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>
12 #include <linux/sched/signal.h>
13 #include <linux/atomic.h>
14 #include <net/sock.h>
15 #include <net/inet_common.h>
16 #include <net/inet_hashtables.h>
17 #include <net/protocol.h>
18 #include <net/tcp.h>
19 #include <net/tcp_states.h>
20 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
21 #include <net/transp_v6.h>
22 #endif
23 #include <net/mptcp.h>
24 #include <net/xfrm.h>
25 #include <asm/ioctls.h>
26 #include "protocol.h"
27 #include "mib.h"
28
29 #define CREATE_TRACE_POINTS
30 #include <trace/events/mptcp.h>
31
32 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
33 struct mptcp6_sock {
34 struct mptcp_sock msk;
35 struct ipv6_pinfo np;
36 };
37 #endif
38
39 enum {
40 MPTCP_CMSG_TS = BIT(0),
41 MPTCP_CMSG_INQ = BIT(1),
42 };
43
44 static struct percpu_counter mptcp_sockets_allocated ____cacheline_aligned_in_smp;
45
46 static void __mptcp_destroy_sock(struct sock *sk);
47 static void mptcp_check_send_data_fin(struct sock *sk);
48
49 DEFINE_PER_CPU(struct mptcp_delegated_action, mptcp_delegated_actions);
50 static struct net_device mptcp_napi_dev;
51
52 /* Returns end sequence number of the receiver's advertised window */
mptcp_wnd_end(const struct mptcp_sock * msk)53 static u64 mptcp_wnd_end(const struct mptcp_sock *msk)
54 {
55 return READ_ONCE(msk->wnd_end);
56 }
57
mptcp_fallback_tcp_ops(const struct sock * sk)58 static const struct proto_ops *mptcp_fallback_tcp_ops(const struct sock *sk)
59 {
60 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
61 if (sk->sk_prot == &tcpv6_prot)
62 return &inet6_stream_ops;
63 #endif
64 WARN_ON_ONCE(sk->sk_prot != &tcp_prot);
65 return &inet_stream_ops;
66 }
67
__mptcp_socket_create(struct mptcp_sock * msk)68 static int __mptcp_socket_create(struct mptcp_sock *msk)
69 {
70 struct mptcp_subflow_context *subflow;
71 struct sock *sk = (struct sock *)msk;
72 struct socket *ssock;
73 int err;
74
75 err = mptcp_subflow_create_socket(sk, sk->sk_family, &ssock);
76 if (err)
77 return err;
78
79 msk->scaling_ratio = tcp_sk(ssock->sk)->scaling_ratio;
80 WRITE_ONCE(msk->first, ssock->sk);
81 subflow = mptcp_subflow_ctx(ssock->sk);
82 list_add(&subflow->node, &msk->conn_list);
83 sock_hold(ssock->sk);
84 subflow->request_mptcp = 1;
85 subflow->subflow_id = msk->subflow_id++;
86
87 /* This is the first subflow, always with id 0 */
88 WRITE_ONCE(subflow->local_id, 0);
89 mptcp_sock_graft(msk->first, sk->sk_socket);
90 iput(SOCK_INODE(ssock));
91
92 return 0;
93 }
94
95 /* If the MPC handshake is not started, returns the first subflow,
96 * eventually allocating it.
97 */
__mptcp_nmpc_sk(struct mptcp_sock * msk)98 struct sock *__mptcp_nmpc_sk(struct mptcp_sock *msk)
99 {
100 struct sock *sk = (struct sock *)msk;
101 int ret;
102
103 if (!((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)))
104 return ERR_PTR(-EINVAL);
105
106 if (!msk->first) {
107 ret = __mptcp_socket_create(msk);
108 if (ret)
109 return ERR_PTR(ret);
110
111 mptcp_sockopt_sync(msk, msk->first);
112 }
113
114 return msk->first;
115 }
116
mptcp_drop(struct sock * sk,struct sk_buff * skb)117 static void mptcp_drop(struct sock *sk, struct sk_buff *skb)
118 {
119 sk_drops_add(sk, skb);
120 __kfree_skb(skb);
121 }
122
mptcp_rmem_fwd_alloc_add(struct sock * sk,int size)123 static void mptcp_rmem_fwd_alloc_add(struct sock *sk, int size)
124 {
125 WRITE_ONCE(mptcp_sk(sk)->rmem_fwd_alloc,
126 mptcp_sk(sk)->rmem_fwd_alloc + size);
127 }
128
mptcp_rmem_charge(struct sock * sk,int size)129 static void mptcp_rmem_charge(struct sock *sk, int size)
130 {
131 mptcp_rmem_fwd_alloc_add(sk, -size);
132 }
133
mptcp_try_coalesce(struct sock * sk,struct sk_buff * to,struct sk_buff * from)134 static bool mptcp_try_coalesce(struct sock *sk, struct sk_buff *to,
135 struct sk_buff *from)
136 {
137 bool fragstolen;
138 int delta;
139
140 if (MPTCP_SKB_CB(from)->offset ||
141 !skb_try_coalesce(to, from, &fragstolen, &delta))
142 return false;
143
144 pr_debug("colesced seq %llx into %llx new len %d new end seq %llx",
145 MPTCP_SKB_CB(from)->map_seq, MPTCP_SKB_CB(to)->map_seq,
146 to->len, MPTCP_SKB_CB(from)->end_seq);
147 MPTCP_SKB_CB(to)->end_seq = MPTCP_SKB_CB(from)->end_seq;
148
149 /* note the fwd memory can reach a negative value after accounting
150 * for the delta, but the later skb free will restore a non
151 * negative one
152 */
153 atomic_add(delta, &sk->sk_rmem_alloc);
154 mptcp_rmem_charge(sk, delta);
155 kfree_skb_partial(from, fragstolen);
156
157 return true;
158 }
159
mptcp_ooo_try_coalesce(struct mptcp_sock * msk,struct sk_buff * to,struct sk_buff * from)160 static bool mptcp_ooo_try_coalesce(struct mptcp_sock *msk, struct sk_buff *to,
161 struct sk_buff *from)
162 {
163 if (MPTCP_SKB_CB(from)->map_seq != MPTCP_SKB_CB(to)->end_seq)
164 return false;
165
166 return mptcp_try_coalesce((struct sock *)msk, to, from);
167 }
168
__mptcp_rmem_reclaim(struct sock * sk,int amount)169 static void __mptcp_rmem_reclaim(struct sock *sk, int amount)
170 {
171 amount >>= PAGE_SHIFT;
172 mptcp_rmem_charge(sk, amount << PAGE_SHIFT);
173 __sk_mem_reduce_allocated(sk, amount);
174 }
175
mptcp_rmem_uncharge(struct sock * sk,int size)176 static void mptcp_rmem_uncharge(struct sock *sk, int size)
177 {
178 struct mptcp_sock *msk = mptcp_sk(sk);
179 int reclaimable;
180
181 mptcp_rmem_fwd_alloc_add(sk, size);
182 reclaimable = msk->rmem_fwd_alloc - sk_unused_reserved_mem(sk);
183
184 /* see sk_mem_uncharge() for the rationale behind the following schema */
185 if (unlikely(reclaimable >= PAGE_SIZE))
186 __mptcp_rmem_reclaim(sk, reclaimable);
187 }
188
mptcp_rfree(struct sk_buff * skb)189 static void mptcp_rfree(struct sk_buff *skb)
190 {
191 unsigned int len = skb->truesize;
192 struct sock *sk = skb->sk;
193
194 atomic_sub(len, &sk->sk_rmem_alloc);
195 mptcp_rmem_uncharge(sk, len);
196 }
197
mptcp_set_owner_r(struct sk_buff * skb,struct sock * sk)198 void mptcp_set_owner_r(struct sk_buff *skb, struct sock *sk)
199 {
200 skb_orphan(skb);
201 skb->sk = sk;
202 skb->destructor = mptcp_rfree;
203 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
204 mptcp_rmem_charge(sk, skb->truesize);
205 }
206
207 /* "inspired" by tcp_data_queue_ofo(), main differences:
208 * - use mptcp seqs
209 * - don't cope with sacks
210 */
mptcp_data_queue_ofo(struct mptcp_sock * msk,struct sk_buff * skb)211 static void mptcp_data_queue_ofo(struct mptcp_sock *msk, struct sk_buff *skb)
212 {
213 struct sock *sk = (struct sock *)msk;
214 struct rb_node **p, *parent;
215 u64 seq, end_seq, max_seq;
216 struct sk_buff *skb1;
217
218 seq = MPTCP_SKB_CB(skb)->map_seq;
219 end_seq = MPTCP_SKB_CB(skb)->end_seq;
220 max_seq = atomic64_read(&msk->rcv_wnd_sent);
221
222 pr_debug("msk=%p seq=%llx limit=%llx empty=%d", msk, seq, max_seq,
223 RB_EMPTY_ROOT(&msk->out_of_order_queue));
224 if (after64(end_seq, max_seq)) {
225 /* out of window */
226 mptcp_drop(sk, skb);
227 pr_debug("oow by %lld, rcv_wnd_sent %llu\n",
228 (unsigned long long)end_seq - (unsigned long)max_seq,
229 (unsigned long long)atomic64_read(&msk->rcv_wnd_sent));
230 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_NODSSWINDOW);
231 return;
232 }
233
234 p = &msk->out_of_order_queue.rb_node;
235 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOQUEUE);
236 if (RB_EMPTY_ROOT(&msk->out_of_order_queue)) {
237 rb_link_node(&skb->rbnode, NULL, p);
238 rb_insert_color(&skb->rbnode, &msk->out_of_order_queue);
239 msk->ooo_last_skb = skb;
240 goto end;
241 }
242
243 /* with 2 subflows, adding at end of ooo queue is quite likely
244 * Use of ooo_last_skb avoids the O(Log(N)) rbtree lookup.
245 */
246 if (mptcp_ooo_try_coalesce(msk, msk->ooo_last_skb, skb)) {
247 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOMERGE);
248 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOQUEUETAIL);
249 return;
250 }
251
252 /* Can avoid an rbtree lookup if we are adding skb after ooo_last_skb */
253 if (!before64(seq, MPTCP_SKB_CB(msk->ooo_last_skb)->end_seq)) {
254 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOQUEUETAIL);
255 parent = &msk->ooo_last_skb->rbnode;
256 p = &parent->rb_right;
257 goto insert;
258 }
259
260 /* Find place to insert this segment. Handle overlaps on the way. */
261 parent = NULL;
262 while (*p) {
263 parent = *p;
264 skb1 = rb_to_skb(parent);
265 if (before64(seq, MPTCP_SKB_CB(skb1)->map_seq)) {
266 p = &parent->rb_left;
267 continue;
268 }
269 if (before64(seq, MPTCP_SKB_CB(skb1)->end_seq)) {
270 if (!after64(end_seq, MPTCP_SKB_CB(skb1)->end_seq)) {
271 /* All the bits are present. Drop. */
272 mptcp_drop(sk, skb);
273 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
274 return;
275 }
276 if (after64(seq, MPTCP_SKB_CB(skb1)->map_seq)) {
277 /* partial overlap:
278 * | skb |
279 * | skb1 |
280 * continue traversing
281 */
282 } else {
283 /* skb's seq == skb1's seq and skb covers skb1.
284 * Replace skb1 with skb.
285 */
286 rb_replace_node(&skb1->rbnode, &skb->rbnode,
287 &msk->out_of_order_queue);
288 mptcp_drop(sk, skb1);
289 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
290 goto merge_right;
291 }
292 } else if (mptcp_ooo_try_coalesce(msk, skb1, skb)) {
293 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOMERGE);
294 return;
295 }
296 p = &parent->rb_right;
297 }
298
299 insert:
300 /* Insert segment into RB tree. */
301 rb_link_node(&skb->rbnode, parent, p);
302 rb_insert_color(&skb->rbnode, &msk->out_of_order_queue);
303
304 merge_right:
305 /* Remove other segments covered by skb. */
306 while ((skb1 = skb_rb_next(skb)) != NULL) {
307 if (before64(end_seq, MPTCP_SKB_CB(skb1)->end_seq))
308 break;
309 rb_erase(&skb1->rbnode, &msk->out_of_order_queue);
310 mptcp_drop(sk, skb1);
311 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
312 }
313 /* If there is no skb after us, we are the last_skb ! */
314 if (!skb1)
315 msk->ooo_last_skb = skb;
316
317 end:
318 skb_condense(skb);
319 mptcp_set_owner_r(skb, sk);
320 }
321
mptcp_rmem_schedule(struct sock * sk,struct sock * ssk,int size)322 static bool mptcp_rmem_schedule(struct sock *sk, struct sock *ssk, int size)
323 {
324 struct mptcp_sock *msk = mptcp_sk(sk);
325 int amt, amount;
326
327 if (size <= msk->rmem_fwd_alloc)
328 return true;
329
330 size -= msk->rmem_fwd_alloc;
331 amt = sk_mem_pages(size);
332 amount = amt << PAGE_SHIFT;
333 if (!__sk_mem_raise_allocated(sk, size, amt, SK_MEM_RECV))
334 return false;
335
336 mptcp_rmem_fwd_alloc_add(sk, amount);
337 return true;
338 }
339
__mptcp_move_skb(struct mptcp_sock * msk,struct sock * ssk,struct sk_buff * skb,unsigned int offset,size_t copy_len)340 static bool __mptcp_move_skb(struct mptcp_sock *msk, struct sock *ssk,
341 struct sk_buff *skb, unsigned int offset,
342 size_t copy_len)
343 {
344 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
345 struct sock *sk = (struct sock *)msk;
346 struct sk_buff *tail;
347 bool has_rxtstamp;
348
349 __skb_unlink(skb, &ssk->sk_receive_queue);
350
351 skb_ext_reset(skb);
352 skb_orphan(skb);
353
354 /* try to fetch required memory from subflow */
355 if (!mptcp_rmem_schedule(sk, ssk, skb->truesize))
356 goto drop;
357
358 has_rxtstamp = TCP_SKB_CB(skb)->has_rxtstamp;
359
360 /* the skb map_seq accounts for the skb offset:
361 * mptcp_subflow_get_mapped_dsn() is based on the current tp->copied_seq
362 * value
363 */
364 MPTCP_SKB_CB(skb)->map_seq = mptcp_subflow_get_mapped_dsn(subflow);
365 MPTCP_SKB_CB(skb)->end_seq = MPTCP_SKB_CB(skb)->map_seq + copy_len;
366 MPTCP_SKB_CB(skb)->offset = offset;
367 MPTCP_SKB_CB(skb)->has_rxtstamp = has_rxtstamp;
368
369 if (MPTCP_SKB_CB(skb)->map_seq == msk->ack_seq) {
370 /* in sequence */
371 msk->bytes_received += copy_len;
372 WRITE_ONCE(msk->ack_seq, msk->ack_seq + copy_len);
373 tail = skb_peek_tail(&sk->sk_receive_queue);
374 if (tail && mptcp_try_coalesce(sk, tail, skb))
375 return true;
376
377 mptcp_set_owner_r(skb, sk);
378 __skb_queue_tail(&sk->sk_receive_queue, skb);
379 return true;
380 } else if (after64(MPTCP_SKB_CB(skb)->map_seq, msk->ack_seq)) {
381 mptcp_data_queue_ofo(msk, skb);
382 return false;
383 }
384
385 /* old data, keep it simple and drop the whole pkt, sender
386 * will retransmit as needed, if needed.
387 */
388 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
389 drop:
390 mptcp_drop(sk, skb);
391 return false;
392 }
393
mptcp_stop_rtx_timer(struct sock * sk)394 static void mptcp_stop_rtx_timer(struct sock *sk)
395 {
396 struct inet_connection_sock *icsk = inet_csk(sk);
397
398 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
399 mptcp_sk(sk)->timer_ival = 0;
400 }
401
mptcp_close_wake_up(struct sock * sk)402 static void mptcp_close_wake_up(struct sock *sk)
403 {
404 if (sock_flag(sk, SOCK_DEAD))
405 return;
406
407 sk->sk_state_change(sk);
408 if (sk->sk_shutdown == SHUTDOWN_MASK ||
409 sk->sk_state == TCP_CLOSE)
410 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_HUP);
411 else
412 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
413 }
414
mptcp_pending_data_fin_ack(struct sock * sk)415 static bool mptcp_pending_data_fin_ack(struct sock *sk)
416 {
417 struct mptcp_sock *msk = mptcp_sk(sk);
418
419 return ((1 << sk->sk_state) &
420 (TCPF_FIN_WAIT1 | TCPF_CLOSING | TCPF_LAST_ACK)) &&
421 msk->write_seq == READ_ONCE(msk->snd_una);
422 }
423
mptcp_check_data_fin_ack(struct sock * sk)424 static void mptcp_check_data_fin_ack(struct sock *sk)
425 {
426 struct mptcp_sock *msk = mptcp_sk(sk);
427
428 /* Look for an acknowledged DATA_FIN */
429 if (mptcp_pending_data_fin_ack(sk)) {
430 WRITE_ONCE(msk->snd_data_fin_enable, 0);
431
432 switch (sk->sk_state) {
433 case TCP_FIN_WAIT1:
434 mptcp_set_state(sk, TCP_FIN_WAIT2);
435 break;
436 case TCP_CLOSING:
437 case TCP_LAST_ACK:
438 mptcp_set_state(sk, TCP_CLOSE);
439 break;
440 }
441
442 mptcp_close_wake_up(sk);
443 }
444 }
445
mptcp_pending_data_fin(struct sock * sk,u64 * seq)446 static bool mptcp_pending_data_fin(struct sock *sk, u64 *seq)
447 {
448 struct mptcp_sock *msk = mptcp_sk(sk);
449
450 if (READ_ONCE(msk->rcv_data_fin) &&
451 ((1 << sk->sk_state) &
452 (TCPF_ESTABLISHED | TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2))) {
453 u64 rcv_data_fin_seq = READ_ONCE(msk->rcv_data_fin_seq);
454
455 if (msk->ack_seq == rcv_data_fin_seq) {
456 if (seq)
457 *seq = rcv_data_fin_seq;
458
459 return true;
460 }
461 }
462
463 return false;
464 }
465
mptcp_set_datafin_timeout(struct sock * sk)466 static void mptcp_set_datafin_timeout(struct sock *sk)
467 {
468 struct inet_connection_sock *icsk = inet_csk(sk);
469 u32 retransmits;
470
471 retransmits = min_t(u32, icsk->icsk_retransmits,
472 ilog2(TCP_RTO_MAX / TCP_RTO_MIN));
473
474 mptcp_sk(sk)->timer_ival = TCP_RTO_MIN << retransmits;
475 }
476
__mptcp_set_timeout(struct sock * sk,long tout)477 static void __mptcp_set_timeout(struct sock *sk, long tout)
478 {
479 mptcp_sk(sk)->timer_ival = tout > 0 ? tout : TCP_RTO_MIN;
480 }
481
mptcp_timeout_from_subflow(const struct mptcp_subflow_context * subflow)482 static long mptcp_timeout_from_subflow(const struct mptcp_subflow_context *subflow)
483 {
484 const struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
485
486 return inet_csk(ssk)->icsk_pending && !subflow->stale_count ?
487 inet_csk(ssk)->icsk_timeout - jiffies : 0;
488 }
489
mptcp_set_timeout(struct sock * sk)490 static void mptcp_set_timeout(struct sock *sk)
491 {
492 struct mptcp_subflow_context *subflow;
493 long tout = 0;
494
495 mptcp_for_each_subflow(mptcp_sk(sk), subflow)
496 tout = max(tout, mptcp_timeout_from_subflow(subflow));
497 __mptcp_set_timeout(sk, tout);
498 }
499
tcp_can_send_ack(const struct sock * ssk)500 static inline bool tcp_can_send_ack(const struct sock *ssk)
501 {
502 return !((1 << inet_sk_state_load(ssk)) &
503 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_TIME_WAIT | TCPF_CLOSE | TCPF_LISTEN));
504 }
505
__mptcp_subflow_send_ack(struct sock * ssk)506 void __mptcp_subflow_send_ack(struct sock *ssk)
507 {
508 if (tcp_can_send_ack(ssk))
509 tcp_send_ack(ssk);
510 }
511
mptcp_subflow_send_ack(struct sock * ssk)512 static void mptcp_subflow_send_ack(struct sock *ssk)
513 {
514 bool slow;
515
516 slow = lock_sock_fast(ssk);
517 __mptcp_subflow_send_ack(ssk);
518 unlock_sock_fast(ssk, slow);
519 }
520
mptcp_send_ack(struct mptcp_sock * msk)521 static void mptcp_send_ack(struct mptcp_sock *msk)
522 {
523 struct mptcp_subflow_context *subflow;
524
525 mptcp_for_each_subflow(msk, subflow)
526 mptcp_subflow_send_ack(mptcp_subflow_tcp_sock(subflow));
527 }
528
mptcp_subflow_cleanup_rbuf(struct sock * ssk)529 static void mptcp_subflow_cleanup_rbuf(struct sock *ssk)
530 {
531 bool slow;
532
533 slow = lock_sock_fast(ssk);
534 if (tcp_can_send_ack(ssk))
535 tcp_cleanup_rbuf(ssk, 1);
536 unlock_sock_fast(ssk, slow);
537 }
538
mptcp_subflow_could_cleanup(const struct sock * ssk,bool rx_empty)539 static bool mptcp_subflow_could_cleanup(const struct sock *ssk, bool rx_empty)
540 {
541 const struct inet_connection_sock *icsk = inet_csk(ssk);
542 u8 ack_pending = READ_ONCE(icsk->icsk_ack.pending);
543 const struct tcp_sock *tp = tcp_sk(ssk);
544
545 return (ack_pending & ICSK_ACK_SCHED) &&
546 ((READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->rcv_wup) >
547 READ_ONCE(icsk->icsk_ack.rcv_mss)) ||
548 (rx_empty && ack_pending &
549 (ICSK_ACK_PUSHED2 | ICSK_ACK_PUSHED)));
550 }
551
mptcp_cleanup_rbuf(struct mptcp_sock * msk)552 static void mptcp_cleanup_rbuf(struct mptcp_sock *msk)
553 {
554 int old_space = READ_ONCE(msk->old_wspace);
555 struct mptcp_subflow_context *subflow;
556 struct sock *sk = (struct sock *)msk;
557 int space = __mptcp_space(sk);
558 bool cleanup, rx_empty;
559
560 cleanup = (space > 0) && (space >= (old_space << 1));
561 rx_empty = !__mptcp_rmem(sk);
562
563 mptcp_for_each_subflow(msk, subflow) {
564 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
565
566 if (cleanup || mptcp_subflow_could_cleanup(ssk, rx_empty))
567 mptcp_subflow_cleanup_rbuf(ssk);
568 }
569 }
570
mptcp_check_data_fin(struct sock * sk)571 static bool mptcp_check_data_fin(struct sock *sk)
572 {
573 struct mptcp_sock *msk = mptcp_sk(sk);
574 u64 rcv_data_fin_seq;
575 bool ret = false;
576
577 /* Need to ack a DATA_FIN received from a peer while this side
578 * of the connection is in ESTABLISHED, FIN_WAIT1, or FIN_WAIT2.
