Commit | Line | Data |
---|---|---|
f870fa0b MM |
1 | // SPDX-License-Identifier: GPL-2.0 |
2 | /* Multipath TCP | |
3 | * | |
4 | * Copyright (c) 2017 - 2019, Intel Corporation. | |
5 | */ | |
6 | ||
7 | #define pr_fmt(fmt) "MPTCP: " fmt | |
8 | ||
9 | #include <linux/kernel.h> | |
10 | #include <linux/module.h> | |
11 | #include <linux/netdevice.h> | |
7a6a6cbc PA |
12 | #include <linux/sched/signal.h> |
13 | #include <linux/atomic.h> | |
f870fa0b MM |
14 | #include <net/sock.h> |
15 | #include <net/inet_common.h> | |
16 | #include <net/inet_hashtables.h> | |
17 | #include <net/protocol.h> | |
18 | #include <net/tcp.h> | |
cf7da0d6 PK |
19 | #if IS_ENABLED(CONFIG_MPTCP_IPV6) |
20 | #include <net/transp_v6.h> | |
21 | #endif | |
f870fa0b MM |
22 | #include <net/mptcp.h> |
23 | #include "protocol.h" | |
fc518953 | 24 | #include "mib.h" |
f870fa0b | 25 | |
2303f994 PK |
26 | #define MPTCP_SAME_STATE TCP_MAX_STATES |
27 | ||
b0519de8 FW |
28 | #if IS_ENABLED(CONFIG_MPTCP_IPV6) |
29 | struct mptcp6_sock { | |
30 | struct mptcp_sock msk; | |
31 | struct ipv6_pinfo np; | |
32 | }; | |
33 | #endif | |
34 | ||
6771bfd9 FW |
35 | struct mptcp_skb_cb { |
36 | u32 offset; | |
37 | }; | |
38 | ||
39 | #define MPTCP_SKB_CB(__skb) ((struct mptcp_skb_cb *)&((__skb)->cb[0])) | |
40 | ||
d027236c PA |
41 | static struct percpu_counter mptcp_sockets_allocated; |
42 | ||
2303f994 PK |
43 | /* If msk has an initial subflow socket, and the MP_CAPABLE handshake has not |
44 | * completed yet or has failed, return the subflow socket. | |
45 | * Otherwise return NULL. | |
46 | */ | |
47 | static struct socket *__mptcp_nmpc_socket(const struct mptcp_sock *msk) | |
48 | { | |
d22f4988 | 49 | if (!msk->subflow || READ_ONCE(msk->can_ack)) |
2303f994 PK |
50 | return NULL; |
51 | ||
52 | return msk->subflow; | |
53 | } | |
54 | ||
8ab183de PA |
55 | static bool __mptcp_needs_tcp_fallback(const struct mptcp_sock *msk) |
56 | { | |
57 | return msk->first && !sk_is_mptcp(msk->first); | |
58 | } | |
59 | ||
0b4f33de FW |
60 | static struct socket *mptcp_is_tcpsk(struct sock *sk) |
61 | { | |
62 | struct socket *sock = sk->sk_socket; | |
63 | ||
64 | if (sock->sk != sk) | |
65 | return NULL; | |
66 | ||
67 | if (unlikely(sk->sk_prot == &tcp_prot)) { | |
68 | /* we are being invoked after mptcp_accept() has | |
69 | * accepted a non-mp-capable flow: sk is a tcp_sk, | |
70 | * not an mptcp one. | |
71 | * | |
72 | * Hand the socket over to tcp so all further socket ops | |
73 | * bypass mptcp. | |
74 | */ | |
75 | sock->ops = &inet_stream_ops; | |
76 | return sock; | |
77 | #if IS_ENABLED(CONFIG_MPTCP_IPV6) | |
78 | } else if (unlikely(sk->sk_prot == &tcpv6_prot)) { | |
79 | sock->ops = &inet6_stream_ops; | |
80 | return sock; | |
81 | #endif | |
82 | } | |
83 | ||
84 | return NULL; | |
85 | } | |
86 | ||
8ab183de | 87 | static struct socket *__mptcp_tcp_fallback(struct mptcp_sock *msk) |
cec37a6e | 88 | { |
0b4f33de FW |
89 | struct socket *sock; |
90 | ||
cec37a6e PK |
91 | sock_owned_by_me((const struct sock *)msk); |
92 | ||
0b4f33de FW |
93 | sock = mptcp_is_tcpsk((struct sock *)msk); |
94 | if (unlikely(sock)) | |
95 | return sock; | |
96 | ||
8ab183de | 97 | if (likely(!__mptcp_needs_tcp_fallback(msk))) |
cec37a6e PK |
98 | return NULL; |
99 | ||
e154659b | 100 | return msk->subflow; |
cec37a6e PK |
101 | } |
102 | ||
2303f994 PK |
103 | static bool __mptcp_can_create_subflow(const struct mptcp_sock *msk) |
104 | { | |
8ab183de | 105 | return !msk->first; |
2303f994 PK |
106 | } |
107 | ||
108 | static struct socket *__mptcp_socket_create(struct mptcp_sock *msk, int state) | |
109 | { | |
110 | struct mptcp_subflow_context *subflow; | |
111 | struct sock *sk = (struct sock *)msk; | |
112 | struct socket *ssock; | |
113 | int err; | |
114 | ||
0b4f33de FW |
115 | ssock = __mptcp_tcp_fallback(msk); |
116 | if (unlikely(ssock)) | |
117 | return ssock; | |
118 | ||
2303f994 PK |
119 | ssock = __mptcp_nmpc_socket(msk); |
120 | if (ssock) | |
121 | goto set_state; | |
122 | ||
123 | if (!__mptcp_can_create_subflow(msk)) | |
124 | return ERR_PTR(-EINVAL); | |
125 | ||
126 | err = mptcp_subflow_create_socket(sk, &ssock); | |
127 | if (err) | |
128 | return ERR_PTR(err); | |
129 | ||
8ab183de | 130 | msk->first = ssock->sk; |
2303f994 PK |
131 | msk->subflow = ssock; |
132 | subflow = mptcp_subflow_ctx(ssock->sk); | |
cec37a6e | 133 | list_add(&subflow->node, &msk->conn_list); |
2303f994 PK |
134 | subflow->request_mptcp = 1; |
135 | ||
136 | set_state: | |
137 | if (state != MPTCP_SAME_STATE) | |
138 | inet_sk_state_store(sk, state); | |
139 | return ssock; | |
140 | } | |
141 | ||
6771bfd9 FW |
142 | static void __mptcp_move_skb(struct mptcp_sock *msk, struct sock *ssk, |
143 | struct sk_buff *skb, | |
144 | unsigned int offset, size_t copy_len) | |
145 | { | |
146 | struct sock *sk = (struct sock *)msk; | |
147 | ||
148 | __skb_unlink(skb, &ssk->sk_receive_queue); | |
600911ff | 149 | skb_set_owner_r(skb, sk); |
6771bfd9 FW |
150 | __skb_queue_tail(&sk->sk_receive_queue, skb); |
151 | ||
152 | msk->ack_seq += copy_len; | |
153 | MPTCP_SKB_CB(skb)->offset = offset; | |
154 | } | |
155 | ||
de06f573 FW |
156 | /* both sockets must be locked */ |
157 | static bool mptcp_subflow_dsn_valid(const struct mptcp_sock *msk, | |
158 | struct sock *ssk) | |
159 | { | |
160 | struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); | |
161 | u64 dsn = mptcp_subflow_get_mapped_dsn(subflow); | |
162 | ||
163 | /* revalidate data sequence number. | |
164 | * | |
165 | * mptcp_subflow_data_available() is usually called | |
166 | * without msk lock. Its unlikely (but possible) | |
167 | * that msk->ack_seq has been advanced since the last | |
168 | * call found in-sequence data. | |
169 | */ | |
170 | if (likely(dsn == msk->ack_seq)) | |
171 | return true; | |
172 | ||
173 | subflow->data_avail = 0; | |
174 | return mptcp_subflow_data_available(ssk); | |
175 | } | |
176 | ||
6771bfd9 FW |
177 | static bool __mptcp_move_skbs_from_subflow(struct mptcp_sock *msk, |
178 | struct sock *ssk, | |
179 | unsigned int *bytes) | |
180 | { | |
181 | struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); | |
600911ff | 182 | struct sock *sk = (struct sock *)msk; |
6771bfd9 FW |
183 | unsigned int moved = 0; |
184 | bool more_data_avail; | |
185 | struct tcp_sock *tp; | |
186 | bool done = false; | |
600911ff | 187 | |
de06f573 FW |
188 | if (!mptcp_subflow_dsn_valid(msk, ssk)) { |
189 | *bytes = 0; | |
190 | return false; | |
191 | } | |
192 | ||
ec33916d FW |
193 | if (!(sk->sk_userlocks & SOCK_RCVBUF_LOCK)) { |
194 | int rcvbuf = max(ssk->sk_rcvbuf, sk->sk_rcvbuf); | |
195 | ||
196 | if (rcvbuf > sk->sk_rcvbuf) | |
197 | sk->sk_rcvbuf = rcvbuf; | |
198 | } | |
6771bfd9 FW |
199 | |
200 | tp = tcp_sk(ssk); | |
201 | do { | |
202 | u32 map_remaining, offset; | |
203 | u32 seq = tp->copied_seq; | |
204 | struct sk_buff *skb; | |
205 | bool fin; | |
206 | ||
207 | /* try to move as much data as available */ | |
208 | map_remaining = subflow->map_data_len - | |
209 | mptcp_subflow_get_map_offset(subflow); | |
210 | ||
211 | skb = skb_peek(&ssk->sk_receive_queue); | |
212 | if (!skb) | |
213 | break; | |
214 | ||
215 | offset = seq - TCP_SKB_CB(skb)->seq; | |
216 | fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN; | |
217 | if (fin) { | |
218 | done = true; | |
219 | seq++; | |
220 | } | |
221 | ||
222 | if (offset < skb->len) { | |
223 | size_t len = skb->len - offset; | |
224 | ||
225 | if (tp->urg_data) | |
226 | done = true; | |
227 | ||
228 | __mptcp_move_skb(msk, ssk, skb, offset, len); | |
229 | seq += len; | |
230 | moved += len; | |
231 | ||
232 | if (WARN_ON_ONCE(map_remaining < len)) | |
233 | break; | |
234 | } else { | |
235 | WARN_ON_ONCE(!fin); | |
236 | sk_eat_skb(ssk, skb); | |
237 | done = true; | |
238 | } | |
239 | ||
240 | WRITE_ONCE(tp->copied_seq, seq); | |
241 | more_data_avail = mptcp_subflow_data_available(ssk); | |
600911ff FW |
242 | |
243 | if (atomic_read(&sk->sk_rmem_alloc) > READ_ONCE(sk->sk_rcvbuf)) { | |
244 | done = true; | |
245 | break; | |
246 | } | |
6771bfd9 FW |
247 | } while (more_data_avail); |
248 | ||
249 | *bytes = moved; | |
250 | ||
251 | return done; | |
252 | } | |
253 | ||
2e52213c FW |
254 | /* In most cases we will be able to lock the mptcp socket. If its already |
255 | * owned, we need to defer to the work queue to avoid ABBA deadlock. | |
256 | */ | |
257 | static bool move_skbs_to_msk(struct mptcp_sock *msk, struct sock *ssk) | |
258 | { | |
259 | struct sock *sk = (struct sock *)msk; | |
260 | unsigned int moved = 0; | |
261 | ||
262 | if (READ_ONCE(sk->sk_lock.owned)) | |
263 | return false; | |
264 | ||
265 | if (unlikely(!spin_trylock_bh(&sk->sk_lock.slock))) | |
266 | return false; | |
267 | ||
268 | /* must re-check after taking the lock */ | |
269 | if (!READ_ONCE(sk->sk_lock.owned)) | |
270 | __mptcp_move_skbs_from_subflow(msk, ssk, &moved); | |
271 | ||
272 | spin_unlock_bh(&sk->sk_lock.slock); | |
273 | ||
274 | return moved > 0; | |
275 | } | |
276 | ||
277 | void mptcp_data_ready(struct sock *sk, struct sock *ssk) | |
