Commit | Line | Data |
---|---|---|
2303f994 PK |
1 | // SPDX-License-Identifier: GPL-2.0 |
2 | /* Multipath TCP | |
3 | * | |
4 | * Copyright (c) 2017 - 2019, Intel Corporation. | |
5 | */ | |
6 | ||
79c0949e PK |
7 | #define pr_fmt(fmt) "MPTCP: " fmt |
8 | ||
2303f994 PK |
9 | #include <linux/kernel.h> |
10 | #include <linux/module.h> | |
11 | #include <linux/netdevice.h> | |
f296234c | 12 | #include <crypto/algapi.h> |
a24d22b2 | 13 | #include <crypto/sha2.h> |
2303f994 PK |
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> | |
cec37a6e PK |
19 | #if IS_ENABLED(CONFIG_MPTCP_IPV6) |
20 | #include <net/ip6_route.h> | |
b19bc294 | 21 | #include <net/transp_v6.h> |
cec37a6e | 22 | #endif |
2303f994 | 23 | #include <net/mptcp.h> |
4596a2c1 | 24 | #include <uapi/linux/mptcp.h> |
2303f994 | 25 | #include "protocol.h" |
fc518953 FW |
26 | #include "mib.h" |
27 | ||
0918e34b | 28 | #include <trace/events/mptcp.h> |
40e0b090 | 29 | #include <trace/events/sock.h> |
0918e34b | 30 | |
b19bc294 PA |
31 | static void mptcp_subflow_ops_undo_override(struct sock *ssk); |
32 | ||
fc518953 FW |
33 | static void SUBFLOW_REQ_INC_STATS(struct request_sock *req, |
34 | enum linux_mptcp_mib_field field) | |
35 | { | |
36 | MPTCP_INC_STATS(sock_net(req_to_sk(req)), field); | |
37 | } | |
2303f994 | 38 | |
79c0949e PK |
39 | static void subflow_req_destructor(struct request_sock *req) |
40 | { | |
41 | struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req); | |
42 | ||
43 | pr_debug("subflow_req=%p", subflow_req); | |
44 | ||
8fd4de12 PA |
45 | if (subflow_req->msk) |
46 | sock_put((struct sock *)subflow_req->msk); | |
47 | ||
2c5ebd00 | 48 | mptcp_token_destroy_request(req); |
79c0949e PK |
49 | } |
50 | ||
f296234c PK |
51 | static void subflow_generate_hmac(u64 key1, u64 key2, u32 nonce1, u32 nonce2, |
52 | void *hmac) | |
53 | { | |
54 | u8 msg[8]; | |
55 | ||
56 | put_unaligned_be32(nonce1, &msg[0]); | |
57 | put_unaligned_be32(nonce2, &msg[4]); | |
58 | ||
59 | mptcp_crypto_hmac_sha(key1, key2, msg, 8, hmac); | |
60 | } | |
61 | ||
4cf8b7e4 PA |
62 | static bool mptcp_can_accept_new_subflow(const struct mptcp_sock *msk) |
63 | { | |
64 | return mptcp_is_fully_established((void *)msk) && | |
4d25247d KM |
65 | ((mptcp_pm_is_userspace(msk) && |
66 | mptcp_userspace_pm_active(msk)) || | |
67 | READ_ONCE(msk->pm.accept_subflow)); | |
4cf8b7e4 PA |
68 | } |
69 | ||
f296234c | 70 | /* validate received token and create truncated hmac and nonce for SYN-ACK */ |
ec20e143 GT |
71 | static void subflow_req_create_thmac(struct mptcp_subflow_request_sock *subflow_req) |
72 | { | |
73 | struct mptcp_sock *msk = subflow_req->msk; | |
74 | u8 hmac[SHA256_DIGEST_SIZE]; | |
75 | ||
76 | get_random_bytes(&subflow_req->local_nonce, sizeof(u32)); | |
77 | ||
78 | subflow_generate_hmac(msk->local_key, msk->remote_key, | |
79 | subflow_req->local_nonce, | |
80 | subflow_req->remote_nonce, hmac); | |
81 | ||
82 | subflow_req->thmac = get_unaligned_be64(hmac); | |
83 | } | |
84 | ||
b5e2e42f | 85 | static struct mptcp_sock *subflow_token_join_request(struct request_sock *req) |
f296234c PK |
86 | { |
87 | struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req); | |
f296234c PK |
88 | struct mptcp_sock *msk; |
89 | int local_id; | |
90 | ||
ea1300b9 | 91 | msk = mptcp_token_get_sock(sock_net(req_to_sk(req)), subflow_req->token); |
f296234c | 92 | if (!msk) { |
fc518953 | 93 | SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINNOTOKEN); |
8fd4de12 | 94 | return NULL; |
f296234c PK |
95 | } |
96 | ||
97 | local_id = mptcp_pm_get_local_id(msk, (struct sock_common *)req); | |
98 | if (local_id < 0) { | |
99 | sock_put((struct sock *)msk); | |
8fd4de12 | 100 | return NULL; |
f296234c PK |
101 | } |
102 | subflow_req->local_id = local_id; | |
103 | ||
8fd4de12 | 104 | return msk; |
f296234c PK |
105 | } |
106 | ||
d8b59efa | 107 | static void subflow_init_req(struct request_sock *req, const struct sock *sk_listener) |
cec37a6e | 108 | { |
cec37a6e | 109 | struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req); |
cec37a6e PK |
110 | |
111 | subflow_req->mp_capable = 0; | |
f296234c | 112 | subflow_req->mp_join = 0; |
06fe1719 | 113 | subflow_req->csum_reqd = mptcp_is_checksum_enabled(sock_net(sk_listener)); |
bab6b88e | 114 | subflow_req->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk_listener)); |
8fd4de12 | 115 | subflow_req->msk = NULL; |
2c5ebd00 | 116 | mptcp_token_init_request(req); |
78d8b7bc FW |
117 | } |
118 | ||
5bc56388 GT |
119 | static bool subflow_use_different_sport(struct mptcp_sock *msk, const struct sock *sk) |
120 | { | |
121 | return inet_sk(sk)->inet_sport != inet_sk((struct sock *)msk)->inet_sport; | |
122 | } | |
123 | ||
dc87efdb FW |
124 | static void subflow_add_reset_reason(struct sk_buff *skb, u8 reason) |
125 | { | |
126 | struct mptcp_ext *mpext = skb_ext_add(skb, SKB_EXT_MPTCP); | |
127 | ||
128 | if (mpext) { | |
129 | memset(mpext, 0, sizeof(*mpext)); | |
130 | mpext->reset_reason = reason; | |
131 | } | |
132 | } | |
133 | ||
3ecfbe3e FW |
134 | /* Init mptcp request socket. |
135 | * | |
136 | * Returns an error code if a JOIN has failed and a TCP reset | |
137 | * should be sent. | |
138 | */ | |
d8b59efa PA |
139 | static int subflow_check_req(struct request_sock *req, |
140 | const struct sock *sk_listener, | |
141 | struct sk_buff *skb) | |
78d8b7bc FW |
142 | { |
143 | struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener); | |
144 | struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req); | |
145 | struct mptcp_options_received mp_opt; | |
74c7dfbe | 146 | bool opt_mp_capable, opt_mp_join; |
78d8b7bc FW |
147 | |
148 | pr_debug("subflow_req=%p, listener=%p", subflow_req, listener); | |
149 | ||
d8b59efa PA |
150 | #ifdef CONFIG_TCP_MD5SIG |
151 | /* no MPTCP if MD5SIG is enabled on this socket or we may run out of | |
152 | * TCP option space. | |
153 | */ | |
154 | if (rcu_access_pointer(tcp_sk(sk_listener)->md5sig_info)) | |
155 | return -EINVAL; | |
156 | #endif | |
78d8b7bc | 157 | |
0799e21b | 158 | mptcp_get_options(skb, &mp_opt); |
78d8b7bc | 159 | |
74c7dfbe PA |
160 | opt_mp_capable = !!(mp_opt.suboptions & OPTIONS_MPTCP_MPC); |
161 | opt_mp_join = !!(mp_opt.suboptions & OPTIONS_MPTCP_MPJ); | |
162 | if (opt_mp_capable) { | |
fc518953 FW |
163 | SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEPASSIVE); |
164 | ||
74c7dfbe | 165 | if (opt_mp_join) |
3ecfbe3e | 166 | return 0; |
74c7dfbe | 167 | } else if (opt_mp_join) { |
fc518953 FW |
168 | SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINSYNRX); |
169 | } | |
f296234c | 170 | |
74c7dfbe | 171 | if (opt_mp_capable && listener->request_mptcp) { |
c68a0cd1 | 172 | int err, retries = MPTCP_TOKEN_MAX_RETRIES; |
535fb815 | 173 | |
c83a47e5 | 174 | subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq; |
535fb815 FW |
175 | again: |
176 | do { | |
177 | get_random_bytes(&subflow_req->local_key, sizeof(subflow_req->local_key)); | |
178 | } while (subflow_req->local_key == 0); | |
79c0949e | 179 | |
c83a47e5 FW |
180 | if (unlikely(req->syncookie)) { |
181 | mptcp_crypto_key_sha(subflow_req->local_key, | |
182 | &subflow_req->token, | |
183 | &subflow_req->idsn); | |
184 | if (mptcp_token_exists(subflow_req->token)) { | |
185 | if (retries-- > 0) | |
186 | goto again; | |
a16195e3 | 187 | SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_TOKENFALLBACKINIT); |
c83a47e5 FW |
188 | } else { |
189 | subflow_req->mp_capable = 1; | |
190 | } | |
3ecfbe3e | 191 | return 0; |
c83a47e5 FW |
192 | } |
193 | ||
79c0949e PK |
194 | err = mptcp_token_new_request(req); |
195 | if (err == 0) | |
196 | subflow_req->mp_capable = 1; | |
535fb815 FW |
197 | else if (retries-- > 0) |
198 | goto again; | |
a16195e3 PA |
199 | else |
200 | SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_TOKENFALLBACKINIT); | |
79c0949e | 201 | |
74c7dfbe | 202 | } else if (opt_mp_join && listener->request_mptcp) { |
ec3edaa7 | 203 | subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq; |
f296234c | 204 | subflow_req->mp_join = 1; |
cfde141e PA |
205 | subflow_req->backup = mp_opt.backup; |
206 | subflow_req->remote_id = mp_opt.join_id; | |
207 | subflow_req->token = mp_opt.token; | |
208 | subflow_req->remote_nonce = mp_opt.nonce; | |
b5e2e42f | 209 | subflow_req->msk = subflow_token_join_request(req); |
9466a1cc | 210 | |
3ecfbe3e | 211 | /* Can't fall back to TCP in this case. */ |
dc87efdb FW |
212 | if (!subflow_req->msk) { |
213 | subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP); | |
3ecfbe3e | 214 | return -EPERM; |
dc87efdb | 215 | } |
3ecfbe3e | 216 | |
5bc56388 GT |
217 | if (subflow_use_different_sport(subflow_req->msk, sk_listener)) { |
218 | pr_debug("syn inet_sport=%d %d", | |
219 | ntohs(inet_sk(sk_listener)->inet_sport), | |
220 | ntohs(inet_sk((struct sock *)subflow_req->msk)->inet_sport)); | |
221 | if (!mptcp_pm_sport_in_anno_list(subflow_req->msk, sk_listener)) { | |
2fbdd9ea | 222 | SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTSYNRX); |
5bc56388 GT |
223 | return -EPERM; |
224 | } | |
2fbdd9ea | 225 | SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTSYNRX); |
5bc56388 GT |
226 | } |
227 | ||
ec20e143 GT |
228 | subflow_req_create_thmac(subflow_req); |
229 | ||
3ecfbe3e | 230 | if (unlikely(req->syncookie)) { |
9466a1cc FW |
231 | if (mptcp_can_accept_new_subflow(subflow_req->msk)) |
232 | subflow_init_req_cookie_join_save(subflow_req, skb); | |
8547ea5f JW |
233 | else |
234 | return -EPERM; | |
9466a1cc FW |
235 | } |
236 | ||
8fd4de12 PA |
237 | pr_debug("token=%u, remote_nonce=%u msk=%p", subflow_req->token, |
238 | subflow_req->remote_nonce, subflow_req->msk); | |
cec37a6e | 239 | } |
3ecfbe3e FW |
240 | |
241 | return 0; | |
cec37a6e PK |
242 | } |
243 | ||
c83a47e5 FW |
244 | int mptcp_subflow_init_cookie_req(struct request_sock *req, |
245 | const struct sock *sk_listener, | |
246 | struct sk_buff *skb) | |
247 | { | |
248 | struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener); | |
249 | struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req); | |
250 | struct mptcp_options_received mp_opt; | |
74c7dfbe | 251 | bool opt_mp_capable, opt_mp_join; |
c83a47e5 FW |
252 | int err; |
253 | ||
d8b59efa | 254 | subflow_init_req(req, sk_listener); |
0799e21b | 255 | mptcp_get_options(skb, &mp_opt); |
c83a47e5 | 256 | |
74c7dfbe PA |
257 | opt_mp_capable = !!(mp_opt.suboptions & OPTIONS_MPTCP_MPC); |
258 | opt_mp_join = !!(mp_opt.suboptions & OPTIONS_MPTCP_MPJ); | |
259 | if (opt_mp_capable && opt_mp_join) | |
c83a47e5 FW |
