subflow.c

   1// SPDX-License-Identifier: GPL-2.0
   2/* Multipath TCP
   3 *
   4 * Copyright (c) 2017 - 2019, Intel Corporation.
   5 */
   6
   7#define pr_fmt(fmt) "MPTCP: " fmt
   8
   9#include <linux/kernel.h>
  10#include <linux/module.h>
  11#include <linux/netdevice.h>
  12#include <crypto/sha2.h>
  13#include <crypto/utils.h>
  14#include <net/sock.h>
  15#include <net/inet_common.h>
  16#include <net/inet_hashtables.h>
  17#include <net/protocol.h>
  18#if IS_ENABLED(CONFIG_MPTCP_IPV6)
  19#include <net/ip6_route.h>
  20#include <net/transp_v6.h>
  21#endif
  22#include <net/mptcp.h>
  23
  24#include "protocol.h"
  25#include "mib.h"
  26
  27#include <trace/events/mptcp.h>
  28#include <trace/events/sock.h>
  29
  30static void mptcp_subflow_ops_undo_override(struct sock *ssk);
  31
  32static void SUBFLOW_REQ_INC_STATS(struct request_sock *req,
  33				  enum linux_mptcp_mib_field field)
  34{
  35	MPTCP_INC_STATS(sock_net(req_to_sk(req)), field);
  36}
  37
  38static void subflow_req_destructor(struct request_sock *req)
  39{
  40	struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
  41
  42	pr_debug("subflow_req=%p\n", subflow_req);
  43
  44	if (subflow_req->msk)
  45		sock_put((struct sock *)subflow_req->msk);
  46
  47	mptcp_token_destroy_request(req);
  48}
  49
  50static void subflow_generate_hmac(u64 key1, u64 key2, u32 nonce1, u32 nonce2,
  51				  void *hmac)
  52{
  53	u8 msg[8];
  54
  55	put_unaligned_be32(nonce1, &msg[0]);
  56	put_unaligned_be32(nonce2, &msg[4]);
  57
  58	mptcp_crypto_hmac_sha(key1, key2, msg, 8, hmac);
  59}
  60
  61static bool mptcp_can_accept_new_subflow(const struct mptcp_sock *msk)
  62{
  63	return mptcp_is_fully_established((void *)msk) &&
  64		((mptcp_pm_is_userspace(msk) &&
  65		  mptcp_userspace_pm_active(msk)) ||
  66		 READ_ONCE(msk->pm.accept_subflow));
  67}
  68
  69/* validate received token and create truncated hmac and nonce for SYN-ACK */
  70static void subflow_req_create_thmac(struct mptcp_subflow_request_sock *subflow_req)
  71{
  72	struct mptcp_sock *msk = subflow_req->msk;
  73	u8 hmac[SHA256_DIGEST_SIZE];
  74
  75	get_random_bytes(&subflow_req->local_nonce, sizeof(u32));
  76
  77	subflow_generate_hmac(READ_ONCE(msk->local_key),
  78			      READ_ONCE(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
  85static struct mptcp_sock *subflow_token_join_request(struct request_sock *req)
  86{
  87	struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
  88	struct mptcp_sock *msk;
  89	int local_id;
  90
  91	msk = mptcp_token_get_sock(sock_net(req_to_sk(req)), subflow_req->token);
  92	if (!msk) {
  93		SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINNOTOKEN);
  94		return NULL;
  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);
 100		return NULL;
 101	}
 102	subflow_req->local_id = local_id;
 103	subflow_req->request_bkup = mptcp_pm_is_backup(msk, (struct sock_common *)req);
 104
 105	return msk;
 106}
 107
 108static void subflow_init_req(struct request_sock *req, const struct sock *sk_listener)
 109{
 110	struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
 111
 112	subflow_req->mp_capable = 0;
 113	subflow_req->mp_join = 0;
 114	subflow_req->csum_reqd = mptcp_is_checksum_enabled(sock_net(sk_listener));
 115	subflow_req->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk_listener));
 116	subflow_req->msk = NULL;
 117	mptcp_token_init_request(req);
 118}
 119
 120static bool subflow_use_different_sport(struct mptcp_sock *msk, const struct sock *sk)
 121{
 122	return inet_sk(sk)->inet_sport != inet_sk((struct sock *)msk)->inet_sport;
 123}
 124
 125static void subflow_add_reset_reason(struct sk_buff *skb, u8 reason)
 126{
 127	struct mptcp_ext *mpext = skb_ext_add(skb, SKB_EXT_MPTCP);
 128
 129	if (mpext) {
 130		memset(mpext, 0, sizeof(*mpext));
 131		mpext->reset_reason = reason;
 132	}
 133}
 134
 135static int subflow_reset_req_endp(struct request_sock *req, struct sk_buff *skb)
 136{
 137	SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEENDPATTEMPT);
 138	subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
 139	return -EPERM;
 140}
 141
 142/* Init mptcp request socket.
 143 *
 144 * Returns an error code if a JOIN has failed and a TCP reset
 145 * should be sent.
 146 */
 147static int subflow_check_req(struct request_sock *req,
 148			     const struct sock *sk_listener,
 149			     struct sk_buff *skb)
 150{
 151	struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
 152	struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
 153	struct mptcp_options_received mp_opt;
 154	bool opt_mp_capable, opt_mp_join;
 155
 156	pr_debug("subflow_req=%p, listener=%p\n", subflow_req, listener);
 157
 158#ifdef CONFIG_TCP_MD5SIG
 159	/* no MPTCP if MD5SIG is enabled on this socket or we may run out of
 160	 * TCP option space.
 161	 */
 162	if (rcu_access_pointer(tcp_sk(sk_listener)->md5sig_info)) {
 163		subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
 164		return -EINVAL;
 165	}
 166#endif
 167
 168	mptcp_get_options(skb, &mp_opt);
 169
 170	opt_mp_capable = !!(mp_opt.suboptions & OPTION_MPTCP_MPC_SYN);
 171	opt_mp_join = !!(mp_opt.suboptions & OPTION_MPTCP_MPJ_SYN);
 172	if (opt_mp_capable) {
 173		SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEPASSIVE);
 174
 175		if (unlikely(listener->pm_listener))
 176			return subflow_reset_req_endp(req, skb);
 177		if (opt_mp_join)
 178			return 0;
 179	} else if (opt_mp_join) {
 180		SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINSYNRX);
 181
 182		if (mp_opt.backup)
 183			SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINSYNBACKUPRX);
 184	} else if (unlikely(listener->pm_listener)) {
 185		return subflow_reset_req_endp(req, skb);
 186	}
 187
 188	if (opt_mp_capable && listener->request_mptcp) {
 189		int err, retries = MPTCP_TOKEN_MAX_RETRIES;
 190
 191		subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
 192again:
 193		do {
 194			get_random_bytes(&subflow_req->local_key, sizeof(subflow_req->local_key));
 195		} while (subflow_req->local_key == 0);
 196
 197		if (unlikely(req->syncookie)) {
 198			mptcp_crypto_key_sha(subflow_req->local_key,
 199					     &subflow_req->token,
 200					     &subflow_req->idsn);
 201			if (mptcp_token_exists(subflow_req->token)) {
 202				if (retries-- > 0)
 203					goto again;
 204				SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_TOKENFALLBACKINIT);
 205			} else {
 206				subflow_req->mp_capable = 1;
 207			}
 208			return 0;
 209		}
 210
 211		err = mptcp_token_new_request(req);
 212		if (err == 0)
 213			subflow_req->mp_capable = 1;
 214		else if (retries-- > 0)
 215			goto again;
 216		else
 217			SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_TOKENFALLBACKINIT);
 218
 219	} else if (opt_mp_join && listener->request_mptcp) {
 220		subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
 221		subflow_req->mp_join = 1;
 222		subflow_req->backup = mp_opt.backup;
 223		subflow_req->remote_id = mp_opt.join_id;
 224		subflow_req->token = mp_opt.token;
 225		subflow_req->remote_nonce = mp_opt.nonce;
 226		subflow_req->msk = subflow_token_join_request(req);
 227
 228		/* Can't fall back to TCP in this case. */
 229		if (!subflow_req->msk) {
 230			subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
 231			return -EPERM;
 232		}
 233
 234		if (subflow_use_different_sport(subflow_req->msk, sk_listener)) {
 235			pr_debug("syn inet_sport=%d %d\n",
 236				 ntohs(inet_sk(sk_listener)->inet_sport),
 237				 ntohs(inet_sk((struct sock *)subflow_req->msk)->inet_sport));
 238			if (!mptcp_pm_sport_in_anno_list(subflow_req->msk, sk_listener)) {
 239				SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTSYNRX);
 240				subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
 241				return -EPERM;
 242			}
 243			SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTSYNRX);
 244		}
 245
 246		subflow_req_create_thmac(subflow_req);
 247
 248		if (unlikely(req->syncookie)) {
 249			if (!mptcp_can_accept_new_subflow(subflow_req->msk)) {
 250				SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINREJECTED);
 251				subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
 252				return -EPERM;
 253			}
 254
 255			subflow_init_req_cookie_join_save(subflow_req, skb);
 256		}
 257
 258		pr_debug("token=%u, remote_nonce=%u msk=%p\n", subflow_req->token,
 259			 subflow_req->remote_nonce, subflow_req->msk);
 260	}
 261
 262	return 0;
 263}
 264
 265int mptcp_subflow_init_cookie_req(struct request_sock *req,
 266				  const struct sock *sk_listener,
 267				  struct sk_buff *skb)
 268{
 269	struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
 270	struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
 271	struct mptcp_options_received mp_opt;
 272	bool opt_mp_capable, opt_mp_join;
 273	int err;
 274
 275	subflow_init_req(req, sk_listener);
 276	mptcp_get_options(skb, &mp_opt);
 277
 278	opt_mp_capable = !!(mp_opt.suboptions & OPTION_MPTCP_MPC_ACK);
 279	opt_mp_join = !!(mp_opt.suboptions & OPTION_MPTCP_MPJ_ACK);