579 * msk->rcv_data_fin was set when parsing the incoming options
580 * at the subflow level and the msk lock was not held, so this
581 * is the first opportunity to act on the DATA_FIN and change
582 * the msk state.
583 *
584 * If we are caught up to the sequence number of the incoming
585 * DATA_FIN, send the DATA_ACK now and do state transition. If
586 * not caught up, do nothing and let the recv code send DATA_ACK
587 * when catching up.
588 */
589
590 if (mptcp_pending_data_fin(sk, &rcv_data_fin_seq)) {
591 WRITE_ONCE(msk->ack_seq, msk->ack_seq + 1);
592 WRITE_ONCE(msk->rcv_data_fin, 0);
593
594 WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | RCV_SHUTDOWN);
595 smp_mb__before_atomic(); /* SHUTDOWN must be visible first */
596
597 switch (sk->sk_state) {
598 case TCP_ESTABLISHED:
599 mptcp_set_state(sk, TCP_CLOSE_WAIT);
600 break;
601 case TCP_FIN_WAIT1:
602 mptcp_set_state(sk, TCP_CLOSING);
603 break;
604 case TCP_FIN_WAIT2:
605 mptcp_set_state(sk, TCP_CLOSE);
606 break;
607 default:
608 /* Other states not expected */
609 WARN_ON_ONCE(1);
610 break;
611 }
612
613 ret = true;
614 if (!__mptcp_check_fallback(msk))
615 mptcp_send_ack(msk);
616 mptcp_close_wake_up(sk);
617 }
618 return ret;
619 }
620
__mptcp_move_skbs_from_subflow(struct mptcp_sock * msk,struct sock * ssk,unsigned int * bytes)621 static bool __mptcp_move_skbs_from_subflow(struct mptcp_sock *msk,
622 struct sock *ssk,
623 unsigned int *bytes)
624 {
625 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
626 struct sock *sk = (struct sock *)msk;
627 unsigned int moved = 0;
628 bool more_data_avail;
629 struct tcp_sock *tp;
630 bool done = false;
631 int sk_rbuf;
632
633 sk_rbuf = READ_ONCE(sk->sk_rcvbuf);
634
635 if (!(sk->sk_userlocks & SOCK_RCVBUF_LOCK)) {
636 int ssk_rbuf = READ_ONCE(ssk->sk_rcvbuf);
637
638 if (unlikely(ssk_rbuf > sk_rbuf)) {
639 WRITE_ONCE(sk->sk_rcvbuf, ssk_rbuf);
640 sk_rbuf = ssk_rbuf;
641 }
642 }
643
644 pr_debug("msk=%p ssk=%p", msk, ssk);
645 tp = tcp_sk(ssk);
646 do {
647 u32 map_remaining, offset;
648 u32 seq = tp->copied_seq;
649 struct sk_buff *skb;
650 bool fin;
651
652 /* try to move as much data as available */
653 map_remaining = subflow->map_data_len -
654 mptcp_subflow_get_map_offset(subflow);
655
656 skb = skb_peek(&ssk->sk_receive_queue);
657 if (!skb) {
658 /* With racing move_skbs_to_msk() and __mptcp_move_skbs(),
659 * a different CPU can have already processed the pending
660 * data, stop here or we can enter an infinite loop
661 */
662 if (!moved)
663 done = true;
664 break;
665 }
666
667 if (__mptcp_check_fallback(msk)) {
668 /* Under fallback skbs have no MPTCP extension and TCP could
669 * collapse them between the dummy map creation and the
670 * current dequeue. Be sure to adjust the map size.
671 */
672 map_remaining = skb->len;
673 subflow->map_data_len = skb->len;
674 }
675
676 offset = seq - TCP_SKB_CB(skb)->seq;
677 fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
678 if (fin) {
679 done = true;
680 seq++;
681 }
682
683 if (offset < skb->len) {
684 size_t len = skb->len - offset;
685
686 if (tp->urg_data)
687 done = true;
688
689 if (__mptcp_move_skb(msk, ssk, skb, offset, len))
690 moved += len;
691 seq += len;
692
693 if (WARN_ON_ONCE(map_remaining < len))
694 break;
695 } else {
696 WARN_ON_ONCE(!fin);
697 sk_eat_skb(ssk, skb);
698 done = true;
699 }
700
701 WRITE_ONCE(tp->copied_seq, seq);
702 more_data_avail = mptcp_subflow_data_available(ssk);
703
704 if (atomic_read(&sk->sk_rmem_alloc) > sk_rbuf) {
705 done = true;
706 break;
707 }
708 } while (more_data_avail);
709
710 *bytes += moved;
711 return done;
712 }
713
__mptcp_ofo_queue(struct mptcp_sock * msk)714 static bool __mptcp_ofo_queue(struct mptcp_sock *msk)
715 {
716 struct sock *sk = (struct sock *)msk;
717 struct sk_buff *skb, *tail;
718 bool moved = false;
719 struct rb_node *p;
720 u64 end_seq;
721
722 p = rb_first(&msk->out_of_order_queue);
723 pr_debug("msk=%p empty=%d", msk, RB_EMPTY_ROOT(&msk->out_of_order_queue));
724 while (p) {
725 skb = rb_to_skb(p);
726 if (after64(MPTCP_SKB_CB(skb)->map_seq, msk->ack_seq))
727 break;
728
729 p = rb_next(p);
730 rb_erase(&skb->rbnode, &msk->out_of_order_queue);
731
732 if (unlikely(!after64(MPTCP_SKB_CB(skb)->end_seq,
733 msk->ack_seq))) {
734 mptcp_drop(sk, skb);
735 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
736 continue;
737 }
738
739 end_seq = MPTCP_SKB_CB(skb)->end_seq;
740 tail = skb_peek_tail(&sk->sk_receive_queue);
741 if (!tail || !mptcp_ooo_try_coalesce(msk, tail, skb)) {
742 int delta = msk->ack_seq - MPTCP_SKB_CB(skb)->map_seq;
743
744 /* skip overlapping data, if any */
745 pr_debug("uncoalesced seq=%llx ack seq=%llx delta=%d",
746 MPTCP_SKB_CB(skb)->map_seq, msk->ack_seq,
747 delta);
748 MPTCP_SKB_CB(skb)->offset += delta;
749 MPTCP_SKB_CB(skb)->map_seq += delta;
750 __skb_queue_tail(&sk->sk_receive_queue, skb);
751 }
752 msk->bytes_received += end_seq - msk->ack_seq;
753 msk->ack_seq = end_seq;
754 moved = true;
755 }
756 return moved;
757 }
758
__mptcp_subflow_error_report(struct sock * sk,struct sock * ssk)759 static bool __mptcp_subflow_error_report(struct sock *sk, struct sock *ssk)
760 {
761 int err = sock_error(ssk);
762 int ssk_state;
763
764 if (!err)
765 return false;
766
767 /* only propagate errors on fallen-back sockets or
768 * on MPC connect
769 */
770 if (sk->sk_state != TCP_SYN_SENT && !__mptcp_check_fallback(mptcp_sk(sk)))
771 return false;
772
773 /* We need to propagate only transition to CLOSE state.
774 * Orphaned socket will see such state change via
775 * subflow_sched_work_if_closed() and that path will properly
776 * destroy the msk as needed.
777 */
778 ssk_state = inet_sk_state_load(ssk);
779 if (ssk_state == TCP_CLOSE && !sock_flag(sk, SOCK_DEAD))
780 mptcp_set_state(sk, ssk_state);
781 WRITE_ONCE(sk->sk_err, -err);
782
783 /* This barrier is coupled with smp_rmb() in mptcp_poll() */
784 smp_wmb();
785 sk_error_report(sk);
786 return true;
787 }
788
__mptcp_error_report(struct sock * sk)789 void __mptcp_error_report(struct sock *sk)
790 {
791 struct mptcp_subflow_context *subflow;
792 struct mptcp_sock *msk = mptcp_sk(sk);
793
794 mptcp_for_each_subflow(msk, subflow)
795 if (__mptcp_subflow_error_report(sk, mptcp_subflow_tcp_sock(subflow)))
796 break;
797 }
798
799 /* In most cases we will be able to lock the mptcp socket. If its already
800 * owned, we need to defer to the work queue to avoid ABBA deadlock.
801 */
move_skbs_to_msk(struct mptcp_sock * msk,struct sock * ssk)802 static bool move_skbs_to_msk(struct mptcp_sock *msk, struct sock *ssk)
803 {
804 struct sock *sk = (struct sock *)msk;
805 unsigned int moved = 0;
806
807 __mptcp_move_skbs_from_subflow(msk, ssk, &moved);
808 __mptcp_ofo_queue(msk);
809 if (unlikely(ssk->sk_err)) {
810 if (!sock_owned_by_user(sk))
811 __mptcp_error_report(sk);
812 else
813 __set_bit(MPTCP_ERROR_REPORT, &msk->cb_flags);
814 }
815
816 /* If the moves have caught up with the DATA_FIN sequence number
817 * it's time to ack the DATA_FIN and change socket state, but
818 * this is not a good place to change state. Let the workqueue
819 * do it.
820 */
821 if (mptcp_pending_data_fin(sk, NULL))
822 mptcp_schedule_work(sk);
823 return moved > 0;
824 }
825
mptcp_data_ready(struct sock * sk,struct sock * ssk)826 void mptcp_data_ready(struct sock *sk, struct sock *ssk)
827 {
828 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
829 struct mptcp_sock *msk = mptcp_sk(sk);
830 int sk_rbuf, ssk_rbuf;
831
832 /* The peer can send data while we are shutting down this
833 * subflow at msk destruction time, but we must avoid enqueuing
834 * more data to the msk receive queue
835 */
836 if (unlikely(subflow->disposable))
837 return;
838
839 ssk_rbuf = READ_ONCE(ssk->sk_rcvbuf);
840 sk_rbuf = READ_ONCE(sk->sk_rcvbuf);
841 if (unlikely(ssk_rbuf > sk_rbuf))
842 sk_rbuf = ssk_rbuf;
843
844 /* over limit? can't append more skbs to msk, Also, no need to wake-up*/
845 if (__mptcp_rmem(sk) > sk_rbuf) {
846 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_RCVPRUNED);
847 return;
848 }
849
850 /* Wake-up the reader only for in-sequence data */
851 mptcp_data_lock(sk);
852 if (move_skbs_to_msk(msk, ssk))
853 sk->sk_data_ready(sk);
854
855 mptcp_data_unlock(sk);
856 }
857
mptcp_subflow_joined(struct mptcp_sock * msk,struct sock * ssk)858 static void mptcp_subflow_joined(struct mptcp_sock *msk, struct sock *ssk)
859 {
860 mptcp_subflow_ctx(ssk)->map_seq = READ_ONCE(msk->ack_seq);
861 WRITE_ONCE(msk->allow_infinite_fallback, false);
862 mptcp_event(MPTCP_EVENT_SUB_ESTABLISHED, msk, ssk, GFP_ATOMIC);
863 }
864
__mptcp_finish_join(struct mptcp_sock * msk,struct sock * ssk)865 static bool __mptcp_finish_join(struct mptcp_sock *msk, struct sock *ssk)
866 {
867 struct sock *sk = (struct sock *)msk;
868
869 if (sk->sk_state != TCP_ESTABLISHED)
870 return false;
871
872 /* attach to msk socket only after we are sure we will deal with it
873 * at close time
874 */
875 if (sk->sk_socket && !ssk->sk_socket)
876 mptcp_sock_graft(ssk, sk->sk_socket);
877
878 mptcp_subflow_ctx(ssk)->subflow_id = msk->subflow_id++;
879 mptcp_sockopt_sync_locked(msk, ssk);
880 mptcp_subflow_joined(msk, ssk);
881 mptcp_stop_tout_timer(sk);
882 __mptcp_propagate_sndbuf(sk, ssk);
883 return true;
884 }
885
__mptcp_flush_join_list(struct sock * sk,struct list_head * join_list)886 static void __mptcp_flush_join_list(struct sock *sk, struct list_head *join_list)
887 {
888 struct mptcp_subflow_context *tmp, *subflow;
889 struct mptcp_sock *msk = mptcp_sk(sk);
890
891 list_for_each_entry_safe(subflow, tmp, join_list, node) {
892 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
893 bool slow = lock_sock_fast(ssk);
894
895 list_move_tail(&subflow->node, &msk->conn_list);
896 if (!__mptcp_finish_join(msk, ssk))
897 mptcp_subflow_reset(ssk);
898 unlock_sock_fast(ssk, slow);
899 }
900 }
901
mptcp_rtx_timer_pending(struct sock * sk)902 static bool mptcp_rtx_timer_pending(struct sock *sk)
903 {
904 return timer_pending(&inet_csk(sk)->icsk_retransmit_timer);
905 }
906
mptcp_reset_rtx_timer(struct sock * sk)907 static void mptcp_reset_rtx_timer(struct sock *sk)
908 {
909 struct inet_connection_sock *icsk = inet_csk(sk);
910 unsigned long tout;
911
912 /* prevent rescheduling on close */
913 if (unlikely(inet_sk_state_load(sk) == TCP_CLOSE))
914 return;
915
916 tout = mptcp_sk(sk)->timer_ival;
917 sk_reset_timer(sk, &icsk->icsk_retransmit_timer, jiffies + tout);
918 }
919
mptcp_schedule_work(struct sock * sk)920 bool mptcp_schedule_work(struct sock *sk)
921 {
922 if (inet_sk_state_load(sk) != TCP_CLOSE &&
923 schedule_work(&mptcp_sk(sk)->work)) {
924 /* each subflow already holds a reference to the sk, and the
925 * workqueue is invoked by a subflow, so sk can't go away here.