260 | return -EINVAL; |
261 | ||
74c7dfbe | 262 | if (opt_mp_capable && listener->request_mptcp) { |
c83a47e5 FW |
263 | if (mp_opt.sndr_key == 0) |
264 | return -EINVAL; | |
265 | ||
266 | subflow_req->local_key = mp_opt.rcvr_key; | |
267 | err = mptcp_token_new_request(req); | |
268 | if (err) | |
269 | return err; | |
270 | ||
271 | subflow_req->mp_capable = 1; | |
272 | subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1; | |
74c7dfbe | 273 | } else if (opt_mp_join && listener->request_mptcp) { |
9466a1cc FW |
274 | if (!mptcp_token_join_cookie_init_state(subflow_req, skb)) |
275 | return -EINVAL; | |
276 | ||
8547ea5f | 277 | subflow_req->mp_join = 1; |
9466a1cc | 278 | subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1; |
c83a47e5 FW |
279 | } |
280 | ||
281 | return 0; | |
282 | } | |
283 | EXPORT_SYMBOL_GPL(mptcp_subflow_init_cookie_req); | |
284 | ||
7ea851d1 FW |
285 | static struct dst_entry *subflow_v4_route_req(const struct sock *sk, |
286 | struct sk_buff *skb, | |
287 | struct flowi *fl, | |
288 | struct request_sock *req) | |
cec37a6e | 289 | { |
7ea851d1 | 290 | struct dst_entry *dst; |
3ecfbe3e | 291 | int err; |
7ea851d1 | 292 | |
cec37a6e | 293 | tcp_rsk(req)->is_mptcp = 1; |
d8b59efa | 294 | subflow_init_req(req, sk); |
cec37a6e | 295 | |
7ea851d1 FW |
296 | dst = tcp_request_sock_ipv4_ops.route_req(sk, skb, fl, req); |
297 | if (!dst) | |
298 | return NULL; | |
cec37a6e | 299 | |
d8b59efa | 300 | err = subflow_check_req(req, sk, skb); |
3ecfbe3e FW |
301 | if (err == 0) |
302 | return dst; | |
cec37a6e | 303 | |
3ecfbe3e FW |
304 | dst_release(dst); |
305 | if (!req->syncookie) | |
306 | tcp_request_sock_ops.send_reset(sk, skb); | |
307 | return NULL; | |
cec37a6e PK |
308 | } |
309 | ||
36b122ba DS |
310 | static void subflow_prep_synack(const struct sock *sk, struct request_sock *req, |
311 | struct tcp_fastopen_cookie *foc, | |
312 | enum tcp_synack_type synack_type) | |
313 | { | |
314 | struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); | |
315 | struct inet_request_sock *ireq = inet_rsk(req); | |
316 | ||
317 | /* clear tstamp_ok, as needed depending on cookie */ | |
318 | if (foc && foc->len > -1) | |
319 | ireq->tstamp_ok = 0; | |
320 | ||
321 | if (synack_type == TCP_SYNACK_FASTOPEN) | |
322 | mptcp_fastopen_subflow_synack_set_params(subflow, req); | |
323 | } | |
324 | ||
325 | static int subflow_v4_send_synack(const struct sock *sk, struct dst_entry *dst, | |
326 | struct flowi *fl, | |
327 | struct request_sock *req, | |
328 | struct tcp_fastopen_cookie *foc, | |
329 | enum tcp_synack_type synack_type, | |
330 | struct sk_buff *syn_skb) | |
331 | { | |
332 | subflow_prep_synack(sk, req, foc, synack_type); | |
333 | ||
334 | return tcp_request_sock_ipv4_ops.send_synack(sk, dst, fl, req, foc, | |
335 | synack_type, syn_skb); | |
336 | } | |
337 | ||
cec37a6e | 338 | #if IS_ENABLED(CONFIG_MPTCP_IPV6) |
36b122ba DS |
339 | static int subflow_v6_send_synack(const struct sock *sk, struct dst_entry *dst, |
340 | struct flowi *fl, | |
341 | struct request_sock *req, | |
342 | struct tcp_fastopen_cookie *foc, | |
343 | enum tcp_synack_type synack_type, | |
344 | struct sk_buff *syn_skb) | |
345 | { | |
346 | subflow_prep_synack(sk, req, foc, synack_type); | |
347 | ||
348 | return tcp_request_sock_ipv6_ops.send_synack(sk, dst, fl, req, foc, | |
349 | synack_type, syn_skb); | |
350 | } | |
351 | ||
7ea851d1 FW |
352 | static struct dst_entry *subflow_v6_route_req(const struct sock *sk, |
353 | struct sk_buff *skb, | |
354 | struct flowi *fl, | |
355 | struct request_sock *req) | |
cec37a6e | 356 | { |
7ea851d1 | 357 | struct dst_entry *dst; |
3ecfbe3e | 358 | int err; |
7ea851d1 | 359 | |
cec37a6e | 360 | tcp_rsk(req)->is_mptcp = 1; |
d8b59efa | 361 | subflow_init_req(req, sk); |
cec37a6e | 362 | |
7ea851d1 FW |
363 | dst = tcp_request_sock_ipv6_ops.route_req(sk, skb, fl, req); |
364 | if (!dst) | |
365 | return NULL; | |
cec37a6e | 366 | |
d8b59efa | 367 | err = subflow_check_req(req, sk, skb); |
3ecfbe3e FW |
368 | if (err == 0) |
369 | return dst; | |
370 | ||
371 | dst_release(dst); | |
372 | if (!req->syncookie) | |
373 | tcp6_request_sock_ops.send_reset(sk, skb); | |
374 | return NULL; | |
cec37a6e PK |
375 | } |
376 | #endif | |
377 | ||
ec3edaa7 PK |
378 | /* validate received truncated hmac and create hmac for third ACK */ |
379 | static bool subflow_thmac_valid(struct mptcp_subflow_context *subflow) | |
380 | { | |
bd697222 | 381 | u8 hmac[SHA256_DIGEST_SIZE]; |
ec3edaa7 PK |
382 | u64 thmac; |
383 | ||
384 | subflow_generate_hmac(subflow->remote_key, subflow->local_key, | |
385 | subflow->remote_nonce, subflow->local_nonce, | |
386 | hmac); | |
387 | ||
388 | thmac = get_unaligned_be64(hmac); | |
389 | pr_debug("subflow=%p, token=%u, thmac=%llu, subflow->thmac=%llu\n", | |
742e2f36 | 390 | subflow, subflow->token, thmac, subflow->thmac); |
ec3edaa7 PK |
391 | |
392 | return thmac == subflow->thmac; | |
393 | } | |
394 | ||
d5824847 PA |
395 | void mptcp_subflow_reset(struct sock *ssk) |
396 | { | |
0e4f35d7 PA |
397 | struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); |
398 | struct sock *sk = subflow->conn; | |
399 | ||
3a236aef PA |
400 | /* mptcp_mp_fail_no_response() can reach here on an already closed |
401 | * socket | |
402 | */ | |
403 | if (ssk->sk_state == TCP_CLOSE) | |
404 | return; | |
405 | ||
ab82e996 FW |
406 | /* must hold: tcp_done() could drop last reference on parent */ |
407 | sock_hold(sk); | |
408 | ||
d5824847 PA |
409 | tcp_send_active_reset(ssk, GFP_ATOMIC); |
410 | tcp_done(ssk); | |
a5cb752b PA |
411 | if (!test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &mptcp_sk(sk)->flags)) |
412 | mptcp_schedule_work(sk); | |
ab82e996 FW |
413 | |
414 | sock_put(sk); | |
d5824847 PA |
415 | } |
416 | ||
5bc56388 GT |
417 | static bool subflow_use_different_dport(struct mptcp_sock *msk, const struct sock *sk) |
418 | { | |
419 | return inet_sk(sk)->inet_dport != inet_sk((struct sock *)msk)->inet_dport; | |
420 | } | |
421 | ||
490274b4 PA |
422 | void __mptcp_set_connected(struct sock *sk) |
423 | { | |
424 | if (sk->sk_state == TCP_SYN_SENT) { | |
425 | inet_sk_state_store(sk, TCP_ESTABLISHED); | |
426 | sk->sk_state_change(sk); | |
427 | } | |
428 | } | |
429 | ||
430 | static void mptcp_set_connected(struct sock *sk) | |
431 | { | |
432 | mptcp_data_lock(sk); | |
433 | if (!sock_owned_by_user(sk)) | |
434 | __mptcp_set_connected(sk); | |
435 | else | |
e9d09bac | 436 | __set_bit(MPTCP_CONNECTED, &mptcp_sk(sk)->cb_flags); |
490274b4 PA |
437 | mptcp_data_unlock(sk); |
438 | } | |
439 | ||
b3ea6b27 PA |
440 | static void subflow_set_remote_key(struct mptcp_sock *msk, |
441 | struct mptcp_subflow_context *subflow, | |
442 | const struct mptcp_options_received *mp_opt) | |
443 | { | |
444 | /* active MPC subflow will reach here multiple times: | |
445 | * at subflow_finish_connect() time and at 4th ack time | |
446 | */ | |
447 | if (subflow->remote_key_valid) | |
448 | return; | |
449 | ||
450 | subflow->remote_key_valid = 1; | |
451 | subflow->remote_key = mp_opt->sndr_key; | |
452 | mptcp_crypto_key_sha(subflow->remote_key, NULL, &subflow->iasn); | |
453 | subflow->iasn++; | |
454 | ||
455 | WRITE_ONCE(msk->remote_key, subflow->remote_key); | |
456 | WRITE_ONCE(msk->ack_seq, subflow->iasn); | |
457 | WRITE_ONCE(msk->can_ack, true); | |
458 | atomic64_set(&msk->rcv_wnd_sent, subflow->iasn); | |
459 | } | |
460 | ||
cec37a6e PK |
461 | static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb) |
462 | { | |
463 | struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); | |
cfde141e | 464 | struct mptcp_options_received mp_opt; |
c3c123d1 | 465 | struct sock *parent = subflow->conn; |
b3ea6b27 | 466 | struct mptcp_sock *msk; |
cec37a6e PK |
467 | |
468 | subflow->icsk_af_ops->sk_rx_dst_set(sk, skb); | |
469 | ||
263e1201 PA |
470 | /* be sure no special action on any packet other than syn-ack */ |
471 | if (subflow->conn_finished) | |
472 | return; | |
473 | ||
b3ea6b27 | 474 | msk = mptcp_sk(parent); |
5cf92bba | 475 | mptcp_propagate_sndbuf(parent, sk); |
b0977bb2 | 476 | subflow->rel_write_seq = 1; |
263e1201 | 477 | subflow->conn_finished = 1; |
e1ff9e82 DC |
478 | subflow->ssn_offset = TCP_SKB_CB(skb)->seq; |
479 | pr_debug("subflow=%p synack seq=%x", subflow, subflow->ssn_offset); | |
263e1201 | 480 | |
0799e21b | 481 | mptcp_get_options(skb, &mp_opt); |
fa25e815 | 482 | if (subflow->request_mptcp) { |
74c7dfbe | 483 | if (!(mp_opt.suboptions & OPTIONS_MPTCP_MPC)) { |
fa25e815 PA |
484 | MPTCP_INC_STATS(sock_net(sk), |
485 | MPTCP_MIB_MPCAPABLEACTIVEFALLBACK); | |
486 | mptcp_do_fallback(sk); | |
b3ea6b27 | 487 | pr_fallback(msk); |
fa25e815 PA |
488 | goto fallback; |
489 | } | |
490 | ||
74c7dfbe | 491 | if (mp_opt.suboptions & OPTION_MPTCP_CSUMREQD) |
b3ea6b27 | 492 | WRITE_ONCE(msk->csum_enabled, true); |
df377be3 | 493 | if (mp_opt.deny_join_id0) |
b3ea6b27 | 494 | WRITE_ONCE(msk->pm.remote_deny_join_id0, true); |
263e1201 | 495 | subflow->mp_capable = 1; |
b3ea6b27 | 496 | subflow_set_remote_key(msk, subflow, &mp_opt); |
5695eb88 | 497 | MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPCAPABLEACTIVEACK); |
fa25e815 | 498 | mptcp_finish_connect(sk); |
490274b4 | 499 | mptcp_set_connected(parent); |
fa25e815 PA |
500 | } else if (subflow->request_join) { |
501 | u8 hmac[SHA256_DIGEST_SIZE]; | |
502 | ||
74c7dfbe | 503 | if (!(mp_opt.suboptions & OPTIONS_MPTCP_MPJ)) { |
dc87efdb | 504 | subflow->reset_reason = MPTCP_RST_EMPTCP; |
fa25e815 | 505 | goto do_reset; |
dc87efdb | 506 | } |
fa25e815 | 507 | |
0460ce22 | 508 | subflow->backup = mp_opt.backup; |
cfde141e PA |
509 | subflow->thmac = mp_opt.thmac; |
510 | subflow->remote_nonce = mp_opt.nonce; | |
8a348392 | 511 | subflow->remote_id = mp_opt.join_id; |
0460ce22 PA |
512 | pr_debug("subflow=%p, thmac=%llu, remote_nonce=%u backup=%d", |
513 | subflow, subflow->thmac, subflow->remote_nonce, | |
514 | subflow->backup); | |
263e1201 | 515 | |
ec3edaa7 | 516 | if (!subflow_thmac_valid(subflow)) { |
fc518953 | 517 | MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINACKMAC); |
dc87efdb | 518 | subflow->reset_reason = MPTCP_RST_EMPTCP; |
ec3edaa7 PK |
519 | goto do_reset; |
520 | } | |
521 | ||
0a4d8e96 JW |
522 | if (!mptcp_finish_join(sk)) |
523 | goto do_reset; | |
524 | ||
ec3edaa7 PK |
525 | subflow_generate_hmac(subflow->local_key, subflow->remote_key, |
526 | subflow->local_nonce, | |
527 | subflow->remote_nonce, | |
bd697222 | 528 | hmac); |
bd697222 | 529 | memcpy(subflow->hmac, hmac, MPTCPOPT_HMAC_LEN); |
ec3edaa7 | 530 | |
