1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * Kernel Connection Multiplexor
   4 *
   5 * Copyright (c) 2016 Tom Herbert <tom@herbertland.com>
   6 */
   7
   8#include <linux/bpf.h>
   9#include <linux/errno.h>
  10#include <linux/errqueue.h>
  11#include <linux/file.h>
  12#include <linux/in.h>
  13#include <linux/kernel.h>
  14#include <linux/module.h>
  15#include <linux/net.h>
  16#include <linux/netdevice.h>
  17#include <linux/poll.h>
  18#include <linux/rculist.h>
  19#include <linux/skbuff.h>
  20#include <linux/socket.h>
  21#include <linux/uaccess.h>
  22#include <linux/workqueue.h>
  23#include <linux/syscalls.h>
  24#include <linux/sched/signal.h>
  25
  26#include <net/kcm.h>
  27#include <net/netns/generic.h>
  28#include <net/sock.h>
  29#include <uapi/linux/kcm.h>
  30
  31unsigned int kcm_net_id;
  32
  33static struct kmem_cache *kcm_psockp __read_mostly;
  34static struct kmem_cache *kcm_muxp __read_mostly;
  35static struct workqueue_struct *kcm_wq;
  36
  37static inline struct kcm_sock *kcm_sk(const struct sock *sk)
  38{
  39	return (struct kcm_sock *)sk;
  40}
  41
  42static inline struct kcm_tx_msg *kcm_tx_msg(struct sk_buff *skb)
  43{
  44	return (struct kcm_tx_msg *)skb->cb;
  45}
  46
  47static void report_csk_error(struct sock *csk, int err)
  48{
  49	csk->sk_err = EPIPE;
  50	csk->sk_error_report(csk);
  51}
  52
  53static void kcm_abort_tx_psock(struct kcm_psock *psock, int err,
  54			       bool wakeup_kcm)
  55{
  56	struct sock *csk = psock->sk;
  57	struct kcm_mux *mux = psock->mux;
  58
  59	/* Unrecoverable error in transmit */
  60
  61	spin_lock_bh(&mux->lock);
  62
  63	if (psock->tx_stopped) {
  64		spin_unlock_bh(&mux->lock);
  65		return;
  66	}
  67
  68	psock->tx_stopped = 1;
  69	KCM_STATS_INCR(psock->stats.tx_aborts);
  70
  71	if (!psock->tx_kcm) {
  72		/* Take off psocks_avail list */
  73		list_del(&psock->psock_avail_list);
  74	} else if (wakeup_kcm) {
  75		/* In this case psock is being aborted while outside of
  76		 * write_msgs and psock is reserved. Schedule tx_work
  77		 * to handle the failure there. Need to commit tx_stopped
  78		 * before queuing work.
  79		 */
  80		smp_mb();
  81
  82		queue_work(kcm_wq, &psock->tx_kcm->tx_work);
  83	}
  84
  85	spin_unlock_bh(&mux->lock);
  86
  87	/* Report error on lower socket */
  88	report_csk_error(csk, err);
  89}
  90
  91/* RX mux lock held. */
  92static void kcm_update_rx_mux_stats(struct kcm_mux *mux,
  93				    struct kcm_psock *psock)
  94{
  95	STRP_STATS_ADD(mux->stats.rx_bytes,
  96		       psock->strp.stats.bytes -
  97		       psock->saved_rx_bytes);
  98	mux->stats.rx_msgs +=
  99		psock->strp.stats.msgs - psock->saved_rx_msgs;
 100	psock->saved_rx_msgs = psock->strp.stats.msgs;
 101	psock->saved_rx_bytes = psock->strp.stats.bytes;
 102}
 103
 104static void kcm_update_tx_mux_stats(struct kcm_mux *mux,
 105				    struct kcm_psock *psock)
 106{
 107	KCM_STATS_ADD(mux->stats.tx_bytes,
 108		      psock->stats.tx_bytes - psock->saved_tx_bytes);
 109	mux->stats.tx_msgs +=
 110		psock->stats.tx_msgs - psock->saved_tx_msgs;
 111	psock->saved_tx_msgs = psock->stats.tx_msgs;
 112	psock->saved_tx_bytes = psock->stats.tx_bytes;
 113}
 114
 115static int kcm_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
 116
 117/* KCM is ready to receive messages on its queue-- either the KCM is new or
 118 * has become unblocked after being blocked on full socket buffer. Queue any
 119 * pending ready messages on a psock. RX mux lock held.
 120 */
 121static void kcm_rcv_ready(struct kcm_sock *kcm)
 122{
 123	struct kcm_mux *mux = kcm->mux;
 124	struct kcm_psock *psock;
 125	struct sk_buff *skb;
 126
 127	if (unlikely(kcm->rx_wait || kcm->rx_psock || kcm->rx_disabled))
 128		return;
 129
 130	while (unlikely((skb = __skb_dequeue(&mux->rx_hold_queue)))) {
 131		if (kcm_queue_rcv_skb(&kcm->sk, skb)) {
 132			/* Assuming buffer limit has been reached */
 133			skb_queue_head(&mux->rx_hold_queue, skb);
 134			WARN_ON(!sk_rmem_alloc_get(&kcm->sk));
 135			return;
 136		}
 137	}
 138
 139	while (!list_empty(&mux->psocks_ready)) {
 140		psock = list_first_entry(&mux->psocks_ready, struct kcm_psock,
 141					 psock_ready_list);
 142
 143		if (kcm_queue_rcv_skb(&kcm->sk, psock->ready_rx_msg)) {
 144			/* Assuming buffer limit has been reached */
 145			WARN_ON(!sk_rmem_alloc_get(&kcm->sk));
 146			return;
 147		}
 148
 149		/* Consumed the ready message on the psock. Schedule rx_work to
 150		 * get more messages.
 151		 */
 152		list_del(&psock->psock_ready_list);
 153		psock->ready_rx_msg = NULL;
 154		/* Commit clearing of ready_rx_msg for queuing work */
 155		smp_mb();
 156
 157		strp_unpause(&psock->strp);
 158		strp_check_rcv(&psock->strp);
 159	}
 160
 161	/* Buffer limit is okay now, add to ready list */
 162	list_add_tail(&kcm->wait_rx_list,
 163		      &kcm->mux->kcm_rx_waiters);
 164	kcm->rx_wait = true;
 165}
 166
 167static void kcm_rfree(struct sk_buff *skb)
 168{
 169	struct sock *sk = skb->sk;
 170	struct kcm_sock *kcm = kcm_sk(sk);
 171	struct kcm_mux *mux = kcm->mux;
 172	unsigned int len = skb->truesize;
 173
 174	sk_mem_uncharge(sk, len);
 175	atomic_sub(len, &sk->sk_rmem_alloc);
 176
 177	/* For reading rx_wait and rx_psock without holding lock */
 178	smp_mb__after_atomic();
 179
 180	if (!kcm->rx_wait && !kcm->rx_psock &&
 181	    sk_rmem_alloc_get(sk) < sk->sk_rcvlowat) {
 182		spin_lock_bh(&mux->rx_lock);
 183		kcm_rcv_ready(kcm);
 184		spin_unlock_bh(&mux->rx_lock);
 185	}
 186}
 187
 188static int kcm_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
 189{
 190	struct sk_buff_head *list = &sk->sk_receive_queue;
 191
 192	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
 193		return -ENOMEM;
 194
 195	if (!sk_rmem_schedule(sk, skb, skb->truesize))
 196		return -ENOBUFS;
 197
 198	skb->dev = NULL;
 199
 200	skb_orphan(skb);
 201	skb->sk = sk;
 202	skb->destructor = kcm_rfree;
 203	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
 204	sk_mem_charge(sk, skb->truesize);
 205
 206	skb_queue_tail(list, skb);
 207
 208	if (!sock_flag(sk, SOCK_DEAD))
 209		sk->sk_data_ready(sk);
 210
 211	return 0;
 212}
 213
 214/* Requeue received messages for a kcm socket to other kcm sockets. This is
 215 * called with a kcm socket is receive disabled.
 216 * RX mux lock held.