926 */
927 sock_hold(sk);
928 return true;
929 }
930 return false;
931 }
932
mptcp_subflow_recv_lookup(const struct mptcp_sock * msk)933 static struct sock *mptcp_subflow_recv_lookup(const struct mptcp_sock *msk)
934 {
935 struct mptcp_subflow_context *subflow;
936
937 msk_owned_by_me(msk);
938
939 mptcp_for_each_subflow(msk, subflow) {
940 if (READ_ONCE(subflow->data_avail))
941 return mptcp_subflow_tcp_sock(subflow);
942 }
943
944 return NULL;
945 }
946
mptcp_skb_can_collapse_to(u64 write_seq,const struct sk_buff * skb,const struct mptcp_ext * mpext)947 static bool mptcp_skb_can_collapse_to(u64 write_seq,
948 const struct sk_buff *skb,
949 const struct mptcp_ext *mpext)
950 {
951 if (!tcp_skb_can_collapse_to(skb))
952 return false;
953
954 /* can collapse only if MPTCP level sequence is in order and this
955 * mapping has not been xmitted yet
956 */
957 return mpext && mpext->data_seq + mpext->data_len == write_seq &&
958 !mpext->frozen;
959 }
960
961 /* we can append data to the given data frag if:
962 * - there is space available in the backing page_frag
963 * - the data frag tail matches the current page_frag free offset
964 * - the data frag end sequence number matches the current write seq
965 */
mptcp_frag_can_collapse_to(const struct mptcp_sock * msk,const struct page_frag * pfrag,const struct mptcp_data_frag * df)966 static bool mptcp_frag_can_collapse_to(const struct mptcp_sock *msk,
967 const struct page_frag *pfrag,
968 const struct mptcp_data_frag *df)
969 {
970 return df && pfrag->page == df->page &&
971 pfrag->size - pfrag->offset > 0 &&
972 pfrag->offset == (df->offset + df->data_len) &&
973 df->data_seq + df->data_len == msk->write_seq;
974 }
975
dfrag_uncharge(struct sock * sk,int len)976 static void dfrag_uncharge(struct sock *sk, int len)
977 {
978 sk_mem_uncharge(sk, len);
979 sk_wmem_queued_add(sk, -len);
980 }
981
dfrag_clear(struct sock * sk,struct mptcp_data_frag * dfrag)982 static void dfrag_clear(struct sock *sk, struct mptcp_data_frag *dfrag)
983 {
984 int len = dfrag->data_len + dfrag->overhead;
985
986 list_del(&dfrag->list);
987 dfrag_uncharge(sk, len);
988 put_page(dfrag->page);
989 }
990
__mptcp_clean_una(struct sock * sk)991 static void __mptcp_clean_una(struct sock *sk)
992 {
993 struct mptcp_sock *msk = mptcp_sk(sk);
994 struct mptcp_data_frag *dtmp, *dfrag;
995 u64 snd_una;
996
997 snd_una = msk->snd_una;
998 list_for_each_entry_safe(dfrag, dtmp, &msk->rtx_queue, list) {
999 if (after64(dfrag->data_seq + dfrag->data_len, snd_una))
1000 break;
1001
1002 if (unlikely(dfrag == msk->first_pending)) {
1003 /* in recovery mode can see ack after the current snd head */
1004 if (WARN_ON_ONCE(!msk->recovery))
1005 break;
1006
1007 WRITE_ONCE(msk->first_pending, mptcp_send_next(sk));
1008 }
1009
1010 dfrag_clear(sk, dfrag);
1011 }
1012
1013 dfrag = mptcp_rtx_head(sk);
1014 if (dfrag && after64(snd_una, dfrag->data_seq)) {
1015 u64 delta = snd_una - dfrag->data_seq;
1016
1017 /* prevent wrap around in recovery mode */
1018 if (unlikely(delta > dfrag->already_sent)) {
1019 if (WARN_ON_ONCE(!msk->recovery))
1020 goto out;
1021 if (WARN_ON_ONCE(delta > dfrag->data_len))
1022 goto out;
1023 dfrag->already_sent += delta - dfrag->already_sent;
1024 }
1025
1026 dfrag->data_seq += delta;
1027 dfrag->offset += delta;
1028 dfrag->data_len -= delta;
1029 dfrag->already_sent -= delta;
1030
1031 dfrag_uncharge(sk, delta);
1032 }
1033
1034 /* all retransmitted data acked, recovery completed */
1035 if (unlikely(msk->recovery) && after64(msk->snd_una, msk->recovery_snd_nxt))
1036 msk->recovery = false;
1037
1038 out:
1039 if (snd_una == READ_ONCE(msk->snd_nxt) &&
1040 snd_una == READ_ONCE(msk->write_seq)) {
1041 if (mptcp_rtx_timer_pending(sk) && !mptcp_data_fin_enabled(msk))
1042 mptcp_stop_rtx_timer(sk);
1043 } else {
1044 mptcp_reset_rtx_timer(sk);
1045 }
1046 }
1047
__mptcp_clean_una_wakeup(struct sock * sk)1048 static void __mptcp_clean_una_wakeup(struct sock *sk)
1049 {
1050 lockdep_assert_held_once(&sk->sk_lock.slock);
1051
1052 __mptcp_clean_una(sk);
1053 mptcp_write_space(sk);
1054 }
1055
mptcp_clean_una_wakeup(struct sock * sk)1056 static void mptcp_clean_una_wakeup(struct sock *sk)
1057 {
1058 mptcp_data_lock(sk);
1059 __mptcp_clean_una_wakeup(sk);
1060 mptcp_data_unlock(sk);
1061 }
1062
mptcp_enter_memory_pressure(struct sock * sk)1063 static void mptcp_enter_memory_pressure(struct sock *sk)
1064 {
1065 struct mptcp_subflow_context *subflow;
1066 struct mptcp_sock *msk = mptcp_sk(sk);
1067 bool first = true;
1068
1069 mptcp_for_each_subflow(msk, subflow) {
1070 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1071
1072 if (first)
1073 tcp_enter_memory_pressure(ssk);
1074 sk_stream_moderate_sndbuf(ssk);
1075
1076 first = false;
1077 }
1078 __mptcp_sync_sndbuf(sk);
1079 }
1080
1081 /* ensure we get enough memory for the frag hdr, beyond some minimal amount of
1082 * data
1083 */
mptcp_page_frag_refill(struct sock * sk,struct page_frag * pfrag)1084 static bool mptcp_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1085 {
1086 if (likely(skb_page_frag_refill(32U + sizeof(struct mptcp_data_frag),
1087 pfrag, sk->sk_allocation)))
1088 return true;
1089
1090 mptcp_enter_memory_pressure(sk);
1091 return false;
1092 }
1093
1094 static struct mptcp_data_frag *
mptcp_carve_data_frag(const struct mptcp_sock * msk,struct page_frag * pfrag,int orig_offset)1095 mptcp_carve_data_frag(const struct mptcp_sock *msk, struct page_frag *pfrag,
1096 int orig_offset)
1097 {
1098 int offset = ALIGN(orig_offset, sizeof(long));
1099 struct mptcp_data_frag *dfrag;
1100
1101 dfrag = (struct mptcp_data_frag *)(page_to_virt(pfrag->page) + offset);
1102 dfrag->data_len = 0;
1103 dfrag->data_seq = msk->write_seq;
1104 dfrag->overhead = offset - orig_offset + sizeof(struct mptcp_data_frag);
1105 dfrag->offset = offset + sizeof(struct mptcp_data_frag);
1106 dfrag->already_sent = 0;
1107 dfrag->page = pfrag->page;
1108
1109 return dfrag;
1110 }
1111
1112 struct mptcp_sendmsg_info {
1113 int mss_now;
1114 int size_goal;
1115 u16 limit;
1116 u16 sent;
1117 unsigned int flags;
1118 bool data_lock_held;
1119 };
1120
mptcp_check_allowed_size(const struct mptcp_sock * msk,struct sock * ssk,u64 data_seq,int avail_size)1121 static int mptcp_check_allowed_size(const struct mptcp_sock *msk, struct sock *ssk,
1122 u64 data_seq, int avail_size)
1123 {
1124 u64 window_end = mptcp_wnd_end(msk);
1125 u64 mptcp_snd_wnd;
1126
1127 if (__mptcp_check_fallback(msk))
1128 return avail_size;
1129
1130 mptcp_snd_wnd = window_end - data_seq;
1131 avail_size = min_t(unsigned int, mptcp_snd_wnd, avail_size);
1132
1133 if (unlikely(tcp_sk(ssk)->snd_wnd < mptcp_snd_wnd)) {
1134 tcp_sk(ssk)->snd_wnd = min_t(u64, U32_MAX, mptcp_snd_wnd);
1135 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_SNDWNDSHARED);
1136 }
1137
1138 return avail_size;
1139 }
1140
__mptcp_add_ext(struct sk_buff * skb,gfp_t gfp)1141 static bool __mptcp_add_ext(struct sk_buff *skb, gfp_t gfp)
1142 {
1143 struct skb_ext *mpext = __skb_ext_alloc(gfp);
1144
1145 if (!mpext)
1146 return false;
1147 __skb_ext_set(skb, SKB_EXT_MPTCP, mpext);
1148 return true;
1149 }
1150
__mptcp_do_alloc_tx_skb(struct sock * sk,gfp_t gfp)1151 static struct sk_buff *__mptcp_do_alloc_tx_skb(struct sock *sk, gfp_t gfp)
1152 {
1153 struct sk_buff *skb;
1154
1155 skb = alloc_skb_fclone(MAX_TCP_HEADER, gfp);
1156 if (likely(skb)) {
1157 if (likely(__mptcp_add_ext(skb, gfp))) {
1158 skb_reserve(skb, MAX_TCP_HEADER);
1159 skb->ip_summed = CHECKSUM_PARTIAL;
1160 INIT_LIST_HEAD(&skb->tcp_tsorted_anchor);
1161 return skb;
1162 }
1163 __kfree_skb(skb);
1164 } else {
1165 mptcp_enter_memory_pressure(sk);
1166 }
1167 return NULL;
1168 }
1169
__mptcp_alloc_tx_skb(struct sock * sk,struct sock * ssk,gfp_t gfp)1170 static struct sk_buff *__mptcp_alloc_tx_skb(struct sock *sk, struct sock *ssk, gfp_t gfp)
1171 {
1172 struct sk_buff *skb;
1173
1174 skb = __mptcp_do_alloc_tx_skb(sk, gfp);
1175 if (!skb)
1176 return NULL;
1177
1178 if (likely(sk_wmem_schedule(ssk, skb->truesize))) {
1179 tcp_skb_entail(ssk, skb);
1180 return skb;
1181 }
1182 tcp_skb_tsorted_anchor_cleanup(skb);
1183 kfree_skb(skb);
1184 return NULL;
1185 }
1186
mptcp_alloc_tx_skb(struct sock * sk,struct sock * ssk,bool data_lock_held)1187 static struct sk_buff *mptcp_alloc_tx_skb(struct sock *sk, struct sock *ssk, bool data_lock_held)
1188 {
1189 gfp_t gfp = data_lock_held ? GFP_ATOMIC : sk->sk_allocation;
1190
1191 return __mptcp_alloc_tx_skb(sk, ssk, gfp);
1192 }
1193
1194 /* note: this always recompute the csum on the whole skb, even
1195 * if we just appended a single frag. More status info needed
1196 */
mptcp_update_data_checksum(struct sk_buff * skb,int added)1197 static void mptcp_update_data_checksum(struct sk_buff *skb, int added)
1198 {
1199 struct mptcp_ext *mpext = mptcp_get_ext(skb);
1200 __wsum csum = ~csum_unfold(mpext->csum);
1201 int offset = skb->len - added;
1202
1203 mpext->csum = csum_fold(csum_block_add(csum, skb_checksum(skb, offset, added, 0), offset));
1204 }
1205
mptcp_update_infinite_map(struct mptcp_sock * msk,struct sock * ssk,struct mptcp_ext * mpext)1206 static void mptcp_update_infinite_map(struct mptcp_sock *msk,
1207 struct sock *ssk,
1208 struct mptcp_ext *mpext)
1209 {
1210 if (!mpext)
1211 return;
1212
1213 mpext->infinite_map = 1;
1214 mpext->data_len = 0;
1215
1216 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPTX);
1217 mptcp_subflow_ctx(ssk)->send_infinite_map = 0;
1218 pr_fallback(msk);
1219 mptcp_do_fallback(ssk);
1220 }
1221
1222 #define MPTCP_MAX_GSO_SIZE (GSO_LEGACY_MAX_SIZE - (MAX_TCP_HEADER + 1))
1223
mptcp_sendmsg_frag(struct sock * sk,struct sock * ssk,struct mptcp_data_frag * dfrag,struct mptcp_sendmsg_info * info)1224 static int mptcp_sendmsg_frag(struct sock *sk, struct sock *ssk,
1225 struct mptcp_data_frag *dfrag,
1226 struct mptcp_sendmsg_info *info)
1227 {
1228 u64 data_seq = dfrag->data_seq + info->sent;
1229 int offset = dfrag->offset + info->sent;
1230 struct mptcp_sock *msk = mptcp_sk(sk);
1231 bool zero_window_probe = false;
1232 struct mptcp_ext *mpext = NULL;
1233 bool can_coalesce = false;
1234 bool reuse_skb = true;
1235 struct sk_buff *skb;
1236 size_t copy;
1237 int i;
1238
1239 pr_debug("msk=%p ssk=%p sending dfrag at seq=%llu len=%u already sent=%u",
1240 msk, ssk, dfrag->data_seq, dfrag->data_len, info->sent);
1241
1242 if (WARN_ON_ONCE(info->sent > info->limit ||
1243 info->limit > dfrag->data_len))
1244 return 0;
1245
1246 if (unlikely(!__tcp_can_send(ssk)))
1247 return -EAGAIN;
1248
1249 /* compute send limit */
1250 if (unlikely(ssk->sk_gso_max_size > MPTCP_MAX_GSO_SIZE))
1251 ssk->sk_gso_max_size = MPTCP_MAX_GSO_SIZE;
1252 info->mss_now = tcp_send_mss(ssk, &info->size_goal, info->flags);
1253 copy = info->size_goal;
1254
1255 skb = tcp_write_queue_tail(ssk);
1256 if (skb && copy > skb->len) {
1257 /* Limit the write to the size available in the
1258 * current skb, if any, so that we create at most a new skb.
1259 * Explicitly tells TCP internals to avoid collapsing on later
1260 * queue management operation, to avoid breaking the ext <->
1261 * SSN association set here
1262 */
1263 mpext = skb_ext_find(skb, SKB_EXT_MPTCP);
1264 if (!mptcp_skb_can_collapse_to(data_seq, skb, mpext)) {
1265 TCP_SKB_CB(skb)->eor = 1;
1266 tcp_mark_push(tcp_sk(ssk), skb);
1267 goto alloc_skb;
1268 }
1269
1270 i = skb_shinfo(skb)->nr_frags;
1271 can_coalesce = skb_can_coalesce(skb, i, dfrag->page, offset);
1272 if (!can_coalesce && i >= READ_ONCE(sysctl_max_skb_frags)) {
1273 tcp_mark_push(tcp_sk(ssk), skb);
1274 goto alloc_skb;
1275 }
1276
1277 copy -= skb->len;
1278 } else {
1279 alloc_skb:
1280 skb = mptcp_alloc_tx_skb(sk, ssk, info->data_lock_held);
1281 if (!skb)
1282 return -ENOMEM;
1283
1284 i = skb_shinfo(skb)->nr_frags;
1285 reuse_skb = false;
1286 mpext = skb_ext_find(skb, SKB_EXT_MPTCP);
1287 }
1288
1289 /* Zero window and all data acked? Probe. */
1290 copy = mptcp_check_allowed_size(msk, ssk, data_seq, copy);
1291 if (copy == 0) {
1292 u64 snd_una = READ_ONCE(msk->snd_una);
1293
1294 if (snd_una != msk->snd_nxt || tcp_write_queue_tail(ssk)) {
1295 tcp_remove_empty_skb(ssk);
1296 return 0;
1297 }
1298
1299 zero_window_probe = true;
1300 data_seq = snd_una - 1;
1301 copy = 1;
1302 }
1303
1304 copy = min_t(size_t, copy, info->limit - info->sent);
1305 if (!sk_wmem_schedule(ssk, copy)) {
1306 tcp_remove_empty_skb(ssk);
1307 return -ENOMEM;
1308 }
1309
1310 if (can_coalesce) {
1311 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1312 } else {
1313 get_page(dfrag->page);
1314 skb_fill_page_desc(skb, i, dfrag->page, offset, copy);
1315 }
1316
1317 skb->len += copy;
1318 skb->data_len += copy;
1319 skb->truesize += copy;
1320 sk_wmem_queued_add(ssk, copy);
1321 sk_mem_charge(ssk, copy);
1322 WRITE_ONCE(tcp_sk(ssk)->write_seq, tcp_sk(ssk)->write_seq + copy);
1323 TCP_SKB_CB(skb)->end_seq += copy;
1324 tcp_skb_pcount_set(skb, 0);
1325
1326 /* on skb reuse we just need to update the DSS len */
1327 if (reuse_skb) {
1328 TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH;
1329 mpext->data_len += copy;
1330 goto out;
1331 }
1332
1333 memset(mpext, 0, sizeof(*mpext));
1334 mpext->data_seq = data_seq;
1335 mpext->subflow_seq = mptcp_subflow_ctx(ssk)->rel_write_seq;
1336 mpext->data_len = copy;
1337 mpext->use_map = 1;
1338 mpext->dsn64 = 1;
1339
1340 pr_debug("data_seq=%llu subflow_seq=%u data_len=%u dsn64=%d",
1341 mpext->data_seq, mpext->subflow_seq, mpext->data_len,
1342 mpext->dsn64);
1343
1344 if (zero_window_probe) {
1345 mptcp_subflow_ctx(ssk)->rel_write_seq += copy;
1346 mpext->frozen = 1;
1347 if (READ_ONCE(msk->csum_enabled))
1348 mptcp_update_data_checksum(skb, copy);
1349 tcp_push_pending_frames(ssk);
1350 return 0;
1351 }
1352 out:
1353 if (READ_ONCE(msk->csum_enabled))
1354 mptcp_update_data_checksum(skb, copy);
1355 if (mptcp_subflow_ctx(ssk)->send_infinite_map)
1356 mptcp_update_infinite_map(msk, ssk, mpext);
1357 trace_mptcp_sendmsg_frag(mpext);
1358 mptcp_subflow_ctx(ssk)->rel_write_seq += copy;
1359 return copy;
1360 }
1361
1362 #define MPTCP_SEND_BURST_SIZE ((1 << 16) - \
1363 sizeof(struct tcphdr) - \
1364 MAX_TCP_OPTION_SPACE - \
1365 sizeof(struct ipv6hdr) - \
1366 sizeof(struct frag_hdr))
1367
1368 struct subflow_send_info {
1369 struct sock *ssk;
1370 u64 linger_time;
1371 };
1372
mptcp_subflow_set_active(struct mptcp_subflow_context * subflow)1373 void mptcp_subflow_set_active(struct mptcp_subflow_context *subflow)
1374 {
1375 if (!subflow->stale)
1376 return;
1377
1378 subflow->stale = 0;
1379 MPTCP_INC_STATS(sock_net(mptcp_subflow_tcp_sock(subflow)), MPTCP_MIB_SUBFLOWRECOVER);
1380 }
1381
mptcp_subflow_active(struct mptcp_subflow_context * subflow)1382 bool mptcp_subflow_active(struct mptcp_subflow_context *subflow)
1383 {
1384 if (unlikely(subflow->stale)) {
1385 u32 rcv_tstamp = READ_ONCE(tcp_sk(mptcp_subflow_tcp_sock(subflow))->rcv_tstamp);
1386
1387 if (subflow->stale_rcv_tstamp == rcv_tstamp)
1388 return false;
1389
1390 mptcp_subflow_set_active(subflow);
1391 }
1392 return __mptcp_subflow_active(subflow);
1393 }
1394
1395 #define SSK_MODE_ACTIVE 0
1396 #define SSK_MODE_BACKUP 1
1397 #define SSK_MODE_MAX 2
1398
1399 /* implement the mptcp packet scheduler;
1400 * returns the subflow that will transmit the next DSS
1401 * additionally updates the rtx timeout
1402 */
mptcp_subflow_get_send(struct mptcp_sock * msk)1403 struct sock *mptcp_subflow_get_send(struct mptcp_sock *msk)
1404 {
1405 struct subflow_send_info send_info[SSK_MODE_MAX];
1406 struct mptcp_subflow_context *subflow;
1407 struct sock *sk = (struct sock *)msk;
1408 u32 pace, burst, wmem;
1409 int i, nr_active = 0;
1410 struct sock *ssk;
1411 u64 linger_time;
1412 long tout = 0;
1413
1414 /* pick the subflow with the lower wmem/wspace ratio */
1415 for (i = 0; i < SSK_MODE_MAX; ++i) {
1416 send_info[i].ssk = NULL;
1417 send_info[i].linger_time = -1;
1418 }
1419
1420 mptcp_for_each_subflow(msk, subflow) {
1421 trace_mptcp_subflow_get_send(subflow);
1422 ssk = mptcp_subflow_tcp_sock(subflow);
1423 if (!mptcp_subflow_active(subflow))
1424 continue;
1425
1426 tout = max(tout, mptcp_timeout_from_subflow(subflow));
1427 nr_active += !subflow->backup;
1428 pace = subflow->avg_pacing_rate;
1429 if (unlikely(!pace)) {
1430 /* init pacing rate from socket */
1431 subflow->avg_pacing_rate = READ_ONCE(ssk->sk_pacing_rate);
1432 pace = subflow->avg_pacing_rate;
1433 if (!pace)
1434 continue;
1435 }
1436
1437 linger_time = div_u64((u64)READ_ONCE(ssk->sk_wmem_queued) << 32, pace);
1438 if (linger_time < send_info[subflow->backup].linger_time) {
1439 send_info[subflow->backup].ssk = ssk;
1440 send_info[subflow->backup].linger_time = linger_time;
1441 }
1442 }
1443 __mptcp_set_timeout(sk, tout);
1444
1445 /* pick the best backup if no other subflow is active */
1446 if (!nr_active)
1447 send_info[SSK_MODE_ACTIVE].ssk = send_info[SSK_MODE_BACKUP].ssk;
1448
1449 /* According to the blest algorithm, to avoid HoL blocking for the
1450 * faster flow, we need to:
1451 * - estimate the faster flow linger time
1452 * - use the above to estimate the amount of byte transferred
1453 * by the faster flow
1454 * - check that the amount of queued data is greter than the above,
1455 * otherwise do not use the picked, slower, subflow
1456 * We select the subflow with the shorter estimated time to flush
1457 * the queued mem, which basically ensure the above. We just need
1458 * to check that subflow has a non empty cwin.
1459 */
1460 ssk = send_info[SSK_MODE_ACTIVE].ssk;
1461 if (!ssk || !sk_stream_memory_free(ssk))
1462 return NULL;
1463
1464 burst = min_t(int, MPTCP_SEND_BURST_SIZE, mptcp_wnd_end(msk) - msk->snd_nxt);
1465 wmem = READ_ONCE(ssk->sk_wmem_queued);
1466 if (!burst)
1467 return ssk;
1468
1469 subflow = mptcp_subflow_ctx(ssk);
1470 subflow->avg_pacing_rate = div_u64((u64)subflow->avg_pacing_rate * wmem +
1471 READ_ONCE(ssk->sk_pacing_rate) * burst,
1472 burst + wmem);
1473 msk->snd_burst = burst;
1474 return ssk;
1475 }
1476
mptcp_push_release(struct sock * ssk,struct mptcp_sendmsg_info * info)1477 static void mptcp_push_release(struct sock *ssk, struct mptcp_sendmsg_info *info)
1478 {
1479 tcp_push(ssk, 0, info->mss_now, tcp_sk(ssk)->nonagle, info->size_goal);
1480 release_sock(ssk);
1481 }
1482
mptcp_update_post_push(struct mptcp_sock * msk,struct mptcp_data_frag * dfrag,u32 sent)1483 static void mptcp_update_post_push(struct mptcp_sock *msk,
1484 struct mptcp_data_frag *dfrag,
1485 u32 sent)
1486 {
1487 u64 snd_nxt_new = dfrag->data_seq;
1488
1489 dfrag->already_sent += sent;
1490
1491 msk->snd_burst -= sent;
1492
1493 snd_nxt_new += dfrag->already_sent;
1494
1495 /* snd_nxt_new can be smaller than snd_nxt in case mptcp
1496 * is recovering after a failover. In that event, this re-sends
1497 * old segments.
1498 *
1499 * Thus compute snd_nxt_new candidate based on
1500 * the dfrag->data_seq that was sent and the data
1501 * that has been handed to the subflow for transmission
1502 * and skip update in case it was old dfrag.