 280	if (opt_mp_capable && opt_mp_join)
 281		return -EINVAL;
 282
 283	if (opt_mp_capable && listener->request_mptcp) {
 284		if (mp_opt.sndr_key == 0)
 285			return -EINVAL;
 286
 287		subflow_req->local_key = mp_opt.rcvr_key;
 288		err = mptcp_token_new_request(req);
 289		if (err)
 290			return err;
 291
 292		subflow_req->mp_capable = 1;
 293		subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
 294	} else if (opt_mp_join && listener->request_mptcp) {
 295		if (!mptcp_token_join_cookie_init_state(subflow_req, skb))
 296			return -EINVAL;
 297
 298		subflow_req->mp_join = 1;
 299		subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
 300	}
 301
 302	return 0;
 303}
 304EXPORT_SYMBOL_GPL(mptcp_subflow_init_cookie_req);
 305
 306static enum sk_rst_reason mptcp_get_rst_reason(const struct sk_buff *skb)
 307{
 308	const struct mptcp_ext *mpext = mptcp_get_ext(skb);
 309
 310	if (!mpext)
 311		return SK_RST_REASON_NOT_SPECIFIED;
 312
 313	return sk_rst_convert_mptcp_reason(mpext->reset_reason);
 314}
 315
 316static struct dst_entry *subflow_v4_route_req(const struct sock *sk,
 317					      struct sk_buff *skb,
 318					      struct flowi *fl,
 319					      struct request_sock *req,
 320					      u32 tw_isn)
 321{
 322	struct dst_entry *dst;
 323	int err;
 324
 325	tcp_rsk(req)->is_mptcp = 1;
 326	subflow_init_req(req, sk);
 327
 328	dst = tcp_request_sock_ipv4_ops.route_req(sk, skb, fl, req, tw_isn);
 329	if (!dst)
 330		return NULL;
 331
 332	err = subflow_check_req(req, sk, skb);
 333	if (err == 0)
 334		return dst;
 335
 336	dst_release(dst);
 337	if (!req->syncookie)
 338		tcp_request_sock_ops.send_reset(sk, skb,
 339						mptcp_get_rst_reason(skb));
 340	return NULL;
 341}
 342
 343static void subflow_prep_synack(const struct sock *sk, struct request_sock *req,
 344				struct tcp_fastopen_cookie *foc,
 345				enum tcp_synack_type synack_type)
 346{
 347	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 348	struct inet_request_sock *ireq = inet_rsk(req);
 349
 350	/* clear tstamp_ok, as needed depending on cookie */
 351	if (foc && foc->len > -1)
 352		ireq->tstamp_ok = 0;
 353
 354	if (synack_type == TCP_SYNACK_FASTOPEN)
 355		mptcp_fastopen_subflow_synack_set_params(subflow, req);
 356}
 357
 358static int subflow_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
 359				  struct flowi *fl,
 360				  struct request_sock *req,
 361				  struct tcp_fastopen_cookie *foc,
 362				  enum tcp_synack_type synack_type,
 363				  struct sk_buff *syn_skb)
 364{
 365	subflow_prep_synack(sk, req, foc, synack_type);
 366
 367	return tcp_request_sock_ipv4_ops.send_synack(sk, dst, fl, req, foc,
 368						     synack_type, syn_skb);
 369}
 370
 371#if IS_ENABLED(CONFIG_MPTCP_IPV6)
 372static int subflow_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
 373				  struct flowi *fl,
 374				  struct request_sock *req,
 375				  struct tcp_fastopen_cookie *foc,
 376				  enum tcp_synack_type synack_type,
 377				  struct sk_buff *syn_skb)
 378{
 379	subflow_prep_synack(sk, req, foc, synack_type);
 380
 381	return tcp_request_sock_ipv6_ops.send_synack(sk, dst, fl, req, foc,
 382						     synack_type, syn_skb);
 383}
 384
 385static struct dst_entry *subflow_v6_route_req(const struct sock *sk,
 386					      struct sk_buff *skb,
 387					      struct flowi *fl,
 388					      struct request_sock *req,
 389					      u32 tw_isn)
 390{
 391	struct dst_entry *dst;
 392	int err;
 393
 394	tcp_rsk(req)->is_mptcp = 1;
 395	subflow_init_req(req, sk);
 396
 397	dst = tcp_request_sock_ipv6_ops.route_req(sk, skb, fl, req, tw_isn);
 398	if (!dst)
 399		return NULL;
 400
 401	err = subflow_check_req(req, sk, skb);
 402	if (err == 0)
 403		return dst;
 404
 405	dst_release(dst);
 406	if (!req->syncookie)
 407		tcp6_request_sock_ops.send_reset(sk, skb,
 408						 mptcp_get_rst_reason(skb));
 409	return NULL;
 410}
 411#endif
 412
 413/* validate received truncated hmac and create hmac for third ACK */
 414static bool subflow_thmac_valid(struct mptcp_subflow_context *subflow)
 415{
 416	u8 hmac[SHA256_DIGEST_SIZE];
 417	u64 thmac;
 418
 419	subflow_generate_hmac(subflow->remote_key, subflow->local_key,
 420			      subflow->remote_nonce, subflow->local_nonce,
 421			      hmac);
 422
 423	thmac = get_unaligned_be64(hmac);
 424	pr_debug("subflow=%p, token=%u, thmac=%llu, subflow->thmac=%llu\n",
 425		 subflow, subflow->token, thmac, subflow->thmac);
 426
 427	return thmac == subflow->thmac;
 428}
 429
 430void mptcp_subflow_reset(struct sock *ssk)
 431{
 432	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 433	struct sock *sk = subflow->conn;
 434
 435	/* mptcp_mp_fail_no_response() can reach here on an already closed
 436	 * socket
 437	 */
 438	if (ssk->sk_state == TCP_CLOSE)
 439		return;
 440
 441	/* must hold: tcp_done() could drop last reference on parent */
 442	sock_hold(sk);
 443
 444	mptcp_send_active_reset_reason(ssk);
 445	tcp_done(ssk);
 446	if (!test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &mptcp_sk(sk)->flags))
 447		mptcp_schedule_work(sk);
 448
 449	sock_put(sk);
 450}
 451
 452static bool subflow_use_different_dport(struct mptcp_sock *msk, const struct sock *sk)
 453{
 454	return inet_sk(sk)->inet_dport != inet_sk((struct sock *)msk)->inet_dport;
 455}
 456
 457void __mptcp_sync_state(struct sock *sk, int state)
 458{
 459	struct mptcp_subflow_context *subflow;
 460	struct mptcp_sock *msk = mptcp_sk(sk);
 461	struct sock *ssk = msk->first;
 462
 463	subflow = mptcp_subflow_ctx(ssk);
 464	__mptcp_propagate_sndbuf(sk, ssk);
 465	if (!msk->rcvspace_init)
 466		mptcp_rcv_space_init(msk, ssk);
 467
 468	if (sk->sk_state == TCP_SYN_SENT) {
 469		/* subflow->idsn is always available is TCP_SYN_SENT state,
 470		 * even for the FASTOPEN scenarios
 471		 */
 472		WRITE_ONCE(msk->write_seq, subflow->idsn + 1);
 473		WRITE_ONCE(msk->snd_nxt, msk->write_seq);
 474		mptcp_set_state(sk, state);
 475		sk->sk_state_change(sk);
 476	}
 477}
 478
 479static void subflow_set_remote_key(struct mptcp_sock *msk,
 480				   struct mptcp_subflow_context *subflow,
 481				   const struct mptcp_options_received *mp_opt)
 482{
 483	/* active MPC subflow will reach here multiple times:
 484	 * at subflow_finish_connect() time and at 4th ack time
 485	 */
 486	if (subflow->remote_key_valid)
 487		return;
 488
 489	subflow->remote_key_valid = 1;
 490	subflow->remote_key = mp_opt->sndr_key;
 491	mptcp_crypto_key_sha(subflow->remote_key, NULL, &subflow->iasn);
 492	subflow->iasn++;
 493
 494	/* for fallback's sake */
 495	subflow->map_seq = subflow->iasn;
 496
 497	WRITE_ONCE(msk->remote_key, subflow->remote_key);
 498	WRITE_ONCE(msk->ack_seq, subflow->iasn);
 499	WRITE_ONCE(msk->can_ack, true);
 500	atomic64_set(&msk->rcv_wnd_sent, subflow->iasn);
 501}
 502
 503static void mptcp_propagate_state(struct sock *sk, struct sock *ssk,
 504				  struct mptcp_subflow_context *subflow,
 505				  const struct mptcp_options_received *mp_opt)
 506{
 507	struct mptcp_sock *msk = mptcp_sk(sk);
 508
 509	mptcp_data_lock(sk);
 510	if (mp_opt) {
 511		/* Options are available only in the non fallback cases
 512		 * avoid updating rx path fields otherwise
 513		 */
 514		WRITE_ONCE(msk->snd_una, subflow->idsn + 1);
 515		WRITE_ONCE(msk->wnd_end, subflow->idsn + 1 + tcp_sk(ssk)->snd_wnd);
 516		subflow_set_remote_key(msk, subflow, mp_opt);
 517	}
 518
 519	if (!sock_owned_by_user(sk)) {
 520		__mptcp_sync_state(sk, ssk->sk_state);
 521	} else {
 522		msk->pending_state = ssk->sk_state;
 523		__set_bit(MPTCP_SYNC_STATE, &msk->cb_flags);
 524	}
 525	mptcp_data_unlock(sk);
 526}
 527
 528static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
 529{
 530	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 531	struct mptcp_options_received mp_opt;
 532	struct sock *parent = subflow->conn;
 533	struct mptcp_sock *msk;
 534
 535	subflow->icsk_af_ops->sk_rx_dst_set(sk, skb);
 536
 537	/* be sure no special action on any packet other than syn-ack */
 538	if (subflow->conn_finished)
 539		return;
 540
 541	msk = mptcp_sk(parent);
 542	subflow->rel_write_seq = 1;
 543	subflow->conn_finished = 1;
 544	subflow->ssn_offset = TCP_SKB_CB(skb)->seq;
 545	pr_debug("subflow=%p synack seq=%x\n", subflow, subflow->ssn_offset);
 546
 547	mptcp_get_options(skb, &mp_opt);
 548	if (subflow->request_mptcp) {
 549		if (!(mp_opt.suboptions & OPTION_MPTCP_MPC_SYNACK)) {
 550			if (!mptcp_try_fallback(sk,
 551						MPTCP_MIB_MPCAPABLEACTIVEFALLBACK)) {
 552				MPTCP_INC_STATS(sock_net(sk),
 553						MPTCP_MIB_FALLBACKFAILED);
 554				goto do_reset;
 555			}
 556
 557			goto fallback;
 558		}
 559
 560		if (mp_opt.suboptions & OPTION_MPTCP_CSUMREQD)
 561			WRITE_ONCE(msk->csum_enabled, true);
 562		if (mp_opt.deny_join_id0)
 563			WRITE_ONCE(msk->pm.remote_deny_join_id0, true);
 564		subflow->mp_capable = 1;
 565		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPCAPABLEACTIVEACK);
 566		mptcp_finish_connect(sk);
 567		mptcp_active_enable(parent);
 568		mptcp_propagate_state(parent, sk, subflow, &mp_opt);
