net/kcm/kcmsock.c

   1 /*
   2  * Kernel Connection Multiplexor
   3  *
   4  * Copyright (c) 2016 Tom Herbert <tom@herbertland.com>
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2
   8  * as published by the Free Software Foundation.
   9  */
  10
  11 #include <linux/bpf.h>
  12 #include <linux/errno.h>
  13 #include <linux/errqueue.h>
  14 #include <linux/file.h>
  15 #include <linux/in.h>
  16 #include <linux/kernel.h>
  17 #include <linux/module.h>
  18 #include <linux/net.h>
  19 #include <linux/netdevice.h>
  20 #include <linux/poll.h>
  21 #include <linux/rculist.h>
  22 #include <linux/skbuff.h>
  23 #include <linux/socket.h>
  24 #include <linux/uaccess.h>
  25 #include <linux/workqueue.h>
  26 #include <linux/syscalls.h>
  27 #include <net/kcm.h>
  28 #include <net/netns/generic.h>
  29 #include <net/sock.h>
  30 #include <uapi/linux/kcm.h>
  31
  32 unsigned int kcm_net_id;
  33
  34 static struct kmem_cache *kcm_psockp __read_mostly;
  35 static struct kmem_cache *kcm_muxp __read_mostly;
  36 static struct workqueue_struct *kcm_wq;
  37
  38 static inline struct kcm_sock *kcm_sk(const struct sock *sk)
  39 {
  40         return (struct kcm_sock *)sk;
  41 }
  42
  43 static inline struct kcm_tx_msg *kcm_tx_msg(struct sk_buff *skb)
  44 {
  45         return (struct kcm_tx_msg *)skb->cb;
  46 }
  47
  48 static void report_csk_error(struct sock *csk, int err)
  49 {
  50         csk->sk_err = EPIPE;
  51         csk->sk_error_report(csk);
  52 }
  53
  54 static void kcm_abort_tx_psock(struct kcm_psock *psock, int err,
  55                                bool wakeup_kcm)
  56 {
  57         struct sock *csk = psock->sk;
  58         struct kcm_mux *mux = psock->mux;
  59
  60         /* Unrecoverable error in transmit */
  61
  62         spin_lock_bh(&mux->lock);
  63
  64         if (psock->tx_stopped) {
  65                 spin_unlock_bh(&mux->lock);
  66                 return;
  67         }
  68
  69         psock->tx_stopped = 1;
  70         KCM_STATS_INCR(psock->stats.tx_aborts);
  71
  72         if (!psock->tx_kcm) {
  73                 /* Take off psocks_avail list */
  74                 list_del(&psock->psock_avail_list);
  75         } else if (wakeup_kcm) {
  76                 /* In this case psock is being aborted while outside of
  77                  * write_msgs and psock is reserved. Schedule tx_work
  78                  * to handle the failure there. Need to commit tx_stopped
  79                  * before queuing work.
  80                  */
  81                 smp_mb();
  82
  83                 queue_work(kcm_wq, &psock->tx_kcm->tx_work);
  84         }
  85
  86         spin_unlock_bh(&mux->lock);
  87
  88         /* Report error on lower socket */
  89         report_csk_error(csk, err);
  90 }
  91
  92 /* RX mux lock held. */
  93 static void kcm_update_rx_mux_stats(struct kcm_mux *mux,
  94                                     struct kcm_psock *psock)
  95 {
  96         STRP_STATS_ADD(mux->stats.rx_bytes,
  97                        psock->strp.stats.rx_bytes -
  98                        psock->saved_rx_bytes);
  99         mux->stats.rx_msgs +=
 100                 psock->strp.stats.rx_msgs - psock->saved_rx_msgs;
 101         psock->saved_rx_msgs = psock->strp.stats.rx_msgs;
 102         psock->saved_rx_bytes = psock->strp.stats.rx_bytes;
 103 }
 104
 105 static void kcm_update_tx_mux_stats(struct kcm_mux *mux,
 106                                     struct kcm_psock *psock)
 107 {
 108         KCM_STATS_ADD(mux->stats.tx_bytes,
 109                       psock->stats.tx_bytes - psock->saved_tx_bytes);
 110         mux->stats.tx_msgs +=
 111                 psock->stats.tx_msgs - psock->saved_tx_msgs;
 112         psock->saved_tx_msgs = psock->stats.tx_msgs;
 113         psock->saved_tx_bytes = psock->stats.tx_bytes;
 114 }
 115
 116 static int kcm_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
 117
 118 /* KCM is ready to receive messages on its queue-- either the KCM is new or
 119  * has become unblocked after being blocked on full socket buffer. Queue any
 120  * pending ready messages on a psock. RX mux lock held.
 121  */
 122 static void kcm_rcv_ready(struct kcm_sock *kcm)
 123 {
 124         struct kcm_mux *mux = kcm->mux;
 125         struct kcm_psock *psock;
 126         struct sk_buff *skb;
 127
 128         if (unlikely(kcm->rx_wait || kcm->rx_psock || kcm->rx_disabled))
 129                 return;
 130
 131         while (unlikely((skb = __skb_dequeue(&mux->rx_hold_queue)))) {
 132                 if (kcm_queue_rcv_skb(&kcm->sk, skb)) {
 133                         /* Assuming buffer limit has been reached */
 134                         skb_queue_head(&mux->rx_hold_queue, skb);
 135                         WARN_ON(!sk_rmem_alloc_get(&kcm->sk));
 136                         return;
 137                 }
 138         }
 139
 140         while (!list_empty(&mux->psocks_ready)) {
 141                 psock = list_first_entry(&mux->psocks_ready, struct kcm_psock,
 142                                          psock_ready_list);
 143
 144                 if (kcm_queue_rcv_skb(&kcm->sk, psock->ready_rx_msg)) {
 145                         /* Assuming buffer limit has been reached */
 146                         WARN_ON(!sk_rmem_alloc_get(&kcm->sk));
 147                         return;
 148                 }
 149
 150                 /* Consumed the ready message on the psock. Schedule rx_work to
 151                  * get more messages.
 152                  */
 153                 list_del(&psock->psock_ready_list);
 154                 psock->ready_rx_msg = NULL;
 155                 /* Commit clearing of ready_rx_msg for queuing work */
 156                 smp_mb();
 157
 158                 strp_unpause(&psock->strp);
 159                 strp_check_rcv(&psock->strp);
 160         }
 161
 162         /* Buffer limit is okay now, add to ready list */
 163         list_add_tail(&kcm->wait_rx_list,
 164                       &kcm->mux->kcm_rx_waiters);
 165         /* paired with lockless reads in kcm_rfree() */
 166         WRITE_ONCE(kcm->rx_wait, true);
 167 }
 168
 169 static void kcm_rfree(struct sk_buff *skb)
 170 {
 171         struct sock *sk = skb->sk;
 172         struct kcm_sock *kcm = kcm_sk(sk);
 173         struct kcm_mux *mux = kcm->mux;
 174         unsigned int len = skb->truesize;
 175
 176         sk_mem_uncharge(sk, len);
 177         atomic_sub(len, &sk->sk_rmem_alloc);
 178
 179         /* For reading rx_wait and rx_psock without holding lock */
 180         smp_mb__after_atomic();
 181
 182         if (!READ_ONCE(kcm->rx_wait) && !READ_ONCE(kcm->rx_psock) &&
 183             sk_rmem_alloc_get(sk) < sk->sk_rcvlowat) {
 184                 spin_lock_bh(&mux->rx_lock);
 185                 kcm_rcv_ready(kcm);
 186                 spin_unlock_bh(&mux->rx_lock);
 187         }
 188 }
 189
 190 static int kcm_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
 191 {
 192         struct sk_buff_head *list = &sk->sk_receive_queue;
 193
 194         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
 195                 return -ENOMEM;
 196
 197         if (!sk_rmem_schedule(sk, skb, skb->truesize))
 198                 return -ENOBUFS;
 199
 200         skb->dev = NULL;
 201
 202         skb_orphan(skb);
 203         skb->sk = sk;
 204         skb->destructor = kcm_rfree;
 205         atomic_add(skb->truesize, &sk->sk_rmem_alloc);
 206         sk_mem_charge(sk, skb->truesize);
 207
 208         skb_queue_tail(list, skb);
 209
 210         if (!sock_flag(sk, SOCK_DEAD))
 211                 sk->sk_data_ready(sk);
 212
 213         return 0;
 214 }
 215
 216 /* Requeue received messages for a kcm socket to other kcm sockets. This is
 217  * called with a kcm socket is receive disabled.
 218  * RX mux lock held.