fa25e815 | 531 | subflow->mp_join = 1; |
fc518953 | 532 | MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKRX); |
5bc56388 | 533 | |
b3ea6b27 | 534 | if (subflow_use_different_dport(msk, sk)) { |
5bc56388 GT |
535 | pr_debug("synack inet_dport=%d %d", |
536 | ntohs(inet_sk(sk)->inet_dport), | |
537 | ntohs(inet_sk(parent)->inet_dport)); | |
2fbdd9ea | 538 | MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINPORTSYNACKRX); |
5bc56388 | 539 | } |
fa25e815 PA |
540 | } else if (mptcp_check_fallback(sk)) { |
541 | fallback: | |
b3ea6b27 | 542 | mptcp_rcv_space_init(msk, sk); |
490274b4 | 543 | mptcp_set_connected(parent); |
cec37a6e | 544 | } |
fa25e815 PA |
545 | return; |
546 | ||
547 | do_reset: | |
dc87efdb | 548 | subflow->reset_transient = 0; |
d5824847 | 549 | mptcp_subflow_reset(sk); |
cec37a6e PK |
550 | } |
551 | ||
4cf86ae8 PA |
552 | static void subflow_set_local_id(struct mptcp_subflow_context *subflow, int local_id) |
553 | { | |
554 | subflow->local_id = local_id; | |
555 | subflow->local_id_valid = 1; | |
556 | } | |
557 | ||
558 | static int subflow_chk_local_id(struct sock *sk) | |
559 | { | |
560 | struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); | |
561 | struct mptcp_sock *msk = mptcp_sk(subflow->conn); | |
562 | int err; | |
563 | ||
564 | if (likely(subflow->local_id_valid)) | |
565 | return 0; | |
566 | ||
567 | err = mptcp_pm_get_local_id(msk, (struct sock_common *)sk); | |
568 | if (err < 0) | |
569 | return err; | |
570 | ||
571 | subflow_set_local_id(subflow, err); | |
572 | return 0; | |
573 | } | |
574 | ||
575 | static int subflow_rebuild_header(struct sock *sk) | |
576 | { | |
577 | int err = subflow_chk_local_id(sk); | |
578 | ||
579 | if (unlikely(err < 0)) | |
580 | return err; | |
581 | ||
582 | return inet_sk_rebuild_header(sk); | |
583 | } | |
584 | ||
585 | #if IS_ENABLED(CONFIG_MPTCP_IPV6) | |
586 | static int subflow_v6_rebuild_header(struct sock *sk) | |
587 | { | |
588 | int err = subflow_chk_local_id(sk); | |
589 | ||
590 | if (unlikely(err < 0)) | |
591 | return err; | |
592 | ||
593 | return inet6_sk_rebuild_header(sk); | |
594 | } | |
595 | #endif | |
596 | ||
34b21d1d | 597 | static struct request_sock_ops mptcp_subflow_v4_request_sock_ops __ro_after_init; |
51fa7f8e | 598 | static struct tcp_request_sock_ops subflow_request_sock_ipv4_ops __ro_after_init; |
cec37a6e PK |
599 | |
600 | static int subflow_v4_conn_request(struct sock *sk, struct sk_buff *skb) | |
601 | { | |
602 | struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); | |
603 | ||
604 | pr_debug("subflow=%p", subflow); | |
605 | ||
606 | /* Never answer to SYNs sent to broadcast or multicast */ | |
607 | if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) | |
608 | goto drop; | |
609 | ||
34b21d1d | 610 | return tcp_conn_request(&mptcp_subflow_v4_request_sock_ops, |
cec37a6e PK |
611 | &subflow_request_sock_ipv4_ops, |
612 | sk, skb); | |
613 | drop: | |
614 | tcp_listendrop(sk); | |
615 | return 0; | |
616 | } | |
617 | ||
d3295fee MB |
618 | static void subflow_v4_req_destructor(struct request_sock *req) |
619 | { | |
620 | subflow_req_destructor(req); | |
621 | tcp_request_sock_ops.destructor(req); | |
622 | } | |
623 | ||
cec37a6e | 624 | #if IS_ENABLED(CONFIG_MPTCP_IPV6) |
34b21d1d | 625 | static struct request_sock_ops mptcp_subflow_v6_request_sock_ops __ro_after_init; |
51fa7f8e FW |
626 | static struct tcp_request_sock_ops subflow_request_sock_ipv6_ops __ro_after_init; |
627 | static struct inet_connection_sock_af_ops subflow_v6_specific __ro_after_init; | |
628 | static struct inet_connection_sock_af_ops subflow_v6m_specific __ro_after_init; | |
822467a4 | 629 | static struct proto tcpv6_prot_override __ro_after_init; |
cec37a6e PK |
630 | |
631 | static int subflow_v6_conn_request(struct sock *sk, struct sk_buff *skb) | |
632 | { | |
633 | struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); | |
634 | ||
635 | pr_debug("subflow=%p", subflow); | |
636 | ||
637 | if (skb->protocol == htons(ETH_P_IP)) | |
638 | return subflow_v4_conn_request(sk, skb); | |
639 | ||
640 | if (!ipv6_unicast_destination(skb)) | |
641 | goto drop; | |
642 | ||
dcc32f4f JK |
643 | if (ipv6_addr_v4mapped(&ipv6_hdr(skb)->saddr)) { |
644 | __IP6_INC_STATS(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS); | |
645 | return 0; | |
646 | } | |
647 | ||
34b21d1d | 648 | return tcp_conn_request(&mptcp_subflow_v6_request_sock_ops, |
cec37a6e PK |
649 | &subflow_request_sock_ipv6_ops, sk, skb); |
650 | ||
651 | drop: | |
652 | tcp_listendrop(sk); | |
653 | return 0; /* don't send reset */ | |
654 | } | |
d3295fee MB |
655 | |
656 | static void subflow_v6_req_destructor(struct request_sock *req) | |
657 | { | |
658 | subflow_req_destructor(req); | |
659 | tcp6_request_sock_ops.destructor(req); | |
660 | } | |
cec37a6e PK |
661 | #endif |
662 | ||
3fff8818 MB |
663 | struct request_sock *mptcp_subflow_reqsk_alloc(const struct request_sock_ops *ops, |
664 | struct sock *sk_listener, | |
665 | bool attach_listener) | |
666 | { | |
34b21d1d MB |
667 | if (ops->family == AF_INET) |
668 | ops = &mptcp_subflow_v4_request_sock_ops; | |
669 | #if IS_ENABLED(CONFIG_MPTCP_IPV6) | |
670 | else if (ops->family == AF_INET6) | |
671 | ops = &mptcp_subflow_v6_request_sock_ops; | |
cec37a6e PK |
672 | #endif |
673 | ||
3fff8818 MB |
674 | return inet_reqsk_alloc(ops, sk_listener, attach_listener); |
675 | } | |
676 | EXPORT_SYMBOL(mptcp_subflow_reqsk_alloc); | |
677 | ||
f296234c PK |
678 | /* validate hmac received in third ACK */ |
679 | static bool subflow_hmac_valid(const struct request_sock *req, | |
cfde141e | 680 | const struct mptcp_options_received *mp_opt) |
f296234c PK |
681 | { |
682 | const struct mptcp_subflow_request_sock *subflow_req; | |
bd697222 | 683 | u8 hmac[SHA256_DIGEST_SIZE]; |
f296234c | 684 | struct mptcp_sock *msk; |
f296234c PK |
685 | |
686 | subflow_req = mptcp_subflow_rsk(req); | |
8fd4de12 | 687 | msk = subflow_req->msk; |
f296234c PK |
688 | if (!msk) |
689 | return false; | |
690 | ||
691 | subflow_generate_hmac(msk->remote_key, msk->local_key, | |
692 | subflow_req->remote_nonce, | |
693 | subflow_req->local_nonce, hmac); | |
694 | ||
8fd4de12 | 695 | return !crypto_memneq(hmac, mp_opt->hmac, MPTCPOPT_HMAC_LEN); |
f296234c PK |
696 | } |
697 | ||
4c8941de PA |
698 | static void subflow_ulp_fallback(struct sock *sk, |
699 | struct mptcp_subflow_context *old_ctx) | |
700 | { | |
701 | struct inet_connection_sock *icsk = inet_csk(sk); | |
702 | ||
703 | mptcp_subflow_tcp_fallback(sk, old_ctx); | |
704 | icsk->icsk_ulp_ops = NULL; | |
705 | rcu_assign_pointer(icsk->icsk_ulp_data, NULL); | |
706 | tcp_sk(sk)->is_mptcp = 0; | |
b19bc294 PA |
707 | |
708 | mptcp_subflow_ops_undo_override(sk); | |
4c8941de PA |
709 | } |
710 | ||
b6985b9b | 711 | void mptcp_subflow_drop_ctx(struct sock *ssk) |
39884604 PA |
712 | { |
713 | struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk); | |
714 | ||
715 | if (!ctx) | |
716 | return; | |
717 | ||
63740448 PA |
718 | list_del(&mptcp_subflow_ctx(ssk)->node); |
719 | if (inet_csk(ssk)->icsk_ulp_ops) { | |
720 | subflow_ulp_fallback(ssk, ctx); | |
721 | if (ctx->conn) | |
722 | sock_put(ctx->conn); | |
723 | } | |
39884604 PA |
724 | |
725 | kfree_rcu(ctx, rcu); | |
726 | } | |
727 | ||
b93df08c | 728 | void mptcp_subflow_fully_established(struct mptcp_subflow_context *subflow, |
b3ea6b27 | 729 | const struct mptcp_options_received *mp_opt) |
b93df08c PA |
730 | { |
731 | struct mptcp_sock *msk = mptcp_sk(subflow->conn); | |
732 | ||
b3ea6b27 | 733 | subflow_set_remote_key(msk, subflow, mp_opt); |
b93df08c | 734 | subflow->fully_established = 1; |
b93df08c | 735 | WRITE_ONCE(msk->fully_established, true); |
dfc8d060 DS |
736 | |
737 | if (subflow->is_mptfo) | |
738 | mptcp_fastopen_gen_msk_ackseq(msk, subflow, mp_opt); | |
b93df08c PA |
739 | } |
740 | ||
cec37a6e PK |
741 | static struct sock *subflow_syn_recv_sock(const struct sock *sk, |
742 | struct sk_buff *skb, | |
743 | struct request_sock *req, | |
744 | struct dst_entry *dst, | |
745 | struct request_sock *req_unhash, | |
746 | bool *own_req) | |
747 | { | |
748 | struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk); | |
cc7972ea | 749 | struct mptcp_subflow_request_sock *subflow_req; |
cfde141e | 750 | struct mptcp_options_received mp_opt; |
9e365ff5 | 751 | bool fallback, fallback_is_fatal; |
3a236aef | 752 | struct mptcp_sock *owner; |
cec37a6e PK |
753 | struct sock *child; |
754 | ||
755 | pr_debug("listener=%p, req=%p, conn=%p", listener, req, listener->conn); | |
756 | ||
74c7dfbe | 757 | /* After child creation we must look for MPC even when options |
9e365ff5 | 758 | * are not parsed |
cfde141e | 759 | */ |
74c7dfbe | 760 | mp_opt.suboptions = 0; |
9e365ff5 PA |
761 | |
762 | /* hopefully temporary handling for MP_JOIN+syncookie */ | |
763 | subflow_req = mptcp_subflow_rsk(req); | |
b7514694 | 764 | fallback_is_fatal = tcp_rsk(req)->is_mptcp && subflow_req->mp_join; |
9e365ff5 PA |
765 | fallback = !tcp_rsk(req)->is_mptcp; |
766 | if (fallback) | |
ae2dd716 FW |
767 | goto create_child; |
768 | ||
d22f4988 | 769 | /* if the sk is MP_CAPABLE, we try to fetch the client key */ |
cc7972ea | 770 | if (subflow_req->mp_capable) { |
06f9a435 PA |
771 | /* we can receive and accept an in-window, out-of-order pkt, |
772 | * which may not carry the MP_CAPABLE opt even on mptcp enabled | |
773 | * paths: always try to extract the peer key, and fallback | |
774 | * for packets missing it. | |
775 | * Even OoO DSS packets coming legitly after dropped or | |
776 | * reordered MPC will cause fallback, but we don't have other | |
777 | * options. | |
778 | */ | |
0799e21b | 779 | mptcp_get_options(skb, &mp_opt); |
a88d0092 | 780 | if (!(mp_opt.suboptions & OPTIONS_MPTCP_MPC)) |
4c8941de | 781 | fallback = true; |
58b09919 | 782 | |
f296234c | 783 | } else if (subflow_req->mp_join) { |
0799e21b | 784 | mptcp_get_options(skb, &mp_opt); |
74c7dfbe PA |
785 | if (!(mp_opt.suboptions & OPTIONS_MPTCP_MPJ) || |
786 | !subflow_hmac_valid(req, &mp_opt) || | |
d3ab7885 | 787 | !mptcp_can_accept_new_subflow(subflow_req->msk)) { |
fc518953 | 788 | SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKMAC); |
9e365ff5 | 789 | fallback = true; |
fc518953 | 790 | } |
cc7972ea | 791 | } |
cec37a6e | 792 | |
d22f4988 | 793 | create_child: |
cec37a6e PK |
794 | child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst, |
795 | req_unhash, own_req); | |
796 | ||
797 | if (child && *own_req) { | |
79c0949e PK |
798 | struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(child); |