 217 */
 218static void requeue_rx_msgs(struct kcm_mux *mux, struct sk_buff_head *head)
 219{
 220	struct sk_buff *skb;
 221	struct kcm_sock *kcm;
 222
 223	while ((skb = __skb_dequeue(head))) {
 224		/* Reset destructor to avoid calling kcm_rcv_ready */
 225		skb->destructor = sock_rfree;
 226		skb_orphan(skb);
 227try_again:
 228		if (list_empty(&mux->kcm_rx_waiters)) {
 229			skb_queue_tail(&mux->rx_hold_queue, skb);
 230			continue;
 231		}
 232
 233		kcm = list_first_entry(&mux->kcm_rx_waiters,
 234				       struct kcm_sock, wait_rx_list);
 235
 236		if (kcm_queue_rcv_skb(&kcm->sk, skb)) {
 237			/* Should mean socket buffer full */
 238			list_del(&kcm->wait_rx_list);
 239			kcm->rx_wait = false;
 240
 241			/* Commit rx_wait to read in kcm_free */
 242			smp_wmb();
 243
 244			goto try_again;
 245		}
 246	}
 247}
 248
 249/* Lower sock lock held */
 250static struct kcm_sock *reserve_rx_kcm(struct kcm_psock *psock,
 251				       struct sk_buff *head)
 252{
 253	struct kcm_mux *mux = psock->mux;
 254	struct kcm_sock *kcm;
 255
 256	WARN_ON(psock->ready_rx_msg);
 257
 258	if (psock->rx_kcm)
 259		return psock->rx_kcm;
 260
 261	spin_lock_bh(&mux->rx_lock);
 262
 263	if (psock->rx_kcm) {
 264		spin_unlock_bh(&mux->rx_lock);
 265		return psock->rx_kcm;
 266	}
 267
 268	kcm_update_rx_mux_stats(mux, psock);
 269
 270	if (list_empty(&mux->kcm_rx_waiters)) {
 271		psock->ready_rx_msg = head;
 272		strp_pause(&psock->strp);
 273		list_add_tail(&psock->psock_ready_list,
 274			      &mux->psocks_ready);
 275		spin_unlock_bh(&mux->rx_lock);
 276		return NULL;
 277	}
 278
 279	kcm = list_first_entry(&mux->kcm_rx_waiters,
 280			       struct kcm_sock, wait_rx_list);
 281	list_del(&kcm->wait_rx_list);
 282	kcm->rx_wait = false;
 283
 284	psock->rx_kcm = kcm;
 285	kcm->rx_psock = psock;
 286
 287	spin_unlock_bh(&mux->rx_lock);
 288
 289	return kcm;
 290}
 291
 292static void kcm_done(struct kcm_sock *kcm);
 293
 294static void kcm_done_work(struct work_struct *w)
 295{
 296	kcm_done(container_of(w, struct kcm_sock, done_work));
 297}
 298
 299/* Lower sock held */
 300static void unreserve_rx_kcm(struct kcm_psock *psock,
 301			     bool rcv_ready)
 302{
 303	struct kcm_sock *kcm = psock->rx_kcm;
 304	struct kcm_mux *mux = psock->mux;
 305
 306	if (!kcm)
 307		return;
 308
 309	spin_lock_bh(&mux->rx_lock);
 310
 311	psock->rx_kcm = NULL;
 312	kcm->rx_psock = NULL;
 313
 314	/* Commit kcm->rx_psock before sk_rmem_alloc_get to sync with
 315	 * kcm_rfree
 316	 */
 317	smp_mb();
 318
 319	if (unlikely(kcm->done)) {
 320		spin_unlock_bh(&mux->rx_lock);
 321
 322		/* Need to run kcm_done in a task since we need to qcquire
 323		 * callback locks which may already be held here.
 324		 */
 325		INIT_WORK(&kcm->done_work, kcm_done_work);
 326		schedule_work(&kcm->done_work);
 327		return;
 328	}
 329
 330	if (unlikely(kcm->rx_disabled)) {
 331		requeue_rx_msgs(mux, &kcm->sk.sk_receive_queue);
 332	} else if (rcv_ready || unlikely(!sk_rmem_alloc_get(&kcm->sk))) {
 333		/* Check for degenerative race with rx_wait that all
 334		 * data was dequeued (accounted for in kcm_rfree).
 335		 */
 336		kcm_rcv_ready(kcm);
 337	}
 338	spin_unlock_bh(&mux->rx_lock);
 339}
 340
 341/* Lower sock lock held */
 342static void psock_data_ready(struct sock *sk)
 343{
 344	struct kcm_psock *psock;
 345
 346	read_lock_bh(&sk->sk_callback_lock);
 347
 348	psock = (struct kcm_psock *)sk->sk_user_data;
 349	if (likely(psock))
 350		strp_data_ready(&psock->strp);
 351
 352	read_unlock_bh(&sk->sk_callback_lock);
 353}
 354
 355/* Called with lower sock held */
 356static void kcm_rcv_strparser(struct strparser *strp, struct sk_buff *skb)
 357{
 358	struct kcm_psock *psock = container_of(strp, struct kcm_psock, strp);
 359	struct kcm_sock *kcm;
 360
 361try_queue:
 362	kcm = reserve_rx_kcm(psock, skb);
 363	if (!kcm) {
 364		 /* Unable to reserve a KCM, message is held in psock and strp
 365		  * is paused.
 366		  */
 367		return;
 368	}
 369
 370	if (kcm_queue_rcv_skb(&kcm->sk, skb)) {
 371		/* Should mean socket buffer full */
 372		unreserve_rx_kcm(psock, false);
 373		goto try_queue;
 374	}
 375}
 376
 377static int kcm_parse_func_strparser(struct strparser *strp, struct sk_buff *skb)
 378{
 379	struct kcm_psock *psock = container_of(strp, struct kcm_psock, strp);
 380	struct bpf_prog *prog = psock->bpf_prog;
 381	int res;
 382
 383	res = bpf_prog_run_pin_on_cpu(prog, skb);
 
 
 384	return res;
 385}
 386
 387static int kcm_read_sock_done(struct strparser *strp, int err)
 388{
 389	struct kcm_psock *psock = container_of(strp, struct kcm_psock, strp);
 390
 391	unreserve_rx_kcm(psock, true);
 392
 393	return err;
 394}
 395
 396static void psock_state_change(struct sock *sk)
 397{
 398	/* TCP only does a EPOLLIN for a half close. Do a EPOLLHUP here
 399	 * since application will normally not poll with EPOLLIN
 400	 * on the TCP sockets.
 401	 */
 402
 403	report_csk_error(sk, EPIPE);
 404}
 405
 406static void psock_write_space(struct sock *sk)
 407{
 408	struct kcm_psock *psock;
 409	struct kcm_mux *mux;
 410	struct kcm_sock *kcm;
 411
 412	read_lock_bh(&sk->sk_callback_lock);
 413
 414	psock = (struct kcm_psock *)sk->sk_user_data;
 415	if (unlikely(!psock))
 416		goto out;
 417	mux = psock->mux;
 418
 419	spin_lock_bh(&mux->lock);
 420
 421	/* Check if the socket is reserved so someone is waiting for sending. */
 422	kcm = psock->tx_kcm;
 423	if (kcm && !unlikely(kcm->tx_stopped))
 424		queue_work(kcm_wq, &kcm->tx_work);
 425
 426	spin_unlock_bh(&mux->lock);
 427out:
 428	read_unlock_bh(&sk->sk_callback_lock);
 429}
 430
 431static void unreserve_psock(struct kcm_sock *kcm);
 432
 433/* kcm sock is locked. */
 434static struct kcm_psock *reserve_psock(struct kcm_sock *kcm)
 435{
 436	struct kcm_mux *mux = kcm->mux;
 437	struct kcm_psock *psock;
 438
 439	psock = kcm->tx_psock;
 440
 441	smp_rmb(); /* Must read tx_psock before tx_wait */
 442
 443	if (psock) {
 444		WARN_ON(kcm->tx_wait);
 445		if (unlikely(psock->tx_stopped))
 446			unreserve_psock(kcm);
 447		else
 448			return kcm->tx_psock;
 449	}
 450
 451	spin_lock_bh(&mux->lock);
 452
 453	/* Check again under lock to see if psock was reserved for this
 454	 * psock via psock_unreserve.
 455	 */
 456	psock = kcm->tx_psock;
 457	if (unlikely(psock)) {
 458		WARN_ON(kcm->tx_wait);
 459		spin_unlock_bh(&mux->lock);
 460		return kcm->tx_psock;
 461	}
 462
 463	if (!list_empty(&mux->psocks_avail)) {
 464		psock = list_first_entry(&mux->psocks_avail,
 465					 struct kcm_psock,
 466					 psock_avail_list);
 467		list_del(&psock->psock_avail_list);
 468		if (kcm->tx_wait) {
 469			list_del(&kcm->wait_psock_list);
 470			kcm->tx_wait = false;
 471		}
 472		kcm->tx_psock = psock;
 473		psock->tx_kcm = kcm;
 474		KCM_STATS_INCR(psock->stats.reserved);
 475	} else if (!kcm->tx_wait) {
 476		list_add_tail(&kcm->wait_psock_list,
 477			      &mux->kcm_tx_waiters);
 478		kcm->tx_wait = true;
 479	}
 480
 481	spin_unlock_bh(&mux->lock);
 482
 483	return psock;
 484}
 485
 486/* mux lock held */
 487static void psock_now_avail(struct kcm_psock *psock)
 488{
 489	struct kcm_mux *mux = psock->mux;
 490	struct kcm_sock *kcm;
 491
 492	if (list_empty(&mux->kcm_tx_waiters)) {
 493		list_add_tail(&psock->psock_avail_list,
 494			      &mux->psocks_avail);
 495	} else {
 496		kcm = list_first_entry(&mux->kcm_tx_waiters,
 497				       struct kcm_sock,
 498				       wait_psock_list);
 499		list_del(&kcm->wait_psock_list);
 500		kcm->tx_wait = false;
 501		psock->tx_kcm = kcm;
 502
 503		/* Commit before changing tx_psock since that is read in
 504		 * reserve_psock before queuing work.
