT(MPTCP_FLUSH_JOIN_LIST))
56a666c4 3433 __mptcp_flush_join_list(sk, &join_list);
2e5de7e0 3434 if (flags & BIT(MPTCP_PUSH_PENDING))
c2e6048f 3435 __mptcp_push_pending(sk, 0);
2d6f5a2b
PA		 3436 if (flags & BIT(MPTCP_RETRANSMIT))
3437 __mptcp_retrans(sk);
c2e6048f
PA		 3438
3439 cond_resched();
6e628cd3
PA		 3440 spin_lock_bh(&sk->sk_lock.slock);
3441 }
c2e6048f 3442
e9d09bac 3443 if (__test_and_clear_bit(MPTCP_CLEAN_UNA, &msk->cb_flags))
417789df 3444 __mptcp_clean_una_wakeup(sk);
17ebf8a4 3445 if (unlikely(msk->cb_flags)) {
4fd19a30 3446 /* be sure to sync the msk state before taking actions
d109a776
PA		 3447 * depending on sk_state (MPTCP_ERROR_REPORT)
3448 * On sk release avoid actions depending on the first subflow
65a569b0 3449 */
4fd19a30
PA		 3450 if (__test_and_clear_bit(MPTCP_SYNC_STATE, &msk->cb_flags) && msk->first)
3451 __mptcp_sync_state(sk, msk->pending_state);
65a569b0
PA		 3452 if (__test_and_clear_bit(MPTCP_ERROR_REPORT, &msk->cb_flags))
3453 __mptcp_error_report(sk);
8005184f
PA		 3454 if (__test_and_clear_bit(MPTCP_SYNC_SNDBUF, &msk->cb_flags))
3455 __mptcp_sync_sndbuf(sk);
65a569b0 3456 }
6e628cd3 3457
87952603 3458 __mptcp_update_rmem(sk);
14c441b5
PA		 3459}
3460
bcd97734
PA		 3461/* MP_JOIN client subflow must wait for 4th ack before sending any data:
3462 * TCP can't schedule delack timer before the subflow is fully established.
3463 * MPTCP uses the delack timer to do 3rd ack retransmissions
3464 */
3465static void schedule_3rdack_retransmission(struct sock *ssk)
3466{
3467 struct inet_connection_sock *icsk = inet_csk(ssk);
3468 struct tcp_sock *tp = tcp_sk(ssk);
3469 unsigned long timeout;
3470
3471 if (mptcp_subflow_ctx(ssk)->fully_established)
3472 return;
3473
3474 /* reschedule with a timeout above RTT, as we must look only for drop */
3475 if (tp->srtt_us)
3476 timeout = usecs_to_jiffies(tp->srtt_us >> (3 - 1));
3477 else
3478 timeout = TCP_TIMEOUT_INIT;
3479 timeout += jiffies;
3480
3481 WARN_ON_ONCE(icsk->icsk_ack.pending & ICSK_ACK_TIMER);
3482 icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
3483 icsk->icsk_ack.timeout = timeout;
3484 sk_reset_timer(ssk, &icsk->icsk_delack_timer, timeout);
3485}
3486
a5efdbce 3487void mptcp_subflow_process_delegated(struct sock *ssk, long status)
b19bc294
PA		 3488{
3489 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
3490 struct sock *sk = subflow->conn;
3491
a5efdbce 3492 if (status & BIT(MPTCP_DELEGATE_SEND)) {
bcd97734
PA		 3493 mptcp_data_lock(sk);
3494 if (!sock_owned_by_user(sk))
73a0052a 3495 __mptcp_subflow_push_pending(sk, ssk, true);
bcd97734 3496 else
e9d09bac 3497 __set_bit(MPTCP_PUSH_PENDING, &mptcp_sk(sk)->cb_flags);
bcd97734 3498 mptcp_data_unlock(sk);
bcd97734 3499 }
8005184f
PA		 3500 if (status & BIT(MPTCP_DELEGATE_SNDBUF)) {
3501 mptcp_data_lock(sk);
3502 if (!sock_owned_by_user(sk))
3503 __mptcp_sync_sndbuf(sk);
3504 else
3505 __set_bit(MPTCP_SYNC_SNDBUF, &mptcp_sk(sk)->cb_flags);
3506 mptcp_data_unlock(sk);
3507 }
a5efdbce 3508 if (status & BIT(MPTCP_DELEGATE_ACK))
bcd97734 3509 schedule_3rdack_retransmission(ssk);
b19bc294
PA		 3510}
3511
2c5ebd00
PA		 3512static int mptcp_hash(struct sock *sk)
3513{
3514 /* should never be called,