101f6f85 FW |
278 | { |
279 | struct mptcp_sock *msk = mptcp_sk(sk); | |
280 | ||
281 | set_bit(MPTCP_DATA_READY, &msk->flags); | |
6771bfd9 | 282 | |
2e52213c FW |
283 | if (atomic_read(&sk->sk_rmem_alloc) < READ_ONCE(sk->sk_rcvbuf) && |
284 | move_skbs_to_msk(msk, ssk)) | |
285 | goto wake; | |
286 | ||
600911ff FW |
287 | /* don't schedule if mptcp sk is (still) over limit */ |
288 | if (atomic_read(&sk->sk_rmem_alloc) > READ_ONCE(sk->sk_rcvbuf)) | |
289 | goto wake; | |
290 | ||
14c441b5 PA |
291 | /* mptcp socket is owned, release_cb should retry */ |
292 | if (!test_and_set_bit(TCP_DELACK_TIMER_DEFERRED, | |
293 | &sk->sk_tsq_flags)) { | |
294 | sock_hold(sk); | |
6771bfd9 | 295 | |
14c441b5 PA |
296 | /* need to try again, its possible release_cb() has already |
297 | * been called after the test_and_set_bit() above. | |
298 | */ | |
299 | move_skbs_to_msk(msk, ssk); | |
300 | } | |
600911ff | 301 | wake: |
101f6f85 FW |
302 | sk->sk_data_ready(sk); |
303 | } | |
304 | ||
ec3edaa7 PK |
305 | static void __mptcp_flush_join_list(struct mptcp_sock *msk) |
306 | { | |
307 | if (likely(list_empty(&msk->join_list))) | |
308 | return; | |
309 | ||
310 | spin_lock_bh(&msk->join_list_lock); | |
311 | list_splice_tail_init(&msk->join_list, &msk->conn_list); | |
312 | spin_unlock_bh(&msk->join_list_lock); | |
313 | } | |
314 | ||
b51f9b80 PA |
315 | static void mptcp_set_timeout(const struct sock *sk, const struct sock *ssk) |
316 | { | |
317 | long tout = ssk && inet_csk(ssk)->icsk_pending ? | |
318 | inet_csk(ssk)->icsk_timeout - jiffies : 0; | |
319 | ||
320 | if (tout <= 0) | |
321 | tout = mptcp_sk(sk)->timer_ival; | |
322 | mptcp_sk(sk)->timer_ival = tout > 0 ? tout : TCP_RTO_MIN; | |
323 | } | |
324 | ||
325 | static bool mptcp_timer_pending(struct sock *sk) | |
326 | { | |
327 | return timer_pending(&inet_csk(sk)->icsk_retransmit_timer); | |
328 | } | |
329 | ||
330 | static void mptcp_reset_timer(struct sock *sk) | |
331 | { | |
332 | struct inet_connection_sock *icsk = inet_csk(sk); | |
333 | unsigned long tout; | |
334 | ||
335 | /* should never be called with mptcp level timer cleared */ | |
336 | tout = READ_ONCE(mptcp_sk(sk)->timer_ival); | |
337 | if (WARN_ON_ONCE(!tout)) | |
338 | tout = TCP_RTO_MIN; | |
339 | sk_reset_timer(sk, &icsk->icsk_retransmit_timer, jiffies + tout); | |
340 | } | |
341 | ||
342 | void mptcp_data_acked(struct sock *sk) | |
343 | { | |
344 | mptcp_reset_timer(sk); | |
3b1d6210 PA |
345 | |
346 | if (!sk_stream_is_writeable(sk) && | |
347 | schedule_work(&mptcp_sk(sk)->work)) | |
348 | sock_hold(sk); | |
b51f9b80 PA |
349 | } |
350 | ||
59832e24 FW |
351 | void mptcp_subflow_eof(struct sock *sk) |
352 | { | |
353 | struct mptcp_sock *msk = mptcp_sk(sk); | |
354 | ||
355 | if (!test_and_set_bit(MPTCP_WORK_EOF, &msk->flags) && | |
356 | schedule_work(&msk->work)) | |
357 | sock_hold(sk); | |
358 | } | |
359 | ||
b51f9b80 PA |
360 | static void mptcp_stop_timer(struct sock *sk) |
361 | { | |
362 | struct inet_connection_sock *icsk = inet_csk(sk); | |
363 | ||
364 | sk_stop_timer(sk, &icsk->icsk_retransmit_timer); | |
365 | mptcp_sk(sk)->timer_ival = 0; | |
366 | } | |
367 | ||
6d0060f6 MM |
368 | static bool mptcp_ext_cache_refill(struct mptcp_sock *msk) |
369 | { | |
370 | if (!msk->cached_ext) | |
371 | msk->cached_ext = __skb_ext_alloc(); | |
372 | ||
373 | return !!msk->cached_ext; | |
374 | } | |
375 | ||
7a6a6cbc PA |
376 | static struct sock *mptcp_subflow_recv_lookup(const struct mptcp_sock *msk) |
377 | { | |
378 | struct mptcp_subflow_context *subflow; | |
379 | struct sock *sk = (struct sock *)msk; | |
380 | ||
381 | sock_owned_by_me(sk); | |
382 | ||
383 | mptcp_for_each_subflow(msk, subflow) { | |
384 | if (subflow->data_avail) | |
385 | return mptcp_subflow_tcp_sock(subflow); | |
386 | } | |
387 | ||
388 | return NULL; | |
389 | } | |
390 | ||
3f8e0aae PA |
391 | static bool mptcp_skb_can_collapse_to(u64 write_seq, |
392 | const struct sk_buff *skb, | |
393 | const struct mptcp_ext *mpext) | |
57040755 PA |
394 | { |
395 | if (!tcp_skb_can_collapse_to(skb)) | |
396 | return false; | |
397 | ||
398 | /* can collapse only if MPTCP level sequence is in order */ | |
3f8e0aae | 399 | return mpext && mpext->data_seq + mpext->data_len == write_seq; |
57040755 PA |
400 | } |
401 | ||
18b683bf PA |
402 | static bool mptcp_frag_can_collapse_to(const struct mptcp_sock *msk, |
403 | const struct page_frag *pfrag, | |
404 | const struct mptcp_data_frag *df) | |
405 | { | |
406 | return df && pfrag->page == df->page && | |
407 | df->data_seq + df->data_len == msk->write_seq; | |
408 | } | |
409 | ||
d027236c PA |
410 | static void dfrag_uncharge(struct sock *sk, int len) |
411 | { | |
412 | sk_mem_uncharge(sk, len); | |
7948f6cc | 413 | sk_wmem_queued_add(sk, -len); |
d027236c PA |
414 | } |
415 | ||
416 | static void dfrag_clear(struct sock *sk, struct mptcp_data_frag *dfrag) | |
18b683bf | 417 | { |
d027236c PA |
418 | int len = dfrag->data_len + dfrag->overhead; |
419 | ||
18b683bf | 420 | list_del(&dfrag->list); |
d027236c | 421 | dfrag_uncharge(sk, len); |
18b683bf PA |
422 | put_page(dfrag->page); |
423 | } | |
424 | ||
425 | static void mptcp_clean_una(struct sock *sk) | |
426 | { | |
427 | struct mptcp_sock *msk = mptcp_sk(sk); | |
428 | struct mptcp_data_frag *dtmp, *dfrag; | |
429 | u64 snd_una = atomic64_read(&msk->snd_una); | |
d027236c | 430 | bool cleaned = false; |
18b683bf PA |
431 | |
432 | list_for_each_entry_safe(dfrag, dtmp, &msk->rtx_queue, list) { | |
433 | if (after64(dfrag->data_seq + dfrag->data_len, snd_una)) | |
434 | break; | |
435 | ||
d027236c PA |
436 | dfrag_clear(sk, dfrag); |
437 | cleaned = true; | |
438 | } | |
439 | ||
7948f6cc FW |
440 | dfrag = mptcp_rtx_head(sk); |
441 | if (dfrag && after64(snd_una, dfrag->data_seq)) { | |
442 | u64 delta = dfrag->data_seq + dfrag->data_len - snd_una; | |
443 | ||
444 | dfrag->data_seq += delta; | |
445 | dfrag->data_len -= delta; | |
446 | ||
447 | dfrag_uncharge(sk, delta); | |
448 | cleaned = true; | |
449 | } | |
450 | ||
d027236c PA |
451 | if (cleaned) { |
452 | sk_mem_reclaim_partial(sk); | |
7948f6cc FW |
453 | |
454 | /* Only wake up writers if a subflow is ready */ | |
455 | if (test_bit(MPTCP_SEND_SPACE, &msk->flags)) | |
456 | sk_stream_write_space(sk); | |
18b683bf PA |
457 | } |
458 | } | |
459 | ||
460 | /* ensure we get enough memory for the frag hdr, beyond some minimal amount of | |
461 | * data | |
462 | */ | |
463 | static bool mptcp_page_frag_refill(struct sock *sk, struct page_frag *pfrag) | |
464 | { | |
465 | if (likely(skb_page_frag_refill(32U + sizeof(struct mptcp_data_frag), | |
466 | pfrag, sk->sk_allocation))) | |
467 | return true; | |
468 | ||
469 | sk->sk_prot->enter_memory_pressure(sk); | |
470 | sk_stream_moderate_sndbuf(sk); | |
471 | return false; | |
472 | } | |
473 | ||
474 | static struct mptcp_data_frag * | |
475 | mptcp_carve_data_frag(const struct mptcp_sock *msk, struct page_frag *pfrag, | |
476 | int orig_offset) | |
477 | { | |
478 | int offset = ALIGN(orig_offset, sizeof(long)); | |
479 | struct mptcp_data_frag *dfrag; | |
480 | ||
481 | dfrag = (struct mptcp_data_frag *)(page_to_virt(pfrag->page) + offset); | |
482 | dfrag->data_len = 0; | |
483 | dfrag->data_seq = msk->write_seq; | |
484 | dfrag->overhead = offset - orig_offset + sizeof(struct mptcp_data_frag); | |
485 | dfrag->offset = offset + sizeof(struct mptcp_data_frag); | |
486 | dfrag->page = pfrag->page; | |
487 | ||
488 | return dfrag; | |
489 | } | |
490 | ||
6d0060f6 | 491 | static int mptcp_sendmsg_frag(struct sock *sk, struct sock *ssk, |
3f8e0aae PA |
492 | struct msghdr *msg, struct mptcp_data_frag *dfrag, |
493 | long *timeo, int *pmss_now, | |
57040755 | 494 | int *ps_goal) |
6d0060f6 | 495 | { |
18b683bf PA |
496 | int mss_now, avail_size, size_goal, offset, ret, frag_truesize = 0; |
497 | bool dfrag_collapsed, can_collapse = false; | |
6d0060f6 MM |
498 | struct mptcp_sock *msk = mptcp_sk(sk); |
499 | struct mptcp_ext *mpext = NULL; | |
3f8e0aae | 500 | bool retransmission = !!dfrag; |
57040755 | 501 | struct sk_buff *skb, *tail; |
6d0060f6 | 502 | struct page_frag *pfrag; |
3f8e0aae PA |
503 | struct page *page; |
504 | u64 *write_seq; | |
6d0060f6 MM |
505 | size_t psize; |
506 | ||
507 | /* use the mptcp page cache so that we can easily move the data | |
508 | * from one substream to another, but do per subflow memory accounting | |
3f8e0aae PA |
509 | * Note: pfrag is used only !retransmission, but the compiler if |
510 | * fooled into a warning if we don't init here | |
6d0060f6 MM |
511 | */ |
512 | pfrag = sk_page_frag(sk); | |
3f8e0aae | 513 | while ((!retransmission && !mptcp_page_frag_refill(ssk, pfrag)) || |
6d0060f6 MM |
514 | !mptcp_ext_cache_refill(msk)) { |
515 | ret = sk_stream_wait_memory(ssk, timeo); | |
516 | if (ret) | |
517 | return ret; | |
18b683bf PA |
518 | |
519 | /* if sk_stream_wait_memory() sleeps snd_una can change | |
520 | * significantly, refresh the rtx queue | |
521 | */ | |
522 | mptcp_clean_una(sk); | |
523 | ||
8ab183de PA |