 505		 */
 506		smp_mb();
 507
 508		kcm->tx_psock = psock;
 509		KCM_STATS_INCR(psock->stats.reserved);
 510		queue_work(kcm_wq, &kcm->tx_work);
 511	}
 512}
 513
 514/* kcm sock is locked. */
 515static void unreserve_psock(struct kcm_sock *kcm)
 516{
 517	struct kcm_psock *psock;
 518	struct kcm_mux *mux = kcm->mux;
 519
 520	spin_lock_bh(&mux->lock);
 521
 522	psock = kcm->tx_psock;
 523
 524	if (WARN_ON(!psock)) {
 525		spin_unlock_bh(&mux->lock);
 526		return;
 527	}
 528
 529	smp_rmb(); /* Read tx_psock before tx_wait */
 530
 531	kcm_update_tx_mux_stats(mux, psock);
 532
 533	WARN_ON(kcm->tx_wait);
 534
 535	kcm->tx_psock = NULL;
 536	psock->tx_kcm = NULL;
 537	KCM_STATS_INCR(psock->stats.unreserved);
 538
 539	if (unlikely(psock->tx_stopped)) {
 540		if (psock->done) {
 541			/* Deferred free */
 542			list_del(&psock->psock_list);
 543			mux->psocks_cnt--;
 544			sock_put(psock->sk);
 545			fput(psock->sk->sk_socket->file);
 546			kmem_cache_free(kcm_psockp, psock);
 547		}
 548
 549		/* Don't put back on available list */
 550
 551		spin_unlock_bh(&mux->lock);
 552
 553		return;
 554	}
 555
 556	psock_now_avail(psock);
 557
 558	spin_unlock_bh(&mux->lock);
 559}
 560
 561static void kcm_report_tx_retry(struct kcm_sock *kcm)
 562{
 563	struct kcm_mux *mux = kcm->mux;
 564
 565	spin_lock_bh(&mux->lock);
 566	KCM_STATS_INCR(mux->stats.tx_retries);
 567	spin_unlock_bh(&mux->lock);
 568}
 569
 570/* Write any messages ready on the kcm socket.  Called with kcm sock lock
 571 * held.  Return bytes actually sent or error.
 572 */
 573static int kcm_write_msgs(struct kcm_sock *kcm)
 574{
 575	struct sock *sk = &kcm->sk;
 576	struct kcm_psock *psock;
 577	struct sk_buff *skb, *head;
 578	struct kcm_tx_msg *txm;
 579	unsigned short fragidx, frag_offset;
 580	unsigned int sent, total_sent = 0;
 581	int ret = 0;
 582
 583	kcm->tx_wait_more = false;
 584	psock = kcm->tx_psock;
 585	if (unlikely(psock && psock->tx_stopped)) {
 586		/* A reserved psock was aborted asynchronously. Unreserve
 587		 * it and we'll retry the message.
 588		 */
 589		unreserve_psock(kcm);
 590		kcm_report_tx_retry(kcm);
 591		if (skb_queue_empty(&sk->sk_write_queue))
 592			return 0;
 593
 594		kcm_tx_msg(skb_peek(&sk->sk_write_queue))->sent = 0;
 595
 596	} else if (skb_queue_empty(&sk->sk_write_queue)) {
 597		return 0;
 598	}
 599
 600	head = skb_peek(&sk->sk_write_queue);
 601	txm = kcm_tx_msg(head);
 602
 603	if (txm->sent) {
 604		/* Send of first skbuff in queue already in progress */
 605		if (WARN_ON(!psock)) {
 606			ret = -EINVAL;
 607			goto out;
 608		}
 609		sent = txm->sent;
 610		frag_offset = txm->frag_offset;
 611		fragidx = txm->fragidx;
 612		skb = txm->frag_skb;
 613
 614		goto do_frag;
 615	}
 616
 617try_again:
 618	psock = reserve_psock(kcm);
 619	if (!psock)
 620		goto out;
 621
 622	do {
 623		skb = head;
 624		txm = kcm_tx_msg(head);
 625		sent = 0;
 626
 627do_frag_list:
 628		if (WARN_ON(!skb_shinfo(skb)->nr_frags)) {
 629			ret = -EINVAL;
 630			goto out;
 631		}
 632
 633		for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags;
 634		     fragidx++) {
 635			skb_frag_t *frag;
 636
 637			frag_offset = 0;
 638do_frag:
 639			frag = &skb_shinfo(skb)->frags[fragidx];
 640			if (WARN_ON(!skb_frag_size(frag))) {
 641				ret = -EINVAL;
 642				goto out;
 643			}
 644
 645			ret = kernel_sendpage(psock->sk->sk_socket,
 646					      skb_frag_page(frag),
 647					      skb_frag_off(frag) + frag_offset,
 648					      skb_frag_size(frag) - frag_offset,
 649					      MSG_DONTWAIT);
 650			if (ret <= 0) {
 651				if (ret == -EAGAIN) {
 652					/* Save state to try again when there's
 653					 * write space on the socket
 654					 */
 655					txm->sent = sent;
 656					txm->frag_offset = frag_offset;
 657					txm->fragidx = fragidx;
 658					txm->frag_skb = skb;
 659
 660					ret = 0;
 661					goto out;
 662				}
 663
 664				/* Hard failure in sending message, abort this
 665				 * psock since it has lost framing
 666				 * synchonization and retry sending the
 667				 * message from the beginning.
 668				 */
 669				kcm_abort_tx_psock(psock, ret ? -ret : EPIPE,
 670						   true);
 671				unreserve_psock(kcm);
 672
 673				txm->sent = 0;
 674				kcm_report_tx_retry(kcm);
 675				ret = 0;
 676
 677				goto try_again;
 678			}
 679
 680			sent += ret;
 681			frag_offset += ret;
 682			KCM_STATS_ADD(psock->stats.tx_bytes, ret);
 683			if (frag_offset < skb_frag_size(frag)) {
 684				/* Not finished with this frag */
 685				goto do_frag;
 686			}
 687		}
 688
 689		if (skb == head) {
 690			if (skb_has_frag_list(skb)) {
 691				skb = skb_shinfo(skb)->frag_list;
 692				goto do_frag_list;
 693			}
 694		} else if (skb->next) {
 695			skb = skb->next;
 696			goto do_frag_list;
 697		}
 698
 699		/* Successfully sent the whole packet, account for it. */
 700		skb_dequeue(&sk->sk_write_queue);
 701		kfree_skb(head);
 702		sk->sk_wmem_queued -= sent;
 703		total_sent += sent;
 704		KCM_STATS_INCR(psock->stats.tx_msgs);
 705	} while ((head = skb_peek(&sk->sk_write_queue)));
 706out:
 707	if (!head) {
 708		/* Done with all queued messages. */
 709		WARN_ON(!skb_queue_empty(&sk->sk_write_queue));
 710		unreserve_psock(kcm);
 711	}
 712
 713	/* Check if write space is available */
 714	sk->sk_write_space(sk);
 715
 716	return total_sent ? : ret;
 717}
 718
 719static void kcm_tx_work(struct work_struct *w)
 720{
 721	struct kcm_sock *kcm = container_of(w, struct kcm_sock, tx_work);
 722	struct sock *sk = &kcm->sk;
 723	int err;
 724
 725	lock_sock(sk);
 726
 727	/* Primarily for SOCK_DGRAM sockets, also handle asynchronous tx
 728	 * aborts
 729	 */
 730	err = kcm_write_msgs(kcm);
 731	if (err < 0) {
 732		/* Hard failure in write, report error on KCM socket */
 733		pr_warn("KCM: Hard failure on kcm_write_msgs %d\n", err);
 734		report_csk_error(&kcm->sk, -err);
 735		goto out;
 736	}
 737
 738	/* Primarily for SOCK_SEQPACKET sockets */
 739	if (likely(sk->sk_socket) &&
 740	    test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
 741		clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 742		sk->sk_write_space(sk);
 743	}
 744
 745out:
 746	release_sock(sk);
 747}
 748
 749static void kcm_push(struct kcm_sock *kcm)
 750{
 751	if (kcm->tx_wait_more)
 752		kcm_write_msgs(kcm);
 753}
 754
 755static ssize_t kcm_sendpage(struct socket *sock, struct page *page,
 756			    int offset, size_t size, int flags)
 757
 758{
 759	struct sock *sk = sock->sk;
 760	struct kcm_sock *kcm = kcm_sk(sk);
 761	struct sk_buff *skb = NULL, *head = NULL;
 762	long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
 763	bool eor;
 764	int err = 0;
 765	int i;
 766
 767	if (flags & MSG_SENDPAGE_NOTLAST)
 768		flags |= MSG_MORE;
 769
 770	/* No MSG_EOR from splice, only look at MSG_MORE */
 771	eor = !(flags & MSG_MORE);
 772
 773	lock_sock(sk);
 774
 775	sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
 776
 777	err = -EPIPE;
 778	if (sk->sk_err)
 779		goto out_error;
 780
 781	if (kcm->seq_skb) {
 782		/* Previously opened message */
 783		head = kcm->seq_skb;
 784		skb = kcm_tx_msg(head)->last_skb;
 785		i = skb_shinfo(skb)->nr_frags;
 786
 787		if (skb_can_coalesce(skb, i, page, offset)) {
 788			skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