 219  */
 220 static void requeue_rx_msgs(struct kcm_mux *mux, struct sk_buff_head *head)
 221 {
 222         struct sk_buff *skb;
 223         struct kcm_sock *kcm;
 224
 225         while ((skb = __skb_dequeue(head))) {
 226                 /* Reset destructor to avoid calling kcm_rcv_ready */
 227                 skb->destructor = sock_rfree;
 228                 skb_orphan(skb);
 229 try_again:
 230                 if (list_empty(&mux->kcm_rx_waiters)) {
 231                         skb_queue_tail(&mux->rx_hold_queue, skb);
 232                         continue;
 233                 }
 234
 235                 kcm = list_first_entry(&mux->kcm_rx_waiters,
 236                                        struct kcm_sock, wait_rx_list);
 237
 238                 if (kcm_queue_rcv_skb(&kcm->sk, skb)) {
 239                         /* Should mean socket buffer full */
 240                         list_del(&kcm->wait_rx_list);
 241                         /* paired with lockless reads in kcm_rfree() */
 242                         WRITE_ONCE(kcm->rx_wait, false);
 243
 244                         /* Commit rx_wait to read in kcm_free */
 245                         smp_wmb();
 246
 247                         goto try_again;
 248                 }
 249         }
 250 }
 251
 252 /* Lower sock lock held */
 253 static struct kcm_sock *reserve_rx_kcm(struct kcm_psock *psock,
 254                                        struct sk_buff *head)
 255 {
 256         struct kcm_mux *mux = psock->mux;
 257         struct kcm_sock *kcm;
 258
 259         WARN_ON(psock->ready_rx_msg);
 260
 261         if (psock->rx_kcm)
 262                 return psock->rx_kcm;
 263
 264         spin_lock_bh(&mux->rx_lock);
 265
 266         if (psock->rx_kcm) {
 267                 spin_unlock_bh(&mux->rx_lock);
 268                 return psock->rx_kcm;
 269         }
 270
 271         kcm_update_rx_mux_stats(mux, psock);
 272
 273         if (list_empty(&mux->kcm_rx_waiters)) {
 274                 psock->ready_rx_msg = head;
 275                 strp_pause(&psock->strp);
 276                 list_add_tail(&psock->psock_ready_list,
 277                               &mux->psocks_ready);
 278                 spin_unlock_bh(&mux->rx_lock);
 279                 return NULL;
 280         }
 281
 282         kcm = list_first_entry(&mux->kcm_rx_waiters,
 283                                struct kcm_sock, wait_rx_list);
 284         list_del(&kcm->wait_rx_list);
 285         /* paired with lockless reads in kcm_rfree() */
 286         WRITE_ONCE(kcm->rx_wait, false);
 287
 288         psock->rx_kcm = kcm;
 289         /* paired with lockless reads in kcm_rfree() */
 290         WRITE_ONCE(kcm->rx_psock, psock);
 291
 292         spin_unlock_bh(&mux->rx_lock);
 293
 294         return kcm;
 295 }
 296
 297 static void kcm_done(struct kcm_sock *kcm);
 298
 299 static void kcm_done_work(struct work_struct *w)
 300 {
 301         kcm_done(container_of(w, struct kcm_sock, done_work));
 302 }
 303
 304 /* Lower sock held */
 305 static void unreserve_rx_kcm(struct kcm_psock *psock,
 306                              bool rcv_ready)
 307 {
 308         struct kcm_sock *kcm = psock->rx_kcm;
 309         struct kcm_mux *mux = psock->mux;
 310
 311         if (!kcm)
 312                 return;
 313
 314         spin_lock_bh(&mux->rx_lock);
 315
 316         psock->rx_kcm = NULL;
 317         /* paired with lockless reads in kcm_rfree() */
 318         WRITE_ONCE(kcm->rx_psock, NULL);
 319
 320         /* Commit kcm->rx_psock before sk_rmem_alloc_get to sync with
 321          * kcm_rfree
 322          */
 323         smp_mb();
 324
 325         if (unlikely(kcm->done)) {
 326                 spin_unlock_bh(&mux->rx_lock);
 327
 328                 /* Need to run kcm_done in a task since we need to qcquire
 329                  * callback locks which may already be held here.
 330                  */
 331                 INIT_WORK(&kcm->done_work, kcm_done_work);
 332                 schedule_work(&kcm->done_work);
 333                 return;
 334         }
 335
 336         if (unlikely(kcm->rx_disabled)) {
 337                 requeue_rx_msgs(mux, &kcm->sk.sk_receive_queue);
 338         } else if (rcv_ready || unlikely(!sk_rmem_alloc_get(&kcm->sk))) {
 339                 /* Check for degenerative race with rx_wait that all
 340                  * data was dequeued (accounted for in kcm_rfree).
 341                  */
 342                 kcm_rcv_ready(kcm);
 343         }
 344         spin_unlock_bh(&mux->rx_lock);
 345 }
 346
 347 /* Lower sock lock held */
 348 static void psock_data_ready(struct sock *sk)
 349 {
 350         struct kcm_psock *psock;
 351
 352         read_lock_bh(&sk->sk_callback_lock);
 353
 354         psock = (struct kcm_psock *)sk->sk_user_data;
 355         if (likely(psock))
 356                 strp_data_ready(&psock->strp);
 357
 358         read_unlock_bh(&sk->sk_callback_lock);
 359 }
 360
 361 /* Called with lower sock held */
 362 static void kcm_rcv_strparser(struct strparser *strp, struct sk_buff *skb)
 363 {
 364         struct kcm_psock *psock = container_of(strp, struct kcm_psock, strp);
 365         struct kcm_sock *kcm;
 366
 367 try_queue:
 368         kcm = reserve_rx_kcm(psock, skb);
 369         if (!kcm) {
 370                  /* Unable to reserve a KCM, message is held in psock and strp
 371                   * is paused.
 372                   */
 373                 return;
 374         }
 375
 376         if (kcm_queue_rcv_skb(&kcm->sk, skb)) {
 377                 /* Should mean socket buffer full */
 378                 unreserve_rx_kcm(psock, false);
 379                 goto try_queue;
 380         }
 381 }
 382
 383 static int kcm_parse_func_strparser(struct strparser *strp, struct sk_buff *skb)
 384 {
 385         struct kcm_psock *psock = container_of(strp, struct kcm_psock, strp);
 386         struct bpf_prog *prog = psock->bpf_prog;
 387
 388         return (*prog->bpf_func)(skb, prog->insnsi);
 389 }
 390
 391 static int kcm_read_sock_done(struct strparser *strp, int err)
 392 {
 393         struct kcm_psock *psock = container_of(strp, struct kcm_psock, strp);
 394
 395         unreserve_rx_kcm(psock, true);
 396
 397         return err;
 398 }
 399
 400 static void psock_state_change(struct sock *sk)
 401 {
 402         /* TCP only does a POLLIN for a half close. Do a POLLHUP here
 403          * since application will normally not poll with POLLIN
 404          * on the TCP sockets.
 405          */
 406
 407         report_csk_error(sk, EPIPE);
 408 }
 409
 410 static void psock_write_space(struct sock *sk)
 411 {
 412         struct kcm_psock *psock;
 413         struct kcm_mux *mux;
 414         struct kcm_sock *kcm;
 415
 416         read_lock_bh(&sk->sk_callback_lock);
 417
 418         psock = (struct kcm_psock *)sk->sk_user_data;
 419         if (unlikely(!psock))
 420                 goto out;
 421         mux = psock->mux;
 422
 423         spin_lock_bh(&mux->lock);
 424
 425         /* Check if the socket is reserved so someone is waiting for sending. */
 426         kcm = psock->tx_kcm;
 427         if (kcm && !unlikely(kcm->tx_stopped))
 428                 queue_work(kcm_wq, &kcm->tx_work);
 429
 430         spin_unlock_bh(&mux->lock);
 431 out:
 432         read_unlock_bh(&sk->sk_callback_lock);
 433 }
 434
 435 static void unreserve_psock(struct kcm_sock *kcm);
 436
 437 /* kcm sock is locked. */
 438 static struct kcm_psock *reserve_psock(struct kcm_sock *kcm)
 439 {
 440         struct kcm_mux *mux = kcm->mux;
 441         struct kcm_psock *psock;
 442
 443         psock = kcm->tx_psock;
 444
 445         smp_rmb(); /* Must read tx_psock before tx_wait */
 446
 447         if (psock) {
 448                 WARN_ON(kcm->tx_wait);
 449                 if (unlikely(psock->tx_stopped))
 450                         unreserve_psock(kcm);
 451                 else
 452                         return kcm->tx_psock;
 453         }
 454
 455         spin_lock_bh(&mux->lock);
 456
 457         /* Check again under lock to see if psock was reserved for this
 458          * psock via psock_unreserve.