799 | ||
90bf4513 PA |
800 | tcp_rsk(req)->drop_req = false; |
801 | ||
4c8941de PA |
802 | /* we need to fallback on ctx allocation failure and on pre-reqs |
803 | * checking above. In the latter scenario we additionally need | |
804 | * to reset the context to non MPTCP status. | |
79c0949e | 805 | */ |
4c8941de | 806 | if (!ctx || fallback) { |
dc87efdb FW |
807 | if (fallback_is_fatal) { |
808 | subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP); | |
729cd643 | 809 | goto dispose_child; |
dc87efdb | 810 | } |
a88d0092 | 811 | goto fallback; |
f296234c | 812 | } |
79c0949e | 813 | |
df00b087 FW |
814 | /* ssk inherits options of listener sk */ |
815 | ctx->setsockopt_seq = listener->setsockopt_seq; | |
816 | ||
79c0949e | 817 | if (ctx->mp_capable) { |
7e8b88ec | 818 | ctx->conn = mptcp_sk_clone_init(listener->conn, &mp_opt, child, req); |
a88d0092 PA |
819 | if (!ctx->conn) |
820 | goto fallback; | |
821 | ||
6f06b4d4 | 822 | ctx->subflow_id = 1; |
a88d0092 | 823 | owner = mptcp_sk(ctx->conn); |
3a236aef | 824 | mptcp_pm_new_connection(owner, child, 1); |
e72e4032 | 825 | |
fca5c82c PA |
826 | /* with OoO packets we can reach here without ingress |
827 | * mpc option | |
828 | */ | |
3a236aef | 829 | if (mp_opt.suboptions & OPTION_MPTCP_MPC_ACK) { |
b93df08c | 830 | mptcp_subflow_fully_established(ctx, &mp_opt); |
7a486c44 | 831 | mptcp_pm_fully_established(owner, child); |
3a236aef PA |
832 | ctx->pm_notified = 1; |
833 | } | |
f296234c | 834 | } else if (ctx->mp_join) { |
8fd4de12 | 835 | owner = subflow_req->msk; |
dc87efdb FW |
836 | if (!owner) { |
837 | subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT); | |
729cd643 | 838 | goto dispose_child; |
dc87efdb | 839 | } |
f296234c | 840 | |
8fd4de12 PA |
841 | /* move the msk reference ownership to the subflow */ |
842 | subflow_req->msk = NULL; | |
f296234c | 843 | ctx->conn = (struct sock *)owner; |
5bc56388 GT |
844 | |
845 | if (subflow_use_different_sport(owner, sk)) { | |
846 | pr_debug("ack inet_sport=%d %d", | |
847 | ntohs(inet_sk(sk)->inet_sport), | |
848 | ntohs(inet_sk((struct sock *)owner)->inet_sport)); | |
2fbdd9ea GT |
849 | if (!mptcp_pm_sport_in_anno_list(owner, sk)) { |
850 | SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTACKRX); | |
9238e900 | 851 | goto dispose_child; |
2fbdd9ea GT |
852 | } |
853 | SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTACKRX); | |
5bc56388 | 854 | } |
9238e900 GT |
855 | |
856 | if (!mptcp_finish_join(child)) | |
857 | goto dispose_child; | |
858 | ||
859 | SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX); | |
860 | tcp_rsk(req)->drop_req = true; | |
cec37a6e PK |
861 | } |
862 | } | |
863 | ||
4c8941de PA |
864 | /* check for expected invariant - should never trigger, just help |
865 | * catching eariler subtle bugs | |
866 | */ | |
ac2b47fb | 867 | WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp && |
4c8941de PA |
868 | (!mptcp_subflow_ctx(child) || |
869 | !mptcp_subflow_ctx(child)->conn)); | |
cec37a6e | 870 | return child; |
f296234c | 871 | |
729cd643 | 872 | dispose_child: |
b6985b9b | 873 | mptcp_subflow_drop_ctx(child); |
729cd643 | 874 | tcp_rsk(req)->drop_req = true; |
729cd643 | 875 | inet_csk_prepare_for_destroy_sock(child); |
f296234c | 876 | tcp_done(child); |
97e61751 | 877 | req->rsk_ops->send_reset(sk, skb); |
729cd643 PA |
878 | |
879 | /* The last child reference will be released by the caller */ | |
880 | return child; | |
a88d0092 PA |
881 | |
882 | fallback: | |
883 | mptcp_subflow_drop_ctx(child); | |
884 | return child; | |
cec37a6e PK |
885 | } |
886 | ||
51fa7f8e | 887 | static struct inet_connection_sock_af_ops subflow_specific __ro_after_init; |
822467a4 | 888 | static struct proto tcp_prot_override __ro_after_init; |
cec37a6e | 889 | |
648ef4b8 MM |
890 | enum mapping_status { |
891 | MAPPING_OK, | |
892 | MAPPING_INVALID, | |
893 | MAPPING_EMPTY, | |
e1ff9e82 | 894 | MAPPING_DATA_FIN, |
31bf11de PA |
895 | MAPPING_DUMMY, |
896 | MAPPING_BAD_CSUM | |
648ef4b8 MM |
897 | }; |
898 | ||
61e71022 | 899 | static void dbg_bad_map(struct mptcp_subflow_context *subflow, u32 ssn) |
648ef4b8 | 900 | { |
61e71022 PA |
901 | pr_debug("Bad mapping: ssn=%d map_seq=%d map_data_len=%d", |
902 | ssn, subflow->map_subflow_seq, subflow->map_data_len); | |
648ef4b8 MM |
903 | } |
904 | ||
905 | static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb) | |
906 | { | |
907 | struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); | |
908 | unsigned int skb_consumed; | |
909 | ||
910 | skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq; | |
911 | if (WARN_ON_ONCE(skb_consumed >= skb->len)) | |
912 | return true; | |
913 | ||
914 | return skb->len - skb_consumed <= subflow->map_data_len - | |
915 | mptcp_subflow_get_map_offset(subflow); | |
916 | } | |
917 | ||
918 | static bool validate_mapping(struct sock *ssk, struct sk_buff *skb) | |
919 | { | |
920 | struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); | |
921 | u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset; | |
922 | ||
923 | if (unlikely(before(ssn, subflow->map_subflow_seq))) { | |
924 | /* Mapping covers data later in the subflow stream, | |
925 | * currently unsupported. | |
926 | */ | |
61e71022 | 927 | dbg_bad_map(subflow, ssn); |
648ef4b8 MM |
928 | return false; |
929 | } | |
930 | if (unlikely(!before(ssn, subflow->map_subflow_seq + | |
931 | subflow->map_data_len))) { | |
932 | /* Mapping does covers past subflow data, invalid */ | |
61e71022 | 933 | dbg_bad_map(subflow, ssn); |
648ef4b8 MM |
934 | return false; |
935 | } | |
936 | return true; | |
937 | } | |
938 | ||
dd8bcd17 PA |
939 | static enum mapping_status validate_data_csum(struct sock *ssk, struct sk_buff *skb, |
940 | bool csum_reqd) | |
941 | { | |
942 | struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); | |
dd8bcd17 | 943 | u32 offset, seq, delta; |
ba2c89e0 | 944 | __sum16 csum; |
dd8bcd17 PA |
945 | int len; |
946 | ||
947 | if (!csum_reqd) | |
948 | return MAPPING_OK; | |
949 | ||
950 | /* mapping already validated on previous traversal */ | |
951 | if (subflow->map_csum_len == subflow->map_data_len) | |
952 | return MAPPING_OK; | |
953 | ||
954 | /* traverse the receive queue, ensuring it contains a full | |
955 | * DSS mapping and accumulating the related csum. | |
956 | * Preserve the accoumlate csum across multiple calls, to compute | |
957 | * the csum only once | |
958 | */ | |
959 | delta = subflow->map_data_len - subflow->map_csum_len; | |
960 | for (;;) { | |
961 | seq = tcp_sk(ssk)->copied_seq + subflow->map_csum_len; | |
962 | offset = seq - TCP_SKB_CB(skb)->seq; | |
963 | ||
964 | /* if the current skb has not been accounted yet, csum its contents | |
965 | * up to the amount covered by the current DSS | |
966 | */ | |
967 | if (offset < skb->len) { | |
968 | __wsum csum; | |
969 | ||
970 | len = min(skb->len - offset, delta); | |
971 | csum = skb_checksum(skb, offset, len, 0); | |
972 | subflow->map_data_csum = csum_block_add(subflow->map_data_csum, csum, | |
973 | subflow->map_csum_len); | |
974 | ||
975 | delta -= len; | |
976 | subflow->map_csum_len += len; | |
977 | } | |
978 | if (delta == 0) | |
979 | break; | |
980 | ||
981 | if (skb_queue_is_last(&ssk->sk_receive_queue, skb)) { | |
982 | /* if this subflow is closed, the partial mapping | |
983 | * will be never completed; flush the pending skbs, so | |
984 | * that subflow_sched_work_if_closed() can kick in | |
985 | */ | |
986 | if (unlikely(ssk->sk_state == TCP_CLOSE)) | |
987 | while ((skb = skb_peek(&ssk->sk_receive_queue))) | |
988 | sk_eat_skb(ssk, skb); | |
989 | ||
990 | /* not enough data to validate the csum */ | |
991 | return MAPPING_EMPTY; | |
992 | } | |
993 | ||
994 | /* the DSS mapping for next skbs will be validated later, | |
995 | * when a get_mapping_status call will process such skb | |
996 | */ | |
997 | skb = skb->next; | |
998 | } | |
999 | ||
1000 | /* note that 'map_data_len' accounts only for the carried data, does | |
1001 | * not include the eventual seq increment due to the data fin, | |
1002 | * while the pseudo header requires the original DSS data len, | |
1003 | * including that | |
1004 | */ | |
8401e87f GT |
1005 | csum = __mptcp_make_csum(subflow->map_seq, |
1006 | subflow->map_subflow_seq, | |
1007 | subflow->map_data_len + subflow->map_data_fin, | |
1008 | subflow->map_data_csum); | |
1009 | if (unlikely(csum)) { | |
fe3ab1cb | 1010 | MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DATACSUMERR); |
31bf11de | 1011 | return MAPPING_BAD_CSUM; |
fe3ab1cb | 1012 | } |
dd8bcd17 | 1013 | |
ae66fb2b | 1014 | subflow->valid_csum_seen = 1; |
dd8bcd17 PA |
1015 | return MAPPING_OK; |
1016 | } | |
1017 | ||
43b54c6e MM |
1018 | static enum mapping_status get_mapping_status(struct sock *ssk, |
1019 | struct mptcp_sock *msk) | |
648ef4b8 MM |
1020 | { |
1021 | struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); | |
dd8bcd17 | 1022 | bool csum_reqd = READ_ONCE(msk->csum_enabled); |
648ef4b8 MM |
1023 | struct mptcp_ext *mpext; |
1024 | struct sk_buff *skb; | |
1025 | u16 data_len; | |
1026 | u64 map_seq; | |
1027 | ||
1028 | skb = skb_peek(&ssk->sk_receive_queue); | |
1029 | if (!skb) | |
1030 | return MAPPING_EMPTY; | |
1031 | ||
e1ff9e82 DC |
1032 | if (mptcp_check_fallback(ssk)) |
1033 | return MAPPING_DUMMY; | |
1034 | ||
648ef4b8 MM |
1035 | mpext = mptcp_get_ext(skb); |
1036 | if (!mpext || !mpext->use_map) { | |
1037 | if (!subflow->map_valid && !skb->len) { | |
1038 | /* the TCP stack deliver 0 len FIN pkt to the receive | |
1039 | * queue, that is the only 0len pkts ever expected here, | |
1040 | * and we can admit no mapping only for 0 len pkts | |
1041 | */ | |
1042 | if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)) | |
1043 | WARN_ONCE(1, "0len seq %d:%d flags %x", | |
1044 | TCP_SKB_CB(skb)->seq, | |
1045 | TCP_SKB_CB(skb)->end_seq, | |
1046 | TCP_SKB_CB(skb)->tcp_flags); | |
1047 | sk_eat_skb(ssk, skb); | |
1048 | return MAPPING_EMPTY; | |
1049 | } | |
1050 | ||
1051 | if (!subflow->map_valid) | |
1052 | return MAPPING_INVALID; | |
1053 | ||
1054 | goto validate_seq; | |
1055 | } | |
1056 | ||
0918e34b | 1057 | trace_get_mapping_status(mpext); |
648ef4b8 MM |
1058 | |
1059 | data_len = mpext->data_len; | |
1060 | if (data_len == 0) { | |
f8d4bcac | 1061 | pr_debug("infinite mapping received"); |
fc518953 | 1062 | MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX); |
f8d4bcac | 1063 | subflow->map_data_len = 0; |
648ef4b8 MM |
1064 | return MAPPING_INVALID; |
1065 | } | |
1066 | ||