3515 * we hash the TCP subflows not the master socket
3516 */
3517 WARN_ON_ONCE(1);
3518 return 0;
3519}
3520
3521static void mptcp_unhash(struct sock *sk)
3522{
3523 /* called from sk_common_release(), but nothing to do here */
3524}
3525
cec37a6e 3526static int mptcp_get_port(struct sock *sk, unsigned short snum)
f870fa0b
MM		 3527{
3528 struct mptcp_sock *msk = mptcp_sk(sk);
f870fa0b 3529
cfb63e50
PA		 3530 pr_debug("msk=%p, ssk=%p", msk, msk->first);
3531 if (WARN_ON_ONCE(!msk->first))
cec37a6e 3532 return -EINVAL;
f870fa0b 3533
cfb63e50 3534 return inet_csk_get_port(msk->first, snum);
cec37a6e 3535}
f870fa0b 3536
cec37a6e
PK		 3537void mptcp_finish_connect(struct sock *ssk)
3538{
3539 struct mptcp_subflow_context *subflow;
3540 struct mptcp_sock *msk;
3541 struct sock *sk;
f870fa0b 3542
cec37a6e 3543 subflow = mptcp_subflow_ctx(ssk);
cec37a6e
PK		 3544 sk = subflow->conn;
3545 msk = mptcp_sk(sk);
3546
648ef4b8
MM		 3547 pr_debug("msk=%p, token=%u", sk, subflow->token);
3548
b3ea6b27 3549 subflow->map_seq = subflow->iasn;
648ef4b8 3550 subflow->map_subflow_seq = 1;
6d0060f6 3551
cec37a6e
PK		 3552 /* the socket is not connected yet, no msk/subflow ops can access/race
3553 * accessing the field below
3554 */
cec37a6e 3555 WRITE_ONCE(msk->local_key, subflow->local_key);
1b1c7a0e 3556
6c714f1b 3557 mptcp_pm_new_connection(msk, ssk, 0);
f870fa0b
MM		 3558}
3559
866f26f2 3560void mptcp_sock_graft(struct sock *sk, struct socket *parent)
cf7da0d6
PK		 3561{
3562 write_lock_bh(&sk->sk_callback_lock);
3563 rcu_assign_pointer(sk->sk_wq, &parent->wq);
3564 sk_set_socket(sk, parent);
3565 sk->sk_uid = SOCK_INODE(parent)->i_uid;
3566 write_unlock_bh(&sk->sk_callback_lock);
3567}
3568
e16163b6 3569bool mptcp_finish_join(struct sock *ssk)
f296234c 3570{
e16163b6 3571 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
f296234c
PK		 3572 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
3573 struct sock *parent = (void *)msk;
3e501490 3574 bool ret = true;
f296234c
PK		 3575
3576 pr_debug("msk=%p, subflow=%p", msk, subflow);
3577
3578 /* mptcp socket already closing? */
dc87efdb
FW		 3579 if (!mptcp_is_fully_established(parent)) {
3580 subflow->reset_reason = MPTCP_RST_EMPTCP;
f296234c 3581 return false;
dc87efdb 3582 }
f296234c 3583
55b47ca7
PA		 3584 /* active subflow, already present inside the conn_list */
3585 if (!list_empty(&subflow->node)) {
3586 mptcp_subflow_joined(msk, ssk);
8005184f 3587 mptcp_propagate_sndbuf(parent, ssk);
55b47ca7
PA		 3588 return true;
3589 }
f296234c 3590
3e501490
PA		 3591 if (!mptcp_pm_allow_new_subflow(msk))
3592 goto err_prohibited;
10f6d46c 3593
55b47ca7 3594 /* If we can't acquire msk socket lock here, let the release callback
3e501490 3595 * handle it
10f6d46c 3596 */
3e501490
PA		 3597 mptcp_data_lock(parent);
3598 if (!sock_owned_by_user(parent)) {
3599 ret = __mptcp_finish_join(msk, ssk);
3600 if (ret) {
3601 sock_hold(ssk);
3602 list_add_tail(&subflow->node, &msk->conn_list);
3603 }
3604 } else {
e16163b6 3605 sock_hold(ssk);
3e501490 3606 list_add_tail(&subflow->node, &msk->join_list);
e9d09bac 3607 __set_bit(MPTCP_FLUSH_JOIN_LIST, &msk->cb_flags);
e16163b6 3608 }
3e501490
PA		 3609 mptcp_data_unlock(parent);
3610