 459          */
 460         psock = kcm->tx_psock;
 461         if (unlikely(psock)) {
 462                 WARN_ON(kcm->tx_wait);
 463                 spin_unlock_bh(&mux->lock);
 464                 return kcm->tx_psock;
 465         }
 466
 467         if (!list_empty(&mux->psocks_avail)) {
 468                 psock = list_first_entry(&mux->psocks_avail,
 469                                          struct kcm_psock,
 470                                          psock_avail_list);
 471                 list_del(&psock->psock_avail_list);
 472                 if (kcm->tx_wait) {
 473                         list_del(&kcm->wait_psock_list);
 474                         kcm->tx_wait = false;
 475                 }
 476                 kcm->tx_psock = psock;
 477                 psock->tx_kcm = kcm;
 478                 KCM_STATS_INCR(psock->stats.reserved);
 479         } else if (!kcm->tx_wait) {
 480                 list_add_tail(&kcm->wait_psock_list,
 481                               &mux->kcm_tx_waiters);
 482                 kcm->tx_wait = true;
 483         }
 484
 485         spin_unlock_bh(&mux->lock);
 486
 487         return psock;
 488 }
 489
 490 /* mux lock held */
 491 static void psock_now_avail(struct kcm_psock *psock)
 492 {
 493         struct kcm_mux *mux = psock->mux;
 494         struct kcm_sock *kcm;
 495
 496         if (list_empty(&mux->kcm_tx_waiters)) {
 497                 list_add_tail(&psock->psock_avail_list,
 498                               &mux->psocks_avail);
 499         } else {
 500                 kcm = list_first_entry(&mux->kcm_tx_waiters,
 501                                        struct kcm_sock,
 502                                        wait_psock_list);
 503                 list_del(&kcm->wait_psock_list);
 504                 kcm->tx_wait = false;
 505                 psock->tx_kcm = kcm;
 506
 507                 /* Commit before changing tx_psock since that is read in
 508                  * reserve_psock before queuing work.
 509                  */
 510                 smp_mb();
 511
 512                 kcm->tx_psock = psock;
 513                 KCM_STATS_INCR(psock->stats.reserved);
 514                 queue_work(kcm_wq, &kcm->tx_work);
 515         }
 516 }
 517
 518 /* kcm sock is locked. */
 519 static void unreserve_psock(struct kcm_sock *kcm)
 520 {
 521         struct kcm_psock *psock;
 522         struct kcm_mux *mux = kcm->mux;
 523
 524         spin_lock_bh(&mux->lock);
 525
 526         psock = kcm->tx_psock;
 527
 528         if (WARN_ON(!psock)) {
 529                 spin_unlock_bh(&mux->lock);
 530                 return;
 531         }
 532
 533         smp_rmb(); /* Read tx_psock before tx_wait */
 534
 535         kcm_update_tx_mux_stats(mux, psock);
 536
 537         WARN_ON(kcm->tx_wait);
 538
 539         kcm->tx_psock = NULL;
 540         psock->tx_kcm = NULL;
 541         KCM_STATS_INCR(psock->stats.unreserved);
 542
 543         if (unlikely(psock->tx_stopped)) {
 544                 if (psock->done) {
 545                         /* Deferred free */
 546                         list_del(&psock->psock_list);
 547                         mux->psocks_cnt--;
 548                         sock_put(psock->sk);
 549                         fput(psock->sk->sk_socket->file);
 550                         kmem_cache_free(kcm_psockp, psock);
 551                 }
 552
 553                 /* Don't put back on available list */
 554
 555                 spin_unlock_bh(&mux->lock);
 556
 557                 return;
 558         }
 559
 560         psock_now_avail(psock);
 561
 562         spin_unlock_bh(&mux->lock);
 563 }
 564
 565 static void kcm_report_tx_retry(struct kcm_sock *kcm)
 566 {
 567         struct kcm_mux *mux = kcm->mux;
 568
 569         spin_lock_bh(&mux->lock);
 570         KCM_STATS_INCR(mux->stats.tx_retries);
 571         spin_unlock_bh(&mux->lock);
 572 }
 573
 574 /* Write any messages ready on the kcm socket.  Called with kcm sock lock
 575  * held.  Return bytes actually sent or error.
 576  */
 577 static int kcm_write_msgs(struct kcm_sock *kcm)
 578 {
 579         struct sock *sk = &kcm->sk;
 580         struct kcm_psock *psock;
 581         struct sk_buff *skb, *head;
 582         struct kcm_tx_msg *txm;
 583         unsigned short fragidx, frag_offset;
 584         unsigned int sent, total_sent = 0;
 585         int ret = 0;
 586
 587         kcm->tx_wait_more = false;
 588         psock = kcm->tx_psock;
 589         if (unlikely(psock && psock->tx_stopped)) {
 590                 /* A reserved psock was aborted asynchronously. Unreserve
 591                  * it and we'll retry the message.
 592                  */
 593                 unreserve_psock(kcm);
 594                 kcm_report_tx_retry(kcm);
 595                 if (skb_queue_empty(&sk->sk_write_queue))
 596                         return 0;
 597
 598                 kcm_tx_msg(skb_peek(&sk->sk_write_queue))->sent = 0;
 599
 600         } else if (skb_queue_empty(&sk->sk_write_queue)) {
 601                 return 0;
 602         }
 603
 604         head = skb_peek(&sk->sk_write_queue);
 605         txm = kcm_tx_msg(head);
 606
 607         if (txm->sent) {
 608                 /* Send of first skbuff in queue already in progress */
 609                 if (WARN_ON(!psock)) {
 610                         ret = -EINVAL;
 611                         goto out;
 612                 }
 613                 sent = txm->sent;
 614                 frag_offset = txm->frag_offset;
 615                 fragidx = txm->fragidx;
 616                 skb = txm->frag_skb;
 617
 618                 goto do_frag;
 619         }
 620
 621 try_again:
 622         psock = reserve_psock(kcm);
 623         if (!psock)
 624                 goto out;
 625
 626         do {
 627                 skb = head;
 628                 txm = kcm_tx_msg(head);
 629                 sent = 0;
 630
 631 do_frag_list:
 632                 if (WARN_ON(!skb_shinfo(skb)->nr_frags)) {
 633                         ret = -EINVAL;
 634                         goto out;
 635                 }
 636
 637                 for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags;
 638                      fragidx++) {
 639                         skb_frag_t *frag;
 640
 641                         frag_offset = 0;
 642 do_frag:
 643                         frag = &skb_shinfo(skb)->frags[fragidx];
 644                         if (WARN_ON(!frag->size)) {
 645                                 ret = -EINVAL;
 646                                 goto out;
 647                         }
 648
 649                         ret = kernel_sendpage(psock->sk->sk_socket,
 650                                               frag->page.p,
 651                                               frag->page_offset + frag_offset,
 652                                               frag->size - frag_offset,
 653                                               MSG_DONTWAIT);
 654                         if (ret <= 0) {
 655                                 if (ret == -EAGAIN) {
 656                                         /* Save state to try again when there's
 657                                          * write space on the socket
 658                                          */
 659                                         txm->sent = sent;
 660                                         txm->frag_offset = frag_offset;
 661                                         txm->fragidx = fragidx;
 662                                         txm->frag_skb = skb;
 663
 664                                         ret = 0;
 665                                         goto out;
 666                                 }
 667
 668                                 /* Hard failure in sending message, abort this
 669                                  * psock since it has lost framing
 670                                  * synchonization and retry sending the
 671                                  * message from the beginning.