 789			skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
 790			goto coalesced;
 791		}
 792
 793		if (i >= MAX_SKB_FRAGS) {
 794			struct sk_buff *tskb;
 795
 796			tskb = alloc_skb(0, sk->sk_allocation);
 797			while (!tskb) {
 798				kcm_push(kcm);
 799				err = sk_stream_wait_memory(sk, &timeo);
 800				if (err)
 801					goto out_error;
 802			}
 803
 804			if (head == skb)
 805				skb_shinfo(head)->frag_list = tskb;
 806			else
 807				skb->next = tskb;
 808
 809			skb = tskb;
 810			skb->ip_summed = CHECKSUM_UNNECESSARY;
 811			i = 0;
 812		}
 813	} else {
 814		/* Call the sk_stream functions to manage the sndbuf mem. */
 815		if (!sk_stream_memory_free(sk)) {
 816			kcm_push(kcm);
 817			set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 818			err = sk_stream_wait_memory(sk, &timeo);
 819			if (err)
 820				goto out_error;
 821		}
 822
 823		head = alloc_skb(0, sk->sk_allocation);
 824		while (!head) {
 825			kcm_push(kcm);
 826			err = sk_stream_wait_memory(sk, &timeo);
 827			if (err)
 828				goto out_error;
 829		}
 830
 831		skb = head;
 832		i = 0;
 833	}
 834
 835	get_page(page);
 836	skb_fill_page_desc(skb, i, page, offset, size);
 837	skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
 838
 839coalesced:
 840	skb->len += size;
 841	skb->data_len += size;
 842	skb->truesize += size;
 843	sk->sk_wmem_queued += size;
 844	sk_mem_charge(sk, size);
 845
 846	if (head != skb) {
 847		head->len += size;
 848		head->data_len += size;
 849		head->truesize += size;
 850	}
 851
 852	if (eor) {
 853		bool not_busy = skb_queue_empty(&sk->sk_write_queue);
 854
 855		/* Message complete, queue it on send buffer */
 856		__skb_queue_tail(&sk->sk_write_queue, head);
 857		kcm->seq_skb = NULL;
 858		KCM_STATS_INCR(kcm->stats.tx_msgs);
 859
 860		if (flags & MSG_BATCH) {
 861			kcm->tx_wait_more = true;
 862		} else if (kcm->tx_wait_more || not_busy) {
 863			err = kcm_write_msgs(kcm);
 864			if (err < 0) {
 865				/* We got a hard error in write_msgs but have
 866				 * already queued this message. Report an error
 867				 * in the socket, but don't affect return value
 868				 * from sendmsg
 869				 */
 870				pr_warn("KCM: Hard failure on kcm_write_msgs\n");
 871				report_csk_error(&kcm->sk, -err);
 872			}
 873		}
 874	} else {
 875		/* Message not complete, save state */
 876		kcm->seq_skb = head;
 877		kcm_tx_msg(head)->last_skb = skb;
 878	}
 879
 880	KCM_STATS_ADD(kcm->stats.tx_bytes, size);
 881
 882	release_sock(sk);
 883	return size;
 884
 885out_error:
 886	kcm_push(kcm);
 887
 888	err = sk_stream_error(sk, flags, err);
 889
 890	/* make sure we wake any epoll edge trigger waiter */
 891	if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 && err == -EAGAIN))
 892		sk->sk_write_space(sk);
 893
 894	release_sock(sk);
 895	return err;
 896}
 897
 898static int kcm_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
 899{
 900	struct sock *sk = sock->sk;
 901	struct kcm_sock *kcm = kcm_sk(sk);
 902	struct sk_buff *skb = NULL, *head = NULL;
 903	size_t copy, copied = 0;
 904	long timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
 905	int eor = (sock->type == SOCK_DGRAM) ?
 906		  !(msg->msg_flags & MSG_MORE) : !!(msg->msg_flags & MSG_EOR);
 907	int err = -EPIPE;
 908
 909	lock_sock(sk);
 910
 911	/* Per tcp_sendmsg this should be in poll */
 912	sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
 913
 914	if (sk->sk_err)
 915		goto out_error;
 916
 917	if (kcm->seq_skb) {
 918		/* Previously opened message */
 919		head = kcm->seq_skb;
 920		skb = kcm_tx_msg(head)->last_skb;
 921		goto start;
 922	}
 923
 924	/* Call the sk_stream functions to manage the sndbuf mem. */
 925	if (!sk_stream_memory_free(sk)) {
 926		kcm_push(kcm);
 927		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 928		err = sk_stream_wait_memory(sk, &timeo);
 929		if (err)
 930			goto out_error;
 931	}
 932
 933	if (msg_data_left(msg)) {
 934		/* New message, alloc head skb */
 935		head = alloc_skb(0, sk->sk_allocation);
 936		while (!head) {
 937			kcm_push(kcm);
 938			err = sk_stream_wait_memory(sk, &timeo);
 939			if (err)
 940				goto out_error;
 941
 942			head = alloc_skb(0, sk->sk_allocation);
 943		}
 944
 945		skb = head;
 946
 947		/* Set ip_summed to CHECKSUM_UNNECESSARY to avoid calling
 948		 * csum_and_copy_from_iter from skb_do_copy_data_nocache.
 949		 */
 950		skb->ip_summed = CHECKSUM_UNNECESSARY;
 951	}
 952
 953start:
 954	while (msg_data_left(msg)) {
 955		bool merge = true;
 956		int i = skb_shinfo(skb)->nr_frags;
 957		struct page_frag *pfrag = sk_page_frag(sk);
 958
 959		if (!sk_page_frag_refill(sk, pfrag))
 960			goto wait_for_memory;
 961
 962		if (!skb_can_coalesce(skb, i, pfrag->page,
 963				      pfrag->offset)) {
 964			if (i == MAX_SKB_FRAGS) {
 965				struct sk_buff *tskb;
 966
 967				tskb = alloc_skb(0, sk->sk_allocation);
 968				if (!tskb)
 969					goto wait_for_memory;
 970
 971				if (head == skb)
 972					skb_shinfo(head)->frag_list = tskb;
 973				else
 974					skb->next = tskb;
 975
 976				skb = tskb;
 977				skb->ip_summed = CHECKSUM_UNNECESSARY;
 978				continue;
 979			}
 980			merge = false;
 981		}
 982
 983		copy = min_t(int, msg_data_left(msg),
 984			     pfrag->size - pfrag->offset);
 985
 986		if (!sk_wmem_schedule(sk, copy))
 987			goto wait_for_memory;
 988
 989		err = skb_copy_to_page_nocache(sk, &msg->msg_iter, skb,
 990					       pfrag->page,
 991					       pfrag->offset,
 992					       copy);
 993		if (err)
 994			goto out_error;
 995
 996		/* Update the skb. */
 997		if (merge) {
 998			skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
 999		} else {
1000			skb_fill_page_desc(skb, i, pfrag->page,
1001					   pfrag->offset, copy);
1002			get_page(pfrag->page);
1003		}
1004
1005		pfrag->offset += copy;
1006		copied += copy;
1007		if (head != skb) {
1008			head->len += copy;
1009			head->data_len += copy;
1010		}
1011
1012		continue;
1013
1014wait_for_memory:
1015		kcm_push(kcm);
1016		err = sk_stream_wait_memory(sk, &timeo);
1017		if (err)
1018			goto out_error;
1019	}
1020
1021	if (eor) {
1022		bool not_busy = skb_queue_empty(&sk->sk_write_queue);
1023
1024		if (head) {
1025			/* Message complete, queue it on send buffer */
1026			__skb_queue_tail(&sk->sk_write_queue, head);
1027			kcm->seq_skb = NULL;
1028			KCM_STATS_INCR(kcm->stats.tx_msgs);
1029		}
1030
1031		if (msg->msg_flags & MSG_BATCH) {
1032			kcm->tx_wait_more = true;
1033		} else if (kcm->tx_wait_more || not_busy) {
1034			err = kcm_write_msgs(kcm);
1035			if (err < 0) {
1036				/* We got a hard error in write_msgs but have
1037				 * already queued this message. Report an error
1038				 * in the socket, but don't affect return value
1039				 * from sendmsg
1040				 */
1041				pr_warn("KCM: Hard failure on kcm_write_msgs\n");
1042				report_csk_error(&kcm->sk, -err);
1043			}
1044		}
1045	} else {
1046		/* Message not complete, save state */
1047partial_message:
1048		if (head) {
1049			kcm->seq_skb = head;
1050			kcm_tx_msg(head)->last_skb = skb;
1051		}
1052	}
1053
1054	KCM_STATS_ADD(kcm->stats.tx_bytes, copied);
1055
1056	release_sock(sk);
1057	return copied;
1058
1059out_error:
1060	kcm_push(kcm);
1061
1062	if (copied && sock->type == SOCK_SEQPACKET) {
1063		/* Wrote some bytes before encountering an
1064		 * error, return partial success.