524 | if (unlikely(__mptcp_needs_tcp_fallback(msk))) |
525 | return 0; | |
6d0060f6 | 526 | } |
3f8e0aae PA |
527 | if (!retransmission) { |
528 | write_seq = &msk->write_seq; | |
529 | page = pfrag->page; | |
530 | } else { | |
531 | write_seq = &dfrag->data_seq; | |
532 | page = dfrag->page; | |
533 | } | |
6d0060f6 MM |
534 | |
535 | /* compute copy limit */ | |
536 | mss_now = tcp_send_mss(ssk, &size_goal, msg->msg_flags); | |
57040755 PA |
537 | *pmss_now = mss_now; |
538 | *ps_goal = size_goal; | |
539 | avail_size = size_goal; | |
540 | skb = tcp_write_queue_tail(ssk); | |
541 | if (skb) { | |
542 | mpext = skb_ext_find(skb, SKB_EXT_MPTCP); | |
543 | ||
544 | /* Limit the write to the size available in the | |
545 | * current skb, if any, so that we create at most a new skb. | |
546 | * Explicitly tells TCP internals to avoid collapsing on later | |
547 | * queue management operation, to avoid breaking the ext <-> | |
548 | * SSN association set here | |
549 | */ | |
550 | can_collapse = (size_goal - skb->len > 0) && | |
3f8e0aae | 551 | mptcp_skb_can_collapse_to(*write_seq, skb, mpext); |
57040755 PA |
552 | if (!can_collapse) |
553 | TCP_SKB_CB(skb)->eor = 1; | |
554 | else | |
555 | avail_size = size_goal - skb->len; | |
556 | } | |
18b683bf | 557 | |
3f8e0aae PA |
558 | if (!retransmission) { |
559 | /* reuse tail pfrag, if possible, or carve a new one from the | |
560 | * page allocator | |
561 | */ | |
562 | dfrag = mptcp_rtx_tail(sk); | |
563 | offset = pfrag->offset; | |
564 | dfrag_collapsed = mptcp_frag_can_collapse_to(msk, pfrag, dfrag); | |
565 | if (!dfrag_collapsed) { | |
566 | dfrag = mptcp_carve_data_frag(msk, pfrag, offset); | |
567 | offset = dfrag->offset; | |
568 | frag_truesize = dfrag->overhead; | |
569 | } | |
570 | psize = min_t(size_t, pfrag->size - offset, avail_size); | |
571 | ||
572 | /* Copy to page */ | |
573 | pr_debug("left=%zu", msg_data_left(msg)); | |
574 | psize = copy_page_from_iter(pfrag->page, offset, | |
575 | min_t(size_t, msg_data_left(msg), | |
576 | psize), | |
577 | &msg->msg_iter); | |
578 | pr_debug("left=%zu", msg_data_left(msg)); | |
579 | if (!psize) | |
580 | return -EINVAL; | |
581 | ||
582 | if (!sk_wmem_schedule(sk, psize + dfrag->overhead)) | |
583 | return -ENOMEM; | |
584 | } else { | |
18b683bf | 585 | offset = dfrag->offset; |
3f8e0aae | 586 | psize = min_t(size_t, dfrag->data_len, avail_size); |
18b683bf | 587 | } |
d027236c | 588 | |
57040755 PA |
589 | /* tell the TCP stack to delay the push so that we can safely |
590 | * access the skb after the sendpages call | |
6d0060f6 | 591 | */ |
3f8e0aae | 592 | ret = do_tcp_sendpages(ssk, page, offset, psize, |
6d0060f6 MM |
593 | msg->msg_flags | MSG_SENDPAGE_NOTLAST); |
594 | if (ret <= 0) | |
595 | return ret; | |
18b683bf PA |
596 | |
597 | frag_truesize += ret; | |
3f8e0aae PA |
598 | if (!retransmission) { |
599 | if (unlikely(ret < psize)) | |
600 | iov_iter_revert(&msg->msg_iter, psize - ret); | |
6d0060f6 | 601 | |
3f8e0aae PA |
602 | /* send successful, keep track of sent data for mptcp-level |
603 | * retransmission | |
604 | */ | |
605 | dfrag->data_len += ret; | |
606 | if (!dfrag_collapsed) { | |
607 | get_page(dfrag->page); | |
608 | list_add_tail(&dfrag->list, &msk->rtx_queue); | |
609 | sk_wmem_queued_add(sk, frag_truesize); | |
610 | } else { | |
611 | sk_wmem_queued_add(sk, ret); | |
612 | } | |
18b683bf | 613 | |
3f8e0aae PA |
614 | /* charge data on mptcp rtx queue to the master socket |
615 | * Note: we charge such data both to sk and ssk | |
616 | */ | |
617 | sk->sk_forward_alloc -= frag_truesize; | |
618 | } | |
d027236c | 619 | |
57040755 PA |
620 | /* if the tail skb extension is still the cached one, collapsing |
621 | * really happened. Note: we can't check for 'same skb' as the sk_buff | |
622 | * hdr on tail can be transmitted, freed and re-allocated by the | |
623 | * do_tcp_sendpages() call | |
624 | */ | |
625 | tail = tcp_write_queue_tail(ssk); | |
626 | if (mpext && tail && mpext == skb_ext_find(tail, SKB_EXT_MPTCP)) { | |
627 | WARN_ON_ONCE(!can_collapse); | |
628 | mpext->data_len += ret; | |
629 | goto out; | |
630 | } | |
631 | ||
6d0060f6 MM |
632 | skb = tcp_write_queue_tail(ssk); |
633 | mpext = __skb_ext_set(skb, SKB_EXT_MPTCP, msk->cached_ext); | |
634 | msk->cached_ext = NULL; | |
635 | ||
636 | memset(mpext, 0, sizeof(*mpext)); | |
3f8e0aae | 637 | mpext->data_seq = *write_seq; |
6d0060f6 MM |
638 | mpext->subflow_seq = mptcp_subflow_ctx(ssk)->rel_write_seq; |
639 | mpext->data_len = ret; | |
640 | mpext->use_map = 1; | |
641 | mpext->dsn64 = 1; | |
642 | ||
643 | pr_debug("data_seq=%llu subflow_seq=%u data_len=%u dsn64=%d", | |
644 | mpext->data_seq, mpext->subflow_seq, mpext->data_len, | |
645 | mpext->dsn64); | |
646 | ||
57040755 | 647 | out: |
3f8e0aae PA |
648 | if (!retransmission) |
649 | pfrag->offset += frag_truesize; | |
650 | *write_seq += ret; | |
6d0060f6 MM |
651 | mptcp_subflow_ctx(ssk)->rel_write_seq += ret; |
652 | ||
6d0060f6 MM |
653 | return ret; |
654 | } | |
655 | ||
f296234c PK |
656 | static struct sock *mptcp_subflow_get_send(struct mptcp_sock *msk) |
657 | { | |
658 | struct mptcp_subflow_context *subflow; | |
659 | struct sock *backup = NULL; | |
660 | ||
661 | sock_owned_by_me((const struct sock *)msk); | |
662 | ||
663 | mptcp_for_each_subflow(msk, subflow) { | |
664 | struct sock *ssk = mptcp_subflow_tcp_sock(subflow); | |
665 | ||
666 | if (!sk_stream_memory_free(ssk)) { | |
667 | struct socket *sock = ssk->sk_socket; | |
668 | ||
669 | if (sock) { | |
670 | clear_bit(MPTCP_SEND_SPACE, &msk->flags); | |
671 | smp_mb__after_atomic(); | |
672 | ||
673 | /* enables sk->write_space() callbacks */ | |
674 | set_bit(SOCK_NOSPACE, &sock->flags); | |
675 | } | |
676 | ||
677 | return NULL; | |
678 | } | |
679 | ||
680 | if (subflow->backup) { | |
681 | if (!backup) | |
682 | backup = ssk; | |
683 | ||
684 | continue; | |
685 | } | |
686 | ||
687 | return ssk; | |
688 | } | |
689 | ||
690 | return backup; | |
691 | } | |
692 | ||
1891c4a0 FW |
693 | static void ssk_check_wmem(struct mptcp_sock *msk, struct sock *ssk) |
694 | { | |
695 | struct socket *sock; | |
696 | ||
697 | if (likely(sk_stream_is_writeable(ssk))) | |
698 | return; | |
699 | ||
700 | sock = READ_ONCE(ssk->sk_socket); | |
701 | ||
702 | if (sock) { | |
703 | clear_bit(MPTCP_SEND_SPACE, &msk->flags); | |
704 | smp_mb__after_atomic(); | |
705 | /* set NOSPACE only after clearing SEND_SPACE flag */ | |
706 | set_bit(SOCK_NOSPACE, &sock->flags); | |
707 | } | |
708 | } | |
709 | ||
f870fa0b MM |
710 | static int mptcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) |
711 | { | |
57040755 | 712 | int mss_now = 0, size_goal = 0, ret = 0; |
f870fa0b | 713 | struct mptcp_sock *msk = mptcp_sk(sk); |
cec37a6e | 714 | struct socket *ssock; |
6d0060f6 | 715 | size_t copied = 0; |
cec37a6e | 716 | struct sock *ssk; |
6d0060f6 | 717 | long timeo; |
f870fa0b MM |
718 | |
719 | if (msg->msg_flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL)) | |
720 | return -EOPNOTSUPP; | |
721 | ||
cec37a6e | 722 | lock_sock(sk); |
1954b860 MM |
723 | |
724 | timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT); | |
725 | ||
726 | if ((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) { | |
727 | ret = sk_stream_wait_connect(sk, &timeo); | |
728 | if (ret) | |
729 | goto out; | |
730 | } | |
731 | ||
e154659b | 732 | fallback: |
cec37a6e | 733 | ssock = __mptcp_tcp_fallback(msk); |
8ab183de | 734 | if (unlikely(ssock)) { |
e154659b | 735 | release_sock(sk); |
cec37a6e PK |
736 | pr_debug("fallback passthrough"); |
737 | ret = sock_sendmsg(ssock, msg); | |
8ab183de | 738 | return ret >= 0 ? ret + copied : (copied ? copied : ret); |
cec37a6e PK |
739 | } |
740 | ||
18b683bf PA |
741 | mptcp_clean_una(sk); |
742 | ||
ec3edaa7 | 743 | __mptcp_flush_join_list(msk); |
f296234c PK |
744 | ssk = mptcp_subflow_get_send(msk); |
745 | while (!sk_stream_memory_free(sk) || !ssk) { | |
746 | ret = sk_stream_wait_memory(sk, &timeo); | |
747 | if (ret) | |
748 | goto out; | |
749 | ||
18b683bf PA |
750 | mptcp_clean_una(sk); |
751 | ||
f296234c PK |
752 | ssk = mptcp_subflow_get_send(msk); |
753 | if (list_empty(&msk->conn_list)) { | |
754 | ret = -ENOTCONN; | |
755 | goto out; | |
756 | } | |
cec37a6e PK |
757 | } |
758 | ||
6d0060f6 | 759 | pr_debug("conn_list->subflow=%p", ssk); |
cec37a6e | 760 | |
6d0060f6 MM |
761 | lock_sock(ssk); |
762 | while (msg_data_left(msg)) { | |
3f8e0aae | 763 | ret = mptcp_sendmsg_frag(sk, ssk, msg, NULL, &timeo, &mss_now, |
57040755 | 764 | &size_goal); |
6d0060f6 MM |
765 | if (ret < 0) |
766 | break; | |
8ab183de | 767 | if (ret == 0 && unlikely(__mptcp_needs_tcp_fallback(msk))) { |
e154659b FW |
768 | /* Can happen for passive sockets: |
769 | * 3WHS negotiated MPTCP, but first packet after is | |
770 | * plain TCP (e.g. due to middlebox filtering unknown | |
771 | * options). | |
772 | * | |
773 | * Fall back to TCP. | |
774 | */ | |
8ab183de | 775 | release_sock(ssk); |
8ab183de PA |
776 | goto fallback; |
777 | } | |
6d0060f6 MM |
778 | |