1503 */
1504 if (likely(after64(snd_nxt_new, msk->snd_nxt))) {
1505 msk->bytes_sent += snd_nxt_new - msk->snd_nxt;
1506 msk->snd_nxt = snd_nxt_new;
1507 }
1508 }
1509
mptcp_check_and_set_pending(struct sock * sk)1510 void mptcp_check_and_set_pending(struct sock *sk)
1511 {
1512 if (mptcp_send_head(sk)) {
1513 mptcp_data_lock(sk);
1514 mptcp_sk(sk)->cb_flags |= BIT(MPTCP_PUSH_PENDING);
1515 mptcp_data_unlock(sk);
1516 }
1517 }
1518
__subflow_push_pending(struct sock * sk,struct sock * ssk,struct mptcp_sendmsg_info * info)1519 static int __subflow_push_pending(struct sock *sk, struct sock *ssk,
1520 struct mptcp_sendmsg_info *info)
1521 {
1522 struct mptcp_sock *msk = mptcp_sk(sk);
1523 struct mptcp_data_frag *dfrag;
1524 int len, copied = 0, err = 0;
1525
1526 while ((dfrag = mptcp_send_head(sk))) {
1527 info->sent = dfrag->already_sent;
1528 info->limit = dfrag->data_len;
1529 len = dfrag->data_len - dfrag->already_sent;
1530 while (len > 0) {
1531 int ret = 0;
1532
1533 ret = mptcp_sendmsg_frag(sk, ssk, dfrag, info);
1534 if (ret <= 0) {
1535 err = copied ? : ret;
1536 goto out;
1537 }
1538
1539 info->sent += ret;
1540 copied += ret;
1541 len -= ret;
1542
1543 mptcp_update_post_push(msk, dfrag, ret);
1544 }
1545 WRITE_ONCE(msk->first_pending, mptcp_send_next(sk));
1546
1547 if (msk->snd_burst <= 0 ||
1548 !sk_stream_memory_free(ssk) ||
1549 !mptcp_subflow_active(mptcp_subflow_ctx(ssk))) {
1550 err = copied;
1551 goto out;
1552 }
1553 mptcp_set_timeout(sk);
1554 }
1555 err = copied;
1556
1557 out:
1558 return err;
1559 }
1560
__mptcp_push_pending(struct sock * sk,unsigned int flags)1561 void __mptcp_push_pending(struct sock *sk, unsigned int flags)
1562 {
1563 struct sock *prev_ssk = NULL, *ssk = NULL;
1564 struct mptcp_sock *msk = mptcp_sk(sk);
1565 struct mptcp_sendmsg_info info = {
1566 .flags = flags,
1567 };
1568 bool do_check_data_fin = false;
1569 int push_count = 1;
1570
1571 while (mptcp_send_head(sk) && (push_count > 0)) {
1572 struct mptcp_subflow_context *subflow;
1573 int ret = 0;
1574
1575 if (mptcp_sched_get_send(msk))
1576 break;
1577
1578 push_count = 0;
1579
1580 mptcp_for_each_subflow(msk, subflow) {
1581 if (READ_ONCE(subflow->scheduled)) {
1582 mptcp_subflow_set_scheduled(subflow, false);
1583
1584 prev_ssk = ssk;
1585 ssk = mptcp_subflow_tcp_sock(subflow);
1586 if (ssk != prev_ssk) {
1587 /* First check. If the ssk has changed since
1588 * the last round, release prev_ssk
1589 */
1590 if (prev_ssk)
1591 mptcp_push_release(prev_ssk, &info);
1592
1593 /* Need to lock the new subflow only if different
1594 * from the previous one, otherwise we are still
1595 * helding the relevant lock
1596 */
1597 lock_sock(ssk);
1598 }
1599
1600 push_count++;
1601
1602 ret = __subflow_push_pending(sk, ssk, &info);
1603 if (ret <= 0) {
1604 if (ret != -EAGAIN ||
1605 (1 << ssk->sk_state) &
1606 (TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2 | TCPF_CLOSE))
1607 push_count--;
1608 continue;
1609 }
1610 do_check_data_fin = true;
1611 }
1612 }
1613 }
1614
1615 /* at this point we held the socket lock for the last subflow we used */
1616 if (ssk)
1617 mptcp_push_release(ssk, &info);
1618
1619 /* ensure the rtx timer is running */
1620 if (!mptcp_rtx_timer_pending(sk))
1621 mptcp_reset_rtx_timer(sk);
1622 if (do_check_data_fin)
1623 mptcp_check_send_data_fin(sk);
1624 }
1625
__mptcp_subflow_push_pending(struct sock * sk,struct sock * ssk,bool first)1626 static void __mptcp_subflow_push_pending(struct sock *sk, struct sock *ssk, bool first)
1627 {
1628 struct mptcp_sock *msk = mptcp_sk(sk);
1629 struct mptcp_sendmsg_info info = {
1630 .data_lock_held = true,
1631 };
1632 bool keep_pushing = true;
1633 struct sock *xmit_ssk;
1634 int copied = 0;
1635
1636 info.flags = 0;
1637 while (mptcp_send_head(sk) && keep_pushing) {
1638 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1639 int ret = 0;
1640
1641 /* check for a different subflow usage only after
1642 * spooling the first chunk of data
1643 */
1644 if (first) {
1645 mptcp_subflow_set_scheduled(subflow, false);
1646 ret = __subflow_push_pending(sk, ssk, &info);
1647 first = false;
1648 if (ret <= 0)
1649 break;
1650 copied += ret;
1651 continue;
1652 }
1653
1654 if (mptcp_sched_get_send(msk))
1655 goto out;
1656
1657 if (READ_ONCE(subflow->scheduled)) {
1658 mptcp_subflow_set_scheduled(subflow, false);
1659 ret = __subflow_push_pending(sk, ssk, &info);
1660 if (ret <= 0)
1661 keep_pushing = false;
1662 copied += ret;
1663 }
1664
1665 mptcp_for_each_subflow(msk, subflow) {
1666 if (READ_ONCE(subflow->scheduled)) {
1667 xmit_ssk = mptcp_subflow_tcp_sock(subflow);
1668 if (xmit_ssk != ssk) {
1669 mptcp_subflow_delegate(subflow,
1670 MPTCP_DELEGATE_SEND);
1671 keep_pushing = false;
1672 }
1673 }
1674 }
1675 }
1676
1677 out:
1678 /* __mptcp_alloc_tx_skb could have released some wmem and we are
1679 * not going to flush it via release_sock()
1680 */
1681 if (copied) {
1682 tcp_push(ssk, 0, info.mss_now, tcp_sk(ssk)->nonagle,
1683 info.size_goal);
1684 if (!mptcp_rtx_timer_pending(sk))
1685 mptcp_reset_rtx_timer(sk);
1686
1687 if (msk->snd_data_fin_enable &&
1688 msk->snd_nxt + 1 == msk->write_seq)
1689 mptcp_schedule_work(sk);
1690 }
1691 }
1692
mptcp_set_nospace(struct sock * sk)1693 static void mptcp_set_nospace(struct sock *sk)
1694 {
1695 /* enable autotune */
1696 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1697
1698 /* will be cleared on avail space */
1699 set_bit(MPTCP_NOSPACE, &mptcp_sk(sk)->flags);
1700 }
1701
1702 static int mptcp_disconnect(struct sock *sk, int flags);
1703
mptcp_sendmsg_fastopen(struct sock * sk,struct msghdr * msg,size_t len,int * copied_syn)1704 static int mptcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg,
1705 size_t len, int *copied_syn)
1706 {
1707 unsigned int saved_flags = msg->msg_flags;
1708 struct mptcp_sock *msk = mptcp_sk(sk);
1709 struct sock *ssk;
1710 int ret;
1711
1712 /* on flags based fastopen the mptcp is supposed to create the
1713 * first subflow right now. Otherwise we are in the defer_connect
1714 * path, and the first subflow must be already present.
1715 * Since the defer_connect flag is cleared after the first succsful
1716 * fastopen attempt, no need to check for additional subflow status.
1717 */
1718 if (msg->msg_flags & MSG_FASTOPEN) {
1719 ssk = __mptcp_nmpc_sk(msk);
1720 if (IS_ERR(ssk))
1721 return PTR_ERR(ssk);
1722 }
1723 if (!msk->first)
1724 return -EINVAL;
1725
1726 ssk = msk->first;
1727
1728 lock_sock(ssk);
1729 msg->msg_flags |= MSG_DONTWAIT;
1730 msk->fastopening = 1;
1731 ret = tcp_sendmsg_fastopen(ssk, msg, copied_syn, len, NULL);
1732 msk->fastopening = 0;
1733 msg->msg_flags = saved_flags;
1734 release_sock(ssk);
1735
1736 /* do the blocking bits of inet_stream_connect outside the ssk socket lock */
1737 if (ret == -EINPROGRESS && !(msg->msg_flags & MSG_DONTWAIT)) {
1738 ret = __inet_stream_connect(sk->sk_socket, msg->msg_name,
1739 msg->msg_namelen, msg->msg_flags, 1);
1740
1741 /* Keep the same behaviour of plain TCP: zero the copied bytes in
1742 * case of any error, except timeout or signal
1743 */
1744 if (ret && ret != -EINPROGRESS && ret != -ERESTARTSYS && ret != -EINTR)
1745 *copied_syn = 0;
1746 } else if (ret && ret != -EINPROGRESS) {
1747 /* The disconnect() op called by tcp_sendmsg_fastopen()/
1748 * __inet_stream_connect() can fail, due to looking check,
1749 * see mptcp_disconnect().
1750 * Attempt it again outside the problematic scope.
1751 */
1752 if (!mptcp_disconnect(sk, 0))
1753 sk->sk_socket->state = SS_UNCONNECTED;
1754 }
1755 inet_clear_bit(DEFER_CONNECT, sk);
1756
1757 return ret;
1758 }
1759
mptcp_sendmsg(struct sock * sk,struct msghdr * msg,size_t len)1760 static int mptcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1761 {
1762 struct mptcp_sock *msk = mptcp_sk(sk);
1763 struct page_frag *pfrag;
1764 size_t copied = 0;
1765 int ret = 0;
1766 long timeo;
1767
1768 /* silently ignore everything else */
1769 msg->msg_flags &= MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL | MSG_FASTOPEN;
1770
1771 lock_sock(sk);
1772
1773 if (unlikely(inet_test_bit(DEFER_CONNECT, sk) ||
1774 msg->msg_flags & MSG_FASTOPEN)) {
1775 int copied_syn = 0;
1776
1777 ret = mptcp_sendmsg_fastopen(sk, msg, len, &copied_syn);
1778 copied += copied_syn;
1779 if (ret == -EINPROGRESS && copied_syn > 0)
1780 goto out;
1781 else if (ret)
1782 goto do_error;
1783 }
1784
1785 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1786
1787 if ((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) {
1788 ret = sk_stream_wait_connect(sk, &timeo);
1789 if (ret)
1790 goto do_error;
1791 }
1792
1793 ret = -EPIPE;
1794 if (unlikely(sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN)))
1795 goto do_error;
1796
1797 pfrag = sk_page_frag(sk);
1798
1799 while (msg_data_left(msg)) {
1800 int total_ts, frag_truesize = 0;
1801 struct mptcp_data_frag *dfrag;
1802 bool dfrag_collapsed;
1803 size_t psize, offset;
1804
1805 /* reuse tail pfrag, if possible, or carve a new one from the
1806 * page allocator
1807 */
1808 dfrag = mptcp_pending_tail(sk);
1809 dfrag_collapsed = mptcp_frag_can_collapse_to(msk, pfrag, dfrag);
1810 if (!dfrag_collapsed) {
1811 if (!sk_stream_memory_free(sk))
1812 goto wait_for_memory;
1813
1814 if (!mptcp_page_frag_refill(sk, pfrag))
1815 goto wait_for_memory;
1816
1817 dfrag = mptcp_carve_data_frag(msk, pfrag, pfrag->offset);
1818 frag_truesize = dfrag->overhead;
1819 }
1820
1821 /* we do not bound vs wspace, to allow a single packet.
1822 * memory accounting will prevent execessive memory usage
1823 * anyway
1824 */
1825 offset = dfrag->offset + dfrag->data_len;
1826 psize = pfrag->size - offset;
1827 psize = min_t(size_t, psize, msg_data_left(msg));
1828 total_ts = psize + frag_truesize;
1829
1830 if (!sk_wmem_schedule(sk, total_ts))
1831 goto wait_for_memory;
1832
1833 if (copy_page_from_iter(dfrag->page, offset, psize,
1834 &msg->msg_iter) != psize) {
1835 ret = -EFAULT;
1836 goto do_error;
1837 }
1838
1839 /* data successfully copied into the write queue */
1840 sk_forward_alloc_add(sk, -total_ts);
1841 copied += psize;
1842 dfrag->data_len += psize;
1843 frag_truesize += psize;
1844 pfrag->offset += frag_truesize;
1845 WRITE_ONCE(msk->write_seq, msk->write_seq + psize);
1846
1847 /* charge data on mptcp pending queue to the msk socket
1848 * Note: we charge such data both to sk and ssk
1849 */
1850 sk_wmem_queued_add(sk, frag_truesize);
1851 if (!dfrag_collapsed) {
1852 get_page(dfrag->page);
1853 list_add_tail(&dfrag->list, &msk->rtx_queue);
1854 if (!msk->first_pending)
1855 WRITE_ONCE(msk->first_pending, dfrag);
1856 }
1857 pr_debug("msk=%p dfrag at seq=%llu len=%u sent=%u new=%d", msk,
1858 dfrag->data_seq, dfrag->data_len, dfrag->already_sent,
1859 !dfrag_collapsed);
1860
1861 continue;
1862
1863 wait_for_memory:
1864 mptcp_set_nospace(sk);
1865 __mptcp_push_pending(sk, msg->msg_flags);
1866 ret = sk_stream_wait_memory(sk, &timeo);
1867 if (ret)
1868 goto do_error;
1869 }
1870
1871 if (copied)
1872 __mptcp_push_pending(sk, msg->msg_flags);
1873
1874 out:
1875 release_sock(sk);
1876 return copied;
1877
1878 do_error:
1879 if (copied)
1880 goto out;
1881
1882 copied = sk_stream_error(sk, msg->msg_flags, ret);
1883 goto out;
1884 }
1885
__mptcp_recvmsg_mskq(struct mptcp_sock * msk,struct msghdr * msg,size_t len,int flags,struct scm_timestamping_internal * tss,int * cmsg_flags)1886 static int __mptcp_recvmsg_mskq(struct mptcp_sock *msk,
1887 struct msghdr *msg,
1888 size_t len, int flags,
1889 struct scm_timestamping_internal *tss,
1890 int *cmsg_flags)
1891 {
1892 struct sk_buff *skb, *tmp;
1893 int copied = 0;
1894
1895 skb_queue_walk_safe(&msk->receive_queue, skb, tmp) {
1896 u32 offset = MPTCP_SKB_CB(skb)->offset;
1897 u32 data_len = skb->len - offset;
1898 u32 count = min_t(size_t, len - copied, data_len);
1899 int err;
1900
1901 if (!(flags & MSG_TRUNC)) {
1902 err = skb_copy_datagram_msg(skb, offset, msg, count);
1903 if (unlikely(err < 0)) {
1904 if (!copied)
1905 return err;
1906 break;
1907 }
1908 }
1909
1910 if (MPTCP_SKB_CB(skb)->has_rxtstamp) {
1911 tcp_update_recv_tstamps(skb, tss);
1912 *cmsg_flags |= MPTCP_CMSG_TS;
1913 }
1914
1915 copied += count;
1916
1917 if (count < data_len) {
1918 if (!(flags & MSG_PEEK)) {
1919 MPTCP_SKB_CB(skb)->offset += count;
1920 MPTCP_SKB_CB(skb)->map_seq += count;
1921 }
1922 break;
1923 }
1924
1925 if (!(flags & MSG_PEEK)) {
1926 /* we will bulk release the skb memory later */
1927 skb->destructor = NULL;
1928 WRITE_ONCE(msk->rmem_released, msk->rmem_released + skb->truesize);
1929 __skb_unlink(skb, &msk->receive_queue);
1930 __kfree_skb(skb);
1931 }
1932
1933 if (copied >= len)
1934 break;
1935 }
1936
1937 return copied;
1938 }
1939
1940 /* receive buffer autotuning. See tcp_rcv_space_adjust for more information.
1941 *
1942 * Only difference: Use highest rtt estimate of the subflows in use.
1943 */
mptcp_rcv_space_adjust(struct mptcp_sock * msk,int copied)1944 static void mptcp_rcv_space_adjust(struct mptcp_sock *msk, int copied)
1945 {
1946 struct mptcp_subflow_context *subflow;
1947 struct sock *sk = (struct sock *)msk;
1948 u8 scaling_ratio = U8_MAX;
1949 u32 time, advmss = 1;
1950 u64 rtt_us, mstamp;
1951
1952 msk_owned_by_me(msk);
1953
1954 if (copied <= 0)
1955 return;
1956
1957 if (!msk->rcvspace_init)
1958 mptcp_rcv_space_init(msk, msk->first);
1959
1960 msk->rcvq_space.copied += copied;
1961
1962 mstamp = div_u64(tcp_clock_ns(), NSEC_PER_USEC);
1963 time = tcp_stamp_us_delta(mstamp, msk->rcvq_space.time);
1964
1965 rtt_us = msk->rcvq_space.rtt_us;
1966 if (rtt_us && time < (rtt_us >> 3))
1967 return;
1968
1969 rtt_us = 0;
1970 mptcp_for_each_subflow(msk, subflow) {
1971 const struct tcp_sock *tp;
1972 u64 sf_rtt_us;
1973 u32 sf_advmss;
1974
1975 tp = tcp_sk(mptcp_subflow_tcp_sock(subflow));
1976
1977 sf_rtt_us = READ_ONCE(tp->rcv_rtt_est.rtt_us);
1978 sf_advmss = READ_ONCE(tp->advmss);
1979
1980 rtt_us = max(sf_rtt_us, rtt_us);
1981 advmss = max(sf_advmss, advmss);
1982 scaling_ratio = min(tp->scaling_ratio, scaling_ratio);
1983 }
1984
1985 msk->rcvq_space.rtt_us = rtt_us;
1986 msk->scaling_ratio = scaling_ratio;
1987 if (time < (rtt_us >> 3) || rtt_us == 0)
1988 return;
1989
1990 if (msk->rcvq_space.copied <= msk->rcvq_space.space)
1991 goto new_measure;
1992
1993 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_moderate_rcvbuf) &&
1994 !(sk->sk_userlocks & SOCK_RCVBUF_LOCK)) {
1995 u64 rcvwin, grow;
1996 int rcvbuf;
1997
1998 rcvwin = ((u64)msk->rcvq_space.copied << 1) + 16 * advmss;
1999
2000 grow = rcvwin * (msk->rcvq_space.copied - msk->rcvq_space.space);
2001
2002 do_div(grow, msk->rcvq_space.space);
2003 rcvwin += (grow << 1);
2004
2005 rcvbuf = min_t(u64, __tcp_space_from_win(scaling_ratio, rcvwin),
2006 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_rmem[2]));
2007
2008 if (rcvbuf > sk->sk_rcvbuf) {
2009 u32 window_clamp;
2010
2011 window_clamp = __tcp_win_from_space(scaling_ratio, rcvbuf);
2012 WRITE_ONCE(sk->sk_rcvbuf, rcvbuf);
2013
2014 /* Make subflows follow along. If we do not do this, we
2015 * get drops at subflow level if skbs can't be moved to
2016 * the mptcp rx queue fast enough (announced rcv_win can
2017 * exceed ssk->sk_rcvbuf).
2018 */
2019 mptcp_for_each_subflow(msk, subflow) {
2020 struct sock *ssk;
2021 bool slow;
2022
2023 ssk = mptcp_subflow_tcp_sock(subflow);
2024 slow = lock_sock_fast(ssk);
2025 WRITE_ONCE(ssk->sk_rcvbuf, rcvbuf);
2026 tcp_sk(ssk)->window_clamp = window_clamp;
2027 tcp_cleanup_rbuf(ssk, 1);
2028 unlock_sock_fast(ssk, slow);
2029 }
2030 }
2031 }
2032
2033 msk->rcvq_space.space = msk->rcvq_space.copied;
2034 new_measure:
2035 msk->rcvq_space.copied = 0;
2036 msk->rcvq_space.time = mstamp;
2037 }
2038
__mptcp_update_rmem(struct sock * sk)2039 static void __mptcp_update_rmem(struct sock *sk)
2040 {
2041 struct mptcp_sock *msk = mptcp_sk(sk);
2042
2043 if (!msk->rmem_released)
2044 return;
2045
2046 atomic_sub(msk->rmem_released, &sk->sk_rmem_alloc);
2047 mptcp_rmem_uncharge(sk, msk->rmem_released);
2048 WRITE_ONCE(msk->rmem_released, 0);
2049 }
2050
__mptcp_splice_receive_queue(struct sock * sk)2051 static void __mptcp_splice_receive_queue(struct sock *sk)
2052 {
2053 struct mptcp_sock *msk = mptcp_sk(sk);
2054
2055 skb_queue_splice_tail_init(&sk->sk_receive_queue, &msk->receive_queue);
2056 }
2057
__mptcp_move_skbs(struct mptcp_sock * msk)2058 static bool __mptcp_move_skbs(struct mptcp_sock *msk)
2059 {
2060 struct sock *sk = (struct sock *)msk;
2061 unsigned int moved = 0;
2062 bool ret, done;
2063
2064 do {
2065 struct sock *ssk = mptcp_subflow_recv_lookup(msk);
2066 bool slowpath;
2067
2068 /* we can have data pending in the subflows only if the msk
2069 * receive buffer was full at subflow_data_ready() time,
2070 * that is an unlikely slow path.