1065		 */
1066		goto partial_message;
1067	}
1068
1069	if (head != kcm->seq_skb)
1070		kfree_skb(head);
1071
1072	err = sk_stream_error(sk, msg->msg_flags, err);
1073
1074	/* make sure we wake any epoll edge trigger waiter */
1075	if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 && err == -EAGAIN))
1076		sk->sk_write_space(sk);
1077
1078	release_sock(sk);
1079	return err;
1080}
1081
1082static struct sk_buff *kcm_wait_data(struct sock *sk, int flags,
1083				     long timeo, int *err)
1084{
1085	struct sk_buff *skb;
1086
1087	while (!(skb = skb_peek(&sk->sk_receive_queue))) {
1088		if (sk->sk_err) {
1089			*err = sock_error(sk);
1090			return NULL;
1091		}
1092
1093		if (sock_flag(sk, SOCK_DONE))
1094			return NULL;
1095
1096		if ((flags & MSG_DONTWAIT) || !timeo) {
1097			*err = -EAGAIN;
1098			return NULL;
1099		}
1100
1101		sk_wait_data(sk, &timeo, NULL);
1102
1103		/* Handle signals */
1104		if (signal_pending(current)) {
1105			*err = sock_intr_errno(timeo);
1106			return NULL;
1107		}
1108	}
1109
1110	return skb;
1111}
1112
1113static int kcm_recvmsg(struct socket *sock, struct msghdr *msg,
1114		       size_t len, int flags)
1115{
1116	struct sock *sk = sock->sk;
1117	struct kcm_sock *kcm = kcm_sk(sk);
1118	int err = 0;
1119	long timeo;
1120	struct strp_msg *stm;
1121	int copied = 0;
1122	struct sk_buff *skb;
1123
1124	timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1125
1126	lock_sock(sk);
1127
1128	skb = kcm_wait_data(sk, flags, timeo, &err);
1129	if (!skb)
1130		goto out;
1131
1132	/* Okay, have a message on the receive queue */
1133
1134	stm = strp_msg(skb);
1135
1136	if (len > stm->full_len)
1137		len = stm->full_len;
1138
1139	err = skb_copy_datagram_msg(skb, stm->offset, msg, len);
1140	if (err < 0)
1141		goto out;
1142
1143	copied = len;
1144	if (likely(!(flags & MSG_PEEK))) {
1145		KCM_STATS_ADD(kcm->stats.rx_bytes, copied);
1146		if (copied < stm->full_len) {
1147			if (sock->type == SOCK_DGRAM) {
1148				/* Truncated message */
1149				msg->msg_flags |= MSG_TRUNC;
1150				goto msg_finished;
1151			}
1152			stm->offset += copied;
1153			stm->full_len -= copied;
1154		} else {
1155msg_finished:
1156			/* Finished with message */
1157			msg->msg_flags |= MSG_EOR;
1158			KCM_STATS_INCR(kcm->stats.rx_msgs);
1159			skb_unlink(skb, &sk->sk_receive_queue);
1160			kfree_skb(skb);
1161		}
1162	}
1163
1164out:
1165	release_sock(sk);
1166
1167	return copied ? : err;
1168}
1169
1170static ssize_t kcm_splice_read(struct socket *sock, loff_t *ppos,
1171			       struct pipe_inode_info *pipe, size_t len,
1172			       unsigned int flags)
1173{
1174	struct sock *sk = sock->sk;
1175	struct kcm_sock *kcm = kcm_sk(sk);
1176	long timeo;
1177	struct strp_msg *stm;
1178	int err = 0;
1179	ssize_t copied;
1180	struct sk_buff *skb;
1181
1182	/* Only support splice for SOCKSEQPACKET */
1183
1184	timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1185
1186	lock_sock(sk);
1187
1188	skb = kcm_wait_data(sk, flags, timeo, &err);
1189	if (!skb)
1190		goto err_out;
1191
1192	/* Okay, have a message on the receive queue */
1193
1194	stm = strp_msg(skb);
1195
1196	if (len > stm->full_len)
1197		len = stm->full_len;
1198
1199	copied = skb_splice_bits(skb, sk, stm->offset, pipe, len, flags);
1200	if (copied < 0) {
1201		err = copied;
1202		goto err_out;
1203	}
1204
1205	KCM_STATS_ADD(kcm->stats.rx_bytes, copied);
1206
1207	stm->offset += copied;
1208	stm->full_len -= copied;
1209
1210	/* We have no way to return MSG_EOR. If all the bytes have been
1211	 * read we still leave the message in the receive socket buffer.
1212	 * A subsequent recvmsg needs to be done to return MSG_EOR and
1213	 * finish reading the message.
1214	 */
1215
1216	release_sock(sk);
1217
1218	return copied;
1219
1220err_out:
1221	release_sock(sk);
1222
1223	return err;
1224}
1225
1226/* kcm sock lock held */
1227static void kcm_recv_disable(struct kcm_sock *kcm)
1228{
1229	struct kcm_mux *mux = kcm->mux;
1230
1231	if (kcm->rx_disabled)
1232		return;
1233
1234	spin_lock_bh(&mux->rx_lock);
1235
1236	kcm->rx_disabled = 1;
1237
1238	/* If a psock is reserved we'll do cleanup in unreserve */
1239	if (!kcm->rx_psock) {
1240		if (kcm->rx_wait) {
1241			list_del(&kcm->wait_rx_list);
1242			kcm->rx_wait = false;
1243		}
1244
1245		requeue_rx_msgs(mux, &kcm->sk.sk_receive_queue);
1246	}
1247
1248	spin_unlock_bh(&mux->rx_lock);
1249}
1250
1251/* kcm sock lock held */
1252static void kcm_recv_enable(struct kcm_sock *kcm)
1253{
1254	struct kcm_mux *mux = kcm->mux;
1255
1256	if (!kcm->rx_disabled)
1257		return;
1258
1259	spin_lock_bh(&mux->rx_lock);
1260
1261	kcm->rx_disabled = 0;
1262	kcm_rcv_ready(kcm);
1263
1264	spin_unlock_bh(&mux->rx_lock);
1265}
1266
1267static int kcm_setsockopt(struct socket *sock, int level, int optname,
1268			  sockptr_t optval, unsigned int optlen)
1269{
1270	struct kcm_sock *kcm = kcm_sk(sock->sk);
1271	int val, valbool;
1272	int err = 0;
1273
1274	if (level != SOL_KCM)
1275		return -ENOPROTOOPT;
1276
1277	if (optlen < sizeof(int))
1278		return -EINVAL;
1279
1280	if (copy_from_sockptr(&val, optval, sizeof(int)))
1281		return -EFAULT;
1282
1283	valbool = val ? 1 : 0;
1284
1285	switch (optname) {
1286	case KCM_RECV_DISABLE:
1287		lock_sock(&kcm->sk);
1288		if (valbool)
1289			kcm_recv_disable(kcm);
1290		else
1291			kcm_recv_enable(kcm);
1292		release_sock(&kcm->sk);
1293		break;
1294	default:
1295		err = -ENOPROTOOPT;
1296	}
1297
1298	return err;
1299}
1300
1301static int kcm_getsockopt(struct socket *sock, int level, int optname,
1302			  char __user *optval, int __user *optlen)
1303{
1304	struct kcm_sock *kcm = kcm_sk(sock->sk);
1305	int val, len;
1306
1307	if (level != SOL_KCM)
1308		return -ENOPROTOOPT;
1309
1310	if (get_user(len, optlen))
1311		return -EFAULT;
1312
1313	len = min_t(unsigned int, len, sizeof(int));
1314	if (len < 0)
1315		return -EINVAL;
1316
1317	switch (optname) {
1318	case KCM_RECV_DISABLE:
1319		val = kcm->rx_disabled;
1320		break;
1321	default:
1322		return -ENOPROTOOPT;
1323	}
1324
1325	if (put_user(len, optlen))
1326		return -EFAULT;
1327	if (copy_to_user(optval, &val, len))
1328		return -EFAULT;
1329	return 0;
1330}
1331
1332static void init_kcm_sock(struct kcm_sock *kcm, struct kcm_mux *mux)
1333{
1334	struct kcm_sock *tkcm;
1335	struct list_head *head;
1336	int index = 0;
1337