779 | copied += ret; | |
780 | } | |
781 | ||
b51f9b80 | 782 | mptcp_set_timeout(sk, ssk); |
57040755 | 783 | if (copied) { |
6d0060f6 | 784 | ret = copied; |
57040755 PA |
785 | tcp_push(ssk, msg->msg_flags, mss_now, tcp_sk(ssk)->nonagle, |
786 | size_goal); | |
b51f9b80 PA |
787 | |
788 | /* start the timer, if it's not pending */ | |
789 | if (!mptcp_timer_pending(sk)) | |
790 | mptcp_reset_timer(sk); | |
57040755 | 791 | } |
6d0060f6 | 792 | |
1891c4a0 | 793 | ssk_check_wmem(msk, ssk); |
6d0060f6 | 794 | release_sock(ssk); |
1954b860 | 795 | out: |
cec37a6e PK |
796 | release_sock(sk); |
797 | return ret; | |
f870fa0b MM |
798 | } |
799 | ||
7a6a6cbc PA |
800 | static void mptcp_wait_data(struct sock *sk, long *timeo) |
801 | { | |
802 | DEFINE_WAIT_FUNC(wait, woken_wake_function); | |
803 | struct mptcp_sock *msk = mptcp_sk(sk); | |
804 | ||
805 | add_wait_queue(sk_sleep(sk), &wait); | |
806 | sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); | |
807 | ||
808 | sk_wait_event(sk, timeo, | |
809 | test_and_clear_bit(MPTCP_DATA_READY, &msk->flags), &wait); | |
810 | ||
811 | sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); | |
812 | remove_wait_queue(sk_sleep(sk), &wait); | |
813 | } | |
814 | ||
6771bfd9 FW |
815 | static int __mptcp_recvmsg_mskq(struct mptcp_sock *msk, |
816 | struct msghdr *msg, | |
817 | size_t len) | |
818 | { | |
819 | struct sock *sk = (struct sock *)msk; | |
820 | struct sk_buff *skb; | |
821 | int copied = 0; | |
822 | ||
823 | while ((skb = skb_peek(&sk->sk_receive_queue)) != NULL) { | |
824 | u32 offset = MPTCP_SKB_CB(skb)->offset; | |
825 | u32 data_len = skb->len - offset; | |
826 | u32 count = min_t(size_t, len - copied, data_len); | |
827 | int err; | |
828 | ||
829 | err = skb_copy_datagram_msg(skb, offset, msg, count); | |
830 | if (unlikely(err < 0)) { | |
831 | if (!copied) | |
832 | return err; | |
833 | break; | |
834 | } | |
835 | ||
836 | copied += count; | |
837 | ||
838 | if (count < data_len) { | |
839 | MPTCP_SKB_CB(skb)->offset += count; | |
840 | break; | |
841 | } | |
842 | ||
843 | __skb_unlink(skb, &sk->sk_receive_queue); | |
844 | __kfree_skb(skb); | |
845 | ||
846 | if (copied >= len) | |
847 | break; | |
848 | } | |
849 | ||
850 | return copied; | |
851 | } | |
852 | ||
853 | static bool __mptcp_move_skbs(struct mptcp_sock *msk) | |
854 | { | |
855 | unsigned int moved = 0; | |
856 | bool done; | |
857 | ||
858 | do { | |
859 | struct sock *ssk = mptcp_subflow_recv_lookup(msk); | |
860 | ||
861 | if (!ssk) | |
862 | break; | |
863 | ||
864 | lock_sock(ssk); | |
865 | done = __mptcp_move_skbs_from_subflow(msk, ssk, &moved); | |
866 | release_sock(ssk); | |
867 | } while (!done); | |
868 | ||
869 | return moved > 0; | |
870 | } | |
871 | ||
f870fa0b MM |
872 | static int mptcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, |
873 | int nonblock, int flags, int *addr_len) | |
874 | { | |
875 | struct mptcp_sock *msk = mptcp_sk(sk); | |
cec37a6e | 876 | struct socket *ssock; |
cec37a6e | 877 | int copied = 0; |
7a6a6cbc PA |
878 | int target; |
879 | long timeo; | |
f870fa0b MM |
880 | |
881 | if (msg->msg_flags & ~(MSG_WAITALL | MSG_DONTWAIT)) | |
882 | return -EOPNOTSUPP; | |
883 | ||
cec37a6e PK |
884 | lock_sock(sk); |
885 | ssock = __mptcp_tcp_fallback(msk); | |
8ab183de PA |
886 | if (unlikely(ssock)) { |
887 | fallback: | |
e154659b | 888 | release_sock(sk); |
cec37a6e PK |
889 | pr_debug("fallback-read subflow=%p", |
890 | mptcp_subflow_ctx(ssock->sk)); | |
891 | copied = sock_recvmsg(ssock, msg, flags); | |
cec37a6e PK |
892 | return copied; |
893 | } | |
894 | ||
7a6a6cbc PA |
895 | timeo = sock_rcvtimeo(sk, nonblock); |
896 | ||
897 | len = min_t(size_t, len, INT_MAX); | |
898 | target = sock_rcvlowat(sk, flags & MSG_WAITALL, len); | |
ec3edaa7 | 899 | __mptcp_flush_join_list(msk); |
7a6a6cbc | 900 | |
6771bfd9 | 901 | while (len > (size_t)copied) { |
7a6a6cbc PA |
902 | int bytes_read; |
903 | ||
6771bfd9 FW |
904 | bytes_read = __mptcp_recvmsg_mskq(msk, msg, len - copied); |
905 | if (unlikely(bytes_read < 0)) { | |
906 | if (!copied) | |
907 | copied = bytes_read; | |
908 | goto out_err; | |
909 | } | |
7a6a6cbc | 910 | |
6771bfd9 | 911 | copied += bytes_read; |
7a6a6cbc | 912 | |
6771bfd9 FW |
913 | if (skb_queue_empty(&sk->sk_receive_queue) && |
914 | __mptcp_move_skbs(msk)) | |
915 | continue; | |
7a6a6cbc PA |
916 | |
917 | /* only the master socket status is relevant here. The exit | |
918 | * conditions mirror closely tcp_recvmsg() | |
919 | */ | |
920 | if (copied >= target) | |
921 | break; | |
922 | ||
923 | if (copied) { | |
924 | if (sk->sk_err || | |
925 | sk->sk_state == TCP_CLOSE || | |
926 | (sk->sk_shutdown & RCV_SHUTDOWN) || | |
927 | !timeo || | |
928 | signal_pending(current)) | |
929 | break; | |
930 | } else { | |
931 | if (sk->sk_err) { | |
932 | copied = sock_error(sk); | |
933 | break; | |
934 | } | |
935 | ||
936 | if (sk->sk_shutdown & RCV_SHUTDOWN) | |
937 | break; | |
938 | ||
939 | if (sk->sk_state == TCP_CLOSE) { | |
940 | copied = -ENOTCONN; | |
941 | break; | |
942 | } | |
943 | ||
944 | if (!timeo) { | |
945 | copied = -EAGAIN; | |
946 | break; | |
947 | } | |
948 | ||
949 | if (signal_pending(current)) { | |
950 | copied = sock_intr_errno(timeo); | |
951 | break; | |
952 | } | |
953 | } | |
954 | ||
955 | pr_debug("block timeout %ld", timeo); | |
7a6a6cbc | 956 | mptcp_wait_data(sk, &timeo); |
8ab183de PA |
957 | if (unlikely(__mptcp_tcp_fallback(msk))) |
958 | goto fallback; | |
cec37a6e PK |
959 | } |
960 | ||
6771bfd9 FW |
961 | if (skb_queue_empty(&sk->sk_receive_queue)) { |
962 | /* entire backlog drained, clear DATA_READY. */ | |
7a6a6cbc | 963 | clear_bit(MPTCP_DATA_READY, &msk->flags); |
cec37a6e | 964 | |
6771bfd9 FW |
965 | /* .. race-breaker: ssk might have gotten new data |
966 | * after last __mptcp_move_skbs() returned false. | |
7a6a6cbc | 967 | */ |
6771bfd9 | 968 | if (unlikely(__mptcp_move_skbs(msk))) |
7a6a6cbc | 969 | set_bit(MPTCP_DATA_READY, &msk->flags); |
6771bfd9 FW |
970 | } else if (unlikely(!test_bit(MPTCP_DATA_READY, &msk->flags))) { |
971 | /* data to read but mptcp_wait_data() cleared DATA_READY */ | |
972 | set_bit(MPTCP_DATA_READY, &msk->flags); | |
7a6a6cbc | 973 | } |
6771bfd9 | 974 | out_err: |
7a6a6cbc | 975 | release_sock(sk); |
cec37a6e PK |
976 | return copied; |
977 | } | |
978 | ||
b51f9b80 PA |
979 | static void mptcp_retransmit_handler(struct sock *sk) |
980 | { | |
981 | struct mptcp_sock *msk = mptcp_sk(sk); | |
982 | ||
3b1d6210 | 983 | if (atomic64_read(&msk->snd_una) == msk->write_seq) { |
b51f9b80 | 984 | mptcp_stop_timer(sk); |
3b1d6210 PA |
985 | } else { |
986 | set_bit(MPTCP_WORK_RTX, &msk->flags); | |
987 | if (schedule_work(&msk->work)) | |
988 | sock_hold(sk); | |
989 | } | |
b51f9b80 PA |
990 | } |
991 | ||
992 | static void mptcp_retransmit_timer(struct timer_list *t) | |
993 | { | |
994 | struct inet_connection_sock *icsk = from_timer(icsk, t, | |
995 | icsk_retransmit_timer); | |
996 | struct sock *sk = &icsk->icsk_inet.sk; | |
997 | ||
998 | bh_lock_sock(sk); | |
999 | if (!sock_owned_by_user(sk)) { | |
1000 | mptcp_retransmit_handler(sk); | |
1001 | } else { | |
1002 | /* delegate our work to tcp_release_cb() */ | |
1003 | if (!test_and_set_bit(TCP_WRITE_TIMER_DEFERRED, | |
1004 | &sk->sk_tsq_flags)) | |
1005 | sock_hold(sk); | |
1006 | } | |
1007 | bh_unlock_sock(sk); | |
1008 | sock_put(sk); | |
1009 | } | |
1010 | ||
3b1d6210 PA |
1011 | /* Find an idle subflow. Return NULL if there is unacked data at tcp |
1012 | * level. | |
1013 | * | |
1014 | * A backup subflow is returned only if that is the only kind available. | |
1015 | */ | |
1016 | static struct sock *mptcp_subflow_get_retrans(const struct mptcp_sock *msk) | |
1017 | { | |
1018 | struct mptcp_subflow_context *subflow; | |
1019 | struct sock *backup = NULL; | |
1020 | ||
1021 | sock_owned_by_me((const struct sock *)msk); | |
1022 | ||
1023 | mptcp_for_each_subflow(msk, subflow) { | |
1024 | struct sock *ssk = mptcp_subflow_tcp_sock(subflow); | |
1025 | ||
1026 | /* still data outstanding at TCP level? Don't retransmit. */ | |
1027 | if (!tcp_write_queue_empty(ssk)) | |
1028 | return NULL; | |
1029 | ||
1030 | if (subflow->backup) { | |
1031 | if (!backup) | |
1032 | backup = ssk; | |
1033 | continue; | |
1034 | } | |
1035 | ||
1036 | return ssk; | |
1037 | } | |
1038 | ||
1039 | return backup; | |
1040 | } | |
1041 | ||
cec37a6e PK |
1042 | /* subflow sockets can be either outgoing (connect) or incoming |
1043 | * (accept). | |
1044 | * | |
1045 | * Outgoing subflows use in-kernel sockets. | |
1046 | * Incoming subflows do not have their own 'struct socket' allocated, | |
1047 | * so we need to use tcp_close() after detaching them from the mptcp | |
1048 | * parent socket. | |
1049 | */ | |
1050 | static void __mptcp_close_ssk(struct sock *sk, struct sock *ssk, | |
1051 | struct mptcp_subflow_context *subflow, | |
1052 | long timeout) | |
1053 | { | |
1054 | struct socket *sock = READ_ONCE(ssk->sk_socket); | |
1055 | ||
1056 | list_del(&subflow->node); | |
1057 | ||