1067 | if (mpext->data_fin == 1) { | |
1068 | if (data_len == 1) { | |
1a49b2c2 MM |
1069 | bool updated = mptcp_update_rcv_data_fin(msk, mpext->data_seq, |
1070 | mpext->dsn64); | |
43b54c6e | 1071 | pr_debug("DATA_FIN with no payload seq=%llu", mpext->data_seq); |
648ef4b8 MM |
1072 | if (subflow->map_valid) { |
1073 | /* A DATA_FIN might arrive in a DSS | |
1074 | * option before the previous mapping | |
1075 | * has been fully consumed. Continue | |
1076 | * handling the existing mapping. | |
1077 | */ | |
1078 | skb_ext_del(skb, SKB_EXT_MPTCP); | |
1079 | return MAPPING_OK; | |
1080 | } else { | |
a5cb752b PA |
1081 | if (updated) |
1082 | mptcp_schedule_work((struct sock *)msk); | |
ef59b195 | 1083 | |
648ef4b8 MM |
1084 | return MAPPING_DATA_FIN; |
1085 | } | |
43b54c6e | 1086 | } else { |
017512a0 | 1087 | u64 data_fin_seq = mpext->data_seq + data_len - 1; |
1a49b2c2 MM |
1088 | |
1089 | /* If mpext->data_seq is a 32-bit value, data_fin_seq | |
1090 | * must also be limited to 32 bits. | |
1091 | */ | |
1092 | if (!mpext->dsn64) | |
1093 | data_fin_seq &= GENMASK_ULL(31, 0); | |
1094 | ||
1095 | mptcp_update_rcv_data_fin(msk, data_fin_seq, mpext->dsn64); | |
1096 | pr_debug("DATA_FIN with mapping seq=%llu dsn64=%d", | |
1097 | data_fin_seq, mpext->dsn64); | |
648ef4b8 MM |
1098 | } |
1099 | ||
1100 | /* Adjust for DATA_FIN using 1 byte of sequence space */ | |
1101 | data_len--; | |
1102 | } | |
1103 | ||
5957a890 | 1104 | map_seq = mptcp_expand_seq(READ_ONCE(msk->ack_seq), mpext->data_seq, mpext->dsn64); |
37198e93 | 1105 | WRITE_ONCE(mptcp_sk(subflow->conn)->use_64bit_ack, !!mpext->dsn64); |
648ef4b8 MM |
1106 | |
1107 | if (subflow->map_valid) { | |
1108 | /* Allow replacing only with an identical map */ | |
1109 | if (subflow->map_seq == map_seq && | |
1110 | subflow->map_subflow_seq == mpext->subflow_seq && | |
dd8bcd17 PA |
1111 | subflow->map_data_len == data_len && |
1112 | subflow->map_csum_reqd == mpext->csum_reqd) { | |
648ef4b8 | 1113 | skb_ext_del(skb, SKB_EXT_MPTCP); |
dd8bcd17 | 1114 | goto validate_csum; |
648ef4b8 MM |
1115 | } |
1116 | ||
1117 | /* If this skb data are fully covered by the current mapping, | |
1118 | * the new map would need caching, which is not supported | |
1119 | */ | |
fc518953 FW |
1120 | if (skb_is_fully_mapped(ssk, skb)) { |
1121 | MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH); | |
648ef4b8 | 1122 | return MAPPING_INVALID; |
fc518953 | 1123 | } |
648ef4b8 MM |
1124 | |
1125 | /* will validate the next map after consuming the current one */ | |
dd8bcd17 | 1126 | goto validate_csum; |
648ef4b8 MM |
1127 | } |
1128 | ||
1129 | subflow->map_seq = map_seq; | |
1130 | subflow->map_subflow_seq = mpext->subflow_seq; | |
1131 | subflow->map_data_len = data_len; | |
1132 | subflow->map_valid = 1; | |
dd8bcd17 | 1133 | subflow->map_data_fin = mpext->data_fin; |
d22f4988 | 1134 | subflow->mpc_map = mpext->mpc_map; |
dd8bcd17 PA |
1135 | subflow->map_csum_reqd = mpext->csum_reqd; |
1136 | subflow->map_csum_len = 0; | |
1137 | subflow->map_data_csum = csum_unfold(mpext->csum); | |
1138 | ||
1139 | /* Cfr RFC 8684 Section 3.3.0 */ | |
1140 | if (unlikely(subflow->map_csum_reqd != csum_reqd)) | |
1141 | return MAPPING_INVALID; | |
1142 | ||
1143 | pr_debug("new map seq=%llu subflow_seq=%u data_len=%u csum=%d:%u", | |
648ef4b8 | 1144 | subflow->map_seq, subflow->map_subflow_seq, |
dd8bcd17 PA |
1145 | subflow->map_data_len, subflow->map_csum_reqd, |
1146 | subflow->map_data_csum); | |
648ef4b8 MM |
1147 | |
1148 | validate_seq: | |
1149 | /* we revalidate valid mapping on new skb, because we must ensure | |
1150 | * the current skb is completely covered by the available mapping | |
1151 | */ | |
06285da9 PA |
1152 | if (!validate_mapping(ssk, skb)) { |
1153 | MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSTCPMISMATCH); | |
648ef4b8 | 1154 | return MAPPING_INVALID; |
06285da9 | 1155 | } |
648ef4b8 MM |
1156 | |
1157 | skb_ext_del(skb, SKB_EXT_MPTCP); | |
dd8bcd17 PA |
1158 | |
1159 | validate_csum: | |
1160 | return validate_data_csum(ssk, skb, csum_reqd); | |
648ef4b8 MM |
1161 | } |
1162 | ||
04e4cd4f | 1163 | static void mptcp_subflow_discard_data(struct sock *ssk, struct sk_buff *skb, |
1d39cd8c | 1164 | u64 limit) |
6719331c PA |
1165 | { |
1166 | struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); | |
04e4cd4f PA |
1167 | bool fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN; |
1168 | u32 incr; | |
1169 | ||
1170 | incr = limit >= skb->len ? skb->len + fin : limit; | |
1171 | ||
1172 | pr_debug("discarding=%d len=%d seq=%d", incr, skb->len, | |
1173 | subflow->map_subflow_seq); | |
06242e44 | 1174 | MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DUPDATA); |
04e4cd4f PA |
1175 | tcp_sk(ssk)->copied_seq += incr; |
1176 | if (!before(tcp_sk(ssk)->copied_seq, TCP_SKB_CB(skb)->end_seq)) | |
1177 | sk_eat_skb(ssk, skb); | |
1178 | if (mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) | |
1179 | subflow->map_valid = 0; | |
6719331c PA |
1180 | } |
1181 | ||
40947e13 FW |
1182 | /* sched mptcp worker to remove the subflow if no more data is pending */ |
1183 | static void subflow_sched_work_if_closed(struct mptcp_sock *msk, struct sock *ssk) | |
1184 | { | |
40947e13 FW |
1185 | if (likely(ssk->sk_state != TCP_CLOSE)) |
1186 | return; | |
1187 | ||
1188 | if (skb_queue_empty(&ssk->sk_receive_queue) && | |
a5cb752b PA |
1189 | !test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags)) |
1190 | mptcp_schedule_work((struct sock *)msk); | |
40947e13 FW |
1191 | } |
1192 | ||
ae66fb2b MM |
1193 | static bool subflow_can_fallback(struct mptcp_subflow_context *subflow) |
1194 | { | |
1195 | struct mptcp_sock *msk = mptcp_sk(subflow->conn); | |
1196 | ||
1197 | if (subflow->mp_join) | |
1198 | return false; | |
1199 | else if (READ_ONCE(msk->csum_enabled)) | |
1200 | return !subflow->valid_csum_seen; | |
1201 | else | |
1202 | return !subflow->fully_established; | |
1203 | } | |
1204 | ||
76a13b31 GT |
1205 | static void mptcp_subflow_fail(struct mptcp_sock *msk, struct sock *ssk) |
1206 | { | |
1207 | struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); | |
1208 | unsigned long fail_tout; | |
1209 | ||
1210 | /* greceful failure can happen only on the MPC subflow */ | |
1211 | if (WARN_ON_ONCE(ssk != READ_ONCE(msk->first))) | |
1212 | return; | |
1213 | ||
1214 | /* since the close timeout take precedence on the fail one, | |
1215 | * no need to start the latter when the first is already set | |
1216 | */ | |
1217 | if (sock_flag((struct sock *)msk, SOCK_DEAD)) | |
1218 | return; | |
1219 | ||
1220 | /* we don't need extreme accuracy here, use a zero fail_tout as special | |
1221 | * value meaning no fail timeout at all; | |
1222 | */ | |
1223 | fail_tout = jiffies + TCP_RTO_MAX; | |
1224 | if (!fail_tout) | |
1225 | fail_tout = 1; | |
1226 | WRITE_ONCE(subflow->fail_tout, fail_tout); | |
1227 | tcp_send_ack(ssk); | |
1228 | ||
1229 | mptcp_reset_timeout(msk, subflow->fail_tout); | |
1230 | } | |
1231 | ||
648ef4b8 MM |
1232 | static bool subflow_check_data_avail(struct sock *ssk) |
1233 | { | |
1234 | struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); | |
1235 | enum mapping_status status; | |
1236 | struct mptcp_sock *msk; | |
1237 | struct sk_buff *skb; | |
1238 | ||
47bebdf3 | 1239 | if (!skb_peek(&ssk->sk_receive_queue)) |
ea56dcb4 | 1240 | WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_NODATA); |
648ef4b8 MM |
1241 | if (subflow->data_avail) |
1242 | return true; | |
1243 | ||
648ef4b8 MM |
1244 | msk = mptcp_sk(subflow->conn); |
1245 | for (;;) { | |
648ef4b8 MM |
1246 | u64 ack_seq; |
1247 | u64 old_ack; | |
1248 | ||
43b54c6e | 1249 | status = get_mapping_status(ssk, msk); |
d96a838a | 1250 | trace_subflow_check_data_avail(status, skb_peek(&ssk->sk_receive_queue)); |
31bf11de PA |
1251 | if (unlikely(status == MAPPING_INVALID || status == MAPPING_DUMMY || |
1252 | status == MAPPING_BAD_CSUM)) | |
dea2b1ea | 1253 | goto fallback; |
648ef4b8 MM |
1254 | |
1255 | if (status != MAPPING_OK) | |
40947e13 | 1256 | goto no_data; |
648ef4b8 MM |
1257 | |
1258 | skb = skb_peek(&ssk->sk_receive_queue); | |
1259 | if (WARN_ON_ONCE(!skb)) | |
40947e13 | 1260 | goto no_data; |
648ef4b8 | 1261 | |
b3ea6b27 PA |
1262 | if (unlikely(!READ_ONCE(msk->can_ack))) |
1263 | goto fallback; | |
d22f4988 | 1264 | |
648ef4b8 MM |
1265 | old_ack = READ_ONCE(msk->ack_seq); |
1266 | ack_seq = mptcp_subflow_get_mapped_dsn(subflow); | |
1267 | pr_debug("msk ack_seq=%llx subflow ack_seq=%llx", old_ack, | |
1268 | ack_seq); | |
99d1055c PA |
1269 | if (unlikely(before64(ack_seq, old_ack))) { |
1270 | mptcp_subflow_discard_data(ssk, skb, old_ack - ack_seq); | |
1271 | continue; | |
47bebdf3 | 1272 | } |
648ef4b8 | 1273 | |
99d1055c PA |
1274 | WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_DATA_AVAIL); |
1275 | break; | |
648ef4b8 MM |
1276 | } |
1277 | return true; | |
1278 | ||
40947e13 FW |
1279 | no_data: |
1280 | subflow_sched_work_if_closed(msk, ssk); | |
1281 | return false; | |
dea2b1ea PA |
1282 | |
1283 | fallback: | |
0348c690 GT |
1284 | if (!__mptcp_check_fallback(msk)) { |
1285 | /* RFC 8684 section 3.7. */ | |
31bf11de PA |
1286 | if (status == MAPPING_BAD_CSUM && |
1287 | (subflow->mp_join || subflow->valid_csum_seen)) { | |
1288 | subflow->send_mp_fail = 1; | |
1289 | ||
7b16871f | 1290 | if (!READ_ONCE(msk->allow_infinite_fallback)) { |
0348c690 GT |
1291 | subflow->reset_transient = 0; |
1292 | subflow->reset_reason = MPTCP_RST_EMIDDLEBOX; | |
f745a3eb | 1293 | goto reset; |
0348c690 | 1294 | } |
f745a3eb PA |
1295 | mptcp_subflow_fail(msk, ssk); |
1296 | WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_DATA_AVAIL); | |
0348c690 GT |
1297 | return true; |
1298 | } | |
1299 | ||
d7e6f583 | 1300 | if (!subflow_can_fallback(subflow) && subflow->map_data_len) { |
0348c690 GT |
1301 | /* fatal protocol error, close the socket. |
1302 | * subflow_error_report() will introduce the appropriate barriers | |
1303 | */ | |
1761fed2 | 1304 | subflow->reset_transient = 0; |
0348c690 | 1305 | subflow->reset_reason = MPTCP_RST_EMPTCP; |
f745a3eb PA |
1306 | |
1307 | reset: | |
9ae8e5ad | 1308 | WRITE_ONCE(ssk->sk_err, EBADMSG); |
f745a3eb PA |
1309 | tcp_set_state(ssk, TCP_CLOSE); |
1310 | while ((skb = skb_peek(&ssk->sk_receive_queue))) | |
1311 | sk_eat_skb(ssk, skb); | |
1761fed2 | 1312 | tcp_send_active_reset(ssk, GFP_ATOMIC); |
0348c690 GT |
1313 | WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_NODATA); |
1314 | return false; | |
478d7700 | 1315 | } |
478d7700 | 1316 | |
d51991e2 | 1317 | mptcp_do_fallback(ssk); |