1338	/* For SOCK_SEQPACKET sock type, datagram_poll checks the sk_state, so
1339	 * we set sk_state, otherwise epoll_wait always returns right away with
1340	 * EPOLLHUP
1341	 */
1342	kcm->sk.sk_state = TCP_ESTABLISHED;
1343
1344	/* Add to mux's kcm sockets list */
1345	kcm->mux = mux;
1346	spin_lock_bh(&mux->lock);
1347
1348	head = &mux->kcm_socks;
1349	list_for_each_entry(tkcm, &mux->kcm_socks, kcm_sock_list) {
1350		if (tkcm->index != index)
1351			break;
1352		head = &tkcm->kcm_sock_list;
1353		index++;
1354	}
1355
1356	list_add(&kcm->kcm_sock_list, head);
1357	kcm->index = index;
1358
1359	mux->kcm_socks_cnt++;
1360	spin_unlock_bh(&mux->lock);
1361
1362	INIT_WORK(&kcm->tx_work, kcm_tx_work);
1363
1364	spin_lock_bh(&mux->rx_lock);
1365	kcm_rcv_ready(kcm);
1366	spin_unlock_bh(&mux->rx_lock);
1367}
1368
1369static int kcm_attach(struct socket *sock, struct socket *csock,
1370		      struct bpf_prog *prog)
1371{
1372	struct kcm_sock *kcm = kcm_sk(sock->sk);
1373	struct kcm_mux *mux = kcm->mux;
1374	struct sock *csk;
1375	struct kcm_psock *psock = NULL, *tpsock;
1376	struct list_head *head;
1377	int index = 0;
1378	static const struct strp_callbacks cb = {
1379		.rcv_msg = kcm_rcv_strparser,
1380		.parse_msg = kcm_parse_func_strparser,
1381		.read_sock_done = kcm_read_sock_done,
1382	};
1383	int err = 0;
1384
1385	csk = csock->sk;
1386	if (!csk)
1387		return -EINVAL;
1388
1389	lock_sock(csk);
1390
1391	/* Only allow TCP sockets to be attached for now */
1392	if ((csk->sk_family != AF_INET && csk->sk_family != AF_INET6) ||
1393	    csk->sk_protocol != IPPROTO_TCP) {
1394		err = -EOPNOTSUPP;
1395		goto out;
1396	}
1397
1398	/* Don't allow listeners or closed sockets */
1399	if (csk->sk_state == TCP_LISTEN || csk->sk_state == TCP_CLOSE) {
1400		err = -EOPNOTSUPP;
1401		goto out;
1402	}
1403
1404	psock = kmem_cache_zalloc(kcm_psockp, GFP_KERNEL);
1405	if (!psock) {
1406		err = -ENOMEM;
1407		goto out;
1408	}
1409
1410	psock->mux = mux;
1411	psock->sk = csk;
1412	psock->bpf_prog = prog;
1413
1414	err = strp_init(&psock->strp, csk, &cb);
1415	if (err) {
1416		kmem_cache_free(kcm_psockp, psock);
1417		goto out;
1418	}
1419
1420	write_lock_bh(&csk->sk_callback_lock);
1421
1422	/* Check if sk_user_data is aready by KCM or someone else.
1423	 * Must be done under lock to prevent race conditions.
1424	 */
1425	if (csk->sk_user_data) {
1426		write_unlock_bh(&csk->sk_callback_lock);
1427		strp_stop(&psock->strp);
1428		strp_done(&psock->strp);
1429		kmem_cache_free(kcm_psockp, psock);
1430		err = -EALREADY;
1431		goto out;
1432	}
1433
1434	psock->save_data_ready = csk->sk_data_ready;
1435	psock->save_write_space = csk->sk_write_space;
1436	psock->save_state_change = csk->sk_state_change;
1437	csk->sk_user_data = psock;
1438	csk->sk_data_ready = psock_data_ready;
1439	csk->sk_write_space = psock_write_space;
1440	csk->sk_state_change = psock_state_change;
1441
1442	write_unlock_bh(&csk->sk_callback_lock);
1443
1444	sock_hold(csk);
1445
1446	/* Finished initialization, now add the psock to the MUX. */
1447	spin_lock_bh(&mux->lock);
1448	head = &mux->psocks;
1449	list_for_each_entry(tpsock, &mux->psocks, psock_list) {
1450		if (tpsock->index != index)
1451			break;
1452		head = &tpsock->psock_list;
1453		index++;
1454	}
1455
1456	list_add(&psock->psock_list, head);
1457	psock->index = index;
1458
1459	KCM_STATS_INCR(mux->stats.psock_attach);
1460	mux->psocks_cnt++;
1461	psock_now_avail(psock);
1462	spin_unlock_bh(&mux->lock);
1463
1464	/* Schedule RX work in case there are already bytes queued */
1465	strp_check_rcv(&psock->strp);
1466
1467out:
1468	release_sock(csk);
1469
1470	return err;
1471}
1472
1473static int kcm_attach_ioctl(struct socket *sock, struct kcm_attach *info)
1474{
1475	struct socket *csock;
1476	struct bpf_prog *prog;
1477	int err;
1478
1479	csock = sockfd_lookup(info->fd, &err);
1480	if (!csock)
1481		return -ENOENT;
1482
1483	prog = bpf_prog_get_type(info->bpf_fd, BPF_PROG_TYPE_SOCKET_FILTER);
1484	if (IS_ERR(prog)) {
1485		err = PTR_ERR(prog);
1486		goto out;
1487	}
1488
1489	err = kcm_attach(sock, csock, prog);
1490	if (err) {
1491		bpf_prog_put(prog);
1492		goto out;
1493	}
1494
1495	/* Keep reference on file also */
1496
1497	return 0;
1498out:
1499	fput(csock->file);
1500	return err;
1501}
1502
1503static void kcm_unattach(struct kcm_psock *psock)
1504{
1505	struct sock *csk = psock->sk;
1506	struct kcm_mux *mux = psock->mux;
1507
1508	lock_sock(csk);
1509
1510	/* Stop getting callbacks from TCP socket. After this there should
1511	 * be no way to reserve a kcm for this psock.
1512	 */
1513	write_lock_bh(&csk->sk_callback_lock);
1514	csk->sk_user_data = NULL;
1515	csk->sk_data_ready = psock->save_data_ready;
1516	csk->sk_write_space = psock->save_write_space;
1517	csk->sk_state_change = psock->save_state_change;
1518	strp_stop(&psock->strp);
1519
1520	if (WARN_ON(psock->rx_kcm)) {
1521		write_unlock_bh(&csk->sk_callback_lock);
1522		release_sock(csk);
1523		return;
1524	}
1525
1526	spin_lock_bh(&mux->rx_lock);
1527
1528	/* Stop receiver activities. After this point psock should not be
1529	 * able to get onto ready list either through callbacks or work.
1530	 */
1531	if (psock->ready_rx_msg) {
1532		list_del(&psock->psock_ready_list);
1533		kfree_skb(psock->ready_rx_msg);
1534		psock->ready_rx_msg = NULL;
1535		KCM_STATS_INCR(mux->stats.rx_ready_drops);
1536	}
1537
1538	spin_unlock_bh(&mux->rx_lock);
1539
1540	write_unlock_bh(&csk->sk_callback_lock);
1541
1542	/* Call strp_done without sock lock */
1543	release_sock(csk);
1544	strp_done(&psock->strp);
1545	lock_sock(csk);
1546
1547	bpf_prog_put(psock->bpf_prog);
1548
1549	spin_lock_bh(&mux->lock);
1550
1551	aggregate_psock_stats(&psock->stats, &mux->aggregate_psock_stats);
1552	save_strp_stats(&psock->strp, &mux->aggregate_strp_stats);
1553
1554	KCM_STATS_INCR(mux->stats.psock_unattach);
1555
1556	if (psock->tx_kcm) {
1557		/* psock was reserved.  Just mark it finished and we will clean
1558		 * up in the kcm paths, we need kcm lock which can not be
1559		 * acquired here.
1560		 */
1561		KCM_STATS_INCR(mux->stats.psock_unattach_rsvd);
1562		spin_unlock_bh(&mux->lock);
1563
1564		/* We are unattaching a socket that is reserved. Abort the
1565		 * socket since we may be out of sync in sending on it. We need
1566		 * to do this without the mux lock.