1058 | if (sock && sock != sk->sk_socket) { | |
1059 | /* outgoing subflow */ | |
1060 | sock_release(sock); | |
1061 | } else { | |
1062 | /* incoming subflow */ | |
1063 | tcp_close(ssk, timeout); | |
1064 | } | |
f870fa0b MM |
1065 | } |
1066 | ||
dc24f8b4 PA |
1067 | static unsigned int mptcp_sync_mss(struct sock *sk, u32 pmtu) |
1068 | { | |
1069 | return 0; | |
1070 | } | |
1071 | ||
59832e24 FW |
1072 | static void mptcp_check_for_eof(struct mptcp_sock *msk) |
1073 | { | |
1074 | struct mptcp_subflow_context *subflow; | |
1075 | struct sock *sk = (struct sock *)msk; | |
1076 | int receivers = 0; | |
1077 | ||
1078 | mptcp_for_each_subflow(msk, subflow) | |
1079 | receivers += !subflow->rx_eof; | |
1080 | ||
1081 | if (!receivers && !(sk->sk_shutdown & RCV_SHUTDOWN)) { | |
1082 | /* hopefully temporary hack: propagate shutdown status | |
1083 | * to msk, when all subflows agree on it | |
1084 | */ | |
1085 | sk->sk_shutdown |= RCV_SHUTDOWN; | |
1086 | ||
1087 | smp_mb__before_atomic(); /* SHUTDOWN must be visible first */ | |
1088 | set_bit(MPTCP_DATA_READY, &msk->flags); | |
1089 | sk->sk_data_ready(sk); | |
1090 | } | |
1091 | } | |
1092 | ||
80992017 PA |
1093 | static void mptcp_worker(struct work_struct *work) |
1094 | { | |
1095 | struct mptcp_sock *msk = container_of(work, struct mptcp_sock, work); | |
3b1d6210 PA |
1096 | struct sock *ssk, *sk = &msk->sk.icsk_inet.sk; |
1097 | int orig_len, orig_offset, ret, mss_now = 0, size_goal = 0; | |
1098 | struct mptcp_data_frag *dfrag; | |
1099 | u64 orig_write_seq; | |
1100 | size_t copied = 0; | |
1101 | struct msghdr msg; | |
1102 | long timeo = 0; | |
80992017 PA |
1103 | |
1104 | lock_sock(sk); | |
3b1d6210 | 1105 | mptcp_clean_una(sk); |
ec3edaa7 | 1106 | __mptcp_flush_join_list(msk); |
6771bfd9 | 1107 | __mptcp_move_skbs(msk); |
3b1d6210 | 1108 | |
59832e24 FW |
1109 | if (test_and_clear_bit(MPTCP_WORK_EOF, &msk->flags)) |
1110 | mptcp_check_for_eof(msk); | |
1111 | ||
3b1d6210 PA |
1112 | if (!test_and_clear_bit(MPTCP_WORK_RTX, &msk->flags)) |
1113 | goto unlock; | |
1114 | ||
1115 | dfrag = mptcp_rtx_head(sk); | |
1116 | if (!dfrag) | |
1117 | goto unlock; | |
1118 | ||
1119 | ssk = mptcp_subflow_get_retrans(msk); | |
1120 | if (!ssk) | |
1121 | goto reset_unlock; | |
1122 | ||
1123 | lock_sock(ssk); | |
1124 | ||
1125 | msg.msg_flags = MSG_DONTWAIT; | |
1126 | orig_len = dfrag->data_len; | |
1127 | orig_offset = dfrag->offset; | |
1128 | orig_write_seq = dfrag->data_seq; | |
1129 | while (dfrag->data_len > 0) { | |
1130 | ret = mptcp_sendmsg_frag(sk, ssk, &msg, dfrag, &timeo, &mss_now, | |
1131 | &size_goal); | |
1132 | if (ret < 0) | |
1133 | break; | |
1134 | ||
fc518953 | 1135 | MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_RETRANSSEGS); |
3b1d6210 PA |
1136 | copied += ret; |
1137 | dfrag->data_len -= ret; | |
1138 | dfrag->offset += ret; | |
1139 | } | |
1140 | if (copied) | |
1141 | tcp_push(ssk, msg.msg_flags, mss_now, tcp_sk(ssk)->nonagle, | |
1142 | size_goal); | |
1143 | ||
1144 | dfrag->data_seq = orig_write_seq; | |
1145 | dfrag->offset = orig_offset; | |
1146 | dfrag->data_len = orig_len; | |
1147 | ||
1148 | mptcp_set_timeout(sk, ssk); | |
1149 | release_sock(ssk); | |
1150 | ||
1151 | reset_unlock: | |
1152 | if (!mptcp_timer_pending(sk)) | |
1153 | mptcp_reset_timer(sk); | |
1154 | ||
1155 | unlock: | |
80992017 PA |
1156 | release_sock(sk); |
1157 | sock_put(sk); | |
1158 | } | |
1159 | ||
784325e9 | 1160 | static int __mptcp_init_sock(struct sock *sk) |
f870fa0b | 1161 | { |
cec37a6e PK |
1162 | struct mptcp_sock *msk = mptcp_sk(sk); |
1163 | ||
ec3edaa7 PK |
1164 | spin_lock_init(&msk->join_list_lock); |
1165 | ||
cec37a6e | 1166 | INIT_LIST_HEAD(&msk->conn_list); |
ec3edaa7 | 1167 | INIT_LIST_HEAD(&msk->join_list); |
18b683bf | 1168 | INIT_LIST_HEAD(&msk->rtx_queue); |
1891c4a0 | 1169 | __set_bit(MPTCP_SEND_SPACE, &msk->flags); |
80992017 | 1170 | INIT_WORK(&msk->work, mptcp_worker); |
cec37a6e | 1171 | |
8ab183de | 1172 | msk->first = NULL; |
dc24f8b4 | 1173 | inet_csk(sk)->icsk_sync_mss = mptcp_sync_mss; |
8ab183de | 1174 | |
1b1c7a0e PK |
1175 | mptcp_pm_data_init(msk); |
1176 | ||
b51f9b80 PA |
1177 | /* re-use the csk retrans timer for MPTCP-level retrans */ |
1178 | timer_setup(&msk->sk.icsk_retransmit_timer, mptcp_retransmit_timer, 0); | |
1179 | ||
f870fa0b MM |
1180 | return 0; |
1181 | } | |
1182 | ||
784325e9 MB |
1183 | static int mptcp_init_sock(struct sock *sk) |
1184 | { | |
fc518953 FW |
1185 | struct net *net = sock_net(sk); |
1186 | int ret; | |
18b683bf | 1187 | |
fc518953 FW |
1188 | if (!mptcp_is_enabled(net)) |
1189 | return -ENOPROTOOPT; | |
1190 | ||
1191 | if (unlikely(!net->mib.mptcp_statistics) && !mptcp_mib_alloc(net)) | |
1192 | return -ENOMEM; | |
1193 | ||
1194 | ret = __mptcp_init_sock(sk); | |
18b683bf PA |
1195 | if (ret) |
1196 | return ret; | |
1197 | ||
d027236c | 1198 | sk_sockets_allocated_inc(sk); |
3f8e0aae | 1199 | sk->sk_sndbuf = sock_net(sk)->ipv4.sysctl_tcp_wmem[2]; |
d027236c | 1200 | |
18b683bf PA |
1201 | return 0; |
1202 | } | |
1203 | ||
1204 | static void __mptcp_clear_xmit(struct sock *sk) | |
1205 | { | |
1206 | struct mptcp_sock *msk = mptcp_sk(sk); | |
1207 | struct mptcp_data_frag *dtmp, *dfrag; | |
1208 | ||
b51f9b80 PA |
1209 | sk_stop_timer(sk, &msk->sk.icsk_retransmit_timer); |
1210 | ||
18b683bf | 1211 | list_for_each_entry_safe(dfrag, dtmp, &msk->rtx_queue, list) |
d027236c | 1212 | dfrag_clear(sk, dfrag); |
784325e9 MB |
1213 | } |
1214 | ||
80992017 PA |
1215 | static void mptcp_cancel_work(struct sock *sk) |
1216 | { | |
1217 | struct mptcp_sock *msk = mptcp_sk(sk); | |
1218 | ||
1219 | if (cancel_work_sync(&msk->work)) | |
1220 | sock_put(sk); | |
1221 | } | |
1222 | ||
76c42a29 MM |
1223 | static void mptcp_subflow_shutdown(struct sock *ssk, int how, |
1224 | bool data_fin_tx_enable, u64 data_fin_tx_seq) | |
21498490 PK |
1225 | { |
1226 | lock_sock(ssk); | |
1227 | ||
1228 | switch (ssk->sk_state) { | |
1229 | case TCP_LISTEN: | |
1230 | if (!(how & RCV_SHUTDOWN)) | |
1231 | break; | |
1232 | /* fall through */ | |
1233 | case TCP_SYN_SENT: | |
1234 | tcp_disconnect(ssk, O_NONBLOCK); | |
1235 | break; | |
1236 | default: | |
76c42a29 MM |
1237 | if (data_fin_tx_enable) { |
1238 | struct mptcp_subflow_context *subflow; | |
1239 | ||
1240 | subflow = mptcp_subflow_ctx(ssk); | |
1241 | subflow->data_fin_tx_seq = data_fin_tx_seq; | |
1242 | subflow->data_fin_tx_enable = 1; | |
1243 | } | |
1244 | ||
21498490 PK |
1245 | ssk->sk_shutdown |= how; |
1246 | tcp_shutdown(ssk, how); | |
1247 | break; | |
1248 | } | |
1249 | ||
1250 | /* Wake up anyone sleeping in poll. */ | |
1251 | ssk->sk_state_change(ssk); | |
1252 | release_sock(ssk); | |
1253 | } | |
1254 | ||
8ab183de | 1255 | /* Called with msk lock held, releases such lock before returning */ |
2c22c06c | 1256 | static void mptcp_close(struct sock *sk, long timeout) |
f870fa0b | 1257 | { |
cec37a6e | 1258 | struct mptcp_subflow_context *subflow, *tmp; |
f870fa0b | 1259 | struct mptcp_sock *msk = mptcp_sk(sk); |
b2c5b614 | 1260 | LIST_HEAD(conn_list); |
76c42a29 | 1261 | u64 data_fin_tx_seq; |
f870fa0b | 1262 | |
2c22c06c FW |
1263 | lock_sock(sk); |
1264 | ||
79c0949e | 1265 | mptcp_token_destroy(msk->token); |
f870fa0b MM |
1266 | inet_sk_state_store(sk, TCP_CLOSE); |
1267 | ||
ec3edaa7 PK |
1268 | __mptcp_flush_join_list(msk); |
1269 | ||
b2c5b614 FW |
1270 | list_splice_init(&msk->conn_list, &conn_list); |
1271 | ||
76c42a29 MM |
1272 | data_fin_tx_seq = msk->write_seq; |
1273 | ||
18b683bf PA |
1274 | __mptcp_clear_xmit(sk); |
1275 | ||
b2c5b614 FW |
1276 | release_sock(sk); |
1277 | ||
1278 | list_for_each_entry_safe(subflow, tmp, &conn_list, node) { | |
cec37a6e PK |
1279 | struct sock *ssk = mptcp_subflow_tcp_sock(subflow); |
1280 | ||
76c42a29 MM |
1281 | subflow->data_fin_tx_seq = data_fin_tx_seq; |
1282 | subflow->data_fin_tx_enable = 1; | |
cec37a6e | 1283 | __mptcp_close_ssk(sk, ssk, subflow, timeout); |
f870fa0b MM |
1284 | } |
1285 | ||
80992017 | 1286 | mptcp_cancel_work(sk); |
926bdeab | 1287 | mptcp_pm_close(msk); |
80992017 | 1288 | |
6771bfd9 FW |
1289 | __skb_queue_purge(&sk->sk_receive_queue); |
1290 | ||
cec37a6e | 1291 | sk_common_release(sk); |
f870fa0b MM |
1292 | } |
1293 | ||
cf7da0d6 PK |
1294 | static void mptcp_copy_inaddrs(struct sock *msk, const struct sock *ssk) |
1295 | { | |
1296 | #if IS_ENABLED(CONFIG_MPTCP_IPV6) | |
1297 | const struct ipv6_pinfo *ssk6 = inet6_sk(ssk); | |
1298 | struct ipv6_pinfo *msk6 = inet6_sk(msk); | |
1299 | ||
1300 | msk->sk_v6_daddr = ssk->sk_v6_daddr; | |
1301 | msk->sk_v6_rcv_saddr = ssk->sk_v6_rcv_saddr; | |
1302 | ||
1303 | if (msk6 && ssk6) { | |
1304 | msk6->saddr = ssk6->saddr; | |
1305 | msk6->flow_label = ssk6->flow_label; | |
1306 | } | |
1307 | #endif | |
1308 | ||
1309 | inet_sk(msk)->inet_num = inet_sk(ssk)->inet_num; | |
1310 | inet_sk(msk)->inet_dport = inet_sk(ssk)->inet_dport; | |
1311 | inet_sk(msk)->inet_sport = inet_sk(ssk)->inet_sport; | |
1312 | inet_sk(msk)->inet_daddr = inet_sk(ssk)->inet_daddr; | |