dc87efdb 3611 if (!ret) {
3e501490 3612err_prohibited:
dc87efdb 3613 subflow->reset_reason = MPTCP_RST_EPROHIBIT;
10f6d46c 3614 return false;
dc87efdb 3615 }
10f6d46c 3616
10f6d46c 3617 return true;
f296234c
PK		 3618}
3619
76e2a55d
PA		 3620static void mptcp_shutdown(struct sock *sk, int how)
3621{
3622 pr_debug("sk=%p, how=%d", sk, how);
3623
3624 if ((how & SEND_SHUTDOWN) && mptcp_close_state(sk))
3625 __mptcp_wr_shutdown(sk);
3626}
3627
6511882c
PA		 3628static int mptcp_forward_alloc_get(const struct sock *sk)
3629{
9531e4a8
ED		 3630 return READ_ONCE(sk->sk_forward_alloc) +
3631 READ_ONCE(mptcp_sk(sk)->rmem_fwd_alloc);
6511882c
PA		 3632}
3633
644807e3
FW		 3634static int mptcp_ioctl_outq(const struct mptcp_sock *msk, u64 v)
3635{
3636 const struct sock *sk = (void *)msk;
3637 u64 delta;
3638
3639 if (sk->sk_state == TCP_LISTEN)
3640 return -EINVAL;
3641
3642 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
3643 return 0;
3644
3645 delta = msk->write_seq - v;
07c2c7a3
PA		 3646 if (__mptcp_check_fallback(msk) && msk->first) {
3647 struct tcp_sock *tp = tcp_sk(msk->first);
3648
3649 /* the first subflow is disconnected after close - see
3650 * __mptcp_close_ssk(). tcp_disconnect() moves the write_seq
3651 * so ignore that status, too.
3652 */
3653 if (!((1 << msk->first->sk_state) &
3654 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE)))
3655 delta += READ_ONCE(tp->write_seq) - tp->snd_una;
3656 }
644807e3
FW		 3657 if (delta > INT_MAX)
3658 delta = INT_MAX;
3659
3660 return (int)delta;
3661}
3662
e1d001fa 3663static int mptcp_ioctl(struct sock *sk, int cmd, int *karg)
644807e3
FW		 3664{
3665 struct mptcp_sock *msk = mptcp_sk(sk);
3666 bool slow;
644807e3
FW		 3667
3668 switch (cmd) {
3669 case SIOCINQ:
3670 if (sk->sk_state == TCP_LISTEN)
3671 return -EINVAL;
3672
3673 lock_sock(sk);
3674 __mptcp_move_skbs(msk);
e1d001fa 3675 *karg = mptcp_inq_hint(sk);
644807e3
FW		 3676 release_sock(sk);
3677 break;
3678 case SIOCOUTQ:
3679 slow = lock_sock_fast(sk);
e1d001fa 3680 *karg = mptcp_ioctl_outq(msk, READ_ONCE(msk->snd_una));
644807e3
FW		 3681 unlock_sock_fast(sk, slow);
3682 break;
3683 case SIOCOUTQNSD:
3684 slow = lock_sock_fast(sk);
e1d001fa 3685 *karg = mptcp_ioctl_outq(msk, msk->snd_nxt);
644807e3
FW		 3686 unlock_sock_fast(sk, slow);
3687 break;
3688 default:
3689 return -ENOIOCTLCMD;
3690 }
3691
e1d001fa 3692 return 0;
644807e3
FW		 3693}
3694
54f1944e
PA		 3695static void mptcp_subflow_early_fallback(struct mptcp_sock *msk,
3696 struct mptcp_subflow_context *subflow)
3697{
3698 subflow->request_mptcp = 0;
3699 __mptcp_do_fallback(msk);
3700}
3701
3702static int mptcp_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
3703{
3704 struct mptcp_subflow_context *subflow;
3705 struct mptcp_sock *msk = mptcp_sk(sk);
54f1944e 3706 int err = -EINVAL;
ccae357c 3707 struct sock *ssk;
54f1944e 3708
3f326a82
PA		 3709 ssk = __mptcp_nmpc_sk(msk);
3710 if (IS_ERR(ssk))
3711 return PTR_ERR(ssk);
54f1944e 3712
c693a851 3713 mptcp_set_state(sk, TCP_SYN_SENT);
ccae357c 3714 subflow = mptcp_subflow_ctx(ssk);
54f1944e
PA		 3715#ifdef CONFIG_TCP_MD5SIG
3716 /* no MPTCP if MD5SIG is enabled on this socket or we may run out of
3717 * TCP option space.