 569	} else if (subflow->request_join) {
 570		u8 hmac[SHA256_DIGEST_SIZE];
 571
 572		if (!(mp_opt.suboptions & OPTION_MPTCP_MPJ_SYNACK)) {
 573			subflow->reset_reason = MPTCP_RST_EMPTCP;
 574			goto do_reset;
 575		}
 576
 577		subflow->backup = mp_opt.backup;
 578		subflow->thmac = mp_opt.thmac;
 579		subflow->remote_nonce = mp_opt.nonce;
 580		WRITE_ONCE(subflow->remote_id, mp_opt.join_id);
 581		pr_debug("subflow=%p, thmac=%llu, remote_nonce=%u backup=%d\n",
 582			 subflow, subflow->thmac, subflow->remote_nonce,
 583			 subflow->backup);
 584
 585		if (!subflow_thmac_valid(subflow)) {
 586			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINACKMAC);
 587			subflow->reset_reason = MPTCP_RST_EMPTCP;
 588			goto do_reset;
 589		}
 590
 591		if (!mptcp_finish_join(sk))
 592			goto do_reset;
 593
 594		subflow_generate_hmac(subflow->local_key, subflow->remote_key,
 595				      subflow->local_nonce,
 596				      subflow->remote_nonce,
 597				      hmac);
 598		memcpy(subflow->hmac, hmac, MPTCPOPT_HMAC_LEN);
 599
 600		subflow->mp_join = 1;
 601		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKRX);
 602
 603		if (subflow->backup)
 604			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKBACKUPRX);
 605
 606		if (subflow_use_different_dport(msk, sk)) {
 607			pr_debug("synack inet_dport=%d %d\n",
 608				 ntohs(inet_sk(sk)->inet_dport),
 609				 ntohs(inet_sk(parent)->inet_dport));
 610			MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINPORTSYNACKRX);
 611		}
 612	} else if (mptcp_check_fallback(sk)) {
 613		/* It looks like MPTCP is blocked, while TCP is not */
 614		if (subflow->mpc_drop)
 615			mptcp_active_disable(parent);
 616fallback:
 617		mptcp_propagate_state(parent, sk, subflow, NULL);
 618	}
 619	return;
 620
 621do_reset:
 622	subflow->reset_transient = 0;
 623	mptcp_subflow_reset(sk);
 624}
 625
 626static void subflow_set_local_id(struct mptcp_subflow_context *subflow, int local_id)
 627{
 628	WARN_ON_ONCE(local_id < 0 || local_id > 255);
 629	WRITE_ONCE(subflow->local_id, local_id);
 630}
 631
 632static int subflow_chk_local_id(struct sock *sk)
 633{
 634	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 635	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 636	int err;
 637
 638	if (likely(subflow->local_id >= 0))
 639		return 0;
 640
 641	err = mptcp_pm_get_local_id(msk, (struct sock_common *)sk);
 642	if (err < 0)
 643		return err;
 644
 645	subflow_set_local_id(subflow, err);
 646	subflow->request_bkup = mptcp_pm_is_backup(msk, (struct sock_common *)sk);
 647
 648	return 0;
 649}
 650
 651static int subflow_rebuild_header(struct sock *sk)
 652{
 653	int err = subflow_chk_local_id(sk);
 654
 655	if (unlikely(err < 0))
 656		return err;
 657
 658	return inet_sk_rebuild_header(sk);
 659}
 660
 661#if IS_ENABLED(CONFIG_MPTCP_IPV6)
 662static int subflow_v6_rebuild_header(struct sock *sk)
 663{
 664	int err = subflow_chk_local_id(sk);
 665
 666	if (unlikely(err < 0))
 667		return err;
 668
 669	return inet6_sk_rebuild_header(sk);
 670}
 671#endif
 672
 673static struct request_sock_ops mptcp_subflow_v4_request_sock_ops __ro_after_init;
 674static struct tcp_request_sock_ops subflow_request_sock_ipv4_ops __ro_after_init;
 675
 676static int subflow_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 677{
 678	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 679
 680	pr_debug("subflow=%p\n", subflow);
 681
 682	/* Never answer to SYNs sent to broadcast or multicast */
 683	if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 684		goto drop;
 685
 686	return tcp_conn_request(&mptcp_subflow_v4_request_sock_ops,
 687				&subflow_request_sock_ipv4_ops,
 688				sk, skb);
 689drop:
 690	tcp_listendrop(sk);
 691	return 0;
 692}
 693
 694static void subflow_v4_req_destructor(struct request_sock *req)
 695{
 696	subflow_req_destructor(req);
 697	tcp_request_sock_ops.destructor(req);
 698}
 699
 700#if IS_ENABLED(CONFIG_MPTCP_IPV6)
 701static struct request_sock_ops mptcp_subflow_v6_request_sock_ops __ro_after_init;
 702static struct tcp_request_sock_ops subflow_request_sock_ipv6_ops __ro_after_init;
 703static struct inet_connection_sock_af_ops subflow_v6_specific __ro_after_init;
 704static struct inet_connection_sock_af_ops subflow_v6m_specific __ro_after_init;
 705static struct proto tcpv6_prot_override __ro_after_init;
 706
 707static int subflow_v6_conn_request(struct sock *sk, struct sk_buff *skb)
 708{
 709	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 710
 711	pr_debug("subflow=%p\n", subflow);
 712
 713	if (skb->protocol == htons(ETH_P_IP))
 714		return subflow_v4_conn_request(sk, skb);
 715
 716	if (!ipv6_unicast_destination(skb))
 717		goto drop;
 718
 719	if (ipv6_addr_v4mapped(&ipv6_hdr(skb)->saddr)) {
 720		__IP6_INC_STATS(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS);
 721		return 0;
 722	}
 723
 724	return tcp_conn_request(&mptcp_subflow_v6_request_sock_ops,
 725				&subflow_request_sock_ipv6_ops, sk, skb);
 726
 727drop:
 728	tcp_listendrop(sk);
 729	return 0; /* don't send reset */
 730}
 731
 732static void subflow_v6_req_destructor(struct request_sock *req)
 733{
 734	subflow_req_destructor(req);
 735	tcp6_request_sock_ops.destructor(req);
 736}
 737#endif
 738
 739struct request_sock *mptcp_subflow_reqsk_alloc(const struct request_sock_ops *ops,
 740					       struct sock *sk_listener,
 741					       bool attach_listener)
 742{
 743	if (ops->family == AF_INET)
 744		ops = &mptcp_subflow_v4_request_sock_ops;
 745#if IS_ENABLED(CONFIG_MPTCP_IPV6)
 746	else if (ops->family == AF_INET6)
 747		ops = &mptcp_subflow_v6_request_sock_ops;
 748#endif
 749
 750	return inet_reqsk_alloc(ops, sk_listener, attach_listener);
 751}
 752EXPORT_SYMBOL(mptcp_subflow_reqsk_alloc);
 753
 754/* validate hmac received in third ACK */
 755static bool subflow_hmac_valid(const struct mptcp_subflow_request_sock *subflow_req,
 756			       const struct mptcp_options_received *mp_opt)
 757{
 758	struct mptcp_sock *msk = subflow_req->msk;
 759	u8 hmac[SHA256_DIGEST_SIZE];
 760
 761	subflow_generate_hmac(READ_ONCE(msk->remote_key),
 762			      READ_ONCE(msk->local_key),
 763			      subflow_req->remote_nonce,
 764			      subflow_req->local_nonce, hmac);
 765
 766	return !crypto_memneq(hmac, mp_opt->hmac, MPTCPOPT_HMAC_LEN);
 767}
 768
 769static void subflow_ulp_fallback(struct sock *sk,
 770				 struct mptcp_subflow_context *old_ctx)
 771{
 772	struct inet_connection_sock *icsk = inet_csk(sk);
 773
 774	mptcp_subflow_tcp_fallback(sk, old_ctx);
 775	icsk->icsk_ulp_ops = NULL;
 776	rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
 777	tcp_sk(sk)->is_mptcp = 0;
 778
 779	mptcp_subflow_ops_undo_override(sk);
 780}
 781
 782void mptcp_subflow_drop_ctx(struct sock *ssk)
 783{
 784	struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
 785
 786	if (!ctx)
 787		return;
 788
 789	list_del(&mptcp_subflow_ctx(ssk)->node);
 790	if (inet_csk(ssk)->icsk_ulp_ops) {
 791		subflow_ulp_fallback(ssk, ctx);
 792		if (ctx->conn)
 793			sock_put(ctx->conn);
 794	}
 795
 796	kfree_rcu(ctx, rcu);
 797}
 798
 799void __mptcp_subflow_fully_established(struct mptcp_sock *msk,
 800				       struct mptcp_subflow_context *subflow,
 801				       const struct mptcp_options_received *mp_opt)
 802{
 803	subflow_set_remote_key(msk, subflow, mp_opt);
 804	WRITE_ONCE(subflow->fully_established, true);
 805	WRITE_ONCE(msk->fully_established, true);
 806}
 807
 808static struct sock *subflow_syn_recv_sock(const struct sock *sk,
 809					  struct sk_buff *skb,
 810					  struct request_sock *req,
 811					  struct dst_entry *dst,
 812					  struct request_sock *req_unhash,
 813					  bool *own_req,
 814					  void (*opt_child_init)(struct sock *newsk,
 815								 const struct sock *sk))
 816{
 817	struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk);
 818	struct mptcp_subflow_request_sock *subflow_req;
 819	struct mptcp_options_received mp_opt;
 820	bool fallback, fallback_is_fatal;
 821	enum sk_rst_reason reason;
 822	struct mptcp_sock *owner;
 823	struct sock *child;
 824
 825	pr_debug("listener=%p, req=%p, conn=%p\n", listener, req, listener->conn);
 826
 827	/* After child creation we must look for MPC even when options
 828	 * are not parsed
 829	 */
 830	mp_opt.suboptions = 0;
 831
 832	/* hopefully temporary handling for MP_JOIN+syncookie */
 833	subflow_req = mptcp_subflow_rsk(req);
 834	fallback_is_fatal = tcp_rsk(req)->is_mptcp && subflow_req->mp_join;
 835	fallback = !tcp_rsk(req)->is_mptcp;
 836	if (fallback)
 837		goto create_child;
 838
 839	/* if the sk is MP_CAPABLE, we try to fetch the client key */
 840	if (subflow_req->mp_capable) {
 841		/* we can receive and accept an in-window, out-of-order pkt,
 842		 * which may not carry the MP_CAPABLE opt even on mptcp enabled
 843		 * paths: always try to extract the peer key, and fallback
 844		 * for packets missing it.
 845		 * Even OoO DSS packets coming legitly after dropped or
 846		 * reordered MPC will cause fallback, but we don't have other
 847		 * options.