 672                                  */
 673                                 kcm_abort_tx_psock(psock, ret ? -ret : EPIPE,
 674                                                    true);
 675                                 unreserve_psock(kcm);
 676
 677                                 txm->sent = 0;
 678                                 kcm_report_tx_retry(kcm);
 679                                 ret = 0;
 680
 681                                 goto try_again;
 682                         }
 683
 684                         sent += ret;
 685                         frag_offset += ret;
 686                         KCM_STATS_ADD(psock->stats.tx_bytes, ret);
 687                         if (frag_offset < frag->size) {
 688                                 /* Not finished with this frag */
 689                                 goto do_frag;
 690                         }
 691                 }
 692
 693                 if (skb == head) {
 694                         if (skb_has_frag_list(skb)) {
 695                                 skb = skb_shinfo(skb)->frag_list;
 696                                 goto do_frag_list;
 697                         }
 698                 } else if (skb->next) {
 699                         skb = skb->next;
 700                         goto do_frag_list;
 701                 }
 702
 703                 /* Successfully sent the whole packet, account for it. */
 704                 skb_dequeue(&sk->sk_write_queue);
 705                 kfree_skb(head);
 706                 sk->sk_wmem_queued -= sent;
 707                 total_sent += sent;
 708                 KCM_STATS_INCR(psock->stats.tx_msgs);
 709         } while ((head = skb_peek(&sk->sk_write_queue)));
 710 out:
 711         if (!head) {
 712                 /* Done with all queued messages. */
 713                 WARN_ON(!skb_queue_empty(&sk->sk_write_queue));
 714                 unreserve_psock(kcm);
 715         }
 716
 717         /* Check if write space is available */
 718         sk->sk_write_space(sk);
 719
 720         return total_sent ? : ret;
 721 }
 722
 723 static void kcm_tx_work(struct work_struct *w)
 724 {
 725         struct kcm_sock *kcm = container_of(w, struct kcm_sock, tx_work);
 726         struct sock *sk = &kcm->sk;
 727         int err;
 728
 729         lock_sock(sk);
 730
 731         /* Primarily for SOCK_DGRAM sockets, also handle asynchronous tx
 732          * aborts
 733          */
 734         err = kcm_write_msgs(kcm);
 735         if (err < 0) {
 736                 /* Hard failure in write, report error on KCM socket */
 737                 pr_warn("KCM: Hard failure on kcm_write_msgs %d\n", err);
 738                 report_csk_error(&kcm->sk, -err);
 739                 goto out;
 740         }
 741
 742         /* Primarily for SOCK_SEQPACKET sockets */
 743         if (likely(sk->sk_socket) &&
 744             test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
 745                 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 746                 sk->sk_write_space(sk);
 747         }
 748
 749 out:
 750         release_sock(sk);
 751 }
 752
 753 static void kcm_push(struct kcm_sock *kcm)
 754 {
 755         if (kcm->tx_wait_more)
 756                 kcm_write_msgs(kcm);
 757 }
 758
 759 static ssize_t kcm_sendpage(struct socket *sock, struct page *page,
 760                             int offset, size_t size, int flags)
 761
 762 {
 763         struct sock *sk = sock->sk;
 764         struct kcm_sock *kcm = kcm_sk(sk);
 765         struct sk_buff *skb = NULL, *head = NULL;
 766         long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
 767         bool eor;
 768         int err = 0;
 769         int i;
 770
 771         if (flags & MSG_SENDPAGE_NOTLAST)
 772                 flags |= MSG_MORE;
 773
 774         /* No MSG_EOR from splice, only look at MSG_MORE */
 775         eor = !(flags & MSG_MORE);
 776
 777         lock_sock(sk);
 778
 779         sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
 780
 781         err = -EPIPE;
 782         if (sk->sk_err)
 783                 goto out_error;
 784
 785         if (kcm->seq_skb) {
 786                 /* Previously opened message */
 787                 head = kcm->seq_skb;
 788                 skb = kcm_tx_msg(head)->last_skb;
 789                 i = skb_shinfo(skb)->nr_frags;
 790
 791                 if (skb_can_coalesce(skb, i, page, offset)) {
 792                         skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
 793                         skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
 794                         goto coalesced;
 795                 }
 796
 797                 if (i >= MAX_SKB_FRAGS) {
 798                         struct sk_buff *tskb;
 799
 800                         tskb = alloc_skb(0, sk->sk_allocation);
 801                         while (!tskb) {
 802                                 kcm_push(kcm);
 803                                 err = sk_stream_wait_memory(sk, &timeo);
 804                                 if (err)
 805                                         goto out_error;
 806                         }
 807
 808                         if (head == skb)
 809                                 skb_shinfo(head)->frag_list = tskb;
 810                         else
 811                                 skb->next = tskb;
 812
 813                         skb = tskb;
 814                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 815                         i = 0;
 816                 }
 817         } else {
 818                 /* Call the sk_stream functions to manage the sndbuf mem. */
 819                 if (!sk_stream_memory_free(sk)) {
 820                         kcm_push(kcm);
 821                         set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 822                         err = sk_stream_wait_memory(sk, &timeo);
 823                         if (err)
 824                                 goto out_error;
 825                 }
 826
 827                 head = alloc_skb(0, sk->sk_allocation);
 828                 while (!head) {
 829                         kcm_push(kcm);
 830                         err = sk_stream_wait_memory(sk, &timeo);
 831                         if (err)
 832                                 goto out_error;
 833                 }
 834
 835                 skb = head;
 836                 i = 0;
 837         }
 838
 839         get_page(page);
 840         skb_fill_page_desc(skb, i, page, offset, size);
 841         skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
 842
 843 coalesced:
 844         skb->len += size;
 845         skb->data_len += size;
 846         skb->truesize += size;
 847         sk->sk_wmem_queued += size;
 848         sk_mem_charge(sk, size);
 849
 850         if (head != skb) {
 851                 head->len += size;
 852                 head->data_len += size;
 853                 head->truesize += size;
 854         }
 855
 856         if (eor) {
 857                 bool not_busy = skb_queue_empty(&sk->sk_write_queue);
 858
 859                 /* Message complete, queue it on send buffer */
 860                 __skb_queue_tail(&sk->sk_write_queue, head);
 861                 kcm->seq_skb = NULL;
 862                 KCM_STATS_INCR(kcm->stats.tx_msgs);
 863
 864                 if (flags & MSG_BATCH) {
 865                         kcm->tx_wait_more = true;
 866                 } else if (kcm->tx_wait_more || not_busy) {
 867                         err = kcm_write_msgs(kcm);
 868                         if (err < 0) {
 869                                 /* We got a hard error in write_msgs but have
 870                                  * already queued this message. Report an error
 871                                  * in the socket, but don't affect return value
 872                                  * from sendmsg
 873                                  */
 874                                 pr_warn("KCM: Hard failure on kcm_write_msgs\n");
 875                                 report_csk_error(&kcm->sk, -err);
 876                         }
 877                 }
 878         } else {
 879                 /* Message not complete, save state */
 880                 kcm->seq_skb = head;
 881                 kcm_tx_msg(head)->last_skb = skb;
 882         }
 883
 884         KCM_STATS_ADD(kcm->stats.tx_bytes, size);
 885
 886         release_sock(sk);
 887         return size;
 888
 889 out_error:
 890         kcm_push(kcm);
 891
 892         err = sk_stream_error(sk, flags, err);
 893
 894         /* make sure we wake any epoll edge trigger waiter */
 895         if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 && err == -EAGAIN))
 896                 sk->sk_write_space(sk);
 897
 898         release_sock(sk);
 899         return err;
 900 }
 901
 902 static int kcm_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
 903 {
 904         struct sock *sk = sock->sk;
 905         struct kcm_sock *kcm = kcm_sk(sk);
 906         struct sk_buff *skb = NULL, *head = NULL;
 907         size_t copy, copied = 0;
 908         long timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
 909         int eor = (sock->type == SOCK_DGRAM) ?
 910                   !(msg->msg_flags & MSG_MORE) : !!(msg->msg_flags & MSG_EOR);
 911         int err = -EPIPE;
 912
 913         lock_sock(sk);
 914
 915         /* Per tcp_sendmsg this should be in poll */
 916         sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
 917
 918         if (sk->sk_err)
 919                 goto out_error;
 920
 921         if (kcm->seq_skb) {
 922                 /* Previously opened message */
 923                 head = kcm->seq_skb;
 924                 skb = kcm_tx_msg(head)->last_skb;
 925                 goto start;
 926         }
 927
 928         /* Call the sk_stream functions to manage the sndbuf mem. */
 929         if (!sk_stream_memory_free(sk)) {
 930                 kcm_push(kcm);
 931                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 932                 err = sk_stream_wait_memory(sk, &timeo);
 933                 if (err)
 934                         goto out_error;
 935         }
 936
 937         if (msg_data_left(msg)) {
 938                 /* New message, alloc head skb */
 939                 head = alloc_skb(0, sk->sk_allocation);
 940                 while (!head) {
 941                         kcm_push(kcm);
 942                         err = sk_stream_wait_memory(sk, &timeo);
 943                         if (err)
 944                                 goto out_error;
 945
 946                         head = alloc_skb(0, sk->sk_allocation);
 947                 }
 948
 949                 skb = head;
 950
 951                 /* Set ip_summed to CHECKSUM_UNNECESSARY to avoid calling
 952                  * csum_and_copy_from_iter from skb_do_copy_data_nocache.