2071 */
2072 if (likely(!ssk))
2073 break;
2074
2075 slowpath = lock_sock_fast(ssk);
2076 mptcp_data_lock(sk);
2077 __mptcp_update_rmem(sk);
2078 done = __mptcp_move_skbs_from_subflow(msk, ssk, &moved);
2079 mptcp_data_unlock(sk);
2080
2081 if (unlikely(ssk->sk_err))
2082 __mptcp_error_report(sk);
2083 unlock_sock_fast(ssk, slowpath);
2084 } while (!done);
2085
2086 /* acquire the data lock only if some input data is pending */
2087 ret = moved > 0;
2088 if (!RB_EMPTY_ROOT(&msk->out_of_order_queue) ||
2089 !skb_queue_empty_lockless(&sk->sk_receive_queue)) {
2090 mptcp_data_lock(sk);
2091 __mptcp_update_rmem(sk);
2092 ret |= __mptcp_ofo_queue(msk);
2093 __mptcp_splice_receive_queue(sk);
2094 mptcp_data_unlock(sk);
2095 }
2096 if (ret)
2097 mptcp_check_data_fin((struct sock *)msk);
2098 return !skb_queue_empty(&msk->receive_queue);
2099 }
2100
mptcp_inq_hint(const struct sock * sk)2101 static unsigned int mptcp_inq_hint(const struct sock *sk)
2102 {
2103 const struct mptcp_sock *msk = mptcp_sk(sk);
2104 const struct sk_buff *skb;
2105
2106 skb = skb_peek(&msk->receive_queue);
2107 if (skb) {
2108 u64 hint_val = msk->ack_seq - MPTCP_SKB_CB(skb)->map_seq;
2109
2110 if (hint_val >= INT_MAX)
2111 return INT_MAX;
2112
2113 return (unsigned int)hint_val;
2114 }
2115
2116 if (sk->sk_state == TCP_CLOSE || (sk->sk_shutdown & RCV_SHUTDOWN))
2117 return 1;
2118
2119 return 0;
2120 }
2121
mptcp_recvmsg(struct sock * sk,struct msghdr * msg,size_t len,int flags,int * addr_len)2122 static int mptcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
2123 int flags, int *addr_len)
2124 {
2125 struct mptcp_sock *msk = mptcp_sk(sk);
2126 struct scm_timestamping_internal tss;
2127 int copied = 0, cmsg_flags = 0;
2128 int target;
2129 long timeo;
2130
2131 /* MSG_ERRQUEUE is really a no-op till we support IP_RECVERR */
2132 if (unlikely(flags & MSG_ERRQUEUE))
2133 return inet_recv_error(sk, msg, len, addr_len);
2134
2135 lock_sock(sk);
2136 if (unlikely(sk->sk_state == TCP_LISTEN)) {
2137 copied = -ENOTCONN;
2138 goto out_err;
2139 }
2140
2141 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
2142
2143 len = min_t(size_t, len, INT_MAX);
2144 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
2145
2146 if (unlikely(msk->recvmsg_inq))
2147 cmsg_flags = MPTCP_CMSG_INQ;
2148
2149 while (copied < len) {
2150 int bytes_read;
2151
2152 bytes_read = __mptcp_recvmsg_mskq(msk, msg, len - copied, flags, &tss, &cmsg_flags);
2153 if (unlikely(bytes_read < 0)) {
2154 if (!copied)
2155 copied = bytes_read;
2156 goto out_err;
2157 }
2158
2159 copied += bytes_read;
2160
2161 /* be sure to advertise window change */
2162 mptcp_cleanup_rbuf(msk);
2163
2164 if (skb_queue_empty(&msk->receive_queue) && __mptcp_move_skbs(msk))
2165 continue;
2166
2167 /* only the master socket status is relevant here. The exit
2168 * conditions mirror closely tcp_recvmsg()
2169 */
2170 if (copied >= target)
2171 break;
2172
2173 if (copied) {
2174 if (sk->sk_err ||
2175 sk->sk_state == TCP_CLOSE ||
2176 (sk->sk_shutdown & RCV_SHUTDOWN) ||
2177 !timeo ||
2178 signal_pending(current))
2179 break;
2180 } else {
2181 if (sk->sk_err) {
2182 copied = sock_error(sk);
2183 break;
2184 }
2185
2186 if (sk->sk_shutdown & RCV_SHUTDOWN) {
2187 /* race breaker: the shutdown could be after the
2188 * previous receive queue check
2189 */
2190 if (__mptcp_move_skbs(msk))
2191 continue;
2192 break;
2193 }
2194
2195 if (sk->sk_state == TCP_CLOSE) {
2196 copied = -ENOTCONN;
2197 break;
2198 }
2199
2200 if (!timeo) {
2201 copied = -EAGAIN;
2202 break;
2203 }
2204
2205 if (signal_pending(current)) {
2206 copied = sock_intr_errno(timeo);
2207 break;
2208 }
2209 }
2210
2211 pr_debug("block timeout %ld", timeo);
2212 sk_wait_data(sk, &timeo, NULL);
2213 }
2214
2215 out_err:
2216 if (cmsg_flags && copied >= 0) {
2217 if (cmsg_flags & MPTCP_CMSG_TS)
2218 tcp_recv_timestamp(msg, sk, &tss);
2219
2220 if (cmsg_flags & MPTCP_CMSG_INQ) {
2221 unsigned int inq = mptcp_inq_hint(sk);
2222
2223 put_cmsg(msg, SOL_TCP, TCP_CM_INQ, sizeof(inq), &inq);
2224 }
2225 }
2226
2227 pr_debug("msk=%p rx queue empty=%d:%d copied=%d",
2228 msk, skb_queue_empty_lockless(&sk->sk_receive_queue),
2229 skb_queue_empty(&msk->receive_queue), copied);
2230 if (!(flags & MSG_PEEK))
2231 mptcp_rcv_space_adjust(msk, copied);
2232
2233 release_sock(sk);
2234 return copied;
2235 }
2236
mptcp_retransmit_timer(struct timer_list * t)2237 static void mptcp_retransmit_timer(struct timer_list *t)
2238 {
2239 struct inet_connection_sock *icsk = from_timer(icsk, t,
2240 icsk_retransmit_timer);
2241 struct sock *sk = &icsk->icsk_inet.sk;
2242 struct mptcp_sock *msk = mptcp_sk(sk);
2243
2244 bh_lock_sock(sk);
2245 if (!sock_owned_by_user(sk)) {
2246 /* we need a process context to retransmit */
2247 if (!test_and_set_bit(MPTCP_WORK_RTX, &msk->flags))
2248 mptcp_schedule_work(sk);
2249 } else {
2250 /* delegate our work to tcp_release_cb() */
2251 __set_bit(MPTCP_RETRANSMIT, &msk->cb_flags);
2252 }
2253 bh_unlock_sock(sk);
2254 sock_put(sk);
2255 }
2256
mptcp_tout_timer(struct timer_list * t)2257 static void mptcp_tout_timer(struct timer_list *t)
2258 {
2259 struct sock *sk = from_timer(sk, t, sk_timer);
2260
2261 mptcp_schedule_work(sk);
2262 sock_put(sk);
2263 }
2264
2265 /* Find an idle subflow. Return NULL if there is unacked data at tcp
2266 * level.
2267 *
2268 * A backup subflow is returned only if that is the only kind available.
2269 */
mptcp_subflow_get_retrans(struct mptcp_sock * msk)2270 struct sock *mptcp_subflow_get_retrans(struct mptcp_sock *msk)
2271 {
2272 struct sock *backup = NULL, *pick = NULL;
2273 struct mptcp_subflow_context *subflow;
2274 int min_stale_count = INT_MAX;
2275
2276 mptcp_for_each_subflow(msk, subflow) {
2277 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2278
2279 if (!__mptcp_subflow_active(subflow))
2280 continue;
2281
2282 /* still data outstanding at TCP level? skip this */
2283 if (!tcp_rtx_and_write_queues_empty(ssk)) {
2284 mptcp_pm_subflow_chk_stale(msk, ssk);
2285 min_stale_count = min_t(int, min_stale_count, subflow->stale_count);
2286 continue;
2287 }
2288
2289 if (subflow->backup) {
2290 if (!backup)
2291 backup = ssk;
2292 continue;
2293 }
2294
2295 if (!pick)
2296 pick = ssk;
2297 }
2298
2299 if (pick)
2300 return pick;
2301
2302 /* use backup only if there are no progresses anywhere */
2303 return min_stale_count > 1 ? backup : NULL;
2304 }
2305
__mptcp_retransmit_pending_data(struct sock * sk)2306 bool __mptcp_retransmit_pending_data(struct sock *sk)
2307 {
2308 struct mptcp_data_frag *cur, *rtx_head;
2309 struct mptcp_sock *msk = mptcp_sk(sk);
2310
2311 if (__mptcp_check_fallback(msk))
2312 return false;
2313
2314 /* the closing socket has some data untransmitted and/or unacked:
2315 * some data in the mptcp rtx queue has not really xmitted yet.
2316 * keep it simple and re-inject the whole mptcp level rtx queue
2317 */
2318 mptcp_data_lock(sk);
2319 __mptcp_clean_una_wakeup(sk);
2320 rtx_head = mptcp_rtx_head(sk);
2321 if (!rtx_head) {
2322 mptcp_data_unlock(sk);
2323 return false;
2324 }
2325
2326 msk->recovery_snd_nxt = msk->snd_nxt;
2327 msk->recovery = true;
2328 mptcp_data_unlock(sk);
2329
2330 msk->first_pending = rtx_head;
2331 msk->snd_burst = 0;
2332
2333 /* be sure to clear the "sent status" on all re-injected fragments */
2334 list_for_each_entry(cur, &msk->rtx_queue, list) {
2335 if (!cur->already_sent)
2336 break;
2337 cur->already_sent = 0;
2338 }
2339
2340 return true;
2341 }
2342
2343 /* flags for __mptcp_close_ssk() */
2344 #define MPTCP_CF_PUSH BIT(1)
2345 #define MPTCP_CF_FASTCLOSE BIT(2)
2346
2347 /* be sure to send a reset only if the caller asked for it, also
2348 * clean completely the subflow status when the subflow reaches
2349 * TCP_CLOSE state
2350 */
__mptcp_subflow_disconnect(struct sock * ssk,struct mptcp_subflow_context * subflow,unsigned int flags)2351 static void __mptcp_subflow_disconnect(struct sock *ssk,
2352 struct mptcp_subflow_context *subflow,
2353 unsigned int flags)
2354 {
2355 if (((1 << ssk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)) ||
2356 (flags & MPTCP_CF_FASTCLOSE)) {
2357 /* The MPTCP code never wait on the subflow sockets, TCP-level
2358 * disconnect should never fail
2359 */
2360 WARN_ON_ONCE(tcp_disconnect(ssk, 0));
2361 mptcp_subflow_ctx_reset(subflow);
2362 } else {
2363 tcp_shutdown(ssk, SEND_SHUTDOWN);
2364 }
2365 }
2366
2367 /* subflow sockets can be either outgoing (connect) or incoming
2368 * (accept).
2369 *
2370 * Outgoing subflows use in-kernel sockets.
2371 * Incoming subflows do not have their own 'struct socket' allocated,
2372 * so we need to use tcp_close() after detaching them from the mptcp
2373 * parent socket.
2374 */
__mptcp_close_ssk(struct sock * sk,struct sock * ssk,struct mptcp_subflow_context * subflow,unsigned int flags)2375 static void __mptcp_close_ssk(struct sock *sk, struct sock *ssk,
2376 struct mptcp_subflow_context *subflow,
2377 unsigned int flags)
2378 {
2379 struct mptcp_sock *msk = mptcp_sk(sk);
2380 bool dispose_it, need_push = false;
2381
2382 /* If the first subflow moved to a close state before accept, e.g. due
2383 * to an incoming reset or listener shutdown, the subflow socket is
2384 * already deleted by inet_child_forget() and the mptcp socket can't
2385 * survive too.
2386 */
2387 if (msk->in_accept_queue && msk->first == ssk &&
2388 (sock_flag(sk, SOCK_DEAD) || sock_flag(ssk, SOCK_DEAD))) {
2389 /* ensure later check in mptcp_worker() will dispose the msk */
2390 mptcp_set_close_tout(sk, tcp_jiffies32 - (TCP_TIMEWAIT_LEN + 1));
2391 sock_set_flag(sk, SOCK_DEAD);
2392 lock_sock_nested(ssk, SINGLE_DEPTH_NESTING);
2393 mptcp_subflow_drop_ctx(ssk);
2394 goto out_release;
2395 }
2396
2397 dispose_it = msk->free_first || ssk != msk->first;
2398 if (dispose_it)
2399 list_del(&subflow->node);
2400
2401 lock_sock_nested(ssk, SINGLE_DEPTH_NESTING);
2402
2403 if ((flags & MPTCP_CF_FASTCLOSE) && !__mptcp_check_fallback(msk)) {
2404 /* be sure to force the tcp_close path
2405 * to generate the egress reset
2406 */
2407 ssk->sk_lingertime = 0;
2408 sock_set_flag(ssk, SOCK_LINGER);
2409 subflow->send_fastclose = 1;
2410 }
2411
2412 need_push = (flags & MPTCP_CF_PUSH) && __mptcp_retransmit_pending_data(sk);
2413 if (!dispose_it) {
2414 __mptcp_subflow_disconnect(ssk, subflow, flags);
2415 release_sock(ssk);
2416
2417 goto out;
2418 }
2419
2420 subflow->disposable = 1;
2421
2422 /* if ssk hit tcp_done(), tcp_cleanup_ulp() cleared the related ops
2423 * the ssk has been already destroyed, we just need to release the
2424 * reference owned by msk;
2425 */
2426 if (!inet_csk(ssk)->icsk_ulp_ops) {
2427 WARN_ON_ONCE(!sock_flag(ssk, SOCK_DEAD));
2428 kfree_rcu(subflow, rcu);
2429 } else {
2430 /* otherwise tcp will dispose of the ssk and subflow ctx */
2431 __tcp_close(ssk, 0);
2432
2433 /* close acquired an extra ref */
2434 __sock_put(ssk);
2435 }
2436
2437 out_release:
2438 __mptcp_subflow_error_report(sk, ssk);
2439 release_sock(ssk);
2440
2441 sock_put(ssk);
2442
2443 if (ssk == msk->first)
2444 WRITE_ONCE(msk->first, NULL);
2445
2446 out:
2447 __mptcp_sync_sndbuf(sk);
2448 if (need_push)
2449 __mptcp_push_pending(sk, 0);
2450
2451 /* Catch every 'all subflows closed' scenario, including peers silently
2452 * closing them, e.g. due to timeout.
2453 * For established sockets, allow an additional timeout before closing,
2454 * as the protocol can still create more subflows.
2455 */
2456 if (list_is_singular(&msk->conn_list) && msk->first &&
2457 inet_sk_state_load(msk->first) == TCP_CLOSE) {
2458 if (sk->sk_state != TCP_ESTABLISHED ||
2459 msk->in_accept_queue || sock_flag(sk, SOCK_DEAD)) {
2460 mptcp_set_state(sk, TCP_CLOSE);
2461 mptcp_close_wake_up(sk);
2462 } else {
2463 mptcp_start_tout_timer(sk);
2464 }
2465 }
2466 }
2467
mptcp_close_ssk(struct sock * sk,struct sock * ssk,struct mptcp_subflow_context * subflow)2468 void mptcp_close_ssk(struct sock *sk, struct sock *ssk,
2469 struct mptcp_subflow_context *subflow)
2470 {
2471 if (sk->sk_state == TCP_ESTABLISHED)
2472 mptcp_event(MPTCP_EVENT_SUB_CLOSED, mptcp_sk(sk), ssk, GFP_KERNEL);
2473
2474 /* subflow aborted before reaching the fully_established status
2475 * attempt the creation of the next subflow
2476 */
2477 mptcp_pm_subflow_check_next(mptcp_sk(sk), ssk, subflow);
2478
2479 __mptcp_close_ssk(sk, ssk, subflow, MPTCP_CF_PUSH);
2480 }
2481
mptcp_sync_mss(struct sock * sk,u32 pmtu)2482 static unsigned int mptcp_sync_mss(struct sock *sk, u32 pmtu)
2483 {
2484 return 0;
2485 }
2486
__mptcp_close_subflow(struct sock * sk)2487 static void __mptcp_close_subflow(struct sock *sk)
2488 {
2489 struct mptcp_subflow_context *subflow, *tmp;
2490 struct mptcp_sock *msk = mptcp_sk(sk);
2491
2492 might_sleep();
2493
2494 mptcp_for_each_subflow_safe(msk, subflow, tmp) {
2495 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2496
2497 if (inet_sk_state_load(ssk) != TCP_CLOSE)
2498 continue;
2499
2500 /* 'subflow_data_ready' will re-sched once rx queue is empty */
2501 if (!skb_queue_empty_lockless(&ssk->sk_receive_queue))
2502 continue;
2503
2504 mptcp_close_ssk(sk, ssk, subflow);
2505 }
2506
2507 }
2508
mptcp_close_tout_expired(const struct sock * sk)2509 static bool mptcp_close_tout_expired(const struct sock *sk)
2510 {
2511 if (!inet_csk(sk)->icsk_mtup.probe_timestamp ||
2512 sk->sk_state == TCP_CLOSE)
2513 return false;
2514
2515 return time_after32(tcp_jiffies32,
2516 inet_csk(sk)->icsk_mtup.probe_timestamp + TCP_TIMEWAIT_LEN);
2517 }
2518
mptcp_check_fastclose(struct mptcp_sock * msk)2519 static void mptcp_check_fastclose(struct mptcp_sock *msk)
2520 {
2521 struct mptcp_subflow_context *subflow, *tmp;
2522 struct sock *sk = (struct sock *)msk;
2523
2524 if (likely(!READ_ONCE(msk->rcv_fastclose)))
2525 return;
2526
2527 mptcp_token_destroy(msk);
2528
2529 mptcp_for_each_subflow_safe(msk, subflow, tmp) {
2530 struct sock *tcp_sk = mptcp_subflow_tcp_sock(subflow);
2531 bool slow;
2532
2533 slow = lock_sock_fast(tcp_sk);
2534 if (tcp_sk->sk_state != TCP_CLOSE) {
2535 tcp_send_active_reset(tcp_sk, GFP_ATOMIC);
2536 tcp_set_state(tcp_sk, TCP_CLOSE);
2537 }
2538 unlock_sock_fast(tcp_sk, slow);
2539 }
2540
2541 /* Mirror the tcp_reset() error propagation */
2542 switch (sk->sk_state) {
2543 case TCP_SYN_SENT:
2544 WRITE_ONCE(sk->sk_err, ECONNREFUSED);
2545 break;
2546 case TCP_CLOSE_WAIT:
2547 WRITE_ONCE(sk->sk_err, EPIPE);
2548 break;
2549 case TCP_CLOSE:
2550 return;
2551 default:
2552 WRITE_ONCE(sk->sk_err, ECONNRESET);
2553 }
2554
2555 mptcp_set_state(sk, TCP_CLOSE);
2556 WRITE_ONCE(sk->sk_shutdown, SHUTDOWN_MASK);
2557 smp_mb__before_atomic(); /* SHUTDOWN must be visible first */
2558 set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags);
2559
2560 /* the calling mptcp_worker will properly destroy the socket */
2561 if (sock_flag(sk, SOCK_DEAD))
2562 return;
2563
2564 sk->sk_state_change(sk);
2565 sk_error_report(sk);
2566 }
2567
__mptcp_retrans(struct sock * sk)2568 static void __mptcp_retrans(struct sock *sk)
2569 {
2570 struct mptcp_sock *msk = mptcp_sk(sk);
2571 struct mptcp_subflow_context *subflow;
2572 struct mptcp_sendmsg_info info = {};
2573 struct mptcp_data_frag *dfrag;
2574 struct sock *ssk;
2575 int ret, err;
2576 u16 len = 0;
2577
2578 mptcp_clean_una_wakeup(sk);
2579
2580 /* first check ssk: need to kick "stale" logic */
2581 err = mptcp_sched_get_retrans(msk);
2582 dfrag = mptcp_rtx_head(sk);
2583 if (!dfrag) {
2584 if (mptcp_data_fin_enabled(msk)) {
2585 struct inet_connection_sock *icsk = inet_csk(sk);
2586
2587 icsk->icsk_retransmits++;
2588 mptcp_set_datafin_timeout(sk);
2589 mptcp_send_ack(msk);
2590
2591 goto reset_timer;
2592 }
2593
2594 if (!mptcp_send_head(sk))
2595 return;
2596
2597 goto reset_timer;
2598 }
2599
2600 if (err)
2601 goto reset_timer;
2602
2603 mptcp_for_each_subflow(msk, subflow) {
2604 if (READ_ONCE(subflow->scheduled)) {
2605 u16 copied = 0;
2606
2607 mptcp_subflow_set_scheduled(subflow, false);
2608
2609 ssk = mptcp_subflow_tcp_sock(subflow);
2610
2611 lock_sock(ssk);
2612
2613 /* limit retransmission to the bytes already sent on some subflows */
2614 info.sent = 0;
2615 info.limit = READ_ONCE(msk->csum_enabled) ? dfrag->data_len :
2616 dfrag->already_sent;