 848		 */
 849		mptcp_get_options(skb, &mp_opt);
 850		if (!(mp_opt.suboptions &
 851		      (OPTION_MPTCP_MPC_SYN | OPTION_MPTCP_MPC_ACK)))
 852			fallback = true;
 853
 854	} else if (subflow_req->mp_join) {
 855		mptcp_get_options(skb, &mp_opt);
 856		if (!(mp_opt.suboptions & OPTION_MPTCP_MPJ_ACK))
 857			fallback = true;
 858	}
 859
 860create_child:
 861	child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
 862						     req_unhash, own_req, opt_child_init);
 863
 864	if (child && *own_req) {
 865		struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(child);
 866
 867		tcp_rsk(req)->drop_req = false;
 868
 869		/* we need to fallback on ctx allocation failure and on pre-reqs
 870		 * checking above. In the latter scenario we additionally need
 871		 * to reset the context to non MPTCP status.
 872		 */
 873		if (!ctx || fallback) {
 874			if (fallback_is_fatal) {
 875				subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
 876				goto dispose_child;
 877			}
 878			goto fallback;
 879		}
 880
 881		/* ssk inherits options of listener sk */
 882		ctx->setsockopt_seq = listener->setsockopt_seq;
 883
 884		if (ctx->mp_capable) {
 885			ctx->conn = mptcp_sk_clone_init(listener->conn, &mp_opt, child, req);
 886			if (!ctx->conn)
 887				goto fallback;
 888
 889			ctx->subflow_id = 1;
 890			owner = mptcp_sk(ctx->conn);
 891
 892			if (mp_opt.deny_join_id0)
 893				WRITE_ONCE(owner->pm.remote_deny_join_id0, true);
 894
 895			mptcp_pm_new_connection(owner, child, 1);
 896
 897			/* with OoO packets we can reach here without ingress
 898			 * mpc option
 899			 */
 900			if (mp_opt.suboptions & OPTION_MPTCP_MPC_ACK) {
 901				mptcp_pm_fully_established(owner, child);
 902				ctx->pm_notified = 1;
 903			}
 904		} else if (ctx->mp_join) {
 905			owner = subflow_req->msk;
 906			if (!owner) {
 907				subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
 908				goto dispose_child;
 909			}
 910
 911			if (!subflow_hmac_valid(subflow_req, &mp_opt)) {
 912				SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKMAC);
 913				subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
 914				goto dispose_child;
 915			}
 916
 917			if (!mptcp_can_accept_new_subflow(owner)) {
 918				SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINREJECTED);
 919				subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
 920				goto dispose_child;
 921			}
 922
 923			/* move the msk reference ownership to the subflow */
 924			subflow_req->msk = NULL;
 925			ctx->conn = (struct sock *)owner;
 926
 927			if (subflow_use_different_sport(owner, sk)) {
 928				pr_debug("ack inet_sport=%d %d\n",
 929					 ntohs(inet_sk(sk)->inet_sport),
 930					 ntohs(inet_sk((struct sock *)owner)->inet_sport));
 931				if (!mptcp_pm_sport_in_anno_list(owner, sk)) {
 932					SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTACKRX);
 933					subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
 934					goto dispose_child;
 935				}
 936				SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTACKRX);
 937			}
 938
 939			if (!mptcp_finish_join(child)) {
 940				struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(child);
 941
 942				subflow_add_reset_reason(skb, subflow->reset_reason);
 943				goto dispose_child;
 944			}
 945
 946			SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX);
 947			tcp_rsk(req)->drop_req = true;
 948		}
 949	}
 950
 951	/* check for expected invariant - should never trigger, just help
 952	 * catching earlier subtle bugs
 953	 */
 954	WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp &&
 955		     (!mptcp_subflow_ctx(child) ||
 956		      !mptcp_subflow_ctx(child)->conn));
 957	return child;
 958
 959dispose_child:
 960	mptcp_subflow_drop_ctx(child);
 961	tcp_rsk(req)->drop_req = true;
 962	inet_csk_prepare_for_destroy_sock(child);
 963	tcp_done(child);
 964	reason = mptcp_get_rst_reason(skb);
 965	req->rsk_ops->send_reset(sk, skb, reason);
 966
 967	/* The last child reference will be released by the caller */
 968	return child;
 969
 970fallback:
 971	if (fallback)
 972		SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEPASSIVEFALLBACK);
 973	mptcp_subflow_drop_ctx(child);
 974	return child;
 975}
 976
 977static struct inet_connection_sock_af_ops subflow_specific __ro_after_init;
 978static struct proto tcp_prot_override __ro_after_init;
 979
 980enum mapping_status {
 981	MAPPING_OK,
 982	MAPPING_INVALID,
 983	MAPPING_EMPTY,
 984	MAPPING_DATA_FIN,
 985	MAPPING_DUMMY,
 986	MAPPING_BAD_CSUM,
 987	MAPPING_NODSS
 988};
 989
 990static void dbg_bad_map(struct mptcp_subflow_context *subflow, u32 ssn)
 991{
 992	pr_debug("Bad mapping: ssn=%d map_seq=%d map_data_len=%d\n",
 993		 ssn, subflow->map_subflow_seq, subflow->map_data_len);
 994}
 995
 996static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb)
 997{
 998	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 999	unsigned int skb_consumed;
1000
1001	skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq;
1002	if (unlikely(skb_consumed >= skb->len)) {
1003		DEBUG_NET_WARN_ON_ONCE(1);
1004		return true;
1005	}
1006
1007	return skb->len - skb_consumed <= subflow->map_data_len -
1008					  mptcp_subflow_get_map_offset(subflow);
1009}
1010
1011static bool validate_mapping(struct sock *ssk, struct sk_buff *skb)
1012{
1013	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1014	u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
1015
1016	if (unlikely(before(ssn, subflow->map_subflow_seq))) {
1017		/* Mapping covers data later in the subflow stream,
1018		 * currently unsupported.
1019		 */
1020		dbg_bad_map(subflow, ssn);
1021		return false;
1022	}
1023	if (unlikely(!before(ssn, subflow->map_subflow_seq +
1024				  subflow->map_data_len))) {
1025		/* Mapping does covers past subflow data, invalid */
1026		dbg_bad_map(subflow, ssn);
1027		return false;
1028	}
1029	return true;
1030}
1031
1032static enum mapping_status validate_data_csum(struct sock *ssk, struct sk_buff *skb,
1033					      bool csum_reqd)
1034{
1035	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1036	u32 offset, seq, delta;
1037	__sum16 csum;
1038	int len;
1039
1040	if (!csum_reqd)
1041		return MAPPING_OK;
1042
1043	/* mapping already validated on previous traversal */
1044	if (subflow->map_csum_len == subflow->map_data_len)
1045		return MAPPING_OK;
1046
1047	/* traverse the receive queue, ensuring it contains a full
1048	 * DSS mapping and accumulating the related csum.
1049	 * Preserve the accoumlate csum across multiple calls, to compute
1050	 * the csum only once
1051	 */
1052	delta = subflow->map_data_len - subflow->map_csum_len;
1053	for (;;) {
1054		seq = tcp_sk(ssk)->copied_seq + subflow->map_csum_len;
1055		offset = seq - TCP_SKB_CB(skb)->seq;
1056
1057		/* if the current skb has not been accounted yet, csum its contents
1058		 * up to the amount covered by the current DSS
1059		 */
1060		if (offset < skb->len) {
1061			__wsum csum;
1062
1063			len = min(skb->len - offset, delta);
1064			csum = skb_checksum(skb, offset, len, 0);
1065			subflow->map_data_csum = csum_block_add(subflow->map_data_csum, csum,
1066								subflow->map_csum_len);
1067
1068			delta -= len;
1069			subflow->map_csum_len += len;
1070		}
1071		if (delta == 0)
1072			break;
1073
1074		if (skb_queue_is_last(&ssk->sk_receive_queue, skb)) {
1075			/* if this subflow is closed, the partial mapping
1076			 * will be never completed; flush the pending skbs, so
1077			 * that subflow_sched_work_if_closed() can kick in
1078			 */
1079			if (unlikely(ssk->sk_state == TCP_CLOSE))
1080				while ((skb = skb_peek(&ssk->sk_receive_queue)))
1081					sk_eat_skb(ssk, skb);
1082
1083			/* not enough data to validate the csum */
1084			return MAPPING_EMPTY;
1085		}
1086
1087		/* the DSS mapping for next skbs will be validated later,
1088		 * when a get_mapping_status call will process such skb
1089		 */
1090		skb = skb->next;
1091	}
1092
1093	/* note that 'map_data_len' accounts only for the carried data, does
1094	 * not include the eventual seq increment due to the data fin,
1095	 * while the pseudo header requires the original DSS data len,
1096	 * including that
1097	 */
1098	csum = __mptcp_make_csum(subflow->map_seq,
1099				 subflow->map_subflow_seq,
1100				 subflow->map_data_len + subflow->map_data_fin,
1101				 subflow->map_data_csum);
1102	if (unlikely(csum)) {
1103		MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DATACSUMERR);
1104		return MAPPING_BAD_CSUM;
1105	}
1106
1107	subflow->valid_csum_seen = 1;
1108	return MAPPING_OK;
1109}
1110
1111static enum mapping_status get_mapping_status(struct sock *ssk,
1112					      struct mptcp_sock *msk)
1113{
1114	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1115	bool csum_reqd = READ_ONCE(msk->csum_enabled);
1116	struct mptcp_ext *mpext;
1117	struct sk_buff *skb;
1118	u16 data_len;
1119	u64 map_seq;
1120
1121	skb = skb_peek(&ssk->sk_receive_queue);
1122	if (!skb)
1123		return MAPPING_EMPTY;
1124
1125	if (mptcp_check_fallback(ssk))
1126		return MAPPING_DUMMY;
1127
1128	mpext = mptcp_get_ext(skb);
1129	if (!mpext || !mpext->use_map) {
1130		if (!subflow->map_valid && !skb->len) {
1131			/* the TCP stack deliver 0 len FIN pkt to the receive
1132			 * queue, that is the only 0len pkts ever expected here,
1133			 * and we can admit no mapping only for 0 len pkts
1134			 */
1135			if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
1136				WARN_ONCE(1, "0len seq %d:%d flags %x",
1137					  TCP_SKB_CB(skb)->seq,
1138					  TCP_SKB_CB(skb)->end_seq,
1139					  TCP_SKB_CB(skb)->tcp_flags);
1140			sk_eat_skb(ssk, skb);
1141			return MAPPING_EMPTY;
1142		}
1143
1144		/* If the required DSS has likely been dropped by a middlebox */
1145		if (!subflow->map_valid)
1146			return MAPPING_NODSS;
1147
1148		goto validate_seq;
1149	}
1150
1151	trace_get_mapping_status(mpext);
1152
1153	data_len = mpext->data_len;
1154	if (data_len == 0) {
1155		pr_debug("infinite mapping received\n");
1156		MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
1157		return MAPPING_INVALID;
1158	}
1159
1160	if (mpext->data_fin == 1) {
1161		u64 data_fin_seq;
1162
1163		if (data_len == 1) {
1164			bool updated = mptcp_update_rcv_data_fin(msk, mpext->data_seq,
1165								 mpext->dsn64);
1166			pr_debug("DATA_FIN with no payload seq=%llu\n", mpext->data_seq);
1167			if (subflow->map_valid) {
1168				/* A DATA_FIN might arrive in a DSS
1169				 * option before the previous mapping
1170				 * has been fully consumed. Continue
1171				 * handling the existing mapping.