dea2b1ea PA |
1318 | } |
1319 | ||
dea2b1ea PA |
1320 | skb = skb_peek(&ssk->sk_receive_queue); |
1321 | subflow->map_valid = 1; | |
1322 | subflow->map_seq = READ_ONCE(msk->ack_seq); | |
1323 | subflow->map_data_len = skb->len; | |
1324 | subflow->map_subflow_seq = tcp_sk(ssk)->copied_seq - subflow->ssn_offset; | |
99d1055c | 1325 | WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_DATA_AVAIL); |
dea2b1ea | 1326 | return true; |
648ef4b8 MM |
1327 | } |
1328 | ||
1329 | bool mptcp_subflow_data_available(struct sock *sk) | |
1330 | { | |
1331 | struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); | |
648ef4b8 MM |
1332 | |
1333 | /* check if current mapping is still valid */ | |
1334 | if (subflow->map_valid && | |
1335 | mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) { | |
1336 | subflow->map_valid = 0; | |
ea56dcb4 | 1337 | WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_NODATA); |
648ef4b8 MM |
1338 | |
1339 | pr_debug("Done with mapping: seq=%u data_len=%u", | |
1340 | subflow->map_subflow_seq, | |
1341 | subflow->map_data_len); | |
1342 | } | |
1343 | ||
47bebdf3 | 1344 | return subflow_check_data_avail(sk); |
648ef4b8 MM |
1345 | } |
1346 | ||
071c8ed6 FW |
1347 | /* If ssk has an mptcp parent socket, use the mptcp rcvbuf occupancy, |
1348 | * not the ssk one. | |
1349 | * | |
1350 | * In mptcp, rwin is about the mptcp-level connection data. | |
1351 | * | |
1352 | * Data that is still on the ssk rx queue can thus be ignored, | |
55320b82 | 1353 | * as far as mptcp peer is concerned that data is still inflight. |
071c8ed6 FW |
1354 | * DSS ACK is updated when skb is moved to the mptcp rx queue. |
1355 | */ | |
1356 | void mptcp_space(const struct sock *ssk, int *space, int *full_space) | |
1357 | { | |
1358 | const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); | |
1359 | const struct sock *sk = subflow->conn; | |
1360 | ||
ea4ca586 | 1361 | *space = __mptcp_space(sk); |
071c8ed6 FW |
1362 | *full_space = tcp_full_space(sk); |
1363 | } | |
1364 | ||
15cc1045 PA |
1365 | void __mptcp_error_report(struct sock *sk) |
1366 | { | |
1367 | struct mptcp_subflow_context *subflow; | |
1368 | struct mptcp_sock *msk = mptcp_sk(sk); | |
1369 | ||
1370 | mptcp_for_each_subflow(msk, subflow) { | |
1371 | struct sock *ssk = mptcp_subflow_tcp_sock(subflow); | |
1372 | int err = sock_error(ssk); | |
1249db44 | 1373 | int ssk_state; |
15cc1045 PA |
1374 | |
1375 | if (!err) | |
1376 | continue; | |
1377 | ||
1378 | /* only propagate errors on fallen-back sockets or | |
1379 | * on MPC connect | |
1380 | */ | |
1381 | if (sk->sk_state != TCP_SYN_SENT && !__mptcp_check_fallback(msk)) | |
1382 | continue; | |
1383 | ||
1249db44 PA |
1384 | /* We need to propagate only transition to CLOSE state. |
1385 | * Orphaned socket will see such state change via | |
1386 | * subflow_sched_work_if_closed() and that path will properly | |
1387 | * destroy the msk as needed. | |
1388 | */ | |
1389 | ssk_state = inet_sk_state_load(ssk); | |
1390 | if (ssk_state == TCP_CLOSE && !sock_flag(sk, SOCK_DEAD)) | |
1391 | inet_sk_state_store(sk, ssk_state); | |
9ae8e5ad | 1392 | WRITE_ONCE(sk->sk_err, -err); |
15cc1045 PA |
1393 | |
1394 | /* This barrier is coupled with smp_rmb() in mptcp_poll() */ | |
1395 | smp_wmb(); | |
e3ae2365 | 1396 | sk_error_report(sk); |
15cc1045 PA |
1397 | break; |
1398 | } | |
1399 | } | |
1400 | ||
1401 | static void subflow_error_report(struct sock *ssk) | |
1402 | { | |
1403 | struct sock *sk = mptcp_subflow_ctx(ssk)->conn; | |
1404 | ||
b7a679ba PA |
1405 | /* bail early if this is a no-op, so that we avoid introducing a |
1406 | * problematic lockdep dependency between TCP accept queue lock | |
1407 | * and msk socket spinlock | |
1408 | */ | |
1409 | if (!sk->sk_socket) | |
1410 | return; | |
1411 | ||
15cc1045 PA |
1412 | mptcp_data_lock(sk); |
1413 | if (!sock_owned_by_user(sk)) | |
1414 | __mptcp_error_report(sk); | |
1415 | else | |
e9d09bac | 1416 | __set_bit(MPTCP_ERROR_REPORT, &mptcp_sk(sk)->cb_flags); |
15cc1045 PA |
1417 | mptcp_data_unlock(sk); |
1418 | } | |
1419 | ||
499ada50 PA |
1420 | static void subflow_data_ready(struct sock *sk) |
1421 | { | |
1422 | struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); | |
1423 | u16 state = 1 << inet_sk_state_load(sk); | |
1424 | struct sock *parent = subflow->conn; | |
1425 | struct mptcp_sock *msk; | |
1426 | ||
40e0b090 PY |
1427 | trace_sk_data_ready(sk); |
1428 | ||
499ada50 PA |
1429 | msk = mptcp_sk(parent); |
1430 | if (state & TCPF_LISTEN) { | |
1431 | /* MPJ subflow are removed from accept queue before reaching here, | |
1432 | * avoid stray wakeups | |
1433 | */ | |
1434 | if (reqsk_queue_empty(&inet_csk(sk)->icsk_accept_queue)) | |
1435 | return; | |
1436 | ||
499ada50 PA |
1437 | parent->sk_data_ready(parent); |
1438 | return; | |
1439 | } | |
1440 | ||
1441 | WARN_ON_ONCE(!__mptcp_check_fallback(msk) && !subflow->mp_capable && | |
1442 | !subflow->mp_join && !(state & TCPF_CLOSE)); | |
1443 | ||
1444 | if (mptcp_subflow_data_available(sk)) | |
1445 | mptcp_data_ready(parent, sk); | |
1446 | else if (unlikely(sk->sk_err)) | |
1447 | subflow_error_report(sk); | |
1448 | } | |
1449 | ||
1450 | static void subflow_write_space(struct sock *ssk) | |
1451 | { | |
1452 | struct sock *sk = mptcp_subflow_ctx(ssk)->conn; | |
1453 | ||
1454 | mptcp_propagate_sndbuf(sk, ssk); | |
1455 | mptcp_write_space(sk); | |
1456 | } | |
1457 | ||
51fa7f8e | 1458 | static const struct inet_connection_sock_af_ops * |
cec37a6e PK |
1459 | subflow_default_af_ops(struct sock *sk) |
1460 | { | |
1461 | #if IS_ENABLED(CONFIG_MPTCP_IPV6) | |
1462 | if (sk->sk_family == AF_INET6) | |
1463 | return &subflow_v6_specific; | |
1464 | #endif | |
1465 | return &subflow_specific; | |
1466 | } | |
1467 | ||
cec37a6e | 1468 | #if IS_ENABLED(CONFIG_MPTCP_IPV6) |
31484d56 GU |
1469 | void mptcpv6_handle_mapped(struct sock *sk, bool mapped) |
1470 | { | |
cec37a6e PK |
1471 | struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); |
1472 | struct inet_connection_sock *icsk = inet_csk(sk); | |
51fa7f8e | 1473 | const struct inet_connection_sock_af_ops *target; |
cec37a6e PK |
1474 | |
1475 | target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk); | |
1476 | ||
1477 | pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d", | |
edc7e489 | 1478 | subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped); |
cec37a6e PK |
1479 | |
1480 | if (likely(icsk->icsk_af_ops == target)) | |
1481 | return; | |
1482 | ||
1483 | subflow->icsk_af_ops = icsk->icsk_af_ops; | |
1484 | icsk->icsk_af_ops = target; | |
cec37a6e | 1485 | } |
31484d56 | 1486 | #endif |
cec37a6e | 1487 | |
1729cf18 GT |
1488 | void mptcp_info2sockaddr(const struct mptcp_addr_info *info, |
1489 | struct sockaddr_storage *addr, | |
1490 | unsigned short family) | |
ec3edaa7 PK |
1491 | { |
1492 | memset(addr, 0, sizeof(*addr)); | |
50a13bc3 | 1493 | addr->ss_family = family; |
ec3edaa7 PK |
1494 | if (addr->ss_family == AF_INET) { |
1495 | struct sockaddr_in *in_addr = (struct sockaddr_in *)addr; | |
1496 | ||
50a13bc3 MB |
1497 | if (info->family == AF_INET) |
1498 | in_addr->sin_addr = info->addr; | |
1499 | #if IS_ENABLED(CONFIG_MPTCP_IPV6) | |
1500 | else if (ipv6_addr_v4mapped(&info->addr6)) | |
1501 | in_addr->sin_addr.s_addr = info->addr6.s6_addr32[3]; | |
1502 | #endif | |
ec3edaa7 PK |
1503 | in_addr->sin_port = info->port; |
1504 | } | |
1505 | #if IS_ENABLED(CONFIG_MPTCP_IPV6) | |
1506 | else if (addr->ss_family == AF_INET6) { | |
1507 | struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr; | |
1508 | ||
50a13bc3 MB |
1509 | if (info->family == AF_INET) |
1510 | ipv6_addr_set_v4mapped(info->addr.s_addr, | |
1511 | &in6_addr->sin6_addr); | |
1512 | else | |
1513 | in6_addr->sin6_addr = info->addr6; | |
ec3edaa7 PK |
1514 | in6_addr->sin6_port = info->port; |
1515 | } | |
1516 | #endif | |
1517 | } | |
1518 | ||
ef0da3b8 | 1519 | int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_addr_info *loc, |
ee285257 | 1520 | const struct mptcp_addr_info *remote) |
ec3edaa7 PK |
1521 | { |
1522 | struct mptcp_sock *msk = mptcp_sk(sk); | |
1523 | struct mptcp_subflow_context *subflow; | |
1524 | struct sockaddr_storage addr; | |
2ff0e566 | 1525 | int remote_id = remote->id; |
6bad912b | 1526 | int local_id = loc->id; |
95d68651 | 1527 | int err = -ENOTCONN; |
ec3edaa7 | 1528 | struct socket *sf; |
6bad912b | 1529 | struct sock *ssk; |
ec3edaa7 PK |
1530 | u32 remote_token; |
1531 | int addrlen; | |
ee285257 GT |
1532 | int ifindex; |
1533 | u8 flags; | |
ec3edaa7 | 1534 | |
b93df08c | 1535 | if (!mptcp_is_fully_established(sk)) |
95d68651 | 1536 | goto err_out; |
ec3edaa7 | 1537 | |
6bc1fe7d | 1538 | err = mptcp_subflow_create_socket(sk, loc->family, &sf); |
ec3edaa7 | 1539 | if (err) |
95d68651 | 1540 | goto err_out; |
ec3edaa7 | 1541 | |
6bad912b PA |
1542 | ssk = sf->sk; |
1543 | subflow = mptcp_subflow_ctx(ssk); | |
1544 | do { | |
1545 | get_random_bytes(&subflow->local_nonce, sizeof(u32)); | |
1546 | } while (!subflow->local_nonce); | |
1547 | ||
4cf86ae8 PA |
1548 | if (local_id) |
1549 | subflow_set_local_id(subflow, local_id); | |
6bad912b | 1550 | |
8b201370 | 1551 | mptcp_pm_get_flags_and_ifindex_by_id(msk, local_id, |
ee285257 | 1552 | &flags, &ifindex); |
b3ea6b27 | 1553 | subflow->remote_key_valid = 1; |
ec3edaa7 PK |
1554 | subflow->remote_key = msk->remote_key; |
1555 | subflow->local_key = msk->local_key; | |
1556 | subflow->token = msk->token; | |
50a13bc3 | 1557 | mptcp_info2sockaddr(loc, &addr, ssk->sk_family); |
ec3edaa7 PK |
1558 | |
1559 | addrlen = sizeof(struct sockaddr_in); | |
1560 | #if IS_ENABLED(CONFIG_MPTCP_IPV6) | |
50a13bc3 | 1561 | if (addr.ss_family == AF_INET6) |
ec3edaa7 PK |
1562 | addrlen = sizeof(struct sockaddr_in6); |
1563 | #endif | |
c9406a23 FW |
1564 | mptcp_sockopt_sync(msk, ssk); |
1565 | ||
daa83ab0 | 1566 | ssk->sk_bound_dev_if = ifindex; |
ec3edaa7 PK |
1567 | err = kernel_bind(sf, (struct sockaddr *)&addr, addrlen); |
1568 | if (err) | |
1569 | goto failed; | |
1570 | ||
1571 | mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL); | |
2ff0e566 GT |
1572 | pr_debug("msk=%p remote_token=%u local_id=%d remote_id=%d", msk, |
1573 | remote_token, local_id, remote_id); | |
ec3edaa7 | 1574 | subflow->remote_token = remote_token; |
2ff0e566 | 1575 | subflow->remote_id = remote_id; |
ec3edaa7 | 1576 | subflow->request_join = 1; |
daa83ab0 | 1577 | subflow->request_bkup = !!(flags & MPTCP_PM_ADDR_FLAG_BACKUP); |
6f06b4d4 | 1578 | subflow->subflow_id = msk->subflow_id++; |