 953                  */
 954                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 955         }
 956
 957 start:
 958         while (msg_data_left(msg)) {
 959                 bool merge = true;
 960                 int i = skb_shinfo(skb)->nr_frags;
 961                 struct page_frag *pfrag = sk_page_frag(sk);
 962
 963                 if (!sk_page_frag_refill(sk, pfrag))
 964                         goto wait_for_memory;
 965
 966                 if (!skb_can_coalesce(skb, i, pfrag->page,
 967                                       pfrag->offset)) {
 968                         if (i == MAX_SKB_FRAGS) {
 969                                 struct sk_buff *tskb;
 970
 971                                 tskb = alloc_skb(0, sk->sk_allocation);
 972                                 if (!tskb)
 973                                         goto wait_for_memory;
 974
 975                                 if (head == skb)
 976                                         skb_shinfo(head)->frag_list = tskb;
 977                                 else
 978                                         skb->next = tskb;
 979
 980                                 skb = tskb;
 981                                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 982                                 continue;
 983                         }
 984                         merge = false;
 985                 }
 986
 987                 copy = min_t(int, msg_data_left(msg),
 988                              pfrag->size - pfrag->offset);
 989
 990                 if (!sk_wmem_schedule(sk, copy))
 991                         goto wait_for_memory;
 992
 993                 err = skb_copy_to_page_nocache(sk, &msg->msg_iter, skb,
 994                                                pfrag->page,
 995                                                pfrag->offset,
 996                                                copy);
 997                 if (err)
 998                         goto out_error;
 999
1000                 /* Update the skb. */
1001                 if (merge) {
1002                         skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1003                 } else {
1004                         skb_fill_page_desc(skb, i, pfrag->page,
1005                                            pfrag->offset, copy);
1006                         get_page(pfrag->page);
1007                 }
1008
1009                 pfrag->offset += copy;
1010                 copied += copy;
1011                 if (head != skb) {
1012                         head->len += copy;
1013                         head->data_len += copy;
1014                 }
1015
1016                 continue;
1017
1018 wait_for_memory:
1019                 kcm_push(kcm);
1020                 err = sk_stream_wait_memory(sk, &timeo);
1021                 if (err)
1022                         goto out_error;
1023         }
1024
1025         if (eor) {
1026                 bool not_busy = skb_queue_empty(&sk->sk_write_queue);
1027
1028                 if (head) {
1029                         /* Message complete, queue it on send buffer */
1030                         __skb_queue_tail(&sk->sk_write_queue, head);
1031                         kcm->seq_skb = NULL;
1032                         KCM_STATS_INCR(kcm->stats.tx_msgs);
1033                 }
1034
1035                 if (msg->msg_flags & MSG_BATCH) {
1036                         kcm->tx_wait_more = true;
1037                 } else if (kcm->tx_wait_more || not_busy) {
1038                         err = kcm_write_msgs(kcm);
1039                         if (err < 0) {
1040                                 /* We got a hard error in write_msgs but have
1041                                  * already queued this message. Report an error
1042                                  * in the socket, but don't affect return value
1043                                  * from sendmsg
1044                                  */
1045                                 pr_warn("KCM: Hard failure on kcm_write_msgs\n");
1046                                 report_csk_error(&kcm->sk, -err);
1047                         }
1048                 }
1049         } else {
1050                 /* Message not complete, save state */
1051 partial_message:
1052                 if (head) {
1053                         kcm->seq_skb = head;
1054                         kcm_tx_msg(head)->last_skb = skb;
1055                 }
1056         }
1057
1058         KCM_STATS_ADD(kcm->stats.tx_bytes, copied);
1059
1060         release_sock(sk);
1061         return copied;
1062
1063 out_error:
1064         kcm_push(kcm);
1065
1066         if (copied && sock->type == SOCK_SEQPACKET) {
1067                 /* Wrote some bytes before encountering an
1068                  * error, return partial success.
1069                  */
1070                 goto partial_message;
1071         }
1072
1073         if (head != kcm->seq_skb)
1074                 kfree_skb(head);
1075
1076         err = sk_stream_error(sk, msg->msg_flags, err);
1077
1078         /* make sure we wake any epoll edge trigger waiter */
1079         if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 && err == -EAGAIN))
1080                 sk->sk_write_space(sk);
1081
1082         release_sock(sk);
1083         return err;
1084 }
1085
1086 static struct sk_buff *kcm_wait_data(struct sock *sk, int flags,
1087                                      long timeo, int *err)
1088 {
1089         struct sk_buff *skb;
1090
1091         while (!(skb = skb_peek(&sk->sk_receive_queue))) {
1092                 if (sk->sk_err) {
1093                         *err = sock_error(sk);
1094                         return NULL;
1095                 }
1096
1097                 if (sock_flag(sk, SOCK_DONE))
1098                         return NULL;
1099
1100                 if ((flags & MSG_DONTWAIT) || !timeo) {
1101                         *err = -EAGAIN;
1102                         return NULL;
1103                 }
1104
1105                 sk_wait_data(sk, &timeo, NULL);
1106
1107                 /* Handle signals */
1108                 if (signal_pending(current)) {
1109                         *err = sock_intr_errno(timeo);
1110                         return NULL;
1111                 }
1112         }
1113
1114         return skb;
1115 }
1116
1117 static int kcm_recvmsg(struct socket *sock, struct msghdr *msg,
1118                        size_t len, int flags)
1119 {
1120         struct sock *sk = sock->sk;
1121         struct kcm_sock *kcm = kcm_sk(sk);
1122         int err = 0;
1123         long timeo;
1124         struct strp_rx_msg *rxm;
1125         int copied = 0;
1126         struct sk_buff *skb;
1127
1128         timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1129
1130         lock_sock(sk);
1131
1132         skb = kcm_wait_data(sk, flags, timeo, &err);
1133         if (!skb)
1134                 goto out;
1135
1136         /* Okay, have a message on the receive queue */
1137
1138         rxm = strp_rx_msg(skb);
1139
1140         if (len > rxm->full_len)
1141                 len = rxm->full_len;
1142
1143         err = skb_copy_datagram_msg(skb, rxm->offset, msg, len);
1144         if (err < 0)
1145                 goto out;
1146
1147         copied = len;
1148         if (likely(!(flags & MSG_PEEK))) {
1149                 KCM_STATS_ADD(kcm->stats.rx_bytes, copied);
1150                 if (copied < rxm->full_len) {
1151                         if (sock->type == SOCK_DGRAM) {
1152                                 /* Truncated message */
1153                                 msg->msg_flags |= MSG_TRUNC;
1154                                 goto msg_finished;
1155                         }
1156                         rxm->offset += copied;
1157                         rxm->full_len -= copied;
1158                 } else {
1159 msg_finished:
1160                         /* Finished with message */
1161                         msg->msg_flags |= MSG_EOR;
1162                         KCM_STATS_INCR(kcm->stats.rx_msgs);
1163                         skb_unlink(skb, &sk->sk_receive_queue);
1164                         kfree_skb(skb);
1165                 }
1166         }
1167
1168 out:
1169         release_sock(sk);
1170
1171         return copied ? : err;
1172 }
1173
1174 static ssize_t kcm_splice_read(struct socket *sock, loff_t *ppos,
1175                                struct pipe_inode_info *pipe, size_t len,
1176                                unsigned int flags)
1177 {
1178         struct sock *sk = sock->sk;
1179         struct kcm_sock *kcm = kcm_sk(sk);
1180         long timeo;
1181         struct strp_rx_msg *rxm;
1182         int err = 0;
1183         ssize_t copied;
1184         struct sk_buff *skb;
1185
1186         /* Only support splice for SOCKSEQPACKET */
1187
1188         timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1189
1190         lock_sock(sk);
1191
1192         skb = kcm_wait_data(sk, flags, timeo, &err);
1193         if (!skb)
1194                 goto err_out;
1195
1196         /* Okay, have a message on the receive queue */
1197
1198         rxm = strp_rx_msg(skb);
1199
1200         if (len > rxm->full_len)
1201                 len = rxm->full_len;
1202
1203         copied = skb_splice_bits(skb, sk, rxm->offset, pipe, len, flags);
1204         if (copied < 0) {
1205                 err = copied;
1206                 goto err_out;
1207         }
1208
1209         KCM_STATS_ADD(kcm->stats.rx_bytes, copied);
1210
1211         rxm->offset += copied;
1212         rxm->full_len -= copied;
1213
1214         /* We have no way to return MSG_EOR. If all the bytes have been
1215          * read we still leave the message in the receive socket buffer.
1216          * A subsequent recvmsg needs to be done to return MSG_EOR and
1217          * finish reading the message.