3718 */
ccae357c 3719 if (rcu_access_pointer(tcp_sk(ssk)->md5sig_info))
54f1944e
PA		 3720 mptcp_subflow_early_fallback(msk, subflow);
3721#endif
ccae357c
PA		 3722 if (subflow->request_mptcp && mptcp_token_new_connect(ssk)) {
3723 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_TOKENFALLBACKINIT);
54f1944e
PA		 3724 mptcp_subflow_early_fallback(msk, subflow);
3725 }
3726 if (likely(!__mptcp_check_fallback(msk)))
3727 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPCAPABLEACTIVE);
3728
3729 /* if reaching here via the fastopen/sendmsg path, the caller already
3730 * acquired the subflow socket lock, too.
3731 */
ccae357c
PA		 3732 if (!msk->fastopening)
3733 lock_sock(ssk);
3734
3735 /* the following mirrors closely a very small chunk of code from
3736 * __inet_stream_connect()
3737 */
3738 if (ssk->sk_state != TCP_CLOSE)
3739 goto out;
3740
3741 if (BPF_CGROUP_PRE_CONNECT_ENABLED(ssk)) {
3742 err = ssk->sk_prot->pre_connect(ssk, uaddr, addr_len);
3743 if (err)
3744 goto out;
3745 }
3746
3747 err = ssk->sk_prot->connect(ssk, uaddr, addr_len);
3748 if (err < 0)
3749 goto out;
3750
08e39c0d 3751 inet_assign_bit(DEFER_CONNECT, sk, inet_test_bit(DEFER_CONNECT, ssk));
ccae357c
PA		 3752
3753out:
3754 if (!msk->fastopening)
3755 release_sock(ssk);
54f1944e
PA		 3756
3757 /* on successful connect, the msk state will be moved to established by
3758 * subflow_finish_connect()
3759 */
ccae357c 3760 if (unlikely(err)) {
131a6277
PA		 3761 /* avoid leaving a dangling token in an unconnected socket */
3762 mptcp_token_destroy(msk);
c693a851 3763 mptcp_set_state(sk, TCP_CLOSE);
54f1944e
PA		 3764 return err;
3765 }
3766
ccae357c 3767 mptcp_copy_inaddrs(sk, ssk);
786fc124 3768 return 0;
54f1944e
PA		 3769}
3770
f870fa0b
MM		 3771static struct proto mptcp_prot = {
3772 .name = "MPTCP",
3773 .owner = THIS_MODULE,
3774 .init = mptcp_init_sock,
54f1944e 3775 .connect = mptcp_connect,
18b683bf 3776 .disconnect = mptcp_disconnect,
f870fa0b 3777 .close = mptcp_close,
717e79c8
PK		 3778 .setsockopt = mptcp_setsockopt,
3779 .getsockopt = mptcp_getsockopt,
76e2a55d 3780 .shutdown = mptcp_shutdown,
79c0949e 3781 .destroy = mptcp_destroy,
f870fa0b 3782 .sendmsg = mptcp_sendmsg,
644807e3 3783 .ioctl = mptcp_ioctl,
f870fa0b 3784 .recvmsg = mptcp_recvmsg,
14c441b5 3785 .release_cb = mptcp_release_cb,
2c5ebd00
PA		 3786 .hash = mptcp_hash,
3787 .unhash = mptcp_unhash,
cec37a6e 3788 .get_port = mptcp_get_port,
6511882c 3789 .forward_alloc_get = mptcp_forward_alloc_get,
29b5e5ef 3790 .stream_memory_free = mptcp_stream_memory_free,
d027236c 3791 .sockets_allocated = &mptcp_sockets_allocated,
0defbb0a 3792
d027236c 3793 .memory_allocated = &tcp_memory_allocated,
0defbb0a
ED		 3794 .per_cpu_fw_alloc = &tcp_memory_per_cpu_fw_alloc,
3795
d027236c 3796 .memory_pressure = &tcp_memory_pressure,
d027236c 3797 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_tcp_wmem),
989ef49b 3798 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_tcp_rmem),
d027236c 3799 .sysctl_mem = sysctl_tcp_mem,
f870fa0b 3800 .obj_size = sizeof(struct mptcp_sock),
2c5ebd00 3801 .slab_flags = SLAB_TYPESAFE_BY_RCU,
f870fa0b
MM		 3802 .no_autobind = true,
3803};
3804
2303f994
PK		 3805static int mptcp_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3806{
3807 struct mptcp_sock *msk = mptcp_sk(sock->sk);
8cf2ebdc 3808 struct sock *ssk, *sk = sock->sk;
8cf2ebdc 3809 int err = -EINVAL;
2303f994 3810
8cf2ebdc 3811 lock_sock(sk);
3f326a82