1313 | inet_sk(msk)->inet_saddr = inet_sk(ssk)->inet_saddr; | |
1314 | inet_sk(msk)->inet_rcv_saddr = inet_sk(ssk)->inet_rcv_saddr; | |
1315 | } | |
1316 | ||
18b683bf PA |
1317 | static int mptcp_disconnect(struct sock *sk, int flags) |
1318 | { | |
42c556fe FW |
1319 | /* Should never be called. |
1320 | * inet_stream_connect() calls ->disconnect, but that | |
1321 | * refers to the subflow socket, not the mptcp one. | |
1322 | */ | |
1323 | WARN_ON_ONCE(1); | |
1324 | return 0; | |
18b683bf PA |
1325 | } |
1326 | ||
b0519de8 FW |
1327 | #if IS_ENABLED(CONFIG_MPTCP_IPV6) |
1328 | static struct ipv6_pinfo *mptcp_inet6_sk(const struct sock *sk) | |
1329 | { | |
1330 | unsigned int offset = sizeof(struct mptcp6_sock) - sizeof(struct ipv6_pinfo); | |
1331 | ||
1332 | return (struct ipv6_pinfo *)(((u8 *)sk) + offset); | |
1333 | } | |
1334 | #endif | |
1335 | ||
fca5c82c PA |
1336 | struct sock *mptcp_sk_clone(const struct sock *sk, |
1337 | const struct tcp_options_received *opt_rx, | |
1338 | struct request_sock *req) | |
b0519de8 | 1339 | { |
58b09919 | 1340 | struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req); |
b0519de8 | 1341 | struct sock *nsk = sk_clone_lock(sk, GFP_ATOMIC); |
58b09919 PA |
1342 | struct mptcp_sock *msk; |
1343 | u64 ack_seq; | |
b0519de8 FW |
1344 | |
1345 | if (!nsk) | |
1346 | return NULL; | |
1347 | ||
1348 | #if IS_ENABLED(CONFIG_MPTCP_IPV6) | |
1349 | if (nsk->sk_family == AF_INET6) | |
1350 | inet_sk(nsk)->pinet6 = mptcp_inet6_sk(nsk); | |
1351 | #endif | |
1352 | ||
58b09919 PA |
1353 | __mptcp_init_sock(nsk); |
1354 | ||
1355 | msk = mptcp_sk(nsk); | |
1356 | msk->local_key = subflow_req->local_key; | |
1357 | msk->token = subflow_req->token; | |
1358 | msk->subflow = NULL; | |
1359 | ||
1360 | if (unlikely(mptcp_token_new_accept(subflow_req->token, nsk))) { | |
9f5ca6a5 | 1361 | nsk->sk_state = TCP_CLOSE; |
58b09919 PA |
1362 | bh_unlock_sock(nsk); |
1363 | ||
1364 | /* we can't call into mptcp_close() here - possible BH context | |
9f5ca6a5 FW |
1365 | * free the sock directly. |
1366 | * sk_clone_lock() sets nsk refcnt to two, hence call sk_free() | |
1367 | * too. | |
58b09919 | 1368 | */ |
9f5ca6a5 | 1369 | sk_common_release(nsk); |
58b09919 PA |
1370 | sk_free(nsk); |
1371 | return NULL; | |
1372 | } | |
1373 | ||
1374 | msk->write_seq = subflow_req->idsn + 1; | |
cc9d2566 | 1375 | atomic64_set(&msk->snd_una, msk->write_seq); |
fca5c82c | 1376 | if (opt_rx->mptcp.mp_capable) { |
58b09919 | 1377 | msk->can_ack = true; |
fca5c82c | 1378 | msk->remote_key = opt_rx->mptcp.sndr_key; |
58b09919 PA |
1379 | mptcp_crypto_key_sha(msk->remote_key, NULL, &ack_seq); |
1380 | ack_seq++; | |
1381 | msk->ack_seq = ack_seq; | |
1382 | } | |
7f20d5fc | 1383 | |
5e20087d | 1384 | sock_reset_flag(nsk, SOCK_RCU_FREE); |
7f20d5fc PA |
1385 | /* will be fully established after successful MPC subflow creation */ |
1386 | inet_sk_state_store(nsk, TCP_SYN_RECV); | |
58b09919 PA |
1387 | bh_unlock_sock(nsk); |
1388 | ||
1389 | /* keep a single reference */ | |
1390 | __sock_put(nsk); | |
b0519de8 FW |
1391 | return nsk; |
1392 | } | |
1393 | ||
cf7da0d6 PK |
1394 | static struct sock *mptcp_accept(struct sock *sk, int flags, int *err, |
1395 | bool kern) | |
1396 | { | |
1397 | struct mptcp_sock *msk = mptcp_sk(sk); | |
1398 | struct socket *listener; | |
1399 | struct sock *newsk; | |
1400 | ||
1401 | listener = __mptcp_nmpc_socket(msk); | |
1402 | if (WARN_ON_ONCE(!listener)) { | |
1403 | *err = -EINVAL; | |
1404 | return NULL; | |
1405 | } | |
1406 | ||
1407 | pr_debug("msk=%p, listener=%p", msk, mptcp_subflow_ctx(listener->sk)); | |
1408 | newsk = inet_csk_accept(listener->sk, flags, err, kern); | |
1409 | if (!newsk) | |
1410 | return NULL; | |
1411 | ||
1412 | pr_debug("msk=%p, subflow is mptcp=%d", msk, sk_is_mptcp(newsk)); | |
1413 | ||
1414 | if (sk_is_mptcp(newsk)) { | |
1415 | struct mptcp_subflow_context *subflow; | |
1416 | struct sock *new_mptcp_sock; | |
1417 | struct sock *ssk = newsk; | |
1418 | ||
1419 | subflow = mptcp_subflow_ctx(newsk); | |
58b09919 | 1420 | new_mptcp_sock = subflow->conn; |
cf7da0d6 | 1421 | |
58b09919 PA |
1422 | /* is_mptcp should be false if subflow->conn is missing, see |
1423 | * subflow_syn_recv_sock() | |
1424 | */ | |
1425 | if (WARN_ON_ONCE(!new_mptcp_sock)) { | |
1426 | tcp_sk(newsk)->is_mptcp = 0; | |
1427 | return newsk; | |
cf7da0d6 PK |
1428 | } |
1429 | ||
58b09919 PA |
1430 | /* acquire the 2nd reference for the owning socket */ |
1431 | sock_hold(new_mptcp_sock); | |
cf7da0d6 | 1432 | |
58b09919 PA |
1433 | local_bh_disable(); |
1434 | bh_lock_sock(new_mptcp_sock); | |
cf7da0d6 | 1435 | msk = mptcp_sk(new_mptcp_sock); |
8ab183de | 1436 | msk->first = newsk; |
cf7da0d6 PK |
1437 | |
1438 | newsk = new_mptcp_sock; | |
1439 | mptcp_copy_inaddrs(newsk, ssk); | |
1440 | list_add(&subflow->node, &msk->conn_list); | |
df1036da | 1441 | inet_sk_state_store(newsk, TCP_ESTABLISHED); |
cf7da0d6 | 1442 | |
cf7da0d6 | 1443 | bh_unlock_sock(new_mptcp_sock); |
fc518953 FW |
1444 | |
1445 | __MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPCAPABLEPASSIVEACK); | |
cf7da0d6 | 1446 | local_bh_enable(); |
fc518953 FW |
1447 | } else { |
1448 | MPTCP_INC_STATS(sock_net(sk), | |
1449 | MPTCP_MIB_MPCAPABLEPASSIVEFALLBACK); | |
cf7da0d6 PK |
1450 | } |
1451 | ||
1452 | return newsk; | |
1453 | } | |
1454 | ||
79c0949e PK |
1455 | static void mptcp_destroy(struct sock *sk) |
1456 | { | |
c9fd9c5f FW |
1457 | struct mptcp_sock *msk = mptcp_sk(sk); |
1458 | ||
1459 | if (msk->cached_ext) | |
1460 | __skb_ext_put(msk->cached_ext); | |
d027236c PA |
1461 | |
1462 | sk_sockets_allocated_dec(sk); | |
79c0949e PK |
1463 | } |
1464 | ||
717e79c8 | 1465 | static int mptcp_setsockopt(struct sock *sk, int level, int optname, |
50e741bb | 1466 | char __user *optval, unsigned int optlen) |
717e79c8 PK |
1467 | { |
1468 | struct mptcp_sock *msk = mptcp_sk(sk); | |
717e79c8 | 1469 | struct socket *ssock; |
717e79c8 PK |
1470 | |
1471 | pr_debug("msk=%p", msk); | |
1472 | ||
1473 | /* @@ the meaning of setsockopt() when the socket is connected and | |
b6e4a1ae MM |
1474 | * there are multiple subflows is not yet defined. It is up to the |
1475 | * MPTCP-level socket to configure the subflows until the subflow | |
1476 | * is in TCP fallback, when TCP socket options are passed through | |
1477 | * to the one remaining subflow. | |
717e79c8 PK |
1478 | */ |
1479 | lock_sock(sk); | |
b6e4a1ae | 1480 | ssock = __mptcp_tcp_fallback(msk); |
e154659b | 1481 | release_sock(sk); |
b6e4a1ae MM |
1482 | if (ssock) |
1483 | return tcp_setsockopt(ssock->sk, level, optname, optval, | |
1484 | optlen); | |
50e741bb | 1485 | |
b6e4a1ae | 1486 | return -EOPNOTSUPP; |
717e79c8 PK |
1487 | } |
1488 | ||
1489 | static int mptcp_getsockopt(struct sock *sk, int level, int optname, | |
50e741bb | 1490 | char __user *optval, int __user *option) |
717e79c8 PK |
1491 | { |
1492 | struct mptcp_sock *msk = mptcp_sk(sk); | |
717e79c8 | 1493 | struct socket *ssock; |
717e79c8 PK |
1494 | |
1495 | pr_debug("msk=%p", msk); | |
1496 | ||
b6e4a1ae MM |
1497 | /* @@ the meaning of setsockopt() when the socket is connected and |
1498 | * there are multiple subflows is not yet defined. It is up to the | |
1499 | * MPTCP-level socket to configure the subflows until the subflow | |
1500 | * is in TCP fallback, when socket options are passed through | |
1501 | * to the one remaining subflow. | |
717e79c8 PK |
1502 | */ |
1503 | lock_sock(sk); | |
b6e4a1ae | 1504 | ssock = __mptcp_tcp_fallback(msk); |
e154659b | 1505 | release_sock(sk); |
b6e4a1ae MM |
1506 | if (ssock) |
1507 | return tcp_getsockopt(ssock->sk, level, optname, optval, | |
1508 | option); | |
50e741bb | 1509 | |
b6e4a1ae | 1510 | return -EOPNOTSUPP; |
717e79c8 PK |
1511 | } |
1512 | ||
b51f9b80 PA |
1513 | #define MPTCP_DEFERRED_ALL (TCPF_DELACK_TIMER_DEFERRED | \ |
1514 | TCPF_WRITE_TIMER_DEFERRED) | |
14c441b5 PA |
1515 | |
1516 | /* this is very alike tcp_release_cb() but we must handle differently a | |
1517 | * different set of events | |
1518 | */ | |
1519 | static void mptcp_release_cb(struct sock *sk) | |
1520 | { | |
1521 | unsigned long flags, nflags; | |
1522 | ||
1523 | do { | |
1524 | flags = sk->sk_tsq_flags; | |
1525 | if (!(flags & MPTCP_DEFERRED_ALL)) | |
1526 | return; | |
1527 | nflags = flags & ~MPTCP_DEFERRED_ALL; | |
1528 | } while (cmpxchg(&sk->sk_tsq_flags, flags, nflags) != flags); | |
1529 | ||
b51f9b80 PA |
1530 | sock_release_ownership(sk); |
1531 | ||
14c441b5 PA |
1532 | if (flags & TCPF_DELACK_TIMER_DEFERRED) { |
1533 | struct mptcp_sock *msk = mptcp_sk(sk); | |
1534 | struct sock *ssk; | |
1535 | ||
1536 | ssk = mptcp_subflow_recv_lookup(msk); | |
1537 | if (!ssk || !schedule_work(&msk->work)) | |
1538 | __sock_put(sk); | |
1539 | } | |
b51f9b80 PA |
1540 | |
1541 | if (flags & TCPF_WRITE_TIMER_DEFERRED) { | |
1542 | mptcp_retransmit_handler(sk); | |
1543 | __sock_put(sk); | |
1544 | } | |
14c441b5 PA |
1545 | } |
1546 | ||
cec37a6e | 1547 | static int mptcp_get_port(struct sock *sk, unsigned short snum) |
f870fa0b MM |
1548 | { |