1218          */
1219
1220         release_sock(sk);
1221
1222         return copied;
1223
1224 err_out:
1225         release_sock(sk);
1226
1227         return err;
1228 }
1229
1230 /* kcm sock lock held */
1231 static void kcm_recv_disable(struct kcm_sock *kcm)
1232 {
1233         struct kcm_mux *mux = kcm->mux;
1234
1235         if (kcm->rx_disabled)
1236                 return;
1237
1238         spin_lock_bh(&mux->rx_lock);
1239
1240         kcm->rx_disabled = 1;
1241
1242         /* If a psock is reserved we'll do cleanup in unreserve */
1243         if (!kcm->rx_psock) {
1244                 if (kcm->rx_wait) {
1245                         list_del(&kcm->wait_rx_list);
1246                         /* paired with lockless reads in kcm_rfree() */
1247                         WRITE_ONCE(kcm->rx_wait, false);
1248                 }
1249
1250                 requeue_rx_msgs(mux, &kcm->sk.sk_receive_queue);
1251         }
1252
1253         spin_unlock_bh(&mux->rx_lock);
1254 }
1255
1256 /* kcm sock lock held */
1257 static void kcm_recv_enable(struct kcm_sock *kcm)
1258 {
1259         struct kcm_mux *mux = kcm->mux;
1260
1261         if (!kcm->rx_disabled)
1262                 return;
1263
1264         spin_lock_bh(&mux->rx_lock);
1265
1266         kcm->rx_disabled = 0;
1267         kcm_rcv_ready(kcm);
1268
1269         spin_unlock_bh(&mux->rx_lock);
1270 }
1271
1272 static int kcm_setsockopt(struct socket *sock, int level, int optname,
1273                           char __user *optval, unsigned int optlen)
1274 {
1275         struct kcm_sock *kcm = kcm_sk(sock->sk);
1276         int val, valbool;
1277         int err = 0;
1278
1279         if (level != SOL_KCM)
1280                 return -ENOPROTOOPT;
1281
1282         if (optlen < sizeof(int))
1283                 return -EINVAL;
1284
1285         if (get_user(val, (int __user *)optval))
1286                 return -EINVAL;
1287
1288         valbool = val ? 1 : 0;
1289
1290         switch (optname) {
1291         case KCM_RECV_DISABLE:
1292                 lock_sock(&kcm->sk);
1293                 if (valbool)
1294                         kcm_recv_disable(kcm);
1295                 else
1296                         kcm_recv_enable(kcm);
1297                 release_sock(&kcm->sk);
1298                 break;
1299         default:
1300                 err = -ENOPROTOOPT;
1301         }
1302
1303         return err;
1304 }
1305
1306 static int kcm_getsockopt(struct socket *sock, int level, int optname,
1307                           char __user *optval, int __user *optlen)
1308 {
1309         struct kcm_sock *kcm = kcm_sk(sock->sk);
1310         int val, len;
1311
1312         if (level != SOL_KCM)
1313                 return -ENOPROTOOPT;
1314
1315         if (get_user(len, optlen))
1316                 return -EFAULT;
1317
1318         len = min_t(unsigned int, len, sizeof(int));
1319         if (len < 0)
1320                 return -EINVAL;
1321
1322         switch (optname) {
1323         case KCM_RECV_DISABLE:
1324                 val = kcm->rx_disabled;
1325                 break;
1326         default:
1327                 return -ENOPROTOOPT;
1328         }
1329
1330         if (put_user(len, optlen))
1331                 return -EFAULT;
1332         if (copy_to_user(optval, &val, len))
1333                 return -EFAULT;
1334         return 0;
1335 }
1336
1337 static void init_kcm_sock(struct kcm_sock *kcm, struct kcm_mux *mux)
1338 {
1339         struct kcm_sock *tkcm;
1340         struct list_head *head;
1341         int index = 0;
1342
1343         /* For SOCK_SEQPACKET sock type, datagram_poll checks the sk_state, so
1344          * we set sk_state, otherwise epoll_wait always returns right away with
1345          * POLLHUP
1346          */
1347         kcm->sk.sk_state = TCP_ESTABLISHED;
1348
1349         /* Add to mux's kcm sockets list */
1350         kcm->mux = mux;
1351         spin_lock_bh(&mux->lock);
1352
1353         head = &mux->kcm_socks;
1354         list_for_each_entry(tkcm, &mux->kcm_socks, kcm_sock_list) {
1355                 if (tkcm->index != index)
1356                         break;
1357                 head = &tkcm->kcm_sock_list;
1358                 index++;
1359         }
1360
1361         list_add(&kcm->kcm_sock_list, head);
1362         kcm->index = index;
1363
1364         mux->kcm_socks_cnt++;
1365         spin_unlock_bh(&mux->lock);
1366
1367         INIT_WORK(&kcm->tx_work, kcm_tx_work);
1368
1369         spin_lock_bh(&mux->rx_lock);
1370         kcm_rcv_ready(kcm);
1371         spin_unlock_bh(&mux->rx_lock);
1372 }
1373
1374 static int kcm_attach(struct socket *sock, struct socket *csock,
1375                       struct bpf_prog *prog)
1376 {
1377         struct kcm_sock *kcm = kcm_sk(sock->sk);
1378         struct kcm_mux *mux = kcm->mux;
1379         struct sock *csk;
1380         struct kcm_psock *psock = NULL, *tpsock;
1381         struct list_head *head;
1382         int index = 0;
1383         struct strp_callbacks cb;
1384         int err = 0;
1385
1386         csk = csock->sk;
1387         if (!csk)
1388                 return -EINVAL;
1389
1390         lock_sock(csk);
1391
1392         /* Only allow TCP sockets to be attached for now */
1393         if ((csk->sk_family != AF_INET && csk->sk_family != AF_INET6) ||
1394             csk->sk_protocol != IPPROTO_TCP) {
1395                 err = -EOPNOTSUPP;
1396                 goto out;
1397         }
1398
1399         /* Don't allow listeners or closed sockets */
1400         if (csk->sk_state == TCP_LISTEN || csk->sk_state == TCP_CLOSE) {
1401                 err = -EOPNOTSUPP;
1402                 goto out;
1403         }
1404
1405         psock = kmem_cache_zalloc(kcm_psockp, GFP_KERNEL);
1406         if (!psock) {
1407                 err = -ENOMEM;
1408                 goto out;
1409         }
1410
1411         psock->mux = mux;
1412         psock->sk = csk;
1413         psock->bpf_prog = prog;
1414
1415         cb.rcv_msg = kcm_rcv_strparser;
1416         cb.abort_parser = NULL;
1417         cb.parse_msg = kcm_parse_func_strparser;
1418         cb.read_sock_done = kcm_read_sock_done;
1419
1420         err = strp_init(&psock->strp, csk, &cb);
1421         if (err) {
1422                 kmem_cache_free(kcm_psockp, psock);
1423                 goto out;
1424         }
1425
1426         write_lock_bh(&csk->sk_callback_lock);
1427
1428         /* Check if sk_user_data is aready by KCM or someone else.
1429          * Must be done under lock to prevent race conditions.
1430          */
1431         if (csk->sk_user_data) {
1432                 write_unlock_bh(&csk->sk_callback_lock);
1433                 strp_stop(&psock->strp);
1434                 strp_done(&psock->strp);
1435                 kmem_cache_free(kcm_psockp, psock);
1436                 err = -EALREADY;
1437                 goto out;
1438         }
1439
1440         psock->save_data_ready = csk->sk_data_ready;
1441         psock->save_write_space = csk->sk_write_space;
1442         psock->save_state_change = csk->sk_state_change;
1443         csk->sk_user_data = psock;
1444         csk->sk_data_ready = psock_data_ready;
1445         csk->sk_write_space = psock_write_space;
1446         csk->sk_state_change = psock_state_change;
1447
1448         write_unlock_bh(&csk->sk_callback_lock);
1449
1450         sock_hold(csk);
1451
1452         /* Finished initialization, now add the psock to the MUX. */
1453         spin_lock_bh(&mux->lock);
1454         head = &mux->psocks;
1455         list_for_each_entry(tpsock, &mux->psocks, psock_list) {
1456                 if (tpsock->index != index)
1457                         break;
1458                 head = &tpsock->psock_list;
1459                 index++;
1460         }
1461
1462         list_add(&psock->psock_list, head);
1463         psock->index = index;
1464
1465         KCM_STATS_INCR(mux->stats.psock_attach);
1466         mux->psocks_cnt++;
1467         psock_now_avail(psock);
1468         spin_unlock_bh(&mux->lock);
1469
1470         /* Schedule RX work in case there are already bytes queued */
1471         strp_check_rcv(&psock->strp);
1472
1473 out:
1474         release_sock(csk);
1475
1476         return err;
1477 }
1478
1479 static int kcm_attach_ioctl(struct socket *sock, struct kcm_attach *info)
1480 {
1481         struct socket *csock;
1482         struct bpf_prog *prog;
1483         int err;
1484
1485         csock = sockfd_lookup(info->fd, &err);
1486         if (!csock)
1487                 return -ENOENT;
1488
1489         prog = bpf_prog_get_type(info->bpf_fd, BPF_PROG_TYPE_SOCKET_FILTER);
1490         if (IS_ERR(prog)) {
1491                 err = PTR_ERR(prog);
1492                 goto out;
1493         }
1494
1495         err = kcm_attach(sock, csock, prog);
1496         if (err) {
1497                 bpf_prog_put(prog);
1498                 goto out;
1499         }
1500
1501         /* Keep reference on file also */
1502
1503         return 0;
1504 out:
1505         fput(csock->file);
1506         return err;
1507 }
1508
1509 static void kcm_unattach(struct kcm_psock *psock)
1510 {
1511         struct sock *csk = psock->sk;
1512         struct kcm_mux *mux = psock->mux;
1513
1514         lock_sock(csk);
1515
1516         /* Stop getting callbacks from TCP socket. After this there should
1517          * be no way to reserve a kcm for this psock.
1518          */
1519         write_lock_bh(&csk->sk_callback_lock);
1520         csk->sk_user_data = NULL;
1521         csk->sk_data_ready = psock->save_data_ready;
1522         csk->sk_write_space = psock->save_write_space;
1523         csk->sk_state_change = psock->save_state_change;
1524         strp_stop(&psock->strp);
1525
1526         if (WARN_ON(psock->rx_kcm)) {
1527                 write_unlock_bh(&csk->sk_callback_lock);
1528                 release_sock(csk);
1529                 return;
1530         }
1531
1532         spin_lock_bh(&mux->rx_lock);
1533
1534         /* Stop receiver activities. After this point psock should not be
1535          * able to get onto ready list either through callbacks or work.