2617 while (info.sent < info.limit) {
2618 ret = mptcp_sendmsg_frag(sk, ssk, dfrag, &info);
2619 if (ret <= 0)
2620 break;
2621
2622 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_RETRANSSEGS);
2623 copied += ret;
2624 info.sent += ret;
2625 }
2626 if (copied) {
2627 len = max(copied, len);
2628 tcp_push(ssk, 0, info.mss_now, tcp_sk(ssk)->nonagle,
2629 info.size_goal);
2630 WRITE_ONCE(msk->allow_infinite_fallback, false);
2631 }
2632
2633 release_sock(ssk);
2634 }
2635 }
2636
2637 msk->bytes_retrans += len;
2638 dfrag->already_sent = max(dfrag->already_sent, len);
2639
2640 reset_timer:
2641 mptcp_check_and_set_pending(sk);
2642
2643 if (!mptcp_rtx_timer_pending(sk))
2644 mptcp_reset_rtx_timer(sk);
2645 }
2646
2647 /* schedule the timeout timer for the relevant event: either close timeout
2648 * or mp_fail timeout. The close timeout takes precedence on the mp_fail one
2649 */
mptcp_reset_tout_timer(struct mptcp_sock * msk,unsigned long fail_tout)2650 void mptcp_reset_tout_timer(struct mptcp_sock *msk, unsigned long fail_tout)
2651 {
2652 struct sock *sk = (struct sock *)msk;
2653 unsigned long timeout, close_timeout;
2654
2655 if (!fail_tout && !inet_csk(sk)->icsk_mtup.probe_timestamp)
2656 return;
2657
2658 close_timeout = inet_csk(sk)->icsk_mtup.probe_timestamp - tcp_jiffies32 + jiffies +
2659 TCP_TIMEWAIT_LEN;
2660
2661 /* the close timeout takes precedence on the fail one, and here at least one of
2662 * them is active
2663 */
2664 timeout = inet_csk(sk)->icsk_mtup.probe_timestamp ? close_timeout : fail_tout;
2665
2666 sk_reset_timer(sk, &sk->sk_timer, timeout);
2667 }
2668
mptcp_mp_fail_no_response(struct mptcp_sock * msk)2669 static void mptcp_mp_fail_no_response(struct mptcp_sock *msk)
2670 {
2671 struct sock *ssk = msk->first;
2672 bool slow;
2673
2674 if (!ssk)
2675 return;
2676
2677 pr_debug("MP_FAIL doesn't respond, reset the subflow");
2678
2679 slow = lock_sock_fast(ssk);
2680 mptcp_subflow_reset(ssk);
2681 WRITE_ONCE(mptcp_subflow_ctx(ssk)->fail_tout, 0);
2682 unlock_sock_fast(ssk, slow);
2683 }
2684
mptcp_do_fastclose(struct sock * sk)2685 static void mptcp_do_fastclose(struct sock *sk)
2686 {
2687 struct mptcp_subflow_context *subflow, *tmp;
2688 struct mptcp_sock *msk = mptcp_sk(sk);
2689
2690 mptcp_set_state(sk, TCP_CLOSE);
2691 mptcp_for_each_subflow_safe(msk, subflow, tmp)
2692 __mptcp_close_ssk(sk, mptcp_subflow_tcp_sock(subflow),
2693 subflow, MPTCP_CF_FASTCLOSE);
2694 }
2695
mptcp_worker(struct work_struct * work)2696 static void mptcp_worker(struct work_struct *work)
2697 {
2698 struct mptcp_sock *msk = container_of(work, struct mptcp_sock, work);
2699 struct sock *sk = (struct sock *)msk;
2700 unsigned long fail_tout;
2701 int state;
2702
2703 lock_sock(sk);
2704 state = sk->sk_state;
2705 if (unlikely((1 << state) & (TCPF_CLOSE | TCPF_LISTEN)))
2706 goto unlock;
2707
2708 mptcp_check_fastclose(msk);
2709
2710 mptcp_pm_nl_work(msk);
2711
2712 mptcp_check_send_data_fin(sk);
2713 mptcp_check_data_fin_ack(sk);
2714 mptcp_check_data_fin(sk);
2715
2716 if (test_and_clear_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags))
2717 __mptcp_close_subflow(sk);
2718
2719 if (mptcp_close_tout_expired(sk)) {
2720 mptcp_do_fastclose(sk);
2721 mptcp_close_wake_up(sk);
2722 }
2723
2724 if (sock_flag(sk, SOCK_DEAD) && sk->sk_state == TCP_CLOSE) {
2725 __mptcp_destroy_sock(sk);
2726 goto unlock;
2727 }
2728
2729 if (test_and_clear_bit(MPTCP_WORK_RTX, &msk->flags))
2730 __mptcp_retrans(sk);
2731
2732 fail_tout = msk->first ? READ_ONCE(mptcp_subflow_ctx(msk->first)->fail_tout) : 0;
2733 if (fail_tout && time_after(jiffies, fail_tout))
2734 mptcp_mp_fail_no_response(msk);
2735
2736 unlock:
2737 release_sock(sk);
2738 sock_put(sk);
2739 }
2740
__mptcp_init_sock(struct sock * sk)2741 static void __mptcp_init_sock(struct sock *sk)
2742 {
2743 struct mptcp_sock *msk = mptcp_sk(sk);
2744
2745 INIT_LIST_HEAD(&msk->conn_list);
2746 INIT_LIST_HEAD(&msk->join_list);
2747 INIT_LIST_HEAD(&msk->rtx_queue);
2748 INIT_WORK(&msk->work, mptcp_worker);
2749 __skb_queue_head_init(&msk->receive_queue);
2750 msk->out_of_order_queue = RB_ROOT;
2751 msk->first_pending = NULL;
2752 msk->rmem_fwd_alloc = 0;
2753 WRITE_ONCE(msk->rmem_released, 0);
2754 msk->timer_ival = TCP_RTO_MIN;
2755
2756 WRITE_ONCE(msk->first, NULL);
2757 inet_csk(sk)->icsk_sync_mss = mptcp_sync_mss;
2758 WRITE_ONCE(msk->csum_enabled, mptcp_is_checksum_enabled(sock_net(sk)));
2759 WRITE_ONCE(msk->allow_infinite_fallback, true);
2760 msk->recovery = false;
2761 msk->subflow_id = 1;
2762
2763 mptcp_pm_data_init(msk);
2764
2765 /* re-use the csk retrans timer for MPTCP-level retrans */
2766 timer_setup(&msk->sk.icsk_retransmit_timer, mptcp_retransmit_timer, 0);
2767 timer_setup(&sk->sk_timer, mptcp_tout_timer, 0);
2768 }
2769
mptcp_ca_reset(struct sock * sk)2770 static void mptcp_ca_reset(struct sock *sk)
2771 {
2772 struct inet_connection_sock *icsk = inet_csk(sk);
2773
2774 tcp_assign_congestion_control(sk);
2775 strcpy(mptcp_sk(sk)->ca_name, icsk->icsk_ca_ops->name);
2776
2777 /* no need to keep a reference to the ops, the name will suffice */
2778 tcp_cleanup_congestion_control(sk);
2779 icsk->icsk_ca_ops = NULL;
2780 }
2781
mptcp_init_sock(struct sock * sk)2782 static int mptcp_init_sock(struct sock *sk)
2783 {
2784 struct net *net = sock_net(sk);
2785 int ret;
2786
2787 __mptcp_init_sock(sk);
2788
2789 if (!mptcp_is_enabled(net))
2790 return -ENOPROTOOPT;
2791
2792 if (unlikely(!net->mib.mptcp_statistics) && !mptcp_mib_alloc(net))
2793 return -ENOMEM;
2794
2795 ret = mptcp_init_sched(mptcp_sk(sk),
2796 mptcp_sched_find(mptcp_get_scheduler(net)));
2797 if (ret)
2798 return ret;
2799
2800 set_bit(SOCK_CUSTOM_SOCKOPT, &sk->sk_socket->flags);
2801
2802 /* fetch the ca name; do it outside __mptcp_init_sock(), so that clone will
2803 * propagate the correct value
2804 */
2805 mptcp_ca_reset(sk);
2806
2807 sk_sockets_allocated_inc(sk);
2808 sk->sk_rcvbuf = READ_ONCE(net->ipv4.sysctl_tcp_rmem[1]);
2809 sk->sk_sndbuf = READ_ONCE(net->ipv4.sysctl_tcp_wmem[1]);
2810
2811 return 0;
2812 }
2813
__mptcp_clear_xmit(struct sock * sk)2814 static void __mptcp_clear_xmit(struct sock *sk)
2815 {
2816 struct mptcp_sock *msk = mptcp_sk(sk);
2817 struct mptcp_data_frag *dtmp, *dfrag;
2818
2819 WRITE_ONCE(msk->first_pending, NULL);
2820 list_for_each_entry_safe(dfrag, dtmp, &msk->rtx_queue, list)
2821 dfrag_clear(sk, dfrag);
2822 }
2823
mptcp_cancel_work(struct sock * sk)2824 void mptcp_cancel_work(struct sock *sk)
2825 {
2826 struct mptcp_sock *msk = mptcp_sk(sk);
2827
2828 if (cancel_work_sync(&msk->work))
2829 __sock_put(sk);
2830 }
2831
mptcp_subflow_shutdown(struct sock * sk,struct sock * ssk,int how)2832 void mptcp_subflow_shutdown(struct sock *sk, struct sock *ssk, int how)
2833 {
2834 lock_sock(ssk);
2835
2836 switch (ssk->sk_state) {
2837 case TCP_LISTEN:
2838 if (!(how & RCV_SHUTDOWN))
2839 break;
2840 fallthrough;
2841 case TCP_SYN_SENT:
2842 WARN_ON_ONCE(tcp_disconnect(ssk, O_NONBLOCK));
2843 break;
2844 default:
2845 if (__mptcp_check_fallback(mptcp_sk(sk))) {
2846 pr_debug("Fallback");
2847 ssk->sk_shutdown |= how;
2848 tcp_shutdown(ssk, how);
2849
2850 /* simulate the data_fin ack reception to let the state
2851 * machine move forward
2852 */
2853 WRITE_ONCE(mptcp_sk(sk)->snd_una, mptcp_sk(sk)->snd_nxt);
2854 mptcp_schedule_work(sk);
2855 } else {
2856 pr_debug("Sending DATA_FIN on subflow %p", ssk);
2857 tcp_send_ack(ssk);
2858 if (!mptcp_rtx_timer_pending(sk))
2859 mptcp_reset_rtx_timer(sk);
2860 }
2861 break;
2862 }
2863
2864 release_sock(ssk);
2865 }
2866
mptcp_set_state(struct sock * sk,int state)2867 void mptcp_set_state(struct sock *sk, int state)
2868 {
2869 int oldstate = sk->sk_state;
2870
2871 switch (state) {
2872 case TCP_ESTABLISHED:
2873 if (oldstate != TCP_ESTABLISHED)
2874 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_CURRESTAB);
2875 break;
2876 case TCP_CLOSE_WAIT:
2877 /* Unlike TCP, MPTCP sk would not have the TCP_SYN_RECV state:
2878 * MPTCP "accepted" sockets will be created later on. So no
2879 * transition from TCP_SYN_RECV to TCP_CLOSE_WAIT.
2880 */
2881 break;
2882 default:
2883 if (oldstate == TCP_ESTABLISHED || oldstate == TCP_CLOSE_WAIT)
2884 MPTCP_DEC_STATS(sock_net(sk), MPTCP_MIB_CURRESTAB);
2885 }
2886
2887 inet_sk_state_store(sk, state);
2888 }
2889
2890 static const unsigned char new_state[16] = {
2891 /* current state: new state: action: */
2892 [0 /* (Invalid) */] = TCP_CLOSE,
2893 [TCP_ESTABLISHED] = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2894 [TCP_SYN_SENT] = TCP_CLOSE,
2895 [TCP_SYN_RECV] = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2896 [TCP_FIN_WAIT1] = TCP_FIN_WAIT1,
2897 [TCP_FIN_WAIT2] = TCP_FIN_WAIT2,
2898 [TCP_TIME_WAIT] = TCP_CLOSE, /* should not happen ! */
2899 [TCP_CLOSE] = TCP_CLOSE,
2900 [TCP_CLOSE_WAIT] = TCP_LAST_ACK | TCP_ACTION_FIN,
2901 [TCP_LAST_ACK] = TCP_LAST_ACK,
2902 [TCP_LISTEN] = TCP_CLOSE,
2903 [TCP_CLOSING] = TCP_CLOSING,
2904 [TCP_NEW_SYN_RECV] = TCP_CLOSE, /* should not happen ! */
2905 };
2906
mptcp_close_state(struct sock * sk)2907 static int mptcp_close_state(struct sock *sk)
2908 {
2909 int next = (int)new_state[sk->sk_state];
2910 int ns = next & TCP_STATE_MASK;
2911
2912 mptcp_set_state(sk, ns);
2913
2914 return next & TCP_ACTION_FIN;
2915 }
2916
mptcp_check_send_data_fin(struct sock * sk)2917 static void mptcp_check_send_data_fin(struct sock *sk)
2918 {
2919 struct mptcp_subflow_context *subflow;
2920 struct mptcp_sock *msk = mptcp_sk(sk);
2921
2922 pr_debug("msk=%p snd_data_fin_enable=%d pending=%d snd_nxt=%llu write_seq=%llu",
2923 msk, msk->snd_data_fin_enable, !!mptcp_send_head(sk),
2924 msk->snd_nxt, msk->write_seq);
2925
2926 /* we still need to enqueue subflows or not really shutting down,
2927 * skip this
2928 */
2929 if (!msk->snd_data_fin_enable || msk->snd_nxt + 1 != msk->write_seq ||
2930 mptcp_send_head(sk))
2931 return;
2932
2933 WRITE_ONCE(msk->snd_nxt, msk->write_seq);
2934
2935 mptcp_for_each_subflow(msk, subflow) {
2936 struct sock *tcp_sk = mptcp_subflow_tcp_sock(subflow);
2937
2938 mptcp_subflow_shutdown(sk, tcp_sk, SEND_SHUTDOWN);
2939 }
2940 }
2941
__mptcp_wr_shutdown(struct sock * sk)2942 static void __mptcp_wr_shutdown(struct sock *sk)
2943 {
2944 struct mptcp_sock *msk = mptcp_sk(sk);
2945
2946 pr_debug("msk=%p snd_data_fin_enable=%d shutdown=%x state=%d pending=%d",
2947 msk, msk->snd_data_fin_enable, sk->sk_shutdown, sk->sk_state,
2948 !!mptcp_send_head(sk));
2949
2950 /* will be ignored by fallback sockets */
2951 WRITE_ONCE(msk->write_seq, msk->write_seq + 1);
2952 WRITE_ONCE(msk->snd_data_fin_enable, 1);
2953
2954 mptcp_check_send_data_fin(sk);
2955 }
2956
__mptcp_destroy_sock(struct sock * sk)2957 static void __mptcp_destroy_sock(struct sock *sk)
2958 {
2959 struct mptcp_sock *msk = mptcp_sk(sk);
2960
2961 pr_debug("msk=%p", msk);
2962
2963 might_sleep();
2964
2965 mptcp_stop_rtx_timer(sk);
2966 sk_stop_timer(sk, &sk->sk_timer);
2967 msk->pm.status = 0;
2968 mptcp_release_sched(msk);
2969
2970 sk->sk_prot->destroy(sk);
2971
2972 WARN_ON_ONCE(msk->rmem_fwd_alloc);
2973 WARN_ON_ONCE(msk->rmem_released);
2974 sk_stream_kill_queues(sk);
2975 xfrm_sk_free_policy(sk);
2976
2977 sock_put(sk);
2978 }
2979
__mptcp_unaccepted_force_close(struct sock * sk)2980 void __mptcp_unaccepted_force_close(struct sock *sk)
2981 {
2982 sock_set_flag(sk, SOCK_DEAD);
2983 mptcp_do_fastclose(sk);
2984 __mptcp_destroy_sock(sk);
2985 }
2986
mptcp_check_readable(struct mptcp_sock * msk)2987 static __poll_t mptcp_check_readable(struct mptcp_sock *msk)
2988 {
2989 /* Concurrent splices from sk_receive_queue into receive_queue will
2990 * always show at least one non-empty queue when checked in this order.
2991 */
2992 if (skb_queue_empty_lockless(&((struct sock *)msk)->sk_receive_queue) &&
2993 skb_queue_empty_lockless(&msk->receive_queue))
2994 return 0;
2995
2996 return EPOLLIN | EPOLLRDNORM;
2997 }
2998
mptcp_check_listen_stop(struct sock * sk)2999 static void mptcp_check_listen_stop(struct sock *sk)
3000 {
3001 struct sock *ssk;
3002
3003 if (inet_sk_state_load(sk) != TCP_LISTEN)
3004 return;
3005
3006 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
3007 ssk = mptcp_sk(sk)->first;
3008 if (WARN_ON_ONCE(!ssk || inet_sk_state_load(ssk) != TCP_LISTEN))
3009 return;
3010
3011 lock_sock_nested(ssk, SINGLE_DEPTH_NESTING);
3012 tcp_set_state(ssk, TCP_CLOSE);
3013 mptcp_subflow_queue_clean(sk, ssk);
3014 inet_csk_listen_stop(ssk);
3015 mptcp_event_pm_listener(ssk, MPTCP_EVENT_LISTENER_CLOSED);
3016 release_sock(ssk);
3017 }
3018
__mptcp_close(struct sock * sk,long timeout)3019 bool __mptcp_close(struct sock *sk, long timeout)
3020 {
3021 struct mptcp_subflow_context *subflow;
3022 struct mptcp_sock *msk = mptcp_sk(sk);
3023 bool do_cancel_work = false;
3024 int subflows_alive = 0;
3025
3026 WRITE_ONCE(sk->sk_shutdown, SHUTDOWN_MASK);
3027
3028 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE)) {
3029 mptcp_check_listen_stop(sk);
3030 mptcp_set_state(sk, TCP_CLOSE);
3031 goto cleanup;
3032 }
3033
3034 if (mptcp_check_readable(msk) || timeout < 0) {
3035 /* If the msk has read data, or the caller explicitly ask it,
3036 * do the MPTCP equivalent of TCP reset, aka MPTCP fastclose
3037 */
3038 mptcp_do_fastclose(sk);
3039 timeout = 0;
3040 } else if (mptcp_close_state(sk)) {
3041 __mptcp_wr_shutdown(sk);
3042 }
3043
3044 sk_stream_wait_close(sk, timeout);
3045
3046 cleanup:
3047 /* orphan all the subflows */
3048 mptcp_for_each_subflow(msk, subflow) {
3049 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
3050 bool slow = lock_sock_fast_nested(ssk);
3051
3052 subflows_alive += ssk->sk_state != TCP_CLOSE;
3053
3054 /* since the close timeout takes precedence on the fail one,
3055 * cancel the latter
3056 */
3057 if (ssk == msk->first)
3058 subflow->fail_tout = 0;
3059
3060 /* detach from the parent socket, but allow data_ready to
3061 * push incoming data into the mptcp stack, to properly ack it
3062 */
3063 ssk->sk_socket = NULL;
3064 ssk->sk_wq = NULL;
3065 unlock_sock_fast(ssk, slow);
3066 }
3067 sock_orphan(sk);
3068
3069 /* all the subflows are closed, only timeout can change the msk
3070 * state, let's not keep resources busy for no reasons
3071 */
3072 if (subflows_alive == 0)
3073 mptcp_set_state(sk, TCP_CLOSE);
3074
3075 sock_hold(sk);
3076 pr_debug("msk=%p state=%d", sk, sk->sk_state);
3077 if (msk->token)
3078 mptcp_event(MPTCP_EVENT_CLOSED, msk, NULL, GFP_KERNEL);
3079
3080 if (sk->sk_state == TCP_CLOSE) {
3081 __mptcp_destroy_sock(sk);
3082 do_cancel_work = true;
3083 } else {
3084 mptcp_start_tout_timer(sk);
3085 }
3086
3087 return do_cancel_work;
3088 }
3089
mptcp_close(struct sock * sk,long timeout)3090 static void mptcp_close(struct sock *sk, long timeout)
3091 {
3092 bool do_cancel_work;
3093
3094 lock_sock(sk);
3095
3096 do_cancel_work = __mptcp_close(sk, timeout);
3097 release_sock(sk);
3098 if (do_cancel_work)
3099 mptcp_cancel_work(sk);
3100
3101 sock_put(sk);
3102 }
3103
mptcp_copy_inaddrs(struct sock * msk,const struct sock * ssk)3104 static void mptcp_copy_inaddrs(struct sock *msk, const struct sock *ssk)
3105 {
3106 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
3107 const struct ipv6_pinfo *ssk6 = inet6_sk(ssk);
3108 struct ipv6_pinfo *msk6 = inet6_sk(msk);
3109
3110 msk->sk_v6_daddr = ssk->sk_v6_daddr;
3111 msk->sk_v6_rcv_saddr = ssk->sk_v6_rcv_saddr;
3112
3113 if (msk6 && ssk6) {
3114 msk6->saddr = ssk6->saddr;
3115 msk6->flow_label = ssk6->flow_label;
3116 }
3117 #endif
3118
3119 inet_sk(msk)->inet_num = inet_sk(ssk)->inet_num;
3120 inet_sk(msk)->inet_dport = inet_sk(ssk)->inet_dport;
3121 inet_sk(msk)->inet_sport = inet_sk(ssk)->inet_sport;
3122 inet_sk(msk)->inet_daddr = inet_sk(ssk)->inet_daddr;
3123 inet_sk(msk)->inet_saddr = inet_sk(ssk)->inet_saddr;
3124 inet_sk(msk)->inet_rcv_saddr = inet_sk(ssk)->inet_rcv_saddr;
3125 }
3126
mptcp_disconnect(struct sock * sk,int flags)3127 static int mptcp_disconnect(struct sock *sk, int flags)
3128 {
3129 struct mptcp_sock *msk = mptcp_sk(sk);
3130
3131 /* We are on the fastopen error path. We can't call straight into the
3132 * subflows cleanup code due to lock nesting (we are already under
3133 * msk->firstsocket lock).