PA		 3812 ssk = __mptcp_nmpc_sk(msk);
3813 if (IS_ERR(ssk)) {
3814 err = PTR_ERR(ssk);
2303f994
PK		 3815 goto unlock;
3816 }
3817
8cf2ebdc
PA		 3818 if (sk->sk_family == AF_INET)
3819 err = inet_bind_sk(ssk, uaddr, addr_len);
3820#if IS_ENABLED(CONFIG_MPTCP_IPV6)
3821 else if (sk->sk_family == AF_INET6)
3822 err = inet6_bind_sk(ssk, uaddr, addr_len);
3823#endif
cf7da0d6 3824 if (!err)
8cf2ebdc 3825 mptcp_copy_inaddrs(sk, ssk);
2303f994
PK		 3826
3827unlock:
8cf2ebdc 3828 release_sock(sk);
2303f994
PK		 3829 return err;
3830}
3831
cf7da0d6
PK		 3832static int mptcp_listen(struct socket *sock, int backlog)
3833{
3834 struct mptcp_sock *msk = mptcp_sk(sock->sk);
cfdcfeed 3835 struct sock *sk = sock->sk;
40f56d0c 3836 struct sock *ssk;
cf7da0d6
PK		 3837 int err;
3838
3839 pr_debug("msk=%p", msk);
3840
cfdcfeed 3841 lock_sock(sk);
0226436a
PA		 3842
3843 err = -EINVAL;
3844 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
3845 goto unlock;
3846
3f326a82
PA		 3847 ssk = __mptcp_nmpc_sk(msk);
3848 if (IS_ERR(ssk)) {
3849 err = PTR_ERR(ssk);
cf7da0d6
PK		 3850 goto unlock;
3851 }
3852
c693a851 3853 mptcp_set_state(sk, TCP_LISTEN);
cfdcfeed 3854 sock_set_flag(sk, SOCK_RCU_FREE);
5e20087d 3855
40f56d0c
PA		 3856 lock_sock(ssk);
3857 err = __inet_listen_sk(ssk, backlog);
3858 release_sock(ssk);
c693a851 3859 mptcp_set_state(sk, inet_sk_state_load(ssk));
40f56d0c 3860
c558246e
MD		 3861 if (!err) {
3862 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
40f56d0c
PA		 3863 mptcp_copy_inaddrs(sk, ssk);
3864 mptcp_event_pm_listener(ssk, MPTCP_EVENT_LISTENER_CREATED);
c558246e 3865 }
cf7da0d6
PK		 3866
3867unlock:
cfdcfeed 3868 release_sock(sk);
cf7da0d6
PK		 3869 return err;
3870}
3871
cf7da0d6
PK		 3872static int mptcp_stream_accept(struct socket *sock, struct socket *newsock,
3873 int flags, bool kern)
3874{
3875 struct mptcp_sock *msk = mptcp_sk(sock->sk);
1f6610b9 3876 struct sock *ssk, *newsk;
cf7da0d6
PK		 3877 int err;
3878
3879 pr_debug("msk=%p", msk);
3880
5b825727 3881 /* Buggy applications can call accept on socket states other then LISTEN
61761231
PA		 3882 * but no need to allocate the first subflow just to error out.
3883 */
1f6610b9
PA		 3884 ssk = READ_ONCE(msk->first);
3885 if (!ssk)
71ba088c 3886 return -EINVAL;
cf7da0d6 3887
8e2b8a9f
DC		 3888 pr_debug("ssk=%p, listener=%p", ssk, mptcp_subflow_ctx(ssk));
3889 newsk = inet_csk_accept(ssk, flags, &err, kern);
e76c8ef5
PA		 3890 if (!newsk)
3891 return err;
3892
8e2b8a9f
DC		 3893 pr_debug("newsk=%p, subflow is mptcp=%d", newsk, sk_is_mptcp(newsk));
3894 if (sk_is_mptcp(newsk)) {
cf7da0d6 3895 struct mptcp_subflow_context *subflow;
8e2b8a9f
DC		 3896 struct sock *new_mptcp_sock;
3897
3898 subflow = mptcp_subflow_ctx(newsk);
3899 new_mptcp_sock = subflow->conn;
3900
3901 /* is_mptcp should be false if subflow->conn is missing, see
3902 * subflow_syn_recv_sock()
3903 */
3904 if (WARN_ON_ONCE(!new_mptcp_sock)) {
3905 tcp_sk(newsk)->is_mptcp = 0;
3906 goto tcpfallback;
3907 }
3908
3909 newsk = new_mptcp_sock;
3910 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_MPCAPABLEPASSIVEACK);
3911
3912 newsk->sk_kern_sock = kern;
3913 lock_sock(newsk);
3914 __inet_accept(sock, newsock, newsk);
0397c6d8 3915
a5ef058d 3916 set_bit(SOCK_CUSTOM_SOCKOPT, &newsock->flags);
8e2b8a9f 3917 msk = mptcp_sk(newsk);
b6985b9b 3918 msk->in_accept_queue = 0;
a5ef058d 3919
cf7da0d6
PK		 3920 /* set ssk->sk_socket of accept()ed flows to mptcp socket.