1536          */
1537         if (psock->ready_rx_msg) {
1538                 list_del(&psock->psock_ready_list);
1539                 kfree_skb(psock->ready_rx_msg);
1540                 psock->ready_rx_msg = NULL;
1541                 KCM_STATS_INCR(mux->stats.rx_ready_drops);
1542         }
1543
1544         spin_unlock_bh(&mux->rx_lock);
1545
1546         write_unlock_bh(&csk->sk_callback_lock);
1547
1548         /* Call strp_done without sock lock */
1549         release_sock(csk);
1550         strp_done(&psock->strp);
1551         lock_sock(csk);
1552
1553         bpf_prog_put(psock->bpf_prog);
1554
1555         spin_lock_bh(&mux->lock);
1556
1557         aggregate_psock_stats(&psock->stats, &mux->aggregate_psock_stats);
1558         save_strp_stats(&psock->strp, &mux->aggregate_strp_stats);
1559
1560         KCM_STATS_INCR(mux->stats.psock_unattach);
1561
1562         if (psock->tx_kcm) {
1563                 /* psock was reserved.  Just mark it finished and we will clean
1564                  * up in the kcm paths, we need kcm lock which can not be
1565                  * acquired here.
1566                  */
1567                 KCM_STATS_INCR(mux->stats.psock_unattach_rsvd);
1568                 spin_unlock_bh(&mux->lock);
1569
1570                 /* We are unattaching a socket that is reserved. Abort the
1571                  * socket since we may be out of sync in sending on it. We need
1572                  * to do this without the mux lock.
1573                  */
1574                 kcm_abort_tx_psock(psock, EPIPE, false);
1575
1576                 spin_lock_bh(&mux->lock);
1577                 if (!psock->tx_kcm) {
1578                         /* psock now unreserved in window mux was unlocked */
1579                         goto no_reserved;
1580                 }
1581                 psock->done = 1;
1582
1583                 /* Commit done before queuing work to process it */
1584                 smp_mb();
1585
1586                 /* Queue tx work to make sure psock->done is handled */
1587                 queue_work(kcm_wq, &psock->tx_kcm->tx_work);
1588                 spin_unlock_bh(&mux->lock);
1589         } else {
1590 no_reserved:
1591                 if (!psock->tx_stopped)
1592                         list_del(&psock->psock_avail_list);
1593                 list_del(&psock->psock_list);
1594                 mux->psocks_cnt--;
1595                 spin_unlock_bh(&mux->lock);
1596
1597                 sock_put(csk);
1598                 fput(csk->sk_socket->file);
1599                 kmem_cache_free(kcm_psockp, psock);
1600         }
1601
1602         release_sock(csk);
1603 }
1604
1605 static int kcm_unattach_ioctl(struct socket *sock, struct kcm_unattach *info)
1606 {
1607         struct kcm_sock *kcm = kcm_sk(sock->sk);
1608         struct kcm_mux *mux = kcm->mux;
1609         struct kcm_psock *psock;
1610         struct socket *csock;
1611         struct sock *csk;
1612         int err;
1613
1614         csock = sockfd_lookup(info->fd, &err);
1615         if (!csock)
1616                 return -ENOENT;
1617
1618         csk = csock->sk;
1619         if (!csk) {
1620                 err = -EINVAL;
1621                 goto out;
1622         }
1623
1624         err = -ENOENT;
1625
1626         spin_lock_bh(&mux->lock);
1627
1628         list_for_each_entry(psock, &mux->psocks, psock_list) {
1629                 if (psock->sk != csk)
1630                         continue;
1631
1632                 /* Found the matching psock */
1633
1634                 if (psock->unattaching || WARN_ON(psock->done)) {
1635                         err = -EALREADY;
1636                         break;
1637                 }
1638
1639                 psock->unattaching = 1;
1640
1641                 spin_unlock_bh(&mux->lock);
1642
1643                 /* Lower socket lock should already be held */
1644                 kcm_unattach(psock);
1645
1646                 err = 0;
1647                 goto out;
1648         }
1649
1650         spin_unlock_bh(&mux->lock);
1651
1652 out:
1653         fput(csock->file);
1654         return err;
1655 }
1656
1657 static struct proto kcm_proto = {
1658         .name   = "KCM",
1659         .owner  = THIS_MODULE,
1660         .obj_size = sizeof(struct kcm_sock),
1661 };
1662
1663 /* Clone a kcm socket. */
1664 static struct file *kcm_clone(struct socket *osock)
1665 {
1666         struct socket *newsock;
1667         struct sock *newsk;
1668         struct file *file;
1669
1670         newsock = sock_alloc();
1671         if (!newsock)
1672                 return ERR_PTR(-ENFILE);
1673
1674         newsock->type = osock->type;
1675         newsock->ops = osock->ops;
1676
1677         __module_get(newsock->ops->owner);
1678
1679         newsk = sk_alloc(sock_net(osock->sk), PF_KCM, GFP_KERNEL,
1680                          &kcm_proto, false);
1681         if (!newsk) {
1682                 sock_release(newsock);
1683                 return ERR_PTR(-ENOMEM);
1684         }
1685         sock_init_data(newsock, newsk);
1686         init_kcm_sock(kcm_sk(newsk), kcm_sk(osock->sk)->mux);
1687
1688         file = sock_alloc_file(newsock, 0, osock->sk->sk_prot_creator->name);
1689         if (IS_ERR(file))
1690                 sock_release(newsock);
1691
1692         return file;
1693 }
1694
1695 static int kcm_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1696 {
1697         int err;
1698
1699         switch (cmd) {
1700         case SIOCKCMATTACH: {
1701                 struct kcm_attach info;
1702
1703                 if (copy_from_user(&info, (void __user *)arg, sizeof(info)))
1704                         return -EFAULT;
1705
1706                 err = kcm_attach_ioctl(sock, &info);
1707
1708                 break;
1709         }
1710         case SIOCKCMUNATTACH: {
1711                 struct kcm_unattach info;
1712
1713                 if (copy_from_user(&info, (void __user *)arg, sizeof(info)))
1714                         return -EFAULT;
1715
1716                 err = kcm_unattach_ioctl(sock, &info);
1717
1718                 break;
1719         }
1720         case SIOCKCMCLONE: {
1721                 struct kcm_clone info;
1722                 struct file *file;
1723
1724                 info.fd = get_unused_fd_flags(0);
1725                 if (unlikely(info.fd < 0))
1726                         return info.fd;
1727
1728                 file = kcm_clone(sock);
1729                 if (IS_ERR(file)) {
1730                         put_unused_fd(info.fd);
1731                         return PTR_ERR(file);
1732                 }
1733                 if (copy_to_user((void __user *)arg, &info,
1734                                  sizeof(info))) {
1735                         put_unused_fd(info.fd);
1736                         fput(file);
1737                         return -EFAULT;
1738                 }
1739                 fd_install(info.fd, file);
1740                 err = 0;
1741                 break;
1742         }
1743         default:
1744                 err = -ENOIOCTLCMD;
1745                 break;
1746         }
1747
1748         return err;
1749 }
1750
1751 static void free_mux(struct rcu_head *rcu)
1752 {
1753         struct kcm_mux *mux = container_of(rcu,
1754             struct kcm_mux, rcu);
1755
1756         kmem_cache_free(kcm_muxp, mux);
1757 }
1758
1759 static void release_mux(struct kcm_mux *mux)
1760 {
1761         struct kcm_net *knet = mux->knet;
1762         struct kcm_psock *psock, *tmp_psock;
1763
1764         /* Release psocks */
1765         list_for_each_entry_safe(psock, tmp_psock,
1766                                  &mux->psocks, psock_list) {
1767                 if (!WARN_ON(psock->unattaching))
1768                         kcm_unattach(psock);
1769         }
1770
1771         if (WARN_ON(mux->psocks_cnt))
1772                 return;
1773
1774         __skb_queue_purge(&mux->rx_hold_queue);
1775
1776         mutex_lock(&knet->mutex);
1777         aggregate_mux_stats(&mux->stats, &knet->aggregate_mux_stats);
1778         aggregate_psock_stats(&mux->aggregate_psock_stats,
1779                               &knet->aggregate_psock_stats);
1780         aggregate_strp_stats(&mux->aggregate_strp_stats,
1781                              &knet->aggregate_strp_stats);
1782         list_del_rcu(&mux->kcm_mux_list);
1783         knet->count--;
1784         mutex_unlock(&knet->mutex);
1785
1786         call_rcu(&mux->rcu, free_mux);
1787 }
1788
1789 static void kcm_done(struct kcm_sock *kcm)
1790 {
1791         struct kcm_mux *mux = kcm->mux;
1792         struct sock *sk = &kcm->sk;
1793         int socks_cnt;
1794
1795         spin_lock_bh(&mux->rx_lock);
1796         if (kcm->rx_psock) {
1797                 /* Cleanup in unreserve_rx_kcm */
1798                 WARN_ON(kcm->done);
1799                 kcm->rx_disabled = 1;
1800                 kcm->done = 1;
1801                 spin_unlock_bh(&mux->rx_lock);
1802                 return;
1803         }
1804
1805         if (kcm->rx_wait) {
1806                 list_del(&kcm->wait_rx_list);
1807                 /* paired with lockless reads in kcm_rfree() */
1808                 WRITE_ONCE(kcm->rx_wait, false);
1809         }
1810         /* Move any pending receive messages to other kcm sockets */
1811         requeue_rx_msgs(mux, &sk->sk_receive_queue);
1812
1813         spin_unlock_bh(&mux->rx_lock);
1814
1815         if (WARN_ON(sk_rmem_alloc_get(sk)))
1816                 return;
1817
1818         /* Detach from MUX */
1819         spin_lock_bh(&mux->lock);
1820
1821         list_del(&kcm->kcm_sock_list);
1822         mux->kcm_socks_cnt--;
1823         socks_cnt = mux->kcm_socks_cnt;
1824
1825         spin_unlock_bh(&mux->lock);
1826
1827         if (!socks_cnt) {
1828                 /* We are done with the mux now. */
1829                 release_mux(mux);
1830         }
1831
1832         WARN_ON(kcm->rx_wait);
1833
1834         sock_put(&kcm->sk);
1835 }
1836
1837 /* Called by kcm_release to close a KCM socket.