1549 | struct mptcp_sock *msk = mptcp_sk(sk); | |
cec37a6e | 1550 | struct socket *ssock; |
f870fa0b | 1551 | |
cec37a6e PK |
1552 | ssock = __mptcp_nmpc_socket(msk); |
1553 | pr_debug("msk=%p, subflow=%p", msk, ssock); | |
1554 | if (WARN_ON_ONCE(!ssock)) | |
1555 | return -EINVAL; | |
f870fa0b | 1556 | |
cec37a6e PK |
1557 | return inet_csk_get_port(ssock->sk, snum); |
1558 | } | |
f870fa0b | 1559 | |
cec37a6e PK |
1560 | void mptcp_finish_connect(struct sock *ssk) |
1561 | { | |
1562 | struct mptcp_subflow_context *subflow; | |
1563 | struct mptcp_sock *msk; | |
1564 | struct sock *sk; | |
6d0060f6 | 1565 | u64 ack_seq; |
f870fa0b | 1566 | |
cec37a6e | 1567 | subflow = mptcp_subflow_ctx(ssk); |
cec37a6e PK |
1568 | sk = subflow->conn; |
1569 | msk = mptcp_sk(sk); | |
1570 | ||
fc518953 FW |
1571 | if (!subflow->mp_capable) { |
1572 | MPTCP_INC_STATS(sock_net(sk), | |
1573 | MPTCP_MIB_MPCAPABLEACTIVEFALLBACK); | |
1574 | return; | |
1575 | } | |
1576 | ||
648ef4b8 MM |
1577 | pr_debug("msk=%p, token=%u", sk, subflow->token); |
1578 | ||
6d0060f6 MM |
1579 | mptcp_crypto_key_sha(subflow->remote_key, NULL, &ack_seq); |
1580 | ack_seq++; | |
648ef4b8 MM |
1581 | subflow->map_seq = ack_seq; |
1582 | subflow->map_subflow_seq = 1; | |
6d0060f6 MM |
1583 | subflow->rel_write_seq = 1; |
1584 | ||
cec37a6e PK |
1585 | /* the socket is not connected yet, no msk/subflow ops can access/race |
1586 | * accessing the field below | |
1587 | */ | |
1588 | WRITE_ONCE(msk->remote_key, subflow->remote_key); | |
1589 | WRITE_ONCE(msk->local_key, subflow->local_key); | |
79c0949e | 1590 | WRITE_ONCE(msk->token, subflow->token); |
6d0060f6 MM |
1591 | WRITE_ONCE(msk->write_seq, subflow->idsn + 1); |
1592 | WRITE_ONCE(msk->ack_seq, ack_seq); | |
d22f4988 | 1593 | WRITE_ONCE(msk->can_ack, 1); |
cc9d2566 | 1594 | atomic64_set(&msk->snd_una, msk->write_seq); |
1b1c7a0e PK |
1595 | |
1596 | mptcp_pm_new_connection(msk, 0); | |
f870fa0b MM |
1597 | } |
1598 | ||
cf7da0d6 PK |
1599 | static void mptcp_sock_graft(struct sock *sk, struct socket *parent) |
1600 | { | |
1601 | write_lock_bh(&sk->sk_callback_lock); | |
1602 | rcu_assign_pointer(sk->sk_wq, &parent->wq); | |
1603 | sk_set_socket(sk, parent); | |
1604 | sk->sk_uid = SOCK_INODE(parent)->i_uid; | |
1605 | write_unlock_bh(&sk->sk_callback_lock); | |
1606 | } | |
1607 | ||
f296234c PK |
1608 | bool mptcp_finish_join(struct sock *sk) |
1609 | { | |
1610 | struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); | |
1611 | struct mptcp_sock *msk = mptcp_sk(subflow->conn); | |
1612 | struct sock *parent = (void *)msk; | |
1613 | struct socket *parent_sock; | |
ec3edaa7 | 1614 | bool ret; |
f296234c PK |
1615 | |
1616 | pr_debug("msk=%p, subflow=%p", msk, subflow); | |
1617 | ||
1618 | /* mptcp socket already closing? */ | |
1619 | if (inet_sk_state_load(parent) != TCP_ESTABLISHED) | |
1620 | return false; | |
1621 | ||
1622 | if (!msk->pm.server_side) | |
1623 | return true; | |
1624 | ||
1625 | /* passive connection, attach to msk socket */ | |
1626 | parent_sock = READ_ONCE(parent->sk_socket); | |
1627 | if (parent_sock && !sk->sk_socket) | |
1628 | mptcp_sock_graft(sk, parent_sock); | |
1629 | ||
ec3edaa7 PK |
1630 | ret = mptcp_pm_allow_new_subflow(msk); |
1631 | if (ret) { | |
1632 | /* active connections are already on conn_list */ | |
1633 | spin_lock_bh(&msk->join_list_lock); | |
1634 | if (!WARN_ON_ONCE(!list_empty(&subflow->node))) | |
1635 | list_add_tail(&subflow->node, &msk->join_list); | |
1636 | spin_unlock_bh(&msk->join_list_lock); | |
1637 | } | |
1638 | return ret; | |
f296234c PK |
1639 | } |
1640 | ||
1641 | bool mptcp_sk_is_subflow(const struct sock *sk) | |
1642 | { | |
1643 | struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); | |
1644 | ||
1645 | return subflow->mp_join == 1; | |
1646 | } | |
1647 | ||
1891c4a0 FW |
1648 | static bool mptcp_memory_free(const struct sock *sk, int wake) |
1649 | { | |
1650 | struct mptcp_sock *msk = mptcp_sk(sk); | |
1651 | ||
1652 | return wake ? test_bit(MPTCP_SEND_SPACE, &msk->flags) : true; | |
1653 | } | |
1654 | ||
f870fa0b MM |
1655 | static struct proto mptcp_prot = { |
1656 | .name = "MPTCP", | |
1657 | .owner = THIS_MODULE, | |
1658 | .init = mptcp_init_sock, | |
18b683bf | 1659 | .disconnect = mptcp_disconnect, |
f870fa0b | 1660 | .close = mptcp_close, |
cf7da0d6 | 1661 | .accept = mptcp_accept, |
717e79c8 PK |
1662 | .setsockopt = mptcp_setsockopt, |
1663 | .getsockopt = mptcp_getsockopt, | |
f870fa0b | 1664 | .shutdown = tcp_shutdown, |
79c0949e | 1665 | .destroy = mptcp_destroy, |
f870fa0b MM |
1666 | .sendmsg = mptcp_sendmsg, |
1667 | .recvmsg = mptcp_recvmsg, | |
14c441b5 | 1668 | .release_cb = mptcp_release_cb, |
f870fa0b MM |
1669 | .hash = inet_hash, |
1670 | .unhash = inet_unhash, | |
cec37a6e | 1671 | .get_port = mptcp_get_port, |
d027236c PA |
1672 | .sockets_allocated = &mptcp_sockets_allocated, |
1673 | .memory_allocated = &tcp_memory_allocated, | |
1674 | .memory_pressure = &tcp_memory_pressure, | |
1891c4a0 | 1675 | .stream_memory_free = mptcp_memory_free, |
d027236c PA |
1676 | .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_tcp_wmem), |
1677 | .sysctl_mem = sysctl_tcp_mem, | |
f870fa0b MM |
1678 | .obj_size = sizeof(struct mptcp_sock), |
1679 | .no_autobind = true, | |
1680 | }; | |
1681 | ||
2303f994 PK |
1682 | static int mptcp_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) |
1683 | { | |
1684 | struct mptcp_sock *msk = mptcp_sk(sock->sk); | |
1685 | struct socket *ssock; | |
cf7da0d6 | 1686 | int err; |
2303f994 PK |
1687 | |
1688 | lock_sock(sock->sk); | |
1689 | ssock = __mptcp_socket_create(msk, MPTCP_SAME_STATE); | |
1690 | if (IS_ERR(ssock)) { | |
1691 | err = PTR_ERR(ssock); | |
1692 | goto unlock; | |
1693 | } | |
1694 | ||
1695 | err = ssock->ops->bind(ssock, uaddr, addr_len); | |
cf7da0d6 PK |
1696 | if (!err) |
1697 | mptcp_copy_inaddrs(sock->sk, ssock->sk); | |
2303f994 PK |
1698 | |
1699 | unlock: | |
1700 | release_sock(sock->sk); | |
1701 | return err; | |
1702 | } | |
1703 | ||
1704 | static int mptcp_stream_connect(struct socket *sock, struct sockaddr *uaddr, | |
1705 | int addr_len, int flags) | |
1706 | { | |
1707 | struct mptcp_sock *msk = mptcp_sk(sock->sk); | |
1708 | struct socket *ssock; | |
1709 | int err; | |
1710 | ||
1711 | lock_sock(sock->sk); | |
1712 | ssock = __mptcp_socket_create(msk, TCP_SYN_SENT); | |
1713 | if (IS_ERR(ssock)) { | |
1714 | err = PTR_ERR(ssock); | |
1715 | goto unlock; | |
1716 | } | |
1717 | ||
cf7da0d6 PK |
1718 | #ifdef CONFIG_TCP_MD5SIG |
1719 | /* no MPTCP if MD5SIG is enabled on this socket or we may run out of | |
1720 | * TCP option space. | |
1721 | */ | |
1722 | if (rcu_access_pointer(tcp_sk(ssock->sk)->md5sig_info)) | |
1723 | mptcp_subflow_ctx(ssock->sk)->request_mptcp = 0; | |
1724 | #endif | |
1725 | ||
2303f994 PK |
1726 | err = ssock->ops->connect(ssock, uaddr, addr_len, flags); |
1727 | inet_sk_state_store(sock->sk, inet_sk_state_load(ssock->sk)); | |
cf7da0d6 | 1728 | mptcp_copy_inaddrs(sock->sk, ssock->sk); |
2303f994 PK |
1729 | |
1730 | unlock: | |
1731 | release_sock(sock->sk); | |
1732 | return err; | |
1733 | } | |
1734 | ||
cf7da0d6 PK |
1735 | static int mptcp_v4_getname(struct socket *sock, struct sockaddr *uaddr, |
1736 | int peer) | |
1737 | { | |
1738 | if (sock->sk->sk_prot == &tcp_prot) { | |
1739 | /* we are being invoked from __sys_accept4, after | |
1740 | * mptcp_accept() has just accepted a non-mp-capable | |
1741 | * flow: sk is a tcp_sk, not an mptcp one. | |
1742 | * | |
1743 | * Hand the socket over to tcp so all further socket ops | |
1744 | * bypass mptcp. | |
1745 | */ | |
1746 | sock->ops = &inet_stream_ops; | |
1747 | } | |
1748 | ||
1749 | return inet_getname(sock, uaddr, peer); | |
1750 | } | |
1751 | ||
1752 | #if IS_ENABLED(CONFIG_MPTCP_IPV6) | |
1753 | static int mptcp_v6_getname(struct socket *sock, struct sockaddr *uaddr, | |
1754 | int peer) | |
1755 | { | |
1756 | if (sock->sk->sk_prot == &tcpv6_prot) { | |
1757 | /* we are being invoked from __sys_accept4 after | |
1758 | * mptcp_accept() has accepted a non-mp-capable | |
1759 | * subflow: sk is a tcp_sk, not mptcp. | |
1760 | * | |
1761 | * Hand the socket over to tcp so all further | |
1762 | * socket ops bypass mptcp. | |
1763 | */ | |
1764 | sock->ops = &inet6_stream_ops; | |
1765 | } | |
1766 | ||
1767 | return inet6_getname(sock, uaddr, peer); | |
1768 | } | |
1769 | #endif | |
1770 | ||
1771 | static int mptcp_listen(struct socket *sock, int backlog) | |
1772 | { | |
1773 | struct mptcp_sock *msk = mptcp_sk(sock->sk); | |
1774 | struct socket *ssock; | |
1775 | int err; | |
1776 | ||
1777 | pr_debug("msk=%p", msk); | |
1778 | ||
1779 | lock_sock(sock->sk); | |
1780 | ssock = __mptcp_socket_create(msk, TCP_LISTEN); | |
1781 | if (IS_ERR(ssock)) { | |
1782 | err = PTR_ERR(ssock); | |
1783 | goto unlock; | |
1784 | } | |
1785 | ||
5e20087d FW |
1786 | sock_set_flag(sock->sk, SOCK_RCU_FREE); |
1787 | ||
cf7da0d6 PK |
1788 | err = ssock->ops->listen(ssock, backlog); |
1789 | inet_sk_state_store(sock->sk, inet_sk_state_load(ssock->sk)); | |