3134 */
3135 if (msk->fastopening)
3136 return -EBUSY;
3137
3138 mptcp_check_listen_stop(sk);
3139 mptcp_set_state(sk, TCP_CLOSE);
3140
3141 mptcp_stop_rtx_timer(sk);
3142 mptcp_stop_tout_timer(sk);
3143
3144 if (msk->token)
3145 mptcp_event(MPTCP_EVENT_CLOSED, msk, NULL, GFP_KERNEL);
3146
3147 /* msk->subflow is still intact, the following will not free the first
3148 * subflow
3149 */
3150 mptcp_destroy_common(msk, MPTCP_CF_FASTCLOSE);
3151 WRITE_ONCE(msk->flags, 0);
3152 msk->cb_flags = 0;
3153 msk->recovery = false;
3154 msk->can_ack = false;
3155 msk->fully_established = false;
3156 msk->rcv_data_fin = false;
3157 msk->snd_data_fin_enable = false;
3158 msk->rcv_fastclose = false;
3159 msk->use_64bit_ack = false;
3160 WRITE_ONCE(msk->csum_enabled, mptcp_is_checksum_enabled(sock_net(sk)));
3161 mptcp_pm_data_reset(msk);
3162 mptcp_ca_reset(sk);
3163 msk->bytes_acked = 0;
3164 msk->bytes_received = 0;
3165 msk->bytes_sent = 0;
3166 msk->bytes_retrans = 0;
3167 msk->rcvspace_init = 0;
3168
3169 WRITE_ONCE(sk->sk_shutdown, 0);
3170 sk_error_report(sk);
3171 return 0;
3172 }
3173
3174 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
mptcp_inet6_sk(const struct sock * sk)3175 static struct ipv6_pinfo *mptcp_inet6_sk(const struct sock *sk)
3176 {
3177 unsigned int offset = sizeof(struct mptcp6_sock) - sizeof(struct ipv6_pinfo);
3178
3179 return (struct ipv6_pinfo *)(((u8 *)sk) + offset);
3180 }
3181
mptcp_copy_ip6_options(struct sock * newsk,const struct sock * sk)3182 static void mptcp_copy_ip6_options(struct sock *newsk, const struct sock *sk)
3183 {
3184 const struct ipv6_pinfo *np = inet6_sk(sk);
3185 struct ipv6_txoptions *opt;
3186 struct ipv6_pinfo *newnp;
3187
3188 newnp = inet6_sk(newsk);
3189
3190 rcu_read_lock();
3191 opt = rcu_dereference(np->opt);
3192 if (opt) {
3193 opt = ipv6_dup_options(newsk, opt);
3194 if (!opt)
3195 net_warn_ratelimited("%s: Failed to copy ip6 options\n", __func__);
3196 }
3197 RCU_INIT_POINTER(newnp->opt, opt);
3198 rcu_read_unlock();
3199 }
3200 #endif
3201
mptcp_copy_ip_options(struct sock * newsk,const struct sock * sk)3202 static void mptcp_copy_ip_options(struct sock *newsk, const struct sock *sk)
3203 {
3204 struct ip_options_rcu *inet_opt, *newopt = NULL;
3205 const struct inet_sock *inet = inet_sk(sk);
3206 struct inet_sock *newinet;
3207
3208 newinet = inet_sk(newsk);
3209
3210 rcu_read_lock();
3211 inet_opt = rcu_dereference(inet->inet_opt);
3212 if (inet_opt) {
3213 newopt = sock_kmalloc(newsk, sizeof(*inet_opt) +
3214 inet_opt->opt.optlen, GFP_ATOMIC);
3215 if (newopt)
3216 memcpy(newopt, inet_opt, sizeof(*inet_opt) +
3217 inet_opt->opt.optlen);
3218 else
3219 net_warn_ratelimited("%s: Failed to copy ip options\n", __func__);
3220 }
3221 RCU_INIT_POINTER(newinet->inet_opt, newopt);
3222 rcu_read_unlock();
3223 }
3224
mptcp_sk_clone_init(const struct sock * sk,const struct mptcp_options_received * mp_opt,struct sock * ssk,struct request_sock * req)3225 struct sock *mptcp_sk_clone_init(const struct sock *sk,
3226 const struct mptcp_options_received *mp_opt,
3227 struct sock *ssk,
3228 struct request_sock *req)
3229 {
3230 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
3231 struct sock *nsk = sk_clone_lock(sk, GFP_ATOMIC);
3232 struct mptcp_subflow_context *subflow;
3233 struct mptcp_sock *msk;
3234
3235 if (!nsk)
3236 return NULL;
3237
3238 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
3239 if (nsk->sk_family == AF_INET6)
3240 inet_sk(nsk)->pinet6 = mptcp_inet6_sk(nsk);
3241 #endif
3242
3243 __mptcp_init_sock(nsk);
3244
3245 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
3246 if (nsk->sk_family == AF_INET6)
3247 mptcp_copy_ip6_options(nsk, sk);
3248 else
3249 #endif
3250 mptcp_copy_ip_options(nsk, sk);
3251
3252 msk = mptcp_sk(nsk);
3253 msk->local_key = subflow_req->local_key;
3254 msk->token = subflow_req->token;
3255 msk->in_accept_queue = 1;
3256 WRITE_ONCE(msk->fully_established, false);
3257 if (mp_opt->suboptions & OPTION_MPTCP_CSUMREQD)
3258 WRITE_ONCE(msk->csum_enabled, true);
3259
3260 msk->write_seq = subflow_req->idsn + 1;
3261 msk->snd_nxt = msk->write_seq;
3262 msk->snd_una = msk->write_seq;
3263 msk->wnd_end = msk->snd_nxt + tcp_sk(ssk)->snd_wnd;
3264 msk->setsockopt_seq = mptcp_sk(sk)->setsockopt_seq;
3265 mptcp_init_sched(msk, mptcp_sk(sk)->sched);
3266
3267 /* passive msk is created after the first/MPC subflow */
3268 msk->subflow_id = 2;
3269
3270 sock_reset_flag(nsk, SOCK_RCU_FREE);
3271 security_inet_csk_clone(nsk, req);
3272
3273 /* this can't race with mptcp_close(), as the msk is
3274 * not yet exposted to user-space
3275 */
3276 mptcp_set_state(nsk, TCP_ESTABLISHED);
3277
3278 /* The msk maintain a ref to each subflow in the connections list */
3279 WRITE_ONCE(msk->first, ssk);
3280 subflow = mptcp_subflow_ctx(ssk);
3281 list_add(&subflow->node, &msk->conn_list);
3282 sock_hold(ssk);
3283
3284 /* new mpc subflow takes ownership of the newly
3285 * created mptcp socket
3286 */
3287 mptcp_token_accept(subflow_req, msk);
3288
3289 /* set msk addresses early to ensure mptcp_pm_get_local_id()
3290 * uses the correct data
3291 */
3292 mptcp_copy_inaddrs(nsk, ssk);
3293 __mptcp_propagate_sndbuf(nsk, ssk);
3294
3295 mptcp_rcv_space_init(msk, ssk);
3296
3297 if (mp_opt->suboptions & OPTION_MPTCP_MPC_ACK)
3298 __mptcp_subflow_fully_established(msk, subflow, mp_opt);
3299 bh_unlock_sock(nsk);
3300
3301 /* note: the newly allocated socket refcount is 2 now */
3302 return nsk;
3303 }
3304
mptcp_rcv_space_init(struct mptcp_sock * msk,const struct sock * ssk)3305 void mptcp_rcv_space_init(struct mptcp_sock *msk, const struct sock *ssk)
3306 {
3307 const struct tcp_sock *tp = tcp_sk(ssk);
3308
3309 msk->rcvspace_init = 1;
3310 msk->rcvq_space.copied = 0;
3311 msk->rcvq_space.rtt_us = 0;
3312
3313 msk->rcvq_space.time = tp->tcp_mstamp;
3314
3315 /* initial rcv_space offering made to peer */
3316 msk->rcvq_space.space = min_t(u32, tp->rcv_wnd,
3317 TCP_INIT_CWND * tp->advmss);
3318 if (msk->rcvq_space.space == 0)
3319 msk->rcvq_space.space = TCP_INIT_CWND * TCP_MSS_DEFAULT;
3320 }
3321
mptcp_destroy_common(struct mptcp_sock * msk,unsigned int flags)3322 void mptcp_destroy_common(struct mptcp_sock *msk, unsigned int flags)
3323 {
3324 struct mptcp_subflow_context *subflow, *tmp;
3325 struct sock *sk = (struct sock *)msk;
3326
3327 __mptcp_clear_xmit(sk);
3328
3329 /* join list will be eventually flushed (with rst) at sock lock release time */
3330 mptcp_for_each_subflow_safe(msk, subflow, tmp)
3331 __mptcp_close_ssk(sk, mptcp_subflow_tcp_sock(subflow), subflow, flags);
3332
3333 /* move to sk_receive_queue, sk_stream_kill_queues will purge it */
3334 mptcp_data_lock(sk);
3335 skb_queue_splice_tail_init(&msk->receive_queue, &sk->sk_receive_queue);
3336 __skb_queue_purge(&sk->sk_receive_queue);
3337 skb_rbtree_purge(&msk->out_of_order_queue);
3338 mptcp_data_unlock(sk);
3339
3340 /* move all the rx fwd alloc into the sk_mem_reclaim_final in
3341 * inet_sock_destruct() will dispose it
3342 */
3343 sk_forward_alloc_add(sk, msk->rmem_fwd_alloc);
3344 WRITE_ONCE(msk->rmem_fwd_alloc, 0);
3345 mptcp_token_destroy(msk);
3346 mptcp_pm_free_anno_list(msk);
3347 mptcp_free_local_addr_list(msk);
3348 }
3349
mptcp_destroy(struct sock * sk)3350 static void mptcp_destroy(struct sock *sk)
3351 {
3352 struct mptcp_sock *msk = mptcp_sk(sk);
3353
3354 /* allow the following to close even the initial subflow */
3355 msk->free_first = 1;
3356 mptcp_destroy_common(msk, 0);
3357 sk_sockets_allocated_dec(sk);
3358 }
3359
__mptcp_data_acked(struct sock * sk)3360 void __mptcp_data_acked(struct sock *sk)
3361 {
3362 if (!sock_owned_by_user(sk))
3363 __mptcp_clean_una(sk);
3364 else
3365 __set_bit(MPTCP_CLEAN_UNA, &mptcp_sk(sk)->cb_flags);
3366
3367 if (mptcp_pending_data_fin_ack(sk))
3368 mptcp_schedule_work(sk);
3369 }
3370
__mptcp_check_push(struct sock * sk,struct sock * ssk)3371 void __mptcp_check_push(struct sock *sk, struct sock *ssk)
3372 {
3373 if (!mptcp_send_head(sk))
3374 return;
3375
3376 if (!sock_owned_by_user(sk))
3377 __mptcp_subflow_push_pending(sk, ssk, false);
3378 else
3379 __set_bit(MPTCP_PUSH_PENDING, &mptcp_sk(sk)->cb_flags);
3380 }
3381
3382 #define MPTCP_FLAGS_PROCESS_CTX_NEED (BIT(MPTCP_PUSH_PENDING) | \
3383 BIT(MPTCP_RETRANSMIT) | \
3384 BIT(MPTCP_FLUSH_JOIN_LIST))
3385
3386 /* processes deferred events and flush wmem */
mptcp_release_cb(struct sock * sk)3387 static void mptcp_release_cb(struct sock *sk)
3388 __must_hold(&sk->sk_lock.slock)
3389 {
3390 struct mptcp_sock *msk = mptcp_sk(sk);
3391
3392 for (;;) {
3393 unsigned long flags = (msk->cb_flags & MPTCP_FLAGS_PROCESS_CTX_NEED);
3394 struct list_head join_list;
3395
3396 if (!flags)
3397 break;
3398
3399 INIT_LIST_HEAD(&join_list);
3400 list_splice_init(&msk->join_list, &join_list);
3401
3402 /* the following actions acquire the subflow socket lock
3403 *
3404 * 1) can't be invoked in atomic scope
3405 * 2) must avoid ABBA deadlock with msk socket spinlock: the RX
3406 * datapath acquires the msk socket spinlock while helding
3407 * the subflow socket lock
3408 */
3409 msk->cb_flags &= ~flags;
3410 spin_unlock_bh(&sk->sk_lock.slock);
3411
3412 if (flags & BIT(MPTCP_FLUSH_JOIN_LIST))
3413 __mptcp_flush_join_list(sk, &join_list);
3414 if (flags & BIT(MPTCP_PUSH_PENDING))
3415 __mptcp_push_pending(sk, 0);
3416 if (flags & BIT(MPTCP_RETRANSMIT))
3417 __mptcp_retrans(sk);
3418
3419 cond_resched();
3420 spin_lock_bh(&sk->sk_lock.slock);
3421 }
3422
3423 if (__test_and_clear_bit(MPTCP_CLEAN_UNA, &msk->cb_flags))
3424 __mptcp_clean_una_wakeup(sk);
3425 if (unlikely(msk->cb_flags)) {
3426 /* be sure to sync the msk state before taking actions
3427 * depending on sk_state (MPTCP_ERROR_REPORT)
3428 * On sk release avoid actions depending on the first subflow
3429 */
3430 if (__test_and_clear_bit(MPTCP_SYNC_STATE, &msk->cb_flags) && msk->first)
3431 __mptcp_sync_state(sk, msk->pending_state);
3432 if (__test_and_clear_bit(MPTCP_ERROR_REPORT, &msk->cb_flags))
3433 __mptcp_error_report(sk);
3434 if (__test_and_clear_bit(MPTCP_SYNC_SNDBUF, &msk->cb_flags))
3435 __mptcp_sync_sndbuf(sk);
3436 }
3437
3438 __mptcp_update_rmem(sk);
3439 }
3440
3441 /* MP_JOIN client subflow must wait for 4th ack before sending any data:
3442 * TCP can't schedule delack timer before the subflow is fully established.
3443 * MPTCP uses the delack timer to do 3rd ack retransmissions
3444 */
schedule_3rdack_retransmission(struct sock * ssk)3445 static void schedule_3rdack_retransmission(struct sock *ssk)
3446 {
3447 struct inet_connection_sock *icsk = inet_csk(ssk);
3448 struct tcp_sock *tp = tcp_sk(ssk);
3449 unsigned long timeout;
3450
3451 if (mptcp_subflow_ctx(ssk)->fully_established)
3452 return;
3453
3454 /* reschedule with a timeout above RTT, as we must look only for drop */
3455 if (tp->srtt_us)
3456 timeout = usecs_to_jiffies(tp->srtt_us >> (3 - 1));
3457 else
3458 timeout = TCP_TIMEOUT_INIT;
3459 timeout += jiffies;
3460
3461 WARN_ON_ONCE(icsk->icsk_ack.pending & ICSK_ACK_TIMER);
3462 icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
3463 icsk->icsk_ack.timeout = timeout;
3464 sk_reset_timer(ssk, &icsk->icsk_delack_timer, timeout);
3465 }
3466
mptcp_subflow_process_delegated(struct sock * ssk,long status)3467 void mptcp_subflow_process_delegated(struct sock *ssk, long status)
3468 {
3469 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
3470 struct sock *sk = subflow->conn;
3471
3472 if (status & BIT(MPTCP_DELEGATE_SEND)) {
3473 mptcp_data_lock(sk);
3474 if (!sock_owned_by_user(sk))
3475 __mptcp_subflow_push_pending(sk, ssk, true);
3476 else
3477 __set_bit(MPTCP_PUSH_PENDING, &mptcp_sk(sk)->cb_flags);
3478 mptcp_data_unlock(sk);
3479 }
3480 if (status & BIT(MPTCP_DELEGATE_SNDBUF)) {
3481 mptcp_data_lock(sk);
3482 if (!sock_owned_by_user(sk))
3483 __mptcp_sync_sndbuf(sk);
3484 else
3485 __set_bit(MPTCP_SYNC_SNDBUF, &mptcp_sk(sk)->cb_flags);
3486 mptcp_data_unlock(sk);
3487 }
3488 if (status & BIT(MPTCP_DELEGATE_ACK))
3489 schedule_3rdack_retransmission(ssk);
3490 }
3491
mptcp_hash(struct sock * sk)3492 static int mptcp_hash(struct sock *sk)
3493 {
3494 /* should never be called,
3495 * we hash the TCP subflows not the master socket
3496 */
3497 WARN_ON_ONCE(1);
3498 return 0;
3499 }
3500
mptcp_unhash(struct sock * sk)3501 static void mptcp_unhash(struct sock *sk)
3502 {
3503 /* called from sk_common_release(), but nothing to do here */
3504 }
3505
mptcp_get_port(struct sock * sk,unsigned short snum)3506 static int mptcp_get_port(struct sock *sk, unsigned short snum)
3507 {
3508 struct mptcp_sock *msk = mptcp_sk(sk);
3509
3510 pr_debug("msk=%p, ssk=%p", msk, msk->first);
3511 if (WARN_ON_ONCE(!msk->first))
3512 return -EINVAL;
3513
3514 return inet_csk_get_port(msk->first, snum);
3515 }
3516
mptcp_finish_connect(struct sock * ssk)3517 void mptcp_finish_connect(struct sock *ssk)
3518 {
3519 struct mptcp_subflow_context *subflow;
3520 struct mptcp_sock *msk;
3521 struct sock *sk;
3522
3523 subflow = mptcp_subflow_ctx(ssk);
3524 sk = subflow->conn;
3525 msk = mptcp_sk(sk);
3526
3527 pr_debug("msk=%p, token=%u", sk, subflow->token);
3528
3529 subflow->map_seq = subflow->iasn;
3530 subflow->map_subflow_seq = 1;
3531
3532 /* the socket is not connected yet, no msk/subflow ops can access/race
3533 * accessing the field below
3534 */
3535 WRITE_ONCE(msk->local_key, subflow->local_key);
3536
3537 mptcp_pm_new_connection(msk, ssk, 0);
3538 }
3539
mptcp_sock_graft(struct sock * sk,struct socket * parent)3540 void mptcp_sock_graft(struct sock *sk, struct socket *parent)
3541 {
3542 write_lock_bh(&sk->sk_callback_lock);
3543 rcu_assign_pointer(sk->sk_wq, &parent->wq);
3544 sk_set_socket(sk, parent);
3545 sk->sk_uid = SOCK_INODE(parent)->i_uid;
3546 write_unlock_bh(&sk->sk_callback_lock);
3547 }
3548
mptcp_finish_join(struct sock * ssk)3549 bool mptcp_finish_join(struct sock *ssk)
3550 {
3551 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
3552 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
3553 struct sock *parent = (void *)msk;
3554 bool ret = true;
3555
3556 pr_debug("msk=%p, subflow=%p", msk, subflow);
3557
3558 /* mptcp socket already closing? */
3559 if (!mptcp_is_fully_established(parent)) {
3560 subflow->reset_reason = MPTCP_RST_EMPTCP;
3561 return false;
3562 }
3563
3564 /* active subflow, already present inside the conn_list */
3565 if (!list_empty(&subflow->node)) {
3566 mptcp_subflow_joined(msk, ssk);
3567 mptcp_propagate_sndbuf(parent, ssk);
3568 return true;
3569 }
3570
3571 if (!mptcp_pm_allow_new_subflow(msk))
3572 goto err_prohibited;
3573
3574 /* If we can't acquire msk socket lock here, let the release callback
3575 * handle it
3576 */
3577 mptcp_data_lock(parent);
3578 if (!sock_owned_by_user(parent)) {
3579 ret = __mptcp_finish_join(msk, ssk);
3580 if (ret) {
3581 sock_hold(ssk);
3582 list_add_tail(&subflow->node, &msk->conn_list);
3583 }
3584 } else {
3585 sock_hold(ssk);
3586 list_add_tail(&subflow->node, &msk->join_list);
3587 __set_bit(MPTCP_FLUSH_JOIN_LIST, &msk->cb_flags);
3588 }
3589 mptcp_data_unlock(parent);
3590
3591 if (!ret) {
3592 err_prohibited:
3593 subflow->reset_reason = MPTCP_RST_EPROHIBIT;
3594 return false;
3595 }
3596
3597 return true;
3598 }
3599
mptcp_shutdown(struct sock * sk,int how)3600 static void mptcp_shutdown(struct sock *sk, int how)
3601 {
3602 pr_debug("sk=%p, how=%d", sk, how);
3603
3604 if ((how & SEND_SHUTDOWN) && mptcp_close_state(sk))
3605 __mptcp_wr_shutdown(sk);
3606 }
3607
mptcp_forward_alloc_get(const struct sock * sk)3608 static int mptcp_forward_alloc_get(const struct sock *sk)
3609 {
3610 return READ_ONCE(sk->sk_forward_alloc) +
3611 READ_ONCE(mptcp_sk(sk)->rmem_fwd_alloc);
3612 }
3613
mptcp_ioctl_outq(const struct mptcp_sock * msk,u64 v)3614 static int mptcp_ioctl_outq(const struct mptcp_sock *msk, u64 v)
3615 {
3616 const struct sock *sk = (void *)msk;
3617 u64 delta;
3618
3619 if (sk->sk_state == TCP_LISTEN)
3620 return -EINVAL;
3621
3622 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
3623 return 0;
3624
3625 delta = msk->write_seq - v;
3626 if (__mptcp_check_fallback(msk) && msk->first) {
3627 struct tcp_sock *tp = tcp_sk(msk->first);
3628
3629 /* the first subflow is disconnected after close - see
3630 * __mptcp_close_ssk(). tcp_disconnect() moves the write_seq
3631 * so ignore that status, too.