50a13bc3 | 1579 | mptcp_info2sockaddr(remote, &addr, ssk->sk_family); |
ec3edaa7 | 1580 | |
3e501490 PA |
1581 | sock_hold(ssk); |
1582 | list_add_tail(&subflow->node, &msk->conn_list); | |
ec3edaa7 PK |
1583 | err = kernel_connect(sf, (struct sockaddr *)&addr, addrlen, O_NONBLOCK); |
1584 | if (err && err != -EINPROGRESS) | |
5b950ff4 | 1585 | goto failed_unlink; |
ec3edaa7 | 1586 | |
866f26f2 PA |
1587 | /* discard the subflow socket */ |
1588 | mptcp_sock_graft(ssk, sk->sk_socket); | |
1589 | iput(SOCK_INODE(sf)); | |
0530020a | 1590 | WRITE_ONCE(msk->allow_infinite_fallback, false); |
b5177ed9 | 1591 | return 0; |
ec3edaa7 | 1592 | |
5b950ff4 | 1593 | failed_unlink: |
5b950ff4 | 1594 | list_del(&subflow->node); |
f0715779 | 1595 | sock_put(mptcp_subflow_tcp_sock(subflow)); |
ec3edaa7 | 1596 | |
ec3edaa7 | 1597 | failed: |
e16163b6 | 1598 | subflow->disposable = 1; |
ec3edaa7 | 1599 | sock_release(sf); |
95d68651 PA |
1600 | |
1601 | err_out: | |
1602 | /* we account subflows before the creation, and this failures will not | |
1603 | * be caught by sk_state_change() | |
1604 | */ | |
1605 | mptcp_pm_close_subflow(msk); | |
ec3edaa7 PK |
1606 | return err; |
1607 | } | |
1608 | ||
3764b0c5 NR |
1609 | static void mptcp_attach_cgroup(struct sock *parent, struct sock *child) |
1610 | { | |
1611 | #ifdef CONFIG_SOCK_CGROUP_DATA | |
1612 | struct sock_cgroup_data *parent_skcd = &parent->sk_cgrp_data, | |
1613 | *child_skcd = &child->sk_cgrp_data; | |
1614 | ||
1615 | /* only the additional subflows created by kworkers have to be modified */ | |
1616 | if (cgroup_id(sock_cgroup_ptr(parent_skcd)) != | |
1617 | cgroup_id(sock_cgroup_ptr(child_skcd))) { | |
1618 | #ifdef CONFIG_MEMCG | |
1619 | struct mem_cgroup *memcg = parent->sk_memcg; | |
1620 | ||
1621 | mem_cgroup_sk_free(child); | |
1622 | if (memcg && css_tryget(&memcg->css)) | |
1623 | child->sk_memcg = memcg; | |
1624 | #endif /* CONFIG_MEMCG */ | |
1625 | ||
1626 | cgroup_sk_free(child_skcd); | |
1627 | *child_skcd = *parent_skcd; | |
1628 | cgroup_sk_clone(child_skcd); | |
1629 | } | |
1630 | #endif /* CONFIG_SOCK_CGROUP_DATA */ | |
1631 | } | |
1632 | ||
b19bc294 PA |
1633 | static void mptcp_subflow_ops_override(struct sock *ssk) |
1634 | { | |
1635 | #if IS_ENABLED(CONFIG_MPTCP_IPV6) | |
1636 | if (ssk->sk_prot == &tcpv6_prot) | |
1637 | ssk->sk_prot = &tcpv6_prot_override; | |
1638 | else | |
1639 | #endif | |
1640 | ssk->sk_prot = &tcp_prot_override; | |
1641 | } | |
1642 | ||
1643 | static void mptcp_subflow_ops_undo_override(struct sock *ssk) | |
1644 | { | |
1645 | #if IS_ENABLED(CONFIG_MPTCP_IPV6) | |
1646 | if (ssk->sk_prot == &tcpv6_prot_override) | |
1647 | ssk->sk_prot = &tcpv6_prot; | |
1648 | else | |
1649 | #endif | |
1650 | ssk->sk_prot = &tcp_prot; | |
1651 | } | |
6bc1fe7d PA |
1652 | |
1653 | int mptcp_subflow_create_socket(struct sock *sk, unsigned short family, | |
1654 | struct socket **new_sock) | |
2303f994 PK |
1655 | { |
1656 | struct mptcp_subflow_context *subflow; | |
1657 | struct net *net = sock_net(sk); | |
1658 | struct socket *sf; | |
1659 | int err; | |
1660 | ||
adf73410 PA |
1661 | /* un-accepted server sockets can reach here - on bad configuration |
1662 | * bail early to avoid greater trouble later | |
1663 | */ | |
1664 | if (unlikely(!sk->sk_socket)) | |
1665 | return -EINVAL; | |
1666 | ||
6bc1fe7d | 1667 | err = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP, &sf); |
2303f994 PK |
1668 | if (err) |
1669 | return err; | |
1670 | ||
ad217100 | 1671 | lock_sock_nested(sf->sk, SINGLE_DEPTH_NESTING); |
2303f994 | 1672 | |
e3d9387f PA |
1673 | err = security_mptcp_add_subflow(sk, sf->sk); |
1674 | if (err) | |
1675 | goto release_ssk; | |
1676 | ||
3764b0c5 NR |
1677 | /* the newly created socket has to be in the same cgroup as its parent */ |
1678 | mptcp_attach_cgroup(sk, sf->sk); | |
1679 | ||
2303f994 PK |
1680 | /* kernel sockets do not by default acquire net ref, but TCP timer |
1681 | * needs it. | |
d1e96cc4 | 1682 | * Update ns_tracker to current stack trace and refcounted tracker. |
2303f994 | 1683 | */ |
d1e96cc4 | 1684 | __netns_tracker_free(net, &sf->sk->ns_tracker, false); |
2303f994 | 1685 | sf->sk->sk_net_refcnt = 1; |
1d2f3d3c | 1686 | get_net_track(net, &sf->sk->ns_tracker, GFP_KERNEL); |
d477eb90 | 1687 | sock_inuse_add(net, 1); |
2303f994 | 1688 | err = tcp_set_ulp(sf->sk, "mptcp"); |
e3d9387f PA |
1689 | |
1690 | release_ssk: | |
2303f994 PK |
1691 | release_sock(sf->sk); |
1692 | ||
b8ad540d WY |
1693 | if (err) { |
1694 | sock_release(sf); | |
2303f994 | 1695 | return err; |
b8ad540d | 1696 | } |
2303f994 | 1697 | |
7d14b0d2 PA |
1698 | /* the newly created socket really belongs to the owning MPTCP master |
1699 | * socket, even if for additional subflows the allocation is performed | |
1700 | * by a kernel workqueue. Adjust inode references, so that the | |
d640516a | 1701 | * procfs/diag interfaces really show this one belonging to the correct |
7d14b0d2 PA |
1702 | * user. |
1703 | */ | |
1704 | SOCK_INODE(sf)->i_ino = SOCK_INODE(sk->sk_socket)->i_ino; | |
1705 | SOCK_INODE(sf)->i_uid = SOCK_INODE(sk->sk_socket)->i_uid; | |
1706 | SOCK_INODE(sf)->i_gid = SOCK_INODE(sk->sk_socket)->i_gid; | |
1707 | ||
2303f994 PK |
1708 | subflow = mptcp_subflow_ctx(sf->sk); |
1709 | pr_debug("subflow=%p", subflow); | |
1710 | ||
1711 | *new_sock = sf; | |
79c0949e | 1712 | sock_hold(sk); |
2303f994 | 1713 | subflow->conn = sk; |
b19bc294 | 1714 | mptcp_subflow_ops_override(sf->sk); |
2303f994 PK |
1715 | |
1716 | return 0; | |
1717 | } | |
1718 | ||
1719 | static struct mptcp_subflow_context *subflow_create_ctx(struct sock *sk, | |
1720 | gfp_t priority) | |
1721 | { | |
1722 | struct inet_connection_sock *icsk = inet_csk(sk); | |
1723 | struct mptcp_subflow_context *ctx; | |
1724 | ||
1725 | ctx = kzalloc(sizeof(*ctx), priority); | |
1726 | if (!ctx) | |
1727 | return NULL; | |
1728 | ||
1729 | rcu_assign_pointer(icsk->icsk_ulp_data, ctx); | |
cec37a6e | 1730 | INIT_LIST_HEAD(&ctx->node); |
b19bc294 | 1731 | INIT_LIST_HEAD(&ctx->delegated_node); |
2303f994 PK |
1732 | |
1733 | pr_debug("subflow=%p", ctx); | |
1734 | ||
1735 | ctx->tcp_sock = sk; | |
1736 | ||
1737 | return ctx; | |
1738 | } | |
1739 | ||
648ef4b8 MM |
1740 | static void __subflow_state_change(struct sock *sk) |
1741 | { | |
1742 | struct socket_wq *wq; | |
1743 | ||
1744 | rcu_read_lock(); | |
1745 | wq = rcu_dereference(sk->sk_wq); | |
1746 | if (skwq_has_sleeper(wq)) | |
1747 | wake_up_interruptible_all(&wq->wait); | |
1748 | rcu_read_unlock(); | |
1749 | } | |
1750 | ||
1751 | static bool subflow_is_done(const struct sock *sk) | |
1752 | { | |
1753 | return sk->sk_shutdown & RCV_SHUTDOWN || sk->sk_state == TCP_CLOSE; | |
1754 | } | |
1755 | ||
1756 | static void subflow_state_change(struct sock *sk) | |
1757 | { | |
1758 | struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); | |
dc093db5 | 1759 | struct sock *parent = subflow->conn; |
81c1d029 | 1760 | struct mptcp_sock *msk; |
648ef4b8 MM |
1761 | |
1762 | __subflow_state_change(sk); | |
1763 | ||
81c1d029 | 1764 | msk = mptcp_sk(parent); |
8fd73804 | 1765 | if (subflow_simultaneous_connect(sk)) { |
5cf92bba | 1766 | mptcp_propagate_sndbuf(parent, sk); |
8fd73804 | 1767 | mptcp_do_fallback(sk); |
81c1d029 PA |
1768 | mptcp_rcv_space_init(msk, sk); |
1769 | pr_fallback(msk); | |
8fd73804 | 1770 | subflow->conn_finished = 1; |
490274b4 | 1771 | mptcp_set_connected(parent); |
8fd73804 DC |
1772 | } |
1773 | ||
648ef4b8 MM |
1774 | /* as recvmsg() does not acquire the subflow socket for ssk selection |
1775 | * a fin packet carrying a DSS can be unnoticed if we don't trigger | |
1776 | * the data available machinery here. | |
1777 | */ | |
e1ff9e82 | 1778 | if (mptcp_subflow_data_available(sk)) |
2e52213c | 1779 | mptcp_data_ready(parent, sk); |
499ada50 PA |
1780 | else if (unlikely(sk->sk_err)) |
1781 | subflow_error_report(sk); | |
648ef4b8 | 1782 | |
40947e13 FW |
1783 | subflow_sched_work_if_closed(mptcp_sk(parent), sk); |
1784 | ||
81c1d029 PA |
1785 | /* when the fallback subflow closes the rx side, trigger a 'dummy' |
1786 | * ingress data fin, so that the msk state will follow along | |
1787 | */ | |
1788 | if (__mptcp_check_fallback(msk) && subflow_is_done(sk) && msk->first == sk && | |
1789 | mptcp_update_rcv_data_fin(msk, READ_ONCE(msk->ack_seq), true)) | |
1790 | mptcp_schedule_work(parent); | |
648ef4b8 MM |
1791 | } |
1792 | ||
2a6a870e PA |
1793 | void mptcp_subflow_queue_clean(struct sock *listener_sk, struct sock *listener_ssk) |
1794 | { | |
1795 | struct request_sock_queue *queue = &inet_csk(listener_ssk)->icsk_accept_queue; | |
1796 | struct mptcp_sock *msk, *next, *head = NULL; | |
1797 | struct request_sock *req; | |
63740448 | 1798 | struct sock *sk; |
2a6a870e PA |
1799 | |
1800 | /* build a list of all unaccepted mptcp sockets */ | |
1801 | spin_lock_bh(&queue->rskq_lock); | |
1802 | for (req = queue->rskq_accept_head; req; req = req->dl_next) { | |
1803 | struct mptcp_subflow_context *subflow; | |
1804 | struct sock *ssk = req->sk; | |
1805 | ||
1806 | if (!sk_is_mptcp(ssk)) | |
1807 | continue; | |
1808 | ||
1809 | subflow = mptcp_subflow_ctx(ssk); | |
1810 | if (!subflow || !subflow->conn) | |
1811 | continue; | |
1812 | ||
1813 | /* skip if already in list */ | |
63740448 PA |
1814 | sk = subflow->conn; |
1815 | msk = mptcp_sk(sk); | |
2a6a870e PA |
1816 | if (msk->dl_next || msk == head) |
1817 | continue; | |
1818 | ||
63740448 | 1819 | sock_hold(sk); |
2a6a870e PA |
1820 | msk->dl_next = head; |
1821 | head = msk; | |
1822 | } | |
1823 | spin_unlock_bh(&queue->rskq_lock); | |
1824 | if (!head) | |
1825 | return; | |
1826 | ||
1827 | /* can't acquire the msk socket lock under the subflow one, | |
1828 | * or will cause ABBA deadlock | |
1829 | */ | |
1830 | release_sock(listener_ssk); | |
1831 | ||
1832 | for (msk = head; msk; msk = next) { | |
63740448 | 1833 | sk = (struct sock *)msk; |
2a6a870e PA |
1834 | |
1835 | lock_sock_nested(sk, SINGLE_DEPTH_NESTING); | |
1836 | next = msk->dl_next; | |
1837 | msk->dl_next = NULL; | |
1838 | ||
63740448 | 1839 | __mptcp_unaccepted_force_close(sk); |
2a6a870e PA |
1840 | release_sock(sk); |
1841 | ||
1842 | /* lockdep will report a false positive ABBA deadlock | |
1843 | * between cancel_work_sync and the listener socket. | |
1844 | * The involved locks belong to different sockets WRT | |
1845 | * the existing AB chain. | |
1846 | * Using a per socket key is problematic as key | |
1847 | * deregistration requires process context and must be | |
1848 | * performed at socket disposal time, in atomic | |