1172				 */
1173				skb_ext_del(skb, SKB_EXT_MPTCP);
1174				return MAPPING_OK;
1175			}
1176
1177			if (updated)
1178				mptcp_schedule_work((struct sock *)msk);
1179
1180			return MAPPING_DATA_FIN;
1181		}
1182
1183		data_fin_seq = mpext->data_seq + data_len - 1;
1184
1185		/* If mpext->data_seq is a 32-bit value, data_fin_seq must also
1186		 * be limited to 32 bits.
1187		 */
1188		if (!mpext->dsn64)
1189			data_fin_seq &= GENMASK_ULL(31, 0);
1190
1191		mptcp_update_rcv_data_fin(msk, data_fin_seq, mpext->dsn64);
1192		pr_debug("DATA_FIN with mapping seq=%llu dsn64=%d\n",
1193			 data_fin_seq, mpext->dsn64);
1194
1195		/* Adjust for DATA_FIN using 1 byte of sequence space */
1196		data_len--;
1197	}
1198
1199	map_seq = mptcp_expand_seq(READ_ONCE(msk->ack_seq), mpext->data_seq, mpext->dsn64);
1200	WRITE_ONCE(mptcp_sk(subflow->conn)->use_64bit_ack, !!mpext->dsn64);
1201
1202	if (subflow->map_valid) {
1203		/* Allow replacing only with an identical map */
1204		if (subflow->map_seq == map_seq &&
1205		    subflow->map_subflow_seq == mpext->subflow_seq &&
1206		    subflow->map_data_len == data_len &&
1207		    subflow->map_csum_reqd == mpext->csum_reqd) {
1208			skb_ext_del(skb, SKB_EXT_MPTCP);
1209			goto validate_csum;
1210		}
1211
1212		/* If this skb data are fully covered by the current mapping,
1213		 * the new map would need caching, which is not supported
1214		 */
1215		if (skb_is_fully_mapped(ssk, skb)) {
1216			MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH);
1217			return MAPPING_INVALID;
1218		}
1219
1220		/* will validate the next map after consuming the current one */
1221		goto validate_csum;
1222	}
1223
1224	subflow->map_seq = map_seq;
1225	subflow->map_subflow_seq = mpext->subflow_seq;
1226	subflow->map_data_len = data_len;
1227	subflow->map_valid = 1;
1228	subflow->map_data_fin = mpext->data_fin;
1229	subflow->mpc_map = mpext->mpc_map;
1230	subflow->map_csum_reqd = mpext->csum_reqd;
1231	subflow->map_csum_len = 0;
1232	subflow->map_data_csum = csum_unfold(mpext->csum);
1233
1234	/* Cfr RFC 8684 Section 3.3.0 */
1235	if (unlikely(subflow->map_csum_reqd != csum_reqd))
1236		return MAPPING_INVALID;
1237
1238	pr_debug("new map seq=%llu subflow_seq=%u data_len=%u csum=%d:%u\n",
1239		 subflow->map_seq, subflow->map_subflow_seq,
1240		 subflow->map_data_len, subflow->map_csum_reqd,
1241		 subflow->map_data_csum);
1242
1243validate_seq:
1244	/* we revalidate valid mapping on new skb, because we must ensure
1245	 * the current skb is completely covered by the available mapping
1246	 */
1247	if (!validate_mapping(ssk, skb)) {
1248		MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSTCPMISMATCH);
1249		return MAPPING_INVALID;
1250	}
1251
1252	skb_ext_del(skb, SKB_EXT_MPTCP);
1253
1254validate_csum:
1255	return validate_data_csum(ssk, skb, csum_reqd);
1256}
1257
1258static void mptcp_subflow_discard_data(struct sock *ssk, struct sk_buff *skb,
1259				       u64 limit)
1260{
1261	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1262	bool fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
1263	struct tcp_sock *tp = tcp_sk(ssk);
1264	u32 offset, incr, avail_len;
1265
1266	offset = tp->copied_seq - TCP_SKB_CB(skb)->seq;
1267	if (WARN_ON_ONCE(offset > skb->len))
1268		goto out;
1269
1270	avail_len = skb->len - offset;
1271	incr = limit >= avail_len ? avail_len + fin : limit;
1272
1273	pr_debug("discarding=%d len=%d offset=%d seq=%d\n", incr, skb->len,
1274		 offset, subflow->map_subflow_seq);
1275	MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DUPDATA);
1276	tcp_sk(ssk)->copied_seq += incr;
1277
1278out:
1279	if (!before(tcp_sk(ssk)->copied_seq, TCP_SKB_CB(skb)->end_seq))
1280		sk_eat_skb(ssk, skb);
1281	if (mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len)
1282		subflow->map_valid = 0;
1283}
1284
1285static bool subflow_is_done(const struct sock *sk)
1286{
1287	return sk->sk_shutdown & RCV_SHUTDOWN || sk->sk_state == TCP_CLOSE;
1288}
1289
1290/* sched mptcp worker for subflow cleanup if no more data is pending */
1291static void subflow_sched_work_if_closed(struct mptcp_sock *msk, struct sock *ssk)
1292{
1293	const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1294	struct sock *sk = (struct sock *)msk;
1295
1296	if (likely(ssk->sk_state != TCP_CLOSE &&
1297		   (ssk->sk_state != TCP_CLOSE_WAIT ||
1298		    inet_sk_state_load(sk) != TCP_ESTABLISHED)))
1299		return;
1300
1301	if (!skb_queue_empty(&ssk->sk_receive_queue))
1302		return;
1303
1304	if (!test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags))
1305		mptcp_schedule_work(sk);
1306
1307	/* when the fallback subflow closes the rx side, trigger a 'dummy'
1308	 * ingress data fin, so that the msk state will follow along
1309	 */
1310	if (__mptcp_check_fallback(msk) && subflow_is_done(ssk) &&
1311	    msk->first == ssk &&
1312	    mptcp_update_rcv_data_fin(msk, subflow->map_seq +
1313				      subflow->map_data_len, true))
1314		mptcp_schedule_work(sk);
1315}
1316
1317static bool mptcp_subflow_fail(struct mptcp_sock *msk, struct sock *ssk)
1318{
1319	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1320	unsigned long fail_tout;
1321
1322	/* we are really failing, prevent any later subflow join */
1323	spin_lock_bh(&msk->fallback_lock);
1324	if (!msk->allow_infinite_fallback) {
1325		spin_unlock_bh(&msk->fallback_lock);
1326		return false;
1327	}
1328	msk->allow_subflows = false;
1329	spin_unlock_bh(&msk->fallback_lock);
1330
1331	/* graceful failure can happen only on the MPC subflow */
1332	if (WARN_ON_ONCE(ssk != READ_ONCE(msk->first)))
1333		return false;
1334
1335	/* since the close timeout take precedence on the fail one,
1336	 * no need to start the latter when the first is already set
1337	 */
1338	if (sock_flag((struct sock *)msk, SOCK_DEAD))
1339		return true;
1340
1341	/* we don't need extreme accuracy here, use a zero fail_tout as special
1342	 * value meaning no fail timeout at all;
1343	 */
1344	fail_tout = jiffies + TCP_RTO_MAX;
1345	if (!fail_tout)
1346		fail_tout = 1;
1347	WRITE_ONCE(subflow->fail_tout, fail_tout);
1348	tcp_send_ack(ssk);
1349
1350	mptcp_reset_tout_timer(msk, subflow->fail_tout);
1351	return true;
1352}
1353
1354static bool subflow_check_data_avail(struct sock *ssk)
1355{
1356	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1357	enum mapping_status status;
1358	struct mptcp_sock *msk;
1359	struct sk_buff *skb;
1360
1361	if (!skb_peek(&ssk->sk_receive_queue))
1362		WRITE_ONCE(subflow->data_avail, false);
1363	if (subflow->data_avail)
1364		return true;
1365
1366	msk = mptcp_sk(subflow->conn);
1367	for (;;) {
1368		u64 ack_seq;
1369		u64 old_ack;
1370
1371		status = get_mapping_status(ssk, msk);
1372		trace_subflow_check_data_avail(status, skb_peek(&ssk->sk_receive_queue));
1373		if (unlikely(status == MAPPING_INVALID || status == MAPPING_DUMMY ||
1374			     status == MAPPING_BAD_CSUM || status == MAPPING_NODSS))
1375			goto fallback;
1376
1377		if (status != MAPPING_OK)
1378			goto no_data;
1379
1380		skb = skb_peek(&ssk->sk_receive_queue);
1381		if (WARN_ON_ONCE(!skb))
1382			goto no_data;
1383
1384		if (unlikely(!READ_ONCE(msk->can_ack)))
1385			goto fallback;
1386
1387		old_ack = READ_ONCE(msk->ack_seq);
1388		ack_seq = mptcp_subflow_get_mapped_dsn(subflow);
1389		pr_debug("msk ack_seq=%llx subflow ack_seq=%llx\n", old_ack,
1390			 ack_seq);
1391		if (unlikely(before64(ack_seq, old_ack))) {
1392			mptcp_subflow_discard_data(ssk, skb, old_ack - ack_seq);
1393			continue;
1394		}
1395
1396		WRITE_ONCE(subflow->data_avail, true);
1397		break;
1398	}
1399	return true;
1400
1401no_data:
1402	subflow_sched_work_if_closed(msk, ssk);
1403	return false;
1404
1405fallback:
1406	if (!__mptcp_check_fallback(msk)) {
1407		/* RFC 8684 section 3.7. */
1408		if (status == MAPPING_BAD_CSUM &&
1409		    (subflow->mp_join || subflow->valid_csum_seen)) {
1410			subflow->send_mp_fail = 1;
1411
1412			if (!mptcp_subflow_fail(msk, ssk)) {
1413				subflow->reset_transient = 0;
1414				subflow->reset_reason = MPTCP_RST_EMIDDLEBOX;
1415				goto reset;
1416			}
1417			WRITE_ONCE(subflow->data_avail, true);
1418			return true;
1419		}
1420
1421		if (!mptcp_try_fallback(ssk, MPTCP_MIB_DSSFALLBACK)) {
1422			/* fatal protocol error, close the socket.