3632 */
3633 if (!((1 << msk->first->sk_state) &
3634 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE)))
3635 delta += READ_ONCE(tp->write_seq) - tp->snd_una;
3636 }
3637 if (delta > INT_MAX)
3638 delta = INT_MAX;
3639
3640 return (int)delta;
3641 }
3642
mptcp_ioctl(struct sock * sk,int cmd,int * karg)3643 static int mptcp_ioctl(struct sock *sk, int cmd, int *karg)
3644 {
3645 struct mptcp_sock *msk = mptcp_sk(sk);
3646 bool slow;
3647
3648 switch (cmd) {
3649 case SIOCINQ:
3650 if (sk->sk_state == TCP_LISTEN)
3651 return -EINVAL;
3652
3653 lock_sock(sk);
3654 __mptcp_move_skbs(msk);
3655 *karg = mptcp_inq_hint(sk);
3656 release_sock(sk);
3657 break;
3658 case SIOCOUTQ:
3659 slow = lock_sock_fast(sk);
3660 *karg = mptcp_ioctl_outq(msk, READ_ONCE(msk->snd_una));
3661 unlock_sock_fast(sk, slow);
3662 break;
3663 case SIOCOUTQNSD:
3664 slow = lock_sock_fast(sk);
3665 *karg = mptcp_ioctl_outq(msk, msk->snd_nxt);
3666 unlock_sock_fast(sk, slow);
3667 break;
3668 default:
3669 return -ENOIOCTLCMD;
3670 }
3671
3672 return 0;
3673 }
3674
mptcp_subflow_early_fallback(struct mptcp_sock * msk,struct mptcp_subflow_context * subflow)3675 static void mptcp_subflow_early_fallback(struct mptcp_sock *msk,
3676 struct mptcp_subflow_context *subflow)
3677 {
3678 subflow->request_mptcp = 0;
3679 __mptcp_do_fallback(msk);
3680 }
3681
mptcp_connect(struct sock * sk,struct sockaddr * uaddr,int addr_len)3682 static int mptcp_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
3683 {
3684 struct mptcp_subflow_context *subflow;
3685 struct mptcp_sock *msk = mptcp_sk(sk);
3686 int err = -EINVAL;
3687 struct sock *ssk;
3688
3689 ssk = __mptcp_nmpc_sk(msk);
3690 if (IS_ERR(ssk))
3691 return PTR_ERR(ssk);
3692
3693 mptcp_set_state(sk, TCP_SYN_SENT);
3694 subflow = mptcp_subflow_ctx(ssk);
3695 #ifdef CONFIG_TCP_MD5SIG
3696 /* no MPTCP if MD5SIG is enabled on this socket or we may run out of
3697 * TCP option space.
3698 */
3699 if (rcu_access_pointer(tcp_sk(ssk)->md5sig_info))
3700 mptcp_subflow_early_fallback(msk, subflow);
3701 #endif
3702 if (subflow->request_mptcp && mptcp_token_new_connect(ssk)) {
3703 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_TOKENFALLBACKINIT);
3704 mptcp_subflow_early_fallback(msk, subflow);
3705 }
3706
3707 WRITE_ONCE(msk->write_seq, subflow->idsn);
3708 WRITE_ONCE(msk->snd_nxt, subflow->idsn);
3709 WRITE_ONCE(msk->snd_una, subflow->idsn);
3710 if (likely(!__mptcp_check_fallback(msk)))
3711 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPCAPABLEACTIVE);
3712
3713 /* if reaching here via the fastopen/sendmsg path, the caller already
3714 * acquired the subflow socket lock, too.
3715 */
3716 if (!msk->fastopening)
3717 lock_sock(ssk);
3718
3719 /* the following mirrors closely a very small chunk of code from
3720 * __inet_stream_connect()
3721 */
3722 if (ssk->sk_state != TCP_CLOSE)
3723 goto out;
3724
3725 if (BPF_CGROUP_PRE_CONNECT_ENABLED(ssk)) {
3726 err = ssk->sk_prot->pre_connect(ssk, uaddr, addr_len);
3727 if (err)
3728 goto out;
3729 }
3730
3731 err = ssk->sk_prot->connect(ssk, uaddr, addr_len);
3732 if (err < 0)
3733 goto out;
3734
3735 inet_assign_bit(DEFER_CONNECT, sk, inet_test_bit(DEFER_CONNECT, ssk));
3736
3737 out:
3738 if (!msk->fastopening)
3739 release_sock(ssk);
3740
3741 /* on successful connect, the msk state will be moved to established by
3742 * subflow_finish_connect()
3743 */
3744 if (unlikely(err)) {
3745 /* avoid leaving a dangling token in an unconnected socket */
3746 mptcp_token_destroy(msk);
3747 mptcp_set_state(sk, TCP_CLOSE);
3748 return err;
3749 }
3750
3751 mptcp_copy_inaddrs(sk, ssk);
3752 return 0;
3753 }
3754
3755 static struct proto mptcp_prot = {
3756 .name = "MPTCP",
3757 .owner = THIS_MODULE,
3758 .init = mptcp_init_sock,
3759 .connect = mptcp_connect,
3760 .disconnect = mptcp_disconnect,
3761 .close = mptcp_close,
3762 .setsockopt = mptcp_setsockopt,
3763 .getsockopt = mptcp_getsockopt,
3764 .shutdown = mptcp_shutdown,
3765 .destroy = mptcp_destroy,
3766 .sendmsg = mptcp_sendmsg,
3767 .ioctl = mptcp_ioctl,
3768 .recvmsg = mptcp_recvmsg,
3769 .release_cb = mptcp_release_cb,
3770 .hash = mptcp_hash,
3771 .unhash = mptcp_unhash,
3772 .get_port = mptcp_get_port,
3773 .forward_alloc_get = mptcp_forward_alloc_get,
3774 .sockets_allocated = &mptcp_sockets_allocated,
3775
3776 .memory_allocated = &tcp_memory_allocated,
3777 .per_cpu_fw_alloc = &tcp_memory_per_cpu_fw_alloc,
3778
3779 .memory_pressure = &tcp_memory_pressure,
3780 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_tcp_wmem),
3781 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_tcp_rmem),
3782 .sysctl_mem = sysctl_tcp_mem,
3783 .obj_size = sizeof(struct mptcp_sock),
3784 .slab_flags = SLAB_TYPESAFE_BY_RCU,
3785 .no_autobind = true,
3786 };
3787
mptcp_bind(struct socket * sock,struct sockaddr * uaddr,int addr_len)3788 static int mptcp_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3789 {
3790 struct mptcp_sock *msk = mptcp_sk(sock->sk);
3791 struct sock *ssk, *sk = sock->sk;
3792 int err = -EINVAL;
3793
3794 lock_sock(sk);
3795 ssk = __mptcp_nmpc_sk(msk);
3796 if (IS_ERR(ssk)) {
3797 err = PTR_ERR(ssk);
3798 goto unlock;
3799 }
3800
3801 if (sk->sk_family == AF_INET)
3802 err = inet_bind_sk(ssk, uaddr, addr_len);
3803 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
3804 else if (sk->sk_family == AF_INET6)
3805 err = inet6_bind_sk(ssk, uaddr, addr_len);
3806 #endif
3807 if (!err)
3808 mptcp_copy_inaddrs(sk, ssk);
3809
3810 unlock:
3811 release_sock(sk);
3812 return err;
3813 }
3814
mptcp_listen(struct socket * sock,int backlog)3815 static int mptcp_listen(struct socket *sock, int backlog)
3816 {
3817 struct mptcp_sock *msk = mptcp_sk(sock->sk);
3818 struct sock *sk = sock->sk;
3819 struct sock *ssk;
3820 int err;
3821
3822 pr_debug("msk=%p", msk);
3823
3824 lock_sock(sk);
3825
3826 err = -EINVAL;
3827 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
3828 goto unlock;
3829
3830 ssk = __mptcp_nmpc_sk(msk);
3831 if (IS_ERR(ssk)) {
3832 err = PTR_ERR(ssk);
3833 goto unlock;
3834 }
3835
3836 mptcp_set_state(sk, TCP_LISTEN);
3837 sock_set_flag(sk, SOCK_RCU_FREE);
3838
3839 lock_sock(ssk);
3840 err = __inet_listen_sk(ssk, backlog);
3841 release_sock(ssk);
3842 mptcp_set_state(sk, inet_sk_state_load(ssk));
3843
3844 if (!err) {
3845 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
3846 mptcp_copy_inaddrs(sk, ssk);
3847 mptcp_event_pm_listener(ssk, MPTCP_EVENT_LISTENER_CREATED);
3848 }
3849
3850 unlock:
3851 release_sock(sk);
3852 return err;
3853 }
3854
mptcp_stream_accept(struct socket * sock,struct socket * newsock,int flags,bool kern)3855 static int mptcp_stream_accept(struct socket *sock, struct socket *newsock,
3856 int flags, bool kern)
3857 {
3858 struct mptcp_sock *msk = mptcp_sk(sock->sk);
3859 struct sock *ssk, *newsk;
3860 int err;
3861
3862 pr_debug("msk=%p", msk);
3863
3864 /* Buggy applications can call accept on socket states other then LISTEN
3865 * but no need to allocate the first subflow just to error out.
3866 */
3867 ssk = READ_ONCE(msk->first);
3868 if (!ssk)
3869 return -EINVAL;
3870
3871 pr_debug("ssk=%p, listener=%p", ssk, mptcp_subflow_ctx(ssk));
3872 newsk = inet_csk_accept(ssk, flags, &err, kern);
3873 if (!newsk)
3874 return err;
3875
3876 pr_debug("newsk=%p, subflow is mptcp=%d", newsk, sk_is_mptcp(newsk));
3877 if (sk_is_mptcp(newsk)) {
3878 struct mptcp_subflow_context *subflow;
3879 struct sock *new_mptcp_sock;
3880
3881 subflow = mptcp_subflow_ctx(newsk);
3882 new_mptcp_sock = subflow->conn;
3883
3884 /* is_mptcp should be false if subflow->conn is missing, see
3885 * subflow_syn_recv_sock()
3886 */
3887 if (WARN_ON_ONCE(!new_mptcp_sock)) {
3888 tcp_sk(newsk)->is_mptcp = 0;
3889 goto tcpfallback;
3890 }
3891
3892 newsk = new_mptcp_sock;
3893 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_MPCAPABLEPASSIVEACK);
3894
3895 newsk->sk_kern_sock = kern;
3896 lock_sock(newsk);
3897 __inet_accept(sock, newsock, newsk);
3898
3899 set_bit(SOCK_CUSTOM_SOCKOPT, &newsock->flags);
3900 msk = mptcp_sk(newsk);
3901 msk->in_accept_queue = 0;
3902
3903 /* set ssk->sk_socket of accept()ed flows to mptcp socket.
3904 * This is needed so NOSPACE flag can be set from tcp stack.
3905 */
3906 mptcp_for_each_subflow(msk, subflow) {
3907 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
3908
3909 if (!ssk->sk_socket)
3910 mptcp_sock_graft(ssk, newsock);
3911 }
3912
3913 /* Do late cleanup for the first subflow as necessary. Also
3914 * deal with bad peers not doing a complete shutdown.
3915 */
3916 if (unlikely(inet_sk_state_load(msk->first) == TCP_CLOSE)) {
3917 __mptcp_close_ssk(newsk, msk->first,
3918 mptcp_subflow_ctx(msk->first), 0);
3919 if (unlikely(list_is_singular(&msk->conn_list)))
3920 mptcp_set_state(newsk, TCP_CLOSE);
3921 }
3922 } else {
3923 tcpfallback:
3924 newsk->sk_kern_sock = kern;
3925 lock_sock(newsk);
3926 __inet_accept(sock, newsock, newsk);
3927 /* we are being invoked after accepting a non-mp-capable
3928 * flow: sk is a tcp_sk, not an mptcp one.
3929 *
3930 * Hand the socket over to tcp so all further socket ops
3931 * bypass mptcp.
3932 */
3933 WRITE_ONCE(newsock->sk->sk_socket->ops,
3934 mptcp_fallback_tcp_ops(newsock->sk));
3935 }
3936 release_sock(newsk);
3937
3938 return 0;
3939 }
3940
mptcp_check_writeable(struct mptcp_sock * msk)3941 static __poll_t mptcp_check_writeable(struct mptcp_sock *msk)
3942 {
3943 struct sock *sk = (struct sock *)msk;
3944
3945 if (sk_stream_is_writeable(sk))
3946 return EPOLLOUT | EPOLLWRNORM;
3947
3948 mptcp_set_nospace(sk);
3949 smp_mb__after_atomic(); /* msk->flags is changed by write_space cb */
3950 if (sk_stream_is_writeable(sk))
3951 return EPOLLOUT | EPOLLWRNORM;
3952
3953 return 0;
3954 }
3955
mptcp_poll(struct file * file,struct socket * sock,struct poll_table_struct * wait)3956 static __poll_t mptcp_poll(struct file *file, struct socket *sock,
3957 struct poll_table_struct *wait)
3958 {
3959 struct sock *sk = sock->sk;
3960 struct mptcp_sock *msk;
3961 __poll_t mask = 0;
3962 u8 shutdown;
3963 int state;
3964
3965 msk = mptcp_sk(sk);
3966 sock_poll_wait(file, sock, wait);
3967
3968 state = inet_sk_state_load(sk);
3969 pr_debug("msk=%p state=%d flags=%lx", msk, state, msk->flags);
3970 if (state == TCP_LISTEN) {
3971 struct sock *ssk = READ_ONCE(msk->first);
3972
3973 if (WARN_ON_ONCE(!ssk))
3974 return 0;
3975
3976 return inet_csk_listen_poll(ssk);
3977 }
3978
3979 shutdown = READ_ONCE(sk->sk_shutdown);
3980 if (shutdown == SHUTDOWN_MASK || state == TCP_CLOSE)
3981 mask |= EPOLLHUP;
3982 if (shutdown & RCV_SHUTDOWN)
3983 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
3984
3985 if (state != TCP_SYN_SENT && state != TCP_SYN_RECV) {
3986 mask |= mptcp_check_readable(msk);
3987 if (shutdown & SEND_SHUTDOWN)
3988 mask |= EPOLLOUT | EPOLLWRNORM;
3989 else
3990 mask |= mptcp_check_writeable(msk);
3991 } else if (state == TCP_SYN_SENT &&
3992 inet_test_bit(DEFER_CONNECT, sk)) {
3993 /* cf tcp_poll() note about TFO */
3994 mask |= EPOLLOUT | EPOLLWRNORM;
3995 }
3996
3997 /* This barrier is coupled with smp_wmb() in __mptcp_error_report() */
3998 smp_rmb();
3999 if (READ_ONCE(sk->sk_err))
4000 mask |= EPOLLERR;
4001
4002 return mask;
4003 }
4004
4005 static const struct proto_ops mptcp_stream_ops = {
4006 .family = PF_INET,
4007 .owner = THIS_MODULE,
4008 .release = inet_release,
4009 .bind = mptcp_bind,
4010 .connect = inet_stream_connect,
4011 .socketpair = sock_no_socketpair,
4012 .accept = mptcp_stream_accept,
4013 .getname = inet_getname,
4014 .poll = mptcp_poll,
4015 .ioctl = inet_ioctl,
4016 .gettstamp = sock_gettstamp,
4017 .listen = mptcp_listen,
4018 .shutdown = inet_shutdown,
4019 .setsockopt = sock_common_setsockopt,
4020 .getsockopt = sock_common_getsockopt,
4021 .sendmsg = inet_sendmsg,
4022 .recvmsg = inet_recvmsg,
4023 .mmap = sock_no_mmap,
4024 };
4025
4026 static struct inet_protosw mptcp_protosw = {
4027 .type = SOCK_STREAM,
4028 .protocol = IPPROTO_MPTCP,
4029 .prot = &mptcp_prot,
4030 .ops = &mptcp_stream_ops,
4031 .flags = INET_PROTOSW_ICSK,
4032 };
4033
mptcp_napi_poll(struct napi_struct * napi,int budget)4034 static int mptcp_napi_poll(struct napi_struct *napi, int budget)
4035 {
4036 struct mptcp_delegated_action *delegated;
4037 struct mptcp_subflow_context *subflow;
4038 int work_done = 0;
4039
4040 delegated = container_of(napi, struct mptcp_delegated_action, napi);
4041 while ((subflow = mptcp_subflow_delegated_next(delegated)) != NULL) {
4042 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
4043
4044 bh_lock_sock_nested(ssk);
4045 if (!sock_owned_by_user(ssk)) {
4046 mptcp_subflow_process_delegated(ssk, xchg(&subflow->delegated_status, 0));
4047 } else {
4048 /* tcp_release_cb_override already processed
4049 * the action or will do at next release_sock().
4050 * In both case must dequeue the subflow here - on the same
4051 * CPU that scheduled it.
4052 */
4053 smp_wmb();
4054 clear_bit(MPTCP_DELEGATE_SCHEDULED, &subflow->delegated_status);
4055 }
4056 bh_unlock_sock(ssk);
4057 sock_put(ssk);
4058
4059 if (++work_done == budget)
4060 return budget;
4061 }
4062
4063 /* always provide a 0 'work_done' argument, so that napi_complete_done
4064 * will not try accessing the NULL napi->dev ptr
4065 */
4066 napi_complete_done(napi, 0);
4067 return work_done;
4068 }
4069
mptcp_proto_init(void)4070 void __init mptcp_proto_init(void)
4071 {
4072 struct mptcp_delegated_action *delegated;
4073 int cpu;
4074
4075 mptcp_prot.h.hashinfo = tcp_prot.h.hashinfo;
4076
4077 if (percpu_counter_init(&mptcp_sockets_allocated, 0, GFP_KERNEL))
4078 panic("Failed to allocate MPTCP pcpu counter\n");
4079
4080 init_dummy_netdev(&mptcp_napi_dev);
4081 for_each_possible_cpu(cpu) {
4082 delegated = per_cpu_ptr(&mptcp_delegated_actions, cpu);
4083 INIT_LIST_HEAD(&delegated->head);
4084 netif_napi_add_tx(&mptcp_napi_dev, &delegated->napi,
4085 mptcp_napi_poll);
4086 napi_enable(&delegated->napi);
4087 }
4088
4089 mptcp_subflow_init();
4090 mptcp_pm_init();
4091 mptcp_sched_init();
4092 mptcp_token_init();
4093
4094 if (proto_register(&mptcp_prot, 1) != 0)
4095 panic("Failed to register MPTCP proto.\n");
4096
4097 inet_register_protosw(&mptcp_protosw);
4098
4099 BUILD_BUG_ON(sizeof(struct mptcp_skb_cb) > sizeof_field(struct sk_buff, cb));
4100 }
4101
4102 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
4103 static const struct proto_ops mptcp_v6_stream_ops = {
4104 .family = PF_INET6,
4105 .owner = THIS_MODULE,
4106 .release = inet6_release,
4107 .bind = mptcp_bind,
4108 .connect = inet_stream_connect,
4109 .socketpair = sock_no_socketpair,
4110 .accept = mptcp_stream_accept,
4111 .getname = inet6_getname,
4112 .poll = mptcp_poll,
4113 .ioctl = inet6_ioctl,
4114 .gettstamp = sock_gettstamp,
4115 .listen = mptcp_listen,
4116 .shutdown = inet_shutdown,
4117 .setsockopt = sock_common_setsockopt,
4118 .getsockopt = sock_common_getsockopt,
4119 .sendmsg = inet6_sendmsg,
4120 .recvmsg = inet6_recvmsg,
4121 .mmap = sock_no_mmap,
4122 #ifdef CONFIG_COMPAT
4123 .compat_ioctl = inet6_compat_ioctl,
4124 #endif
4125 };
4126
4127 static struct proto mptcp_v6_prot;
4128
4129 static struct inet_protosw mptcp_v6_protosw = {
4130 .type = SOCK_STREAM,
4131 .protocol = IPPROTO_MPTCP,
4132 .prot = &mptcp_v6_prot,
4133 .ops = &mptcp_v6_stream_ops,
4134 .flags = INET_PROTOSW_ICSK,
4135 };
4136
mptcp_proto_v6_init(void)4137 int __init mptcp_proto_v6_init(void)
4138 {
4139 int err;
4140
4141 mptcp_v6_prot = mptcp_prot;
4142 strcpy(mptcp_v6_prot.name, "MPTCPv6");
4143 mptcp_v6_prot.slab = NULL;
4144 mptcp_v6_prot.obj_size = sizeof(struct mptcp6_sock);
4145 mptcp_v6_prot.ipv6_pinfo_offset = offsetof(struct mptcp6_sock, np);
4146
4147 err = proto_register(&mptcp_v6_prot, 1);
4148 if (err)
4149 return err;
4150
4151 err = inet6_register_protosw(&mptcp_v6_protosw);
4152 if (err)
4153 proto_unregister(&mptcp_v6_prot);
4154
4155 return err;
4156 }
4157 #endif
4158