1567		 */
1568		kcm_abort_tx_psock(psock, EPIPE, false);
1569
1570		spin_lock_bh(&mux->lock);
1571		if (!psock->tx_kcm) {
1572			/* psock now unreserved in window mux was unlocked */
1573			goto no_reserved;
1574		}
1575		psock->done = 1;
1576
1577		/* Commit done before queuing work to process it */
1578		smp_mb();
1579
1580		/* Queue tx work to make sure psock->done is handled */
1581		queue_work(kcm_wq, &psock->tx_kcm->tx_work);
1582		spin_unlock_bh(&mux->lock);
1583	} else {
1584no_reserved:
1585		if (!psock->tx_stopped)
1586			list_del(&psock->psock_avail_list);
1587		list_del(&psock->psock_list);
1588		mux->psocks_cnt--;
1589		spin_unlock_bh(&mux->lock);
1590
1591		sock_put(csk);
1592		fput(csk->sk_socket->file);
1593		kmem_cache_free(kcm_psockp, psock);
1594	}
1595
1596	release_sock(csk);
1597}
1598
1599static int kcm_unattach_ioctl(struct socket *sock, struct kcm_unattach *info)
1600{
1601	struct kcm_sock *kcm = kcm_sk(sock->sk);
1602	struct kcm_mux *mux = kcm->mux;
1603	struct kcm_psock *psock;
1604	struct socket *csock;
1605	struct sock *csk;
1606	int err;
1607
1608	csock = sockfd_lookup(info->fd, &err);
1609	if (!csock)
1610		return -ENOENT;
1611
1612	csk = csock->sk;
1613	if (!csk) {
1614		err = -EINVAL;
1615		goto out;
1616	}
1617
1618	err = -ENOENT;
1619
1620	spin_lock_bh(&mux->lock);
1621
1622	list_for_each_entry(psock, &mux->psocks, psock_list) {
1623		if (psock->sk != csk)
1624			continue;
1625
1626		/* Found the matching psock */
1627
1628		if (psock->unattaching || WARN_ON(psock->done)) {
1629			err = -EALREADY;
1630			break;
1631		}
1632
1633		psock->unattaching = 1;
1634
1635		spin_unlock_bh(&mux->lock);
1636
1637		/* Lower socket lock should already be held */
1638		kcm_unattach(psock);
1639
1640		err = 0;
1641		goto out;
1642	}
1643
1644	spin_unlock_bh(&mux->lock);
1645
1646out:
1647	fput(csock->file);
1648	return err;
1649}
1650
1651static struct proto kcm_proto = {
1652	.name	= "KCM",
1653	.owner	= THIS_MODULE,
1654	.obj_size = sizeof(struct kcm_sock),
1655};
1656
1657/* Clone a kcm socket. */
1658static struct file *kcm_clone(struct socket *osock)
1659{
1660	struct socket *newsock;
1661	struct sock *newsk;
1662
1663	newsock = sock_alloc();
1664	if (!newsock)
1665		return ERR_PTR(-ENFILE);
1666
1667	newsock->type = osock->type;
1668	newsock->ops = osock->ops;
1669
1670	__module_get(newsock->ops->owner);
1671
1672	newsk = sk_alloc(sock_net(osock->sk), PF_KCM, GFP_KERNEL,
1673			 &kcm_proto, false);
1674	if (!newsk) {
1675		sock_release(newsock);
1676		return ERR_PTR(-ENOMEM);
1677	}
1678	sock_init_data(newsock, newsk);
1679	init_kcm_sock(kcm_sk(newsk), kcm_sk(osock->sk)->mux);
1680
1681	return sock_alloc_file(newsock, 0, osock->sk->sk_prot_creator->name);
1682}
1683
1684static int kcm_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1685{
1686	int err;
1687
1688	switch (cmd) {
1689	case SIOCKCMATTACH: {
1690		struct kcm_attach info;
1691
1692		if (copy_from_user(&info, (void __user *)arg, sizeof(info)))
1693			return -EFAULT;
1694
1695		err = kcm_attach_ioctl(sock, &info);
1696
1697		break;
1698	}
1699	case SIOCKCMUNATTACH: {
1700		struct kcm_unattach info;
1701
1702		if (copy_from_user(&info, (void __user *)arg, sizeof(info)))
1703			return -EFAULT;
1704
1705		err = kcm_unattach_ioctl(sock, &info);
1706
1707		break;
1708	}
1709	case SIOCKCMCLONE: {
1710		struct kcm_clone info;
1711		struct file *file;
1712
1713		info.fd = get_unused_fd_flags(0);
1714		if (unlikely(info.fd < 0))
1715			return info.fd;
1716
1717		file = kcm_clone(sock);
1718		if (IS_ERR(file)) {
1719			put_unused_fd(info.fd);
1720			return PTR_ERR(file);
1721		}
1722		if (copy_to_user((void __user *)arg, &info,
1723				 sizeof(info))) {
1724			put_unused_fd(info.fd);
1725			fput(file);
1726			return -EFAULT;
1727		}
1728		fd_install(info.fd, file);
1729		err = 0;
1730		break;
1731	}
1732	default:
1733		err = -ENOIOCTLCMD;
1734		break;
1735	}
1736
1737	return err;
1738}
1739
1740static void free_mux(struct rcu_head *rcu)
1741{
1742	struct kcm_mux *mux = container_of(rcu,
1743	    struct kcm_mux, rcu);
1744
1745	kmem_cache_free(kcm_muxp, mux);
1746}
1747
1748static void release_mux(struct kcm_mux *mux)
1749{
1750	struct kcm_net *knet = mux->knet;
1751	struct kcm_psock *psock, *tmp_psock;
1752
1753	/* Release psocks */
1754	list_for_each_entry_safe(psock, tmp_psock,
1755				 &mux->psocks, psock_list) {
1756		if (!WARN_ON(psock->unattaching))
1757			kcm_unattach(psock);
1758	}
1759
1760	if (WARN_ON(mux->psocks_cnt))
1761		return;
1762
1763	__skb_queue_purge(&mux->rx_hold_queue);
1764
1765	mutex_lock(&knet->mutex);
1766	aggregate_mux_stats(&mux->stats, &knet->aggregate_mux_stats);
1767	aggregate_psock_stats(&mux->aggregate_psock_stats,
1768			      &knet->aggregate_psock_stats);
1769	aggregate_strp_stats(&mux->aggregate_strp_stats,
1770			     &knet->aggregate_strp_stats);
1771	list_del_rcu(&mux->kcm_mux_list);
1772	knet->count--;
1773	mutex_unlock(&knet->mutex);
1774
1775	call_rcu(&mux->rcu, free_mux);
1776}
1777
1778static void kcm_done(struct kcm_sock *kcm)
1779{
1780	struct kcm_mux *mux = kcm->mux;
1781	struct sock *sk = &kcm->sk;
1782	int socks_cnt;
1783
1784	spin_lock_bh(&mux->rx_lock);
1785	if (kcm->rx_psock) {
1786		/* Cleanup in unreserve_rx_kcm */
1787		WARN_ON(kcm->done);
1788		kcm->rx_disabled = 1;
1789		kcm->done = 1;
1790		spin_unlock_bh(&mux->rx_lock);
1791		return;
1792	}
1793
1794	if (kcm->rx_wait) {
1795		list_del(&kcm->wait_rx_list);
1796		kcm->rx_wait = false;
1797	}
1798	/* Move any pending receive messages to other kcm sockets */
1799	requeue_rx_msgs(mux, &sk->sk_receive_queue);
1800
1801	spin_unlock_bh(&mux->rx_lock);
1802
1803	if (WARN_ON(sk_rmem_alloc_get(sk)))
1804		return;
1805
1806	/* Detach from MUX */
1807	spin_lock_bh(&mux->lock);
1808
1809	list_del(&kcm->kcm_sock_list);
1810	mux->kcm_socks_cnt--;
1811	socks_cnt = mux->kcm_socks_cnt;
1812
1813	spin_unlock_bh(&mux->lock);
1814
1815	if (!socks_cnt) {
1816		/* We are done with the mux now. */
1817		release_mux(mux);
1818	}
1819
1820	WARN_ON(kcm->rx_wait);
1821
1822	sock_put(&kcm->sk);
1823}
1824
1825/* Called by kcm_release to close a KCM socket.
1826 * If this is the last KCM socket on the MUX, destroy the MUX.
1827 */
1828static int kcm_release(struct socket *sock)
1829{
1830	struct sock *sk = sock->sk;
1831	struct kcm_sock *kcm;
1832	struct kcm_mux *mux;
1833	struct kcm_psock *psock;
1834
1835	if (!sk)
1836		return 0;
1837
1838	kcm = kcm_sk(sk);
1839	mux = kcm->mux;
1840
1841	sock_orphan(sk);
1842	kfree_skb(kcm->seq_skb);
1843
1844	lock_sock(sk);
1845	/* Purge queue under lock to avoid race condition with tx_work trying
1846	 * to act when queue is nonempty. If tx_work runs after this point
1847	 * it will just return.
1848	 */
1849	__skb_queue_purge(&sk->sk_write_queue);
1850
1851	/* Set tx_stopped. This is checked when psock is bound to a kcm and we
1852	 * get a writespace callback. This prevents further work being queued
1853	 * from the callback (unbinding the psock occurs after canceling work.