1423			 * subflow_error_report() will introduce the appropriate barriers
1424			 */
1425			subflow->reset_transient = 0;
1426			subflow->reset_reason = status == MAPPING_NODSS ?
1427						MPTCP_RST_EMIDDLEBOX :
1428						MPTCP_RST_EMPTCP;
1429
1430reset:
1431			WRITE_ONCE(ssk->sk_err, EBADMSG);
1432			tcp_set_state(ssk, TCP_CLOSE);
1433			while ((skb = skb_peek(&ssk->sk_receive_queue)))
1434				sk_eat_skb(ssk, skb);
1435			mptcp_send_active_reset_reason(ssk);
1436			WRITE_ONCE(subflow->data_avail, false);
1437			return false;
1438		}
1439	}
1440
1441	skb = skb_peek(&ssk->sk_receive_queue);
1442	subflow->map_valid = 1;
1443	subflow->map_data_len = skb->len;
1444	subflow->map_subflow_seq = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
1445	subflow->map_seq = __mptcp_expand_seq(subflow->map_seq,
1446					      subflow->iasn +
1447					      TCP_SKB_CB(skb)->seq -
1448					      subflow->ssn_offset - 1);
1449	WRITE_ONCE(subflow->data_avail, true);
1450	return true;
1451}
1452
1453bool mptcp_subflow_data_available(struct sock *sk)
1454{
1455	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1456
1457	/* check if current mapping is still valid */
1458	if (subflow->map_valid &&
1459	    mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) {
1460		subflow->map_valid = 0;
1461		WRITE_ONCE(subflow->data_avail, false);
1462
1463		pr_debug("Done with mapping: seq=%u data_len=%u\n",
1464			 subflow->map_subflow_seq,
1465			 subflow->map_data_len);
1466	}
1467
1468	return subflow_check_data_avail(sk);
1469}
1470
1471/* If ssk has an mptcp parent socket, use the mptcp rcvbuf occupancy,
1472 * not the ssk one.
1473 *
1474 * In mptcp, rwin is about the mptcp-level connection data.
1475 *
1476 * Data that is still on the ssk rx queue can thus be ignored,
1477 * as far as mptcp peer is concerned that data is still inflight.
1478 * DSS ACK is updated when skb is moved to the mptcp rx queue.
1479 */
1480void mptcp_space(const struct sock *ssk, int *space, int *full_space)
1481{
1482	const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1483	const struct sock *sk = subflow->conn;
1484
1485	*space = __mptcp_space(sk);
1486	*full_space = mptcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf));
1487}
1488
1489static void subflow_error_report(struct sock *ssk)
1490{
1491	struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1492
1493	/* bail early if this is a no-op, so that we avoid introducing a
1494	 * problematic lockdep dependency between TCP accept queue lock
1495	 * and msk socket spinlock
1496	 */
1497	if (!sk->sk_socket)
1498		return;
1499
1500	mptcp_data_lock(sk);
1501	if (!sock_owned_by_user(sk))
1502		__mptcp_error_report(sk);
1503	else
1504		__set_bit(MPTCP_ERROR_REPORT,  &mptcp_sk(sk)->cb_flags);
1505	mptcp_data_unlock(sk);
1506}
1507
1508static void subflow_data_ready(struct sock *sk)
1509{
1510	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1511	u16 state = 1 << inet_sk_state_load(sk);
1512	struct sock *parent = subflow->conn;
1513	struct mptcp_sock *msk;
1514
1515	trace_sk_data_ready(sk);
1516
1517	msk = mptcp_sk(parent);
1518	if (state & TCPF_LISTEN) {
1519		/* MPJ subflow are removed from accept queue before reaching here,
1520		 * avoid stray wakeups
1521		 */
1522		if (reqsk_queue_empty(&inet_csk(sk)->icsk_accept_queue))
1523			return;
1524
1525		parent->sk_data_ready(parent);
1526		return;
1527	}
1528
1529	WARN_ON_ONCE(!__mptcp_check_fallback(msk) && !subflow->mp_capable &&
1530		     !subflow->mp_join && !(state & TCPF_CLOSE));
1531
1532	if (mptcp_subflow_data_available(sk)) {
1533		mptcp_data_ready(parent, sk);
1534
1535		/* subflow-level lowat test are not relevant.
1536		 * respect the msk-level threshold eventually mandating an immediate ack
1537		 */
1538		if (mptcp_data_avail(msk) < parent->sk_rcvlowat &&
1539		    (tcp_sk(sk)->rcv_nxt - tcp_sk(sk)->rcv_wup) > inet_csk(sk)->icsk_ack.rcv_mss)
1540			inet_csk(sk)->icsk_ack.pending |= ICSK_ACK_NOW;
1541	} else if (unlikely(sk->sk_err)) {
1542		subflow_error_report(sk);
1543	}
1544}
1545
1546static void subflow_write_space(struct sock *ssk)
1547{
1548	struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1549
1550	mptcp_propagate_sndbuf(sk, ssk);
1551	mptcp_write_space(sk);
1552}
1553
1554static const struct inet_connection_sock_af_ops *
1555subflow_default_af_ops(struct sock *sk)
1556{
1557#if IS_ENABLED(CONFIG_MPTCP_IPV6)
1558	if (sk->sk_family == AF_INET6)
1559		return &subflow_v6_specific;
1560#endif
1561	return &subflow_specific;
1562}
1563
1564#if IS_ENABLED(CONFIG_MPTCP_IPV6)
1565void mptcpv6_handle_mapped(struct sock *sk, bool mapped)
1566{
1567	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1568	struct inet_connection_sock *icsk = inet_csk(sk);
1569	const struct inet_connection_sock_af_ops *target;
1570
1571	target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk);
1572
1573	pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d\n",
1574		 subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped);
1575
1576	if (likely(icsk->icsk_af_ops == target))
1577		return;
1578
1579	subflow->icsk_af_ops = icsk->icsk_af_ops;
1580	icsk->icsk_af_ops = target;
1581}
1582#endif
1583
1584void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
1585			 struct sockaddr_storage *addr,
1586			 unsigned short family)
1587{
1588	memset(addr, 0, sizeof(*addr));
1589	addr->ss_family = family;
1590	if (addr->ss_family == AF_INET) {
1591		struct sockaddr_in *in_addr = (struct sockaddr_in *)addr;
1592
1593		if (info->family == AF_INET)
1594			in_addr->sin_addr = info->addr;
1595#if IS_ENABLED(CONFIG_MPTCP_IPV6)
1596		else if (ipv6_addr_v4mapped(&info->addr6))
1597			in_addr->sin_addr.s_addr = info->addr6.s6_addr32[3];
1598#endif
1599		in_addr->sin_port = info->port;
1600	}
1601#if IS_ENABLED(CONFIG_MPTCP_IPV6)
1602	else if (addr->ss_family == AF_INET6) {
1603		struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr;
1604
1605		if (info->family == AF_INET)
1606			ipv6_addr_set_v4mapped(info->addr.s_addr,
1607					       &in6_addr->sin6_addr);
1608		else
1609			in6_addr->sin6_addr = info->addr6;
1610		in6_addr->sin6_port = info->port;
1611	}
1612#endif
1613}
1614
1615int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_pm_local *local,
1616			    const struct mptcp_addr_info *remote)
1617{
1618	struct mptcp_sock *msk = mptcp_sk(sk);
1619	struct mptcp_subflow_context *subflow;
1620	int local_id = local->addr.id;
1621	struct sockaddr_storage addr;
1622	int remote_id = remote->id;
1623	int err = -ENOTCONN;
1624	struct socket *sf;
1625	struct sock *ssk;
1626	u32 remote_token;
1627	int addrlen;
1628
1629	/* The userspace PM sent the request too early? */
1630	if (!mptcp_is_fully_established(sk))
1631		goto err_out;
1632
1633	err = mptcp_subflow_create_socket(sk, local->addr.family, &sf);
1634	if (err) {
1635		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNTXCREATSKERR);
1636		pr_debug("msk=%p local=%d remote=%d create sock error: %d\n",
1637			 msk, local_id, remote_id, err);
1638		goto err_out;
1639	}
1640
1641	ssk = sf->sk;
1642	subflow = mptcp_subflow_ctx(ssk);
1643	do {
1644		get_random_bytes(&subflow->local_nonce, sizeof(u32));
1645	} while (!subflow->local_nonce);
1646
1647	/* if 'IPADDRANY', the ID will be set later, after the routing */
1648	if (local->addr.family == AF_INET) {
1649		if (!local->addr.addr.s_addr)
1650			local_id = -1;
1651#if IS_ENABLED(CONFIG_MPTCP_IPV6)
1652	} else if (sk->sk_family == AF_INET6) {
1653		if (ipv6_addr_any(&local->addr.addr6))
1654			local_id = -1;
1655#endif
1656	}
1657
1658	if (local_id >= 0)
1659		subflow_set_local_id(subflow, local_id);
1660
1661	subflow->remote_key_valid = 1;
1662	subflow->remote_key = READ_ONCE(msk->remote_key);
1663	subflow->local_key = READ_ONCE(msk->local_key);
1664	subflow->token = msk->token;
1665	mptcp_info2sockaddr(&local->addr, &addr, ssk->sk_family);
1666
1667	addrlen = sizeof(struct sockaddr_in);
1668#if IS_ENABLED(CONFIG_MPTCP_IPV6)
1669	if (addr.ss_family == AF_INET6)
1670		addrlen = sizeof(struct sockaddr_in6);
1671#endif
1672	ssk->sk_bound_dev_if = local->ifindex;
1673	err = kernel_bind(sf, (struct sockaddr_unsized *)&addr, addrlen);
1674	if (err) {
1675		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNTXBINDERR);
1676		pr_debug("msk=%p local=%d remote=%d bind error: %d\n",
1677			 msk, local_id, remote_id, err);
1678		goto failed;