3921 * This is needed so NOSPACE flag can be set from tcp stack.
3922 */
190f8b06 3923 mptcp_for_each_subflow(msk, subflow) {
cf7da0d6
PK		 3924 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
3925
3926 if (!ssk->sk_socket)
3927 mptcp_sock_graft(ssk, newsock);
3928 }
63740448
PA		 3929
3930 /* Do late cleanup for the first subflow as necessary. Also
3931 * deal with bad peers not doing a complete shutdown.
3932 */
39880bd8 3933 if (unlikely(inet_sk_state_load(msk->first) == TCP_CLOSE)) {
63740448
PA		 3934 __mptcp_close_ssk(newsk, msk->first,
3935 mptcp_subflow_ctx(msk->first), 0);
39880bd8 3936 if (unlikely(list_is_singular(&msk->conn_list)))
c693a851 3937 mptcp_set_state(newsk, TCP_CLOSE);
63740448 3938 }
8e2b8a9f
DC		 3939 } else {
3940 MPTCP_INC_STATS(sock_net(ssk),
3941 MPTCP_MIB_MPCAPABLEPASSIVEFALLBACK);
3942tcpfallback:
3943 newsk->sk_kern_sock = kern;
3944 lock_sock(newsk);
3945 __inet_accept(sock, newsock, newsk);
3946 /* we are being invoked after accepting a non-mp-capable
3947 * flow: sk is a tcp_sk, not an mptcp one.
3948 *
3949 * Hand the socket over to tcp so all further socket ops
3950 * bypass mptcp.
3951 */
3952 WRITE_ONCE(newsock->sk->sk_socket->ops,
3953 mptcp_fallback_tcp_ops(newsock->sk));
cf7da0d6 3954 }
e76c8ef5 3955 release_sock(newsk);
cf7da0d6 3956
e76c8ef5 3957 return 0;
cf7da0d6
PK		 3958}
3959
8edf0864
FW		 3960static __poll_t mptcp_check_writeable(struct mptcp_sock *msk)
3961{
3962 struct sock *sk = (struct sock *)msk;
8edf0864 3963
29b5e5ef 3964 if (__mptcp_stream_is_writeable(sk, 1))
8edf0864
FW		 3965 return EPOLLOUT | EPOLLWRNORM;
3966
037db6ea
PA		 3967 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
3968 smp_mb__after_atomic(); /* NOSPACE is changed by mptcp_write_space() */
29b5e5ef 3969 if (__mptcp_stream_is_writeable(sk, 1))
6e628cd3 3970 return EPOLLOUT | EPOLLWRNORM;
8edf0864 3971
6e628cd3 3972 return 0;
8edf0864
FW		 3973}
3974
2303f994
PK		 3975static __poll_t mptcp_poll(struct file *file, struct socket *sock,
3976 struct poll_table_struct *wait)
3977{
1891c4a0 3978 struct sock *sk = sock->sk;
8ab183de 3979 struct mptcp_sock *msk;
2303f994 3980 __poll_t mask = 0;
6b9831bf 3981 u8 shutdown;
8a05661b 3982 int state;
2303f994 3983
1891c4a0 3984 msk = mptcp_sk(sk);
1891c4a0 3985 sock_poll_wait(file, sock, wait);
1891c4a0 3986
8a05661b 3987 state = inet_sk_state_load(sk);
6719331c 3988 pr_debug("msk=%p state=%d flags=%lx", msk, state, msk->flags);
71ba088c 3989 if (state == TCP_LISTEN) {
5426a4ef 3990 struct sock *ssk = READ_ONCE(msk->first);
5b825727 3991
5426a4ef 3992 if (WARN_ON_ONCE(!ssk))
71ba088c
PA		 3993 return 0;
3994
5426a4ef 3995 return inet_csk_listen_poll(ssk);
71ba088c 3996 }
8a05661b 3997
6b9831bf
PA		 3998 shutdown = READ_ONCE(sk->sk_shutdown);