1854	 */
1855	kcm->tx_stopped = 1;
1856
1857	release_sock(sk);
1858
1859	spin_lock_bh(&mux->lock);
1860	if (kcm->tx_wait) {
1861		/* Take of tx_wait list, after this point there should be no way
1862		 * that a psock will be assigned to this kcm.
1863		 */
1864		list_del(&kcm->wait_psock_list);
1865		kcm->tx_wait = false;
1866	}
1867	spin_unlock_bh(&mux->lock);
1868
1869	/* Cancel work. After this point there should be no outside references
1870	 * to the kcm socket.
1871	 */
1872	cancel_work_sync(&kcm->tx_work);
1873
1874	lock_sock(sk);
1875	psock = kcm->tx_psock;
1876	if (psock) {
1877		/* A psock was reserved, so we need to kill it since it
1878		 * may already have some bytes queued from a message. We
1879		 * need to do this after removing kcm from tx_wait list.
1880		 */
1881		kcm_abort_tx_psock(psock, EPIPE, false);
1882		unreserve_psock(kcm);
1883	}
1884	release_sock(sk);
1885
1886	WARN_ON(kcm->tx_wait);
1887	WARN_ON(kcm->tx_psock);
1888
1889	sock->sk = NULL;
1890
1891	kcm_done(kcm);
1892
1893	return 0;
1894}
1895
1896static const struct proto_ops kcm_dgram_ops = {
1897	.family =	PF_KCM,
1898	.owner =	THIS_MODULE,
1899	.release =	kcm_release,
1900	.bind =		sock_no_bind,
1901	.connect =	sock_no_connect,
1902	.socketpair =	sock_no_socketpair,
1903	.accept =	sock_no_accept,
1904	.getname =	sock_no_getname,
1905	.poll =		datagram_poll,
1906	.ioctl =	kcm_ioctl,
1907	.listen =	sock_no_listen,
1908	.shutdown =	sock_no_shutdown,
1909	.setsockopt =	kcm_setsockopt,
1910	.getsockopt =	kcm_getsockopt,
1911	.sendmsg =	kcm_sendmsg,
1912	.recvmsg =	kcm_recvmsg,
1913	.mmap =		sock_no_mmap,
1914	.sendpage =	kcm_sendpage,
1915};
1916
1917static const struct proto_ops kcm_seqpacket_ops = {
1918	.family =	PF_KCM,
1919	.owner =	THIS_MODULE,
1920	.release =	kcm_release,
1921	.bind =		sock_no_bind,
1922	.connect =	sock_no_connect,
1923	.socketpair =	sock_no_socketpair,
1924	.accept =	sock_no_accept,
1925	.getname =	sock_no_getname,
1926	.poll =		datagram_poll,
1927	.ioctl =	kcm_ioctl,
1928	.listen =	sock_no_listen,
1929	.shutdown =	sock_no_shutdown,
1930	.setsockopt =	kcm_setsockopt,
1931	.getsockopt =	kcm_getsockopt,
1932	.sendmsg =	kcm_sendmsg,
1933	.recvmsg =	kcm_recvmsg,
1934	.mmap =		sock_no_mmap,
1935	.sendpage =	kcm_sendpage,
1936	.splice_read =	kcm_splice_read,
1937};
1938
1939/* Create proto operation for kcm sockets */
1940static int kcm_create(struct net *net, struct socket *sock,
1941		      int protocol, int kern)
1942{
1943	struct kcm_net *knet = net_generic(net, kcm_net_id);
1944	struct sock *sk;
1945	struct kcm_mux *mux;
1946
1947	switch (sock->type) {
1948	case SOCK_DGRAM:
1949		sock->ops = &kcm_dgram_ops;
1950		break;
1951	case SOCK_SEQPACKET:
1952		sock->ops = &kcm_seqpacket_ops;
1953		break;
1954	default:
1955		return -ESOCKTNOSUPPORT;
1956	}
1957
1958	if (protocol != KCMPROTO_CONNECTED)
1959		return -EPROTONOSUPPORT;
1960
1961	sk = sk_alloc(net, PF_KCM, GFP_KERNEL, &kcm_proto, kern);
1962	if (!sk)
1963		return -ENOMEM;
1964
1965	/* Allocate a kcm mux, shared between KCM sockets */
1966	mux = kmem_cache_zalloc(kcm_muxp, GFP_KERNEL);
1967	if (!mux) {
1968		sk_free(sk);
1969		return -ENOMEM;
1970	}
1971
1972	spin_lock_init(&mux->lock);
1973	spin_lock_init(&mux->rx_lock);
1974	INIT_LIST_HEAD(&mux->kcm_socks);
1975	INIT_LIST_HEAD(&mux->kcm_rx_waiters);
1976	INIT_LIST_HEAD(&mux->kcm_tx_waiters);
1977
1978	INIT_LIST_HEAD(&mux->psocks);
1979	INIT_LIST_HEAD(&mux->psocks_ready);
1980	INIT_LIST_HEAD(&mux->psocks_avail);
1981
1982	mux->knet = knet;
1983
1984	/* Add new MUX to list */
1985	mutex_lock(&knet->mutex);
1986	list_add_rcu(&mux->kcm_mux_list, &knet->mux_list);
1987	knet->count++;
1988	mutex_unlock(&knet->mutex);
1989
1990	skb_queue_head_init(&mux->rx_hold_queue);
1991
1992	/* Init KCM socket */
1993	sock_init_data(sock, sk);
1994	init_kcm_sock(kcm_sk(sk), mux);
1995
1996	return 0;
1997}
1998
1999static const struct net_proto_family kcm_family_ops = {
2000	.family = PF_KCM,
2001	.create = kcm_create,
2002	.owner  = THIS_MODULE,
2003};
2004
2005static __net_init int kcm_init_net(struct net *net)
2006{
2007	struct kcm_net *knet = net_generic(net, kcm_net_id);
2008
2009	INIT_LIST_HEAD_RCU(&knet->mux_list);
2010	mutex_init(&knet->mutex);
2011
2012	return 0;
2013}
2014
2015static __net_exit void kcm_exit_net(struct net *net)
2016{
2017	struct kcm_net *knet = net_generic(net, kcm_net_id);
2018
2019	/* All KCM sockets should be closed at this point, which should mean
2020	 * that all multiplexors and psocks have been destroyed.
2021	 */
2022	WARN_ON(!list_empty(&knet->mux_list));
2023}
2024
2025static struct pernet_operations kcm_net_ops = {
2026	.init = kcm_init_net,
2027	.exit = kcm_exit_net,
2028	.id   = &kcm_net_id,
2029	.size = sizeof(struct kcm_net),
2030};
2031
2032static int __init kcm_init(void)
2033{
2034	int err = -ENOMEM;
2035
2036	kcm_muxp = kmem_cache_create("kcm_mux_cache",
2037				     sizeof(struct kcm_mux), 0,
2038				     SLAB_HWCACHE_ALIGN, NULL);
2039	if (!kcm_muxp)
2040		goto fail;
2041
2042	kcm_psockp = kmem_cache_create("kcm_psock_cache",
2043				       sizeof(struct kcm_psock), 0,
2044					SLAB_HWCACHE_ALIGN, NULL);
2045	if (!kcm_psockp)
2046		goto fail;
2047
2048	kcm_wq = create_singlethread_workqueue("kkcmd");
2049	if (!kcm_wq)
2050		goto fail;
2051
2052	err = proto_register(&kcm_proto, 1);
2053	if (err)
2054		goto fail;
2055
2056	err = register_pernet_device(&kcm_net_ops);
2057	if (err)
2058		goto net_ops_fail;
2059
2060	err = sock_register(&kcm_family_ops);
2061	if (err)
2062		goto sock_register_fail;
2063
2064	err = kcm_proc_init();
2065	if (err)
2066		goto proc_init_fail;
2067
2068	return 0;
2069
2070proc_init_fail:
2071	sock_unregister(PF_KCM);
2072
2073sock_register_fail:
2074	unregister_pernet_device(&kcm_net_ops);
2075
2076net_ops_fail:
2077	proto_unregister(&kcm_proto);
2078
2079fail:
2080	kmem_cache_destroy(kcm_muxp);
2081	kmem_cache_destroy(kcm_psockp);
2082
2083	if (kcm_wq)
2084		destroy_workqueue(kcm_wq);
2085
2086	return err;
2087}
2088
2089static void __exit kcm_exit(void)
2090{
2091	kcm_proc_exit();
2092	sock_unregister(PF_KCM);
2093	unregister_pernet_device(&kcm_net_ops);
2094	proto_unregister(&kcm_proto);
2095	destroy_workqueue(kcm_wq);
2096
2097	kmem_cache_destroy(kcm_muxp);
2098	kmem_cache_destroy(kcm_psockp);
2099}
2100
2101module_init(kcm_init);
2102module_exit(kcm_exit);
2103
2104MODULE_LICENSE("GPL");
2105MODULE_ALIAS_NETPROTO(PF_KCM);