1679	}
1680
1681	mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL);
1682	pr_debug("msk=%p remote_token=%u local_id=%d remote_id=%d\n", msk,
1683		 remote_token, local_id, remote_id);
1684	subflow->remote_token = remote_token;
1685	WRITE_ONCE(subflow->remote_id, remote_id);
1686	subflow->request_join = 1;
1687	subflow->request_bkup = !!(local->flags & MPTCP_PM_ADDR_FLAG_BACKUP);
1688	subflow->subflow_id = msk->subflow_id++;
1689	mptcp_info2sockaddr(remote, &addr, ssk->sk_family);
1690
1691	sock_hold(ssk);
1692	list_add_tail(&subflow->node, &msk->conn_list);
1693	err = kernel_connect(sf, (struct sockaddr_unsized *)&addr, addrlen, O_NONBLOCK);
1694	if (err && err != -EINPROGRESS) {
1695		MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNTXCONNECTERR);
1696		pr_debug("msk=%p local=%d remote=%d connect error: %d\n",
1697			 msk, local_id, remote_id, err);
1698		goto failed_unlink;
1699	}
1700
1701	MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNTX);
1702
1703	/* discard the subflow socket */
1704	mptcp_sock_graft(ssk, sk->sk_socket);
1705	iput(SOCK_INODE(sf));
1706	mptcp_stop_tout_timer(sk);
1707	return 0;
1708
1709failed_unlink:
1710	list_del(&subflow->node);
1711	sock_put(mptcp_subflow_tcp_sock(subflow));
1712
1713failed:
1714	subflow->disposable = 1;
1715	sock_release(sf);
1716
1717err_out:
1718	/* we account subflows before the creation, and this failures will not
1719	 * be caught by sk_state_change()
1720	 */
1721	mptcp_pm_close_subflow(msk);
1722	return err;
1723}
1724
1725void __mptcp_inherit_memcg(struct sock *sk, struct sock *ssk, gfp_t gfp)
1726{
1727	/* Only if the msk has been accepted already (and not orphaned).*/
1728	if (!mem_cgroup_sockets_enabled || !sk->sk_socket)
1729		return;
1730
1731	mem_cgroup_sk_inherit(sk, ssk);
1732	__sk_charge(ssk, gfp);
1733}
1734
1735void __mptcp_inherit_cgrp_data(struct sock *sk, struct sock *ssk)
1736{
1737#ifdef CONFIG_SOCK_CGROUP_DATA
1738	struct sock_cgroup_data *sk_cd = &sk->sk_cgrp_data,
1739				*ssk_cd = &ssk->sk_cgrp_data;
1740
1741	/* only the additional subflows created by kworkers have to be modified */
1742	if (cgroup_id(sock_cgroup_ptr(sk_cd)) !=
1743	    cgroup_id(sock_cgroup_ptr(ssk_cd))) {
1744		cgroup_sk_free(ssk_cd);
1745		*ssk_cd = *sk_cd;
1746		cgroup_sk_clone(sk_cd);
1747	}
1748#endif /* CONFIG_SOCK_CGROUP_DATA */
1749}
1750
1751static void mptcp_attach_cgroup(struct sock *parent, struct sock *child)
1752{
1753	__mptcp_inherit_cgrp_data(parent, child);
1754	if (mem_cgroup_sockets_enabled)
1755		mem_cgroup_sk_inherit(parent, child);
1756}
1757
1758static void mptcp_subflow_ops_override(struct sock *ssk)
1759{
1760#if IS_ENABLED(CONFIG_MPTCP_IPV6)
1761	if (ssk->sk_prot == &tcpv6_prot)
1762		ssk->sk_prot = &tcpv6_prot_override;
1763	else
1764#endif
1765		ssk->sk_prot = &tcp_prot_override;
1766}
1767
1768static void mptcp_subflow_ops_undo_override(struct sock *ssk)
1769{
1770#if IS_ENABLED(CONFIG_MPTCP_IPV6)
1771	if (ssk->sk_prot == &tcpv6_prot_override)
1772		ssk->sk_prot = &tcpv6_prot;
1773	else
1774#endif
1775		ssk->sk_prot = &tcp_prot;
1776}
1777
1778int mptcp_subflow_create_socket(struct sock *sk, unsigned short family,
1779				struct socket **new_sock)
1780{
1781	struct mptcp_subflow_context *subflow;
1782	struct net *net = sock_net(sk);
1783	struct socket *sf;
1784	int err;
1785
1786	/* un-accepted server sockets can reach here - on bad configuration
1787	 * bail early to avoid greater trouble later
1788	 */
1789	if (unlikely(!sk->sk_socket))
1790		return -EINVAL;
1791
1792	err = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP, &sf);
1793	if (err)
1794		return err;
1795
1796	lock_sock_nested(sf->sk, SINGLE_DEPTH_NESTING);
1797
1798	err = security_mptcp_add_subflow(sk, sf->sk);
1799	if (err)
1800		goto err_free;
1801
1802	/* the newly created socket has to be in the same cgroup as its parent */
1803	mptcp_attach_cgroup(sk, sf->sk);
1804
1805	/* kernel sockets do not by default acquire net ref, but TCP timer
1806	 * needs it.
1807	 * Update ns_tracker to current stack trace and refcounted tracker.
1808	 */
1809	sk_net_refcnt_upgrade(sf->sk);
1810	err = tcp_set_ulp(sf->sk, "mptcp");
1811	if (err)
1812		goto err_free;
1813
1814	mptcp_sockopt_sync_locked(mptcp_sk(sk), sf->sk);
1815	release_sock(sf->sk);
1816
1817	/* the newly created socket really belongs to the owning MPTCP
1818	 * socket, even if for additional subflows the allocation is performed
1819	 * by a kernel workqueue. Adjust inode references, so that the
1820	 * procfs/diag interfaces really show this one belonging to the correct
1821	 * user.
1822	 */
1823	SOCK_INODE(sf)->i_ino = SOCK_INODE(sk->sk_socket)->i_ino;
1824	SOCK_INODE(sf)->i_uid = SOCK_INODE(sk->sk_socket)->i_uid;
1825	SOCK_INODE(sf)->i_gid = SOCK_INODE(sk->sk_socket)->i_gid;
1826
1827	subflow = mptcp_subflow_ctx(sf->sk);
1828	pr_debug("subflow=%p\n", subflow);
1829
1830	*new_sock = sf;
1831	sock_hold(sk);
1832	subflow->conn = sk;
1833	mptcp_subflow_ops_override(sf->sk);
1834
1835	return 0;
1836
1837err_free:
1838	release_sock(sf->sk);
1839	sock_release(sf);
1840	return err;
1841}
1842
1843static struct mptcp_subflow_context *subflow_create_ctx(struct sock *sk,
1844							gfp_t priority)
1845{
1846	struct inet_connection_sock *icsk = inet_csk(sk);
1847	struct mptcp_subflow_context *ctx;
1848
1849	ctx = kzalloc(sizeof(*ctx), priority);
1850	if (!ctx)
1851		return NULL;
1852
1853	rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
1854	INIT_LIST_HEAD(&ctx->node);
1855	INIT_LIST_HEAD(&ctx->delegated_node);
1856
1857	pr_debug("subflow=%p\n", ctx);
1858
1859	ctx->tcp_sock = sk;
1860	WRITE_ONCE(ctx->local_id, -1);
1861
1862	return ctx;
1863}
1864
1865static void __subflow_state_change(struct sock *sk)
1866{
1867	struct socket_wq *wq;
1868
1869	rcu_read_lock();
1870	wq = rcu_dereference(sk->sk_wq);
1871	if (skwq_has_sleeper(wq))
1872		wake_up_interruptible_all(&wq->wait);
1873	rcu_read_unlock();
1874}
1875
1876static void subflow_state_change(struct sock *sk)
1877{
1878	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1879	struct sock *parent = subflow->conn;
1880
1881	__subflow_state_change(sk);
1882
1883	/* as recvmsg() does not acquire the subflow socket for ssk selection
1884	 * a fin packet carrying a DSS can be unnoticed if we don't trigger
1885	 * the data available machinery here.
1886	 */
1887	if (mptcp_subflow_data_available(sk))
1888		mptcp_data_ready(parent, sk);
1889	else if (unlikely(sk->sk_err))
1890		subflow_error_report(sk);
1891
1892	subflow_sched_work_if_closed(mptcp_sk(parent), sk);
1893}
1894
1895void mptcp_subflow_queue_clean(struct sock *listener_sk, struct sock *listener_ssk)
1896{
1897	struct request_sock_queue *queue = &inet_csk(listener_ssk)->icsk_accept_queue;
1898	struct request_sock *req, *head, *tail;
1899	struct mptcp_subflow_context *subflow;
1900	struct sock *sk, *ssk;
1901
1902	/* Due to lock dependencies no relevant lock can be acquired under rskq_lock.
1903	 * Splice the req list, so that accept() can not reach the pending ssk after
1904	 * the listener socket is released below.
1905	 */
1906	spin_lock_bh(&queue->rskq_lock);
1907	head = queue->rskq_accept_head;
1908	tail = queue->rskq_accept_tail;
1909	queue->rskq_accept_head = NULL;
1910	queue->rskq_accept_tail = NULL;
1911	spin_unlock_bh(&queue->rskq_lock);
1912
1913	if (!head)
1914		return;
1915
1916	/* can't acquire the msk socket lock under the subflow one,
1917	 * or will cause ABBA deadlock
1918	 */
1919	release_sock(listener_ssk);
1920
1921	for (req = head; req; req = req->dl_next) {
1922		ssk = req->sk;
1923		if (!sk_is_mptcp(ssk))
1924			continue;
1925
1926		subflow = mptcp_subflow_ctx(ssk);
1927		if (!subflow || !subflow->conn)
1928			continue;
1929
1930		sk = subflow->conn;
1931		sock_hold(sk);
1932
1933		lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
1934		__mptcp_unaccepted_force_close(sk);
1935		release_sock(sk);
1936
1937		/* lockdep will report a false positive ABBA deadlock
1938		 * between cancel_work_sync and the listener socket.
1939		 * The involved locks belong to different sockets WRT
1940		 * the existing AB chain.
1941		 * Using a per socket key is problematic as key
1942		 * deregistration requires process context and must be
1943		 * performed at socket disposal time, in atomic
1944		 * context.
1945		 * Just tell lockdep to consider the listener socket
1946		 * released here.