3999 if (shutdown == SHUTDOWN_MASK || state == TCP_CLOSE)
4000 mask |= EPOLLHUP;
4001 if (shutdown & RCV_SHUTDOWN)
4002 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
4003
8a05661b 4004 if (state != TCP_SYN_SENT && state != TCP_SYN_RECV) {
5684ab1a 4005 mask |= mptcp_check_readable(sk);
6b9831bf
PA		 4006 if (shutdown & SEND_SHUTDOWN)
4007 mask |= EPOLLOUT | EPOLLWRNORM;
4008 else
4009 mask |= mptcp_check_writeable(msk);
08e39c0d
ED		 4010 } else if (state == TCP_SYN_SENT &&
4011 inet_test_bit(DEFER_CONNECT, sk)) {
a42cf9d1
BH		 4012 /* cf tcp_poll() note about TFO */
4013 mask |= EPOLLOUT | EPOLLWRNORM;
8a05661b 4014 }
1891c4a0 4015
d89e3ed7 4016 /* This barrier is coupled with smp_wmb() in __mptcp_error_report() */
15cc1045 4017 smp_rmb();
9ae8e5ad 4018 if (READ_ONCE(sk->sk_err))
15cc1045
PA		 4019 mask |= EPOLLERR;
4020
2303f994
PK		 4021 return mask;
4022}
4023
e42f1ac6
FW		 4024static const struct proto_ops mptcp_stream_ops = {
4025 .family = PF_INET,
4026 .owner = THIS_MODULE,
0a3cc579 4027 .release = inet_release,
e42f1ac6 4028 .bind = mptcp_bind,
786fc124 4029 .connect = inet_stream_connect,
e42f1ac6
FW		 4030 .socketpair = sock_no_socketpair,
4031 .accept = mptcp_stream_accept,
d2f77c53 4032 .getname = inet_getname,
e42f1ac6
FW		 4033 .poll = mptcp_poll,
4034 .ioctl = inet_ioctl,
4035 .gettstamp = sock_gettstamp,
4036 .listen = mptcp_listen,
76e2a55d 4037 .shutdown = inet_shutdown,
e42f1ac6
FW		 4038 .setsockopt = sock_common_setsockopt,
4039 .getsockopt = sock_common_getsockopt,
4040 .sendmsg = inet_sendmsg,
4041 .recvmsg = inet_recvmsg,
4042 .mmap = sock_no_mmap,
5684ab1a 4043 .set_rcvlowat = mptcp_set_rcvlowat,
e42f1ac6 4044};
2303f994 4045
f870fa0b
MM		 4046static struct inet_protosw mptcp_protosw = {
4047 .type = SOCK_STREAM,
4048 .protocol = IPPROTO_MPTCP,
4049 .prot = &mptcp_prot,
2303f994
PK		 4050 .ops = &mptcp_stream_ops,
4051 .flags = INET_PROTOSW_ICSK,
f870fa0b
MM		 4052};
4053
b19bc294
PA		 4054static int mptcp_napi_poll(struct napi_struct *napi, int budget)
4055{
4056 struct mptcp_delegated_action *delegated;
4057 struct mptcp_subflow_context *subflow;
4058 int work_done = 0;
4059
4060 delegated = container_of(napi, struct mptcp_delegated_action, napi);
4061 while ((subflow = mptcp_subflow_delegated_next(delegated)) != NULL) {
4062 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
4063
4064 bh_lock_sock_nested(ssk);
a5efdbce
PA		 4065 if (!sock_owned_by_user(ssk)) {
4066 mptcp_subflow_process_delegated(ssk, xchg(&subflow->delegated_status, 0));
4067 } else {
4068 /* tcp_release_cb_override already processed
4069 * the action or will do at next release_sock().
4070 * In both case must dequeue the subflow here - on the same
4071 * CPU that scheduled it.