1849 | * context. | |
1850 | * Just tell lockdep to consider the listener socket | |
1851 | * released here. | |
1852 | */ | |
1853 | mutex_release(&listener_sk->sk_lock.dep_map, _RET_IP_); | |
1854 | mptcp_cancel_work(sk); | |
1855 | mutex_acquire(&listener_sk->sk_lock.dep_map, 0, 0, _RET_IP_); | |
1856 | ||
1857 | sock_put(sk); | |
1858 | } | |
1859 | ||
1860 | /* we are still under the listener msk socket lock */ | |
1861 | lock_sock_nested(listener_ssk, SINGLE_DEPTH_NESTING); | |
1862 | } | |
1863 | ||
2303f994 PK |
1864 | static int subflow_ulp_init(struct sock *sk) |
1865 | { | |
cec37a6e | 1866 | struct inet_connection_sock *icsk = inet_csk(sk); |
2303f994 PK |
1867 | struct mptcp_subflow_context *ctx; |
1868 | struct tcp_sock *tp = tcp_sk(sk); | |
1869 | int err = 0; | |
1870 | ||
1871 | /* disallow attaching ULP to a socket unless it has been | |
1872 | * created with sock_create_kern() | |
1873 | */ | |
1874 | if (!sk->sk_kern_sock) { | |
1875 | err = -EOPNOTSUPP; | |
1876 | goto out; | |
1877 | } | |
1878 | ||
1879 | ctx = subflow_create_ctx(sk, GFP_KERNEL); | |
1880 | if (!ctx) { | |
1881 | err = -ENOMEM; | |
1882 | goto out; | |
1883 | } | |
1884 | ||
1885 | pr_debug("subflow=%p, family=%d", ctx, sk->sk_family); | |
1886 | ||
1887 | tp->is_mptcp = 1; | |
cec37a6e PK |
1888 | ctx->icsk_af_ops = icsk->icsk_af_ops; |
1889 | icsk->icsk_af_ops = subflow_default_af_ops(sk); | |
648ef4b8 | 1890 | ctx->tcp_state_change = sk->sk_state_change; |
15cc1045 | 1891 | ctx->tcp_error_report = sk->sk_error_report; |
952382c6 FW |
1892 | |
1893 | WARN_ON_ONCE(sk->sk_data_ready != sock_def_readable); | |
1894 | WARN_ON_ONCE(sk->sk_write_space != sk_stream_write_space); | |
1895 | ||
648ef4b8 MM |
1896 | sk->sk_data_ready = subflow_data_ready; |
1897 | sk->sk_write_space = subflow_write_space; | |
1898 | sk->sk_state_change = subflow_state_change; | |
15cc1045 | 1899 | sk->sk_error_report = subflow_error_report; |
2303f994 PK |
1900 | out: |
1901 | return err; | |
1902 | } | |
1903 | ||
e16163b6 | 1904 | static void subflow_ulp_release(struct sock *ssk) |
2303f994 | 1905 | { |
e16163b6 PA |
1906 | struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk); |
1907 | bool release = true; | |
1908 | struct sock *sk; | |
2303f994 PK |
1909 | |
1910 | if (!ctx) | |
1911 | return; | |
1912 | ||
e16163b6 PA |
1913 | sk = ctx->conn; |
1914 | if (sk) { | |
1915 | /* if the msk has been orphaned, keep the ctx | |
0597d0f8 PA |
1916 | * alive, will be freed by __mptcp_close_ssk(), |
1917 | * when the subflow is still unaccepted | |
e16163b6 | 1918 | */ |
0597d0f8 | 1919 | release = ctx->disposable || list_empty(&ctx->node); |
b6985b9b PA |
1920 | |
1921 | /* inet_child_forget() does not call sk_state_change(), | |
1922 | * explicitly trigger the socket close machinery | |
1923 | */ | |
1924 | if (!release && !test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, | |
1925 | &mptcp_sk(sk)->flags)) | |
1926 | mptcp_schedule_work(sk); | |
e16163b6 PA |
1927 | sock_put(sk); |
1928 | } | |
79c0949e | 1929 | |
b19bc294 | 1930 | mptcp_subflow_ops_undo_override(ssk); |
e16163b6 PA |
1931 | if (release) |
1932 | kfree_rcu(ctx, rcu); | |
2303f994 PK |
1933 | } |
1934 | ||
cec37a6e PK |
1935 | static void subflow_ulp_clone(const struct request_sock *req, |
1936 | struct sock *newsk, | |
1937 | const gfp_t priority) | |
1938 | { | |
1939 | struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req); | |
1940 | struct mptcp_subflow_context *old_ctx = mptcp_subflow_ctx(newsk); | |
1941 | struct mptcp_subflow_context *new_ctx; | |
1942 | ||
f296234c PK |
1943 | if (!tcp_rsk(req)->is_mptcp || |
1944 | (!subflow_req->mp_capable && !subflow_req->mp_join)) { | |
648ef4b8 | 1945 | subflow_ulp_fallback(newsk, old_ctx); |
cec37a6e PK |
1946 | return; |
1947 | } | |
1948 | ||
1949 | new_ctx = subflow_create_ctx(newsk, priority); | |
edc7e489 | 1950 | if (!new_ctx) { |
648ef4b8 | 1951 | subflow_ulp_fallback(newsk, old_ctx); |
cec37a6e PK |
1952 | return; |
1953 | } | |
1954 | ||
1955 | new_ctx->conn_finished = 1; | |
1956 | new_ctx->icsk_af_ops = old_ctx->icsk_af_ops; | |
648ef4b8 | 1957 | new_ctx->tcp_state_change = old_ctx->tcp_state_change; |
15cc1045 | 1958 | new_ctx->tcp_error_report = old_ctx->tcp_error_report; |
58b09919 PA |
1959 | new_ctx->rel_write_seq = 1; |
1960 | new_ctx->tcp_sock = newsk; | |
1961 | ||
f296234c PK |
1962 | if (subflow_req->mp_capable) { |
1963 | /* see comments in subflow_syn_recv_sock(), MPTCP connection | |
1964 | * is fully established only after we receive the remote key | |
1965 | */ | |
1966 | new_ctx->mp_capable = 1; | |
f296234c PK |
1967 | new_ctx->local_key = subflow_req->local_key; |
1968 | new_ctx->token = subflow_req->token; | |
1969 | new_ctx->ssn_offset = subflow_req->ssn_offset; | |
1970 | new_ctx->idsn = subflow_req->idsn; | |
4cf86ae8 PA |
1971 | |
1972 | /* this is the first subflow, id is always 0 */ | |
1973 | new_ctx->local_id_valid = 1; | |
f296234c | 1974 | } else if (subflow_req->mp_join) { |
ec3edaa7 | 1975 | new_ctx->ssn_offset = subflow_req->ssn_offset; |
f296234c PK |
1976 | new_ctx->mp_join = 1; |
1977 | new_ctx->fully_established = 1; | |
b3ea6b27 | 1978 | new_ctx->remote_key_valid = 1; |
f296234c | 1979 | new_ctx->backup = subflow_req->backup; |
2ff0e566 | 1980 | new_ctx->remote_id = subflow_req->remote_id; |
f296234c PK |
1981 | new_ctx->token = subflow_req->token; |
1982 | new_ctx->thmac = subflow_req->thmac; | |
4cf86ae8 PA |
1983 | |
1984 | /* the subflow req id is valid, fetched via subflow_check_req() | |
1985 | * and subflow_token_join_request() | |
1986 | */ | |
1987 | subflow_set_local_id(new_ctx, subflow_req->local_id); | |
f296234c | 1988 | } |
cec37a6e PK |
1989 | } |
1990 | ||
b19bc294 PA |
1991 | static void tcp_release_cb_override(struct sock *ssk) |
1992 | { | |
1993 | struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); | |
1994 | ||
1995 | if (mptcp_subflow_has_delegated_action(subflow)) | |
1996 | mptcp_subflow_process_delegated(ssk); | |
1997 | ||
1998 | tcp_release_cb(ssk); | |
1999 | } | |
2000 | ||
2303f994 PK |
2001 | static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = { |
2002 | .name = "mptcp", | |
2003 | .owner = THIS_MODULE, | |
2004 | .init = subflow_ulp_init, | |
2005 | .release = subflow_ulp_release, | |
cec37a6e | 2006 | .clone = subflow_ulp_clone, |
2303f994 PK |
2007 | }; |
2008 | ||
cec37a6e PK |
2009 | static int subflow_ops_init(struct request_sock_ops *subflow_ops) |
2010 | { | |
2011 | subflow_ops->obj_size = sizeof(struct mptcp_subflow_request_sock); | |
cec37a6e PK |
2012 | |
2013 | subflow_ops->slab = kmem_cache_create(subflow_ops->slab_name, | |
2014 | subflow_ops->obj_size, 0, | |
2015 | SLAB_ACCOUNT | | |
2016 | SLAB_TYPESAFE_BY_RCU, | |
2017 | NULL); | |
2018 | if (!subflow_ops->slab) | |
2019 | return -ENOMEM; | |
2020 | ||
2021 | return 0; | |
2022 | } | |
2023 | ||
d39dceca | 2024 | void __init mptcp_subflow_init(void) |
2303f994 | 2025 | { |
34b21d1d MB |
2026 | mptcp_subflow_v4_request_sock_ops = tcp_request_sock_ops; |
2027 | mptcp_subflow_v4_request_sock_ops.slab_name = "request_sock_subflow_v4"; | |
d3295fee MB |
2028 | mptcp_subflow_v4_request_sock_ops.destructor = subflow_v4_req_destructor; |
2029 | ||
34b21d1d MB |
2030 | if (subflow_ops_init(&mptcp_subflow_v4_request_sock_ops) != 0) |
2031 | panic("MPTCP: failed to init subflow v4 request sock ops\n"); | |
cec37a6e PK |
2032 | |
2033 | subflow_request_sock_ipv4_ops = tcp_request_sock_ipv4_ops; | |
7ea851d1 | 2034 | subflow_request_sock_ipv4_ops.route_req = subflow_v4_route_req; |
36b122ba | 2035 | subflow_request_sock_ipv4_ops.send_synack = subflow_v4_send_synack; |
cec37a6e PK |
2036 | |
2037 | subflow_specific = ipv4_specific; | |
2038 | subflow_specific.conn_request = subflow_v4_conn_request; | |
2039 | subflow_specific.syn_recv_sock = subflow_syn_recv_sock; | |
2040 | subflow_specific.sk_rx_dst_set = subflow_finish_connect; | |
4cf86ae8 | 2041 | subflow_specific.rebuild_header = subflow_rebuild_header; |
cec37a6e | 2042 | |
b19bc294 PA |
2043 | tcp_prot_override = tcp_prot; |
2044 | tcp_prot_override.release_cb = tcp_release_cb_override; | |
2045 | ||
cec37a6e | 2046 | #if IS_ENABLED(CONFIG_MPTCP_IPV6) |
34b21d1d MB |
2047 | /* In struct mptcp_subflow_request_sock, we assume the TCP request sock |
2048 | * structures for v4 and v6 have the same size. It should not changed in | |
2049 | * the future but better to make sure to be warned if it is no longer | |
2050 | * the case. | |
2051 | */ | |
2052 | BUILD_BUG_ON(sizeof(struct tcp_request_sock) != sizeof(struct tcp6_request_sock)); | |
2053 | ||
2054 | mptcp_subflow_v6_request_sock_ops = tcp6_request_sock_ops; | |
2055 | mptcp_subflow_v6_request_sock_ops.slab_name = "request_sock_subflow_v6"; | |
d3295fee MB |
2056 | mptcp_subflow_v6_request_sock_ops.destructor = subflow_v6_req_destructor; |
2057 | ||
34b21d1d MB |
2058 | if (subflow_ops_init(&mptcp_subflow_v6_request_sock_ops) != 0) |
2059 | panic("MPTCP: failed to init subflow v6 request sock ops\n"); | |
2060 | ||
cec37a6e | 2061 | subflow_request_sock_ipv6_ops = tcp_request_sock_ipv6_ops; |
7ea851d1 | 2062 | subflow_request_sock_ipv6_ops.route_req = subflow_v6_route_req; |
36b122ba | 2063 | subflow_request_sock_ipv6_ops.send_synack = subflow_v6_send_synack; |
cec37a6e PK |
2064 | |
2065 | subflow_v6_specific = ipv6_specific; | |
2066 | subflow_v6_specific.conn_request = subflow_v6_conn_request; | |
2067 | subflow_v6_specific.syn_recv_sock = subflow_syn_recv_sock; | |
2068 | subflow_v6_specific.sk_rx_dst_set = subflow_finish_connect; | |
4cf86ae8 | 2069 | subflow_v6_specific.rebuild_header = subflow_v6_rebuild_header; |
cec37a6e PK |
2070 | |
2071 | subflow_v6m_specific = subflow_v6_specific; | |
2072 | subflow_v6m_specific.queue_xmit = ipv4_specific.queue_xmit; | |
2073 | subflow_v6m_specific.send_check = ipv4_specific.send_check; | |
2074 | subflow_v6m_specific.net_header_len = ipv4_specific.net_header_len; | |
2075 | subflow_v6m_specific.mtu_reduced = ipv4_specific.mtu_reduced; | |
2076 | subflow_v6m_specific.net_frag_header_len = 0; | |
4cf86ae8 | 2077 | subflow_v6m_specific.rebuild_header = subflow_rebuild_header; |
b19bc294 PA |
2078 | |
2079 | tcpv6_prot_override = tcpv6_prot; | |
2080 | tcpv6_prot_override.release_cb = tcp_release_cb_override; | |
cec37a6e PK |
2081 | #endif |
2082 | ||
5147dfb5 DC |
2083 | mptcp_diag_subflow_init(&subflow_ulp_ops); |
2084 | ||
2303f994 PK |
2085 | if (tcp_register_ulp(&subflow_ulp_ops) != 0) |
2086 | panic("MPTCP: failed to register subflows to ULP\n"); | |
2087 | } |