1947		 */
1948		mutex_release(&listener_sk->sk_lock.dep_map, _RET_IP_);
1949		mptcp_cancel_work(sk);
1950		mutex_acquire(&listener_sk->sk_lock.dep_map, 0, 0, _RET_IP_);
1951
1952		sock_put(sk);
1953	}
1954
1955	/* we are still under the listener msk socket lock */
1956	lock_sock_nested(listener_ssk, SINGLE_DEPTH_NESTING);
1957
1958	/* restore the listener queue, to let the TCP code clean it up */
1959	spin_lock_bh(&queue->rskq_lock);
1960	WARN_ON_ONCE(queue->rskq_accept_head);
1961	queue->rskq_accept_head = head;
1962	queue->rskq_accept_tail = tail;
1963	spin_unlock_bh(&queue->rskq_lock);
1964}
1965
1966static int subflow_ulp_init(struct sock *sk)
1967{
1968	struct inet_connection_sock *icsk = inet_csk(sk);
1969	struct mptcp_subflow_context *ctx;
1970	struct tcp_sock *tp = tcp_sk(sk);
1971	int err = 0;
1972
1973	/* disallow attaching ULP to a socket unless it has been
1974	 * created with sock_create_kern()
1975	 */
1976	if (!sk->sk_kern_sock) {
1977		err = -EOPNOTSUPP;
1978		goto out;
1979	}
1980
1981	ctx = subflow_create_ctx(sk, GFP_KERNEL);
1982	if (!ctx) {
1983		err = -ENOMEM;
1984		goto out;
1985	}
1986
1987	pr_debug("subflow=%p, family=%d\n", ctx, sk->sk_family);
1988
1989	tp->is_mptcp = 1;
1990	ctx->icsk_af_ops = icsk->icsk_af_ops;
1991	icsk->icsk_af_ops = subflow_default_af_ops(sk);
1992	ctx->tcp_state_change = sk->sk_state_change;
1993	ctx->tcp_error_report = sk->sk_error_report;
1994
1995	WARN_ON_ONCE(sk->sk_data_ready != sock_def_readable);
1996	WARN_ON_ONCE(sk->sk_write_space != sk_stream_write_space);
1997
1998	sk->sk_data_ready = subflow_data_ready;
1999	sk->sk_write_space = subflow_write_space;
2000	sk->sk_state_change = subflow_state_change;
2001	sk->sk_error_report = subflow_error_report;
2002out:
2003	return err;
2004}
2005
2006static void subflow_ulp_release(struct sock *ssk)
2007{
2008	struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
2009	bool release = true;
2010	struct sock *sk;
2011
2012	if (!ctx)
2013		return;
2014
2015	sk = ctx->conn;
2016	if (sk) {
2017		/* if the msk has been orphaned, keep the ctx
2018		 * alive, will be freed by __mptcp_close_ssk(),
2019		 * when the subflow is still unaccepted
2020		 */
2021		release = ctx->disposable || list_empty(&ctx->node);
2022
2023		/* inet_child_forget() does not call sk_state_change(),
2024		 * explicitly trigger the socket close machinery
2025		 */
2026		if (!release && !test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW,
2027						  &mptcp_sk(sk)->flags))
2028			mptcp_schedule_work(sk);
2029		sock_put(sk);
2030	}
2031
2032	mptcp_subflow_ops_undo_override(ssk);
2033	if (release)
2034		kfree_rcu(ctx, rcu);
2035}
2036
2037static void subflow_ulp_clone(const struct request_sock *req,
2038			      struct sock *newsk,
2039			      const gfp_t priority)
2040{
2041	struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
2042	struct mptcp_subflow_context *old_ctx = mptcp_subflow_ctx(newsk);
2043	struct mptcp_subflow_context *new_ctx;
2044
2045	if (!tcp_rsk(req)->is_mptcp ||
2046	    (!subflow_req->mp_capable && !subflow_req->mp_join)) {
2047		subflow_ulp_fallback(newsk, old_ctx);
2048		return;
2049	}
2050
2051	new_ctx = subflow_create_ctx(newsk, priority);
2052	if (!new_ctx) {
2053		subflow_ulp_fallback(newsk, old_ctx);
2054		return;
2055	}
2056
2057	new_ctx->conn_finished = 1;
2058	new_ctx->icsk_af_ops = old_ctx->icsk_af_ops;
2059	new_ctx->tcp_state_change = old_ctx->tcp_state_change;
2060	new_ctx->tcp_error_report = old_ctx->tcp_error_report;
2061	new_ctx->rel_write_seq = 1;
2062
2063	if (subflow_req->mp_capable) {
2064		/* see comments in subflow_syn_recv_sock(), MPTCP connection
2065		 * is fully established only after we receive the remote key
2066		 */
2067		new_ctx->mp_capable = 1;
2068		new_ctx->local_key = subflow_req->local_key;
2069		new_ctx->token = subflow_req->token;
2070		new_ctx->ssn_offset = subflow_req->ssn_offset;
2071		new_ctx->idsn = subflow_req->idsn;
2072
2073		/* this is the first subflow, id is always 0 */
2074		subflow_set_local_id(new_ctx, 0);
2075	} else if (subflow_req->mp_join) {
2076		new_ctx->ssn_offset = subflow_req->ssn_offset;
2077		new_ctx->mp_join = 1;
2078		WRITE_ONCE(new_ctx->fully_established, true);
2079		new_ctx->remote_key_valid = 1;
2080		new_ctx->backup = subflow_req->backup;
2081		new_ctx->request_bkup = subflow_req->request_bkup;
2082		WRITE_ONCE(new_ctx->remote_id, subflow_req->remote_id);
2083		new_ctx->token = subflow_req->token;
2084		new_ctx->thmac = subflow_req->thmac;
2085
2086		/* the subflow req id is valid, fetched via subflow_check_req()
2087		 * and subflow_token_join_request()
2088		 */
2089		subflow_set_local_id(new_ctx, subflow_req->local_id);
2090	}
2091}
2092
2093static void tcp_release_cb_override(struct sock *ssk)
2094{
2095	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
2096	long status;
2097
2098	/* process and clear all the pending actions, but leave the subflow into
2099	 * the napi queue. To respect locking, only the same CPU that originated
2100	 * the action can touch the list. mptcp_napi_poll will take care of it.
2101	 */
2102	status = set_mask_bits(&subflow->delegated_status, MPTCP_DELEGATE_ACTIONS_MASK, 0);
2103	if (status)
2104		mptcp_subflow_process_delegated(ssk, status);
2105
2106	tcp_release_cb(ssk);
2107}
2108
2109static int tcp_abort_override(struct sock *ssk, int err)
2110{
2111	/* closing a listener subflow requires a great deal of care.
2112	 * keep it simple and just prevent such operation
2113	 */
2114	if (inet_sk_state_load(ssk) == TCP_LISTEN)
2115		return -EINVAL;
2116
2117	return tcp_abort(ssk, err);
2118}
2119
2120static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = {
2121	.name		= "mptcp",
2122	.owner		= THIS_MODULE,
2123	.init		= subflow_ulp_init,
2124	.release	= subflow_ulp_release,
2125	.clone		= subflow_ulp_clone,
2126};
2127
2128static int subflow_ops_init(struct request_sock_ops *subflow_ops)
2129{
2130	subflow_ops->obj_size = sizeof(struct mptcp_subflow_request_sock);
2131
2132	subflow_ops->slab = kmem_cache_create(subflow_ops->slab_name,
2133					      subflow_ops->obj_size, 0,
2134					      SLAB_ACCOUNT |
2135					      SLAB_TYPESAFE_BY_RCU,
2136					      NULL);
2137	if (!subflow_ops->slab)
2138		return -ENOMEM;
2139
2140	return 0;
2141}
2142
2143void __init mptcp_subflow_init(void)
2144{
2145	mptcp_subflow_v4_request_sock_ops = tcp_request_sock_ops;
2146	mptcp_subflow_v4_request_sock_ops.slab_name = "request_sock_subflow_v4";
2147	mptcp_subflow_v4_request_sock_ops.destructor = subflow_v4_req_destructor;
2148
2149	if (subflow_ops_init(&mptcp_subflow_v4_request_sock_ops) != 0)
2150		panic("MPTCP: failed to init subflow v4 request sock ops\n");
2151
2152	subflow_request_sock_ipv4_ops = tcp_request_sock_ipv4_ops;
2153	subflow_request_sock_ipv4_ops.route_req = subflow_v4_route_req;
2154	subflow_request_sock_ipv4_ops.send_synack = subflow_v4_send_synack;
2155
2156	subflow_specific = ipv4_specific;
2157	subflow_specific.conn_request = subflow_v4_conn_request;
2158	subflow_specific.syn_recv_sock = subflow_syn_recv_sock;
2159	subflow_specific.sk_rx_dst_set = subflow_finish_connect;
2160	subflow_specific.rebuild_header = subflow_rebuild_header;
2161
2162	tcp_prot_override = tcp_prot;
2163	tcp_prot_override.release_cb = tcp_release_cb_override;
2164	tcp_prot_override.diag_destroy = tcp_abort_override;
2165#ifdef CONFIG_BPF_SYSCALL
2166	/* Disable sockmap processing for subflows */
2167	tcp_prot_override.psock_update_sk_prot = NULL;
2168#endif
2169
2170	mptcp_diag_subflow_init(&subflow_ulp_ops);
2171
2172	if (tcp_register_ulp(&subflow_ulp_ops) != 0)
2173		panic("MPTCP: failed to register subflows to ULP\n");
2174}
2175
2176#if IS_ENABLED(CONFIG_MPTCP_IPV6)
2177void __init mptcp_subflow_v6_init(void)
2178{
2179	/* In struct mptcp_subflow_request_sock, we assume the TCP request sock
2180	 * structures for v4 and v6 have the same size. It should not changed in
2181	 * the future but better to make sure to be warned if it is no longer
2182	 * the case.
2183	 */
2184	BUILD_BUG_ON(sizeof(struct tcp_request_sock) != sizeof(struct tcp6_request_sock));
2185
2186	mptcp_subflow_v6_request_sock_ops = tcp6_request_sock_ops;
2187	mptcp_subflow_v6_request_sock_ops.slab_name = "request_sock_subflow_v6";
2188	mptcp_subflow_v6_request_sock_ops.destructor = subflow_v6_req_destructor;
2189
2190	if (subflow_ops_init(&mptcp_subflow_v6_request_sock_ops) != 0)
2191		panic("MPTCP: failed to init subflow v6 request sock ops\n");
2192
2193	subflow_request_sock_ipv6_ops = tcp_request_sock_ipv6_ops;
2194	subflow_request_sock_ipv6_ops.route_req = subflow_v6_route_req;
2195	subflow_request_sock_ipv6_ops.send_synack = subflow_v6_send_synack;
2196
2197	subflow_v6_specific = ipv6_specific;
2198	subflow_v6_specific.conn_request = subflow_v6_conn_request;
2199	subflow_v6_specific.syn_recv_sock = subflow_syn_recv_sock;
2200	subflow_v6_specific.sk_rx_dst_set = subflow_finish_connect;
2201	subflow_v6_specific.rebuild_header = subflow_v6_rebuild_header;
2202
2203	subflow_v6m_specific = subflow_v6_specific;
2204	subflow_v6m_specific.queue_xmit = ipv4_specific.queue_xmit;
2205	subflow_v6m_specific.send_check = ipv4_specific.send_check;
2206	subflow_v6m_specific.net_header_len = ipv4_specific.net_header_len;
2207	subflow_v6m_specific.mtu_reduced = ipv4_specific.mtu_reduced;
2208	subflow_v6m_specific.rebuild_header = subflow_rebuild_header;
2209
2210	tcpv6_prot_override = tcpv6_prot;
2211	tcpv6_prot_override.release_cb = tcp_release_cb_override;
2212	tcpv6_prot_override.diag_destroy = tcp_abort_override;
2213#ifdef CONFIG_BPF_SYSCALL
2214	/* Disable sockmap processing for subflows */
2215	tcpv6_prot_override.psock_update_sk_prot = NULL;
2216#endif
2217}
2218#endif