4072 */
4073 smp_wmb();
4074 clear_bit(MPTCP_DELEGATE_SCHEDULED, &subflow->delegated_status);
4075 }
b19bc294
PA		 4076 bh_unlock_sock(ssk);
4077 sock_put(ssk);
4078
4079 if (++work_done == budget)
4080 return budget;
4081 }
4082
4083 /* always provide a 0 'work_done' argument, so that napi_complete_done
4084 * will not try accessing the NULL napi->dev ptr
4085 */
4086 napi_complete_done(napi, 0);
4087 return work_done;
4088}
4089
d39dceca 4090void __init mptcp_proto_init(void)
f870fa0b 4091{
b19bc294
PA		 4092 struct mptcp_delegated_action *delegated;
4093 int cpu;
4094
2303f994 4095 mptcp_prot.h.hashinfo = tcp_prot.h.hashinfo;
2303f994 4096
d027236c
PA		 4097 if (percpu_counter_init(&mptcp_sockets_allocated, 0, GFP_KERNEL))
4098 panic("Failed to allocate MPTCP pcpu counter\n");
4099
b19bc294
PA		 4100 init_dummy_netdev(&mptcp_napi_dev);
4101 for_each_possible_cpu(cpu) {
4102 delegated = per_cpu_ptr(&mptcp_delegated_actions, cpu);
4103 INIT_LIST_HEAD(&delegated->head);
16d083e2
JK		 4104 netif_napi_add_tx(&mptcp_napi_dev, &delegated->napi,
4105 mptcp_napi_poll);
b19bc294
PA		 4106 napi_enable(&delegated->napi);
4107 }
4108
2303f994 4109 mptcp_subflow_init();
1b1c7a0e 4110 mptcp_pm_init();
ed1ad86b 4111 mptcp_sched_init();
2c5ebd00 4112 mptcp_token_init();
2303f994 4113
f870fa0b
MM		 4114 if (proto_register(&mptcp_prot, 1) != 0)
4115 panic("Failed to register MPTCP proto.\n");
4116
4117 inet_register_protosw(&mptcp_protosw);
6771bfd9
FW		 4118
4119 BUILD_BUG_ON(sizeof(struct mptcp_skb_cb) > sizeof_field(struct sk_buff, cb));
f870fa0b
MM		 4120}
4121
4122#if IS_ENABLED(CONFIG_MPTCP_IPV6)
e42f1ac6
FW		 4123static const struct proto_ops mptcp_v6_stream_ops = {
4124 .family = PF_INET6,
4125 .owner = THIS_MODULE,
0a3cc579 4126 .release = inet6_release,
e42f1ac6 4127 .bind = mptcp_bind,
786fc124 4128 .connect = inet_stream_connect,
e42f1ac6
FW		 4129 .socketpair = sock_no_socketpair,
4130 .accept = mptcp_stream_accept,
d2f77c53 4131 .getname = inet6_getname,
e42f1ac6
FW		 4132 .poll = mptcp_poll,
4133 .ioctl = inet6_ioctl,
4134 .gettstamp = sock_gettstamp,
4135 .listen = mptcp_listen,
76e2a55d 4136 .shutdown = inet_shutdown,
e42f1ac6
FW		 4137 .setsockopt = sock_common_setsockopt,
4138 .getsockopt = sock_common_getsockopt,
4139 .sendmsg = inet6_sendmsg,
4140 .recvmsg = inet6_recvmsg,
4141 .mmap = sock_no_mmap,
e42f1ac6 4142#ifdef CONFIG_COMPAT
3986912f 4143 .compat_ioctl = inet6_compat_ioctl,
e42f1ac6 4144#endif
5684ab1a 4145 .set_rcvlowat = mptcp_set_rcvlowat,
e42f1ac6
FW		 4146};
4147
f870fa0b
MM		 4148static struct proto mptcp_v6_prot;
4149
4150static struct inet_protosw mptcp_v6_protosw = {
4151 .type = SOCK_STREAM,
4152 .protocol = IPPROTO_MPTCP,
4153 .prot = &mptcp_v6_prot,
2303f994 4154 .ops = &mptcp_v6_stream_ops,
f870fa0b
MM		 4155 .flags = INET_PROTOSW_ICSK,
4156};
4157
d39dceca 4158int __init mptcp_proto_v6_init(void)
f870fa0b
MM		 4159{
4160 int err;
4161
4162 mptcp_v6_prot = mptcp_prot;
4163 strcpy(mptcp_v6_prot.name, "MPTCPv6");
4164 mptcp_v6_prot.slab = NULL;
b0519de8 4165 mptcp_v6_prot.obj_size = sizeof(struct mptcp6_sock);
f5f80e32 4166 mptcp_v6_prot.ipv6_pinfo_offset = offsetof(struct mptcp6_sock, np);
f870fa0b
MM		 4167
4168 err = proto_register(&mptcp_v6_prot, 1);
4169 if (err)
4170 return err;
4171
4172 err = inet6_register_protosw(&mptcp_v6_protosw);
4173 if (err)
4174 proto_unregister(&mptcp_v6_prot);
4175
4176 return err;
4177}
4178#endif
Linux 6.x block layer and io_uring tree(s)