1790 | if (!err) | |
1791 | mptcp_copy_inaddrs(sock->sk, ssock->sk); | |
1792 | ||
1793 | unlock: | |
1794 | release_sock(sock->sk); | |
1795 | return err; | |
1796 | } | |
1797 | ||
1798 | static bool is_tcp_proto(const struct proto *p) | |
1799 | { | |
1800 | #if IS_ENABLED(CONFIG_MPTCP_IPV6) | |
1801 | return p == &tcp_prot || p == &tcpv6_prot; | |
1802 | #else | |
1803 | return p == &tcp_prot; | |
1804 | #endif | |
1805 | } | |
1806 | ||
1807 | static int mptcp_stream_accept(struct socket *sock, struct socket *newsock, | |
1808 | int flags, bool kern) | |
1809 | { | |
1810 | struct mptcp_sock *msk = mptcp_sk(sock->sk); | |
1811 | struct socket *ssock; | |
1812 | int err; | |
1813 | ||
1814 | pr_debug("msk=%p", msk); | |
1815 | ||
1816 | lock_sock(sock->sk); | |
1817 | if (sock->sk->sk_state != TCP_LISTEN) | |
1818 | goto unlock_fail; | |
1819 | ||
1820 | ssock = __mptcp_nmpc_socket(msk); | |
1821 | if (!ssock) | |
1822 | goto unlock_fail; | |
1823 | ||
1824 | sock_hold(ssock->sk); | |
1825 | release_sock(sock->sk); | |
1826 | ||
1827 | err = ssock->ops->accept(sock, newsock, flags, kern); | |
1828 | if (err == 0 && !is_tcp_proto(newsock->sk->sk_prot)) { | |
1829 | struct mptcp_sock *msk = mptcp_sk(newsock->sk); | |
1830 | struct mptcp_subflow_context *subflow; | |
1831 | ||
1832 | /* set ssk->sk_socket of accept()ed flows to mptcp socket. | |
1833 | * This is needed so NOSPACE flag can be set from tcp stack. | |
1834 | */ | |
ec3edaa7 | 1835 | __mptcp_flush_join_list(msk); |
cf7da0d6 PK |
1836 | list_for_each_entry(subflow, &msk->conn_list, node) { |
1837 | struct sock *ssk = mptcp_subflow_tcp_sock(subflow); | |
1838 | ||
1839 | if (!ssk->sk_socket) | |
1840 | mptcp_sock_graft(ssk, newsock); | |
1841 | } | |
cf7da0d6 PK |
1842 | } |
1843 | ||
1844 | sock_put(ssock->sk); | |
1845 | return err; | |
1846 | ||
1847 | unlock_fail: | |
1848 | release_sock(sock->sk); | |
1849 | return -EINVAL; | |
1850 | } | |
1851 | ||
2303f994 PK |
1852 | static __poll_t mptcp_poll(struct file *file, struct socket *sock, |
1853 | struct poll_table_struct *wait) | |
1854 | { | |
1891c4a0 | 1855 | struct sock *sk = sock->sk; |
8ab183de | 1856 | struct mptcp_sock *msk; |
1891c4a0 | 1857 | struct socket *ssock; |
2303f994 PK |
1858 | __poll_t mask = 0; |
1859 | ||
1891c4a0 FW |
1860 | msk = mptcp_sk(sk); |
1861 | lock_sock(sk); | |
0b4f33de FW |
1862 | ssock = __mptcp_tcp_fallback(msk); |
1863 | if (!ssock) | |
1864 | ssock = __mptcp_nmpc_socket(msk); | |
1891c4a0 FW |
1865 | if (ssock) { |
1866 | mask = ssock->ops->poll(file, ssock, wait); | |
1867 | release_sock(sk); | |
1868 | return mask; | |
1869 | } | |
1870 | ||
1871 | release_sock(sk); | |
1872 | sock_poll_wait(file, sock, wait); | |
1873 | lock_sock(sk); | |
1874 | ||
1875 | if (test_bit(MPTCP_DATA_READY, &msk->flags)) | |
1876 | mask = EPOLLIN | EPOLLRDNORM; | |
1877 | if (sk_stream_is_writeable(sk) && | |
1878 | test_bit(MPTCP_SEND_SPACE, &msk->flags)) | |
1879 | mask |= EPOLLOUT | EPOLLWRNORM; | |
1880 | if (sk->sk_shutdown & RCV_SHUTDOWN) | |
1881 | mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP; | |
1882 | ||
1883 | release_sock(sk); | |
1884 | ||
2303f994 PK |
1885 | return mask; |
1886 | } | |
1887 | ||
21498490 PK |
1888 | static int mptcp_shutdown(struct socket *sock, int how) |
1889 | { | |
1890 | struct mptcp_sock *msk = mptcp_sk(sock->sk); | |
1891 | struct mptcp_subflow_context *subflow; | |
0b4f33de | 1892 | struct socket *ssock; |
21498490 PK |
1893 | int ret = 0; |
1894 | ||
1895 | pr_debug("sk=%p, how=%d", msk, how); | |
1896 | ||
1897 | lock_sock(sock->sk); | |
0b4f33de FW |
1898 | ssock = __mptcp_tcp_fallback(msk); |
1899 | if (ssock) { | |
1900 | release_sock(sock->sk); | |
1901 | return inet_shutdown(ssock, how); | |
1902 | } | |
21498490 PK |
1903 | |
1904 | if (how == SHUT_WR || how == SHUT_RDWR) | |
1905 | inet_sk_state_store(sock->sk, TCP_FIN_WAIT1); | |
1906 | ||
1907 | how++; | |
1908 | ||
1909 | if ((how & ~SHUTDOWN_MASK) || !how) { | |
1910 | ret = -EINVAL; | |
1911 | goto out_unlock; | |
1912 | } | |
1913 | ||
1914 | if (sock->state == SS_CONNECTING) { | |
1915 | if ((1 << sock->sk->sk_state) & | |
1916 | (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE)) | |
1917 | sock->state = SS_DISCONNECTING; | |
1918 | else | |
1919 | sock->state = SS_CONNECTED; | |
1920 | } | |
1921 | ||
ec3edaa7 | 1922 | __mptcp_flush_join_list(msk); |
21498490 PK |
1923 | mptcp_for_each_subflow(msk, subflow) { |
1924 | struct sock *tcp_sk = mptcp_subflow_tcp_sock(subflow); | |
1925 | ||
76c42a29 | 1926 | mptcp_subflow_shutdown(tcp_sk, how, 1, msk->write_seq); |
21498490 PK |
1927 | } |
1928 | ||
1929 | out_unlock: | |
1930 | release_sock(sock->sk); | |
1931 | ||
1932 | return ret; | |
1933 | } | |
1934 | ||
e42f1ac6 FW |
1935 | static const struct proto_ops mptcp_stream_ops = { |
1936 | .family = PF_INET, | |
1937 | .owner = THIS_MODULE, | |
1938 | .release = inet_release, | |
1939 | .bind = mptcp_bind, | |
1940 | .connect = mptcp_stream_connect, | |
1941 | .socketpair = sock_no_socketpair, | |
1942 | .accept = mptcp_stream_accept, | |
1943 | .getname = mptcp_v4_getname, | |
1944 | .poll = mptcp_poll, | |
1945 | .ioctl = inet_ioctl, | |
1946 | .gettstamp = sock_gettstamp, | |
1947 | .listen = mptcp_listen, | |
1948 | .shutdown = mptcp_shutdown, | |
1949 | .setsockopt = sock_common_setsockopt, | |
1950 | .getsockopt = sock_common_getsockopt, | |
1951 | .sendmsg = inet_sendmsg, | |
1952 | .recvmsg = inet_recvmsg, | |
1953 | .mmap = sock_no_mmap, | |
1954 | .sendpage = inet_sendpage, | |
1955 | #ifdef CONFIG_COMPAT | |
1956 | .compat_setsockopt = compat_sock_common_setsockopt, | |
1957 | .compat_getsockopt = compat_sock_common_getsockopt, | |
1958 | #endif | |
1959 | }; | |
2303f994 | 1960 | |
f870fa0b MM |
1961 | static struct inet_protosw mptcp_protosw = { |
1962 | .type = SOCK_STREAM, | |
1963 | .protocol = IPPROTO_MPTCP, | |
1964 | .prot = &mptcp_prot, | |
2303f994 PK |
1965 | .ops = &mptcp_stream_ops, |
1966 | .flags = INET_PROTOSW_ICSK, | |
f870fa0b MM |
1967 | }; |
1968 | ||
784325e9 | 1969 | void mptcp_proto_init(void) |
f870fa0b | 1970 | { |
2303f994 | 1971 | mptcp_prot.h.hashinfo = tcp_prot.h.hashinfo; |
2303f994 | 1972 | |
d027236c PA |
1973 | if (percpu_counter_init(&mptcp_sockets_allocated, 0, GFP_KERNEL)) |
1974 | panic("Failed to allocate MPTCP pcpu counter\n"); | |
1975 | ||
2303f994 | 1976 | mptcp_subflow_init(); |
1b1c7a0e | 1977 | mptcp_pm_init(); |
2303f994 | 1978 | |
f870fa0b MM |
1979 | if (proto_register(&mptcp_prot, 1) != 0) |
1980 | panic("Failed to register MPTCP proto.\n"); | |
1981 | ||
1982 | inet_register_protosw(&mptcp_protosw); | |
6771bfd9 FW |
1983 | |
1984 | BUILD_BUG_ON(sizeof(struct mptcp_skb_cb) > sizeof_field(struct sk_buff, cb)); | |
f870fa0b MM |
1985 | } |
1986 | ||
1987 | #if IS_ENABLED(CONFIG_MPTCP_IPV6) | |
e42f1ac6 FW |
1988 | static const struct proto_ops mptcp_v6_stream_ops = { |
1989 | .family = PF_INET6, | |
1990 | .owner = THIS_MODULE, | |
1991 | .release = inet6_release, | |
1992 | .bind = mptcp_bind, | |
1993 | .connect = mptcp_stream_connect, | |
1994 | .socketpair = sock_no_socketpair, | |
1995 | .accept = mptcp_stream_accept, | |
1996 | .getname = mptcp_v6_getname, | |
1997 | .poll = mptcp_poll, | |
1998 | .ioctl = inet6_ioctl, | |
1999 | .gettstamp = sock_gettstamp, | |
2000 | .listen = mptcp_listen, | |
2001 | .shutdown = mptcp_shutdown, | |
2002 | .setsockopt = sock_common_setsockopt, | |
2003 | .getsockopt = sock_common_getsockopt, | |
2004 | .sendmsg = inet6_sendmsg, | |
2005 | .recvmsg = inet6_recvmsg, | |
2006 | .mmap = sock_no_mmap, | |
2007 | .sendpage = inet_sendpage, | |
2008 | #ifdef CONFIG_COMPAT | |
2009 | .compat_setsockopt = compat_sock_common_setsockopt, | |
2010 | .compat_getsockopt = compat_sock_common_getsockopt, | |
2011 | #endif | |
2012 | }; | |
2013 | ||
f870fa0b MM |
2014 | static struct proto mptcp_v6_prot; |
2015 | ||
79c0949e PK |
2016 | static void mptcp_v6_destroy(struct sock *sk) |
2017 | { | |
2018 | mptcp_destroy(sk); | |
2019 | inet6_destroy_sock(sk); | |
2020 | } | |
2021 | ||
f870fa0b MM |
2022 | static struct inet_protosw mptcp_v6_protosw = { |
2023 | .type = SOCK_STREAM, | |
2024 | .protocol = IPPROTO_MPTCP, | |
2025 | .prot = &mptcp_v6_prot, | |
2303f994 | 2026 | .ops = &mptcp_v6_stream_ops, |
f870fa0b MM |
2027 | .flags = INET_PROTOSW_ICSK, |
2028 | }; | |
2029 | ||
784325e9 | 2030 | int mptcp_proto_v6_init(void) |
f870fa0b MM |
2031 | { |
2032 | int err; | |
2033 | ||
2034 | mptcp_v6_prot = mptcp_prot; | |
2035 | strcpy(mptcp_v6_prot.name, "MPTCPv6"); | |
2036 | mptcp_v6_prot.slab = NULL; | |
79c0949e | 2037 | mptcp_v6_prot.destroy = mptcp_v6_destroy; |
b0519de8 | 2038 | mptcp_v6_prot.obj_size = sizeof(struct mptcp6_sock); |
f870fa0b MM |
2039 | |
2040 | err = proto_register(&mptcp_v6_prot, 1); | |
2041 | if (err) | |
2042 | return err; | |
2043 | ||
2044 | err = inet6_register_protosw(&mptcp_v6_protosw); | |
2045 | if (err) | |
2046 | proto_unregister(&mptcp_v6_prot); | |
2047 | ||
2048 | return err; | |
2049 | } | |
2050 | #endif |