1838  * If this is the last KCM socket on the MUX, destroy the MUX.
1839  */
1840 static int kcm_release(struct socket *sock)
1841 {
1842         struct sock *sk = sock->sk;
1843         struct kcm_sock *kcm;
1844         struct kcm_mux *mux;
1845         struct kcm_psock *psock;
1846
1847         if (!sk)
1848                 return 0;
1849
1850         kcm = kcm_sk(sk);
1851         mux = kcm->mux;
1852
1853         sock_orphan(sk);
1854         kfree_skb(kcm->seq_skb);
1855
1856         lock_sock(sk);
1857         /* Purge queue under lock to avoid race condition with tx_work trying
1858          * to act when queue is nonempty. If tx_work runs after this point
1859          * it will just return.
1860          */
1861         __skb_queue_purge(&sk->sk_write_queue);
1862
1863         /* Set tx_stopped. This is checked when psock is bound to a kcm and we
1864          * get a writespace callback. This prevents further work being queued
1865          * from the callback (unbinding the psock occurs after canceling work.
1866          */
1867         kcm->tx_stopped = 1;
1868
1869         release_sock(sk);
1870
1871         spin_lock_bh(&mux->lock);
1872         if (kcm->tx_wait) {
1873                 /* Take of tx_wait list, after this point there should be no way
1874                  * that a psock will be assigned to this kcm.
1875                  */
1876                 list_del(&kcm->wait_psock_list);
1877                 kcm->tx_wait = false;
1878         }
1879         spin_unlock_bh(&mux->lock);
1880
1881         /* Cancel work. After this point there should be no outside references
1882          * to the kcm socket.
1883          */
1884         cancel_work_sync(&kcm->tx_work);
1885
1886         lock_sock(sk);
1887         psock = kcm->tx_psock;
1888         if (psock) {
1889                 /* A psock was reserved, so we need to kill it since it
1890                  * may already have some bytes queued from a message. We
1891                  * need to do this after removing kcm from tx_wait list.
1892                  */
1893                 kcm_abort_tx_psock(psock, EPIPE, false);
1894                 unreserve_psock(kcm);
1895         }
1896         release_sock(sk);
1897
1898         WARN_ON(kcm->tx_wait);
1899         WARN_ON(kcm->tx_psock);
1900
1901         sock->sk = NULL;
1902
1903         kcm_done(kcm);
1904
1905         return 0;
1906 }
1907
1908 static const struct proto_ops kcm_dgram_ops = {
1909         .family =       PF_KCM,
1910         .owner =        THIS_MODULE,
1911         .release =      kcm_release,
1912         .bind =         sock_no_bind,
1913         .connect =      sock_no_connect,
1914         .socketpair =   sock_no_socketpair,
1915         .accept =       sock_no_accept,
1916         .getname =      sock_no_getname,
1917         .poll =         datagram_poll,
1918         .ioctl =        kcm_ioctl,
1919         .listen =       sock_no_listen,
1920         .shutdown =     sock_no_shutdown,
1921         .setsockopt =   kcm_setsockopt,
1922         .getsockopt =   kcm_getsockopt,
1923         .sendmsg =      kcm_sendmsg,
1924         .recvmsg =      kcm_recvmsg,
1925         .mmap =         sock_no_mmap,
1926         .sendpage =     kcm_sendpage,
1927 };
1928
1929 static const struct proto_ops kcm_seqpacket_ops = {
1930         .family =       PF_KCM,
1931         .owner =        THIS_MODULE,
1932         .release =      kcm_release,
1933         .bind =         sock_no_bind,
1934         .connect =      sock_no_connect,
1935         .socketpair =   sock_no_socketpair,
1936         .accept =       sock_no_accept,
1937         .getname =      sock_no_getname,
1938         .poll =         datagram_poll,
1939         .ioctl =        kcm_ioctl,
1940         .listen =       sock_no_listen,
1941         .shutdown =     sock_no_shutdown,
1942         .setsockopt =   kcm_setsockopt,
1943         .getsockopt =   kcm_getsockopt,
1944         .sendmsg =      kcm_sendmsg,
1945         .recvmsg =      kcm_recvmsg,
1946         .mmap =         sock_no_mmap,
1947         .sendpage =     kcm_sendpage,
1948         .splice_read =  kcm_splice_read,
1949 };
1950
1951 /* Create proto operation for kcm sockets */
1952 static int kcm_create(struct net *net, struct socket *sock,
1953                       int protocol, int kern)
1954 {
1955         struct kcm_net *knet = net_generic(net, kcm_net_id);
1956         struct sock *sk;
1957         struct kcm_mux *mux;
1958
1959         switch (sock->type) {
1960         case SOCK_DGRAM:
1961                 sock->ops = &kcm_dgram_ops;
1962                 break;
1963         case SOCK_SEQPACKET:
1964                 sock->ops = &kcm_seqpacket_ops;
1965                 break;
1966         default:
1967                 return -ESOCKTNOSUPPORT;
1968         }
1969
1970         if (protocol != KCMPROTO_CONNECTED)
1971                 return -EPROTONOSUPPORT;
1972
1973         sk = sk_alloc(net, PF_KCM, GFP_KERNEL, &kcm_proto, kern);
1974         if (!sk)
1975                 return -ENOMEM;
1976
1977         /* Allocate a kcm mux, shared between KCM sockets */
1978         mux = kmem_cache_zalloc(kcm_muxp, GFP_KERNEL);
1979         if (!mux) {
1980                 sk_free(sk);
1981                 return -ENOMEM;
1982         }
1983
1984         spin_lock_init(&mux->lock);
1985         spin_lock_init(&mux->rx_lock);
1986         INIT_LIST_HEAD(&mux->kcm_socks);
1987         INIT_LIST_HEAD(&mux->kcm_rx_waiters);
1988         INIT_LIST_HEAD(&mux->kcm_tx_waiters);
1989
1990         INIT_LIST_HEAD(&mux->psocks);
1991         INIT_LIST_HEAD(&mux->psocks_ready);
1992         INIT_LIST_HEAD(&mux->psocks_avail);
1993
1994         mux->knet = knet;
1995
1996         /* Add new MUX to list */
1997         mutex_lock(&knet->mutex);
1998         list_add_rcu(&mux->kcm_mux_list, &knet->mux_list);
1999         knet->count++;
2000         mutex_unlock(&knet->mutex);
2001
2002         skb_queue_head_init(&mux->rx_hold_queue);
2003
2004         /* Init KCM socket */
2005         sock_init_data(sock, sk);
2006         init_kcm_sock(kcm_sk(sk), mux);
2007
2008         return 0;
2009 }
2010
2011 static struct net_proto_family kcm_family_ops = {
2012         .family = PF_KCM,
2013         .create = kcm_create,
2014         .owner  = THIS_MODULE,
2015 };
2016
2017 static __net_init int kcm_init_net(struct net *net)
2018 {
2019         struct kcm_net *knet = net_generic(net, kcm_net_id);
2020
2021         INIT_LIST_HEAD_RCU(&knet->mux_list);
2022         mutex_init(&knet->mutex);
2023
2024         return 0;
2025 }
2026
2027 static __net_exit void kcm_exit_net(struct net *net)
2028 {
2029         struct kcm_net *knet = net_generic(net, kcm_net_id);
2030
2031         /* All KCM sockets should be closed at this point, which should mean
2032          * that all multiplexors and psocks have been destroyed.
2033          */
2034         WARN_ON(!list_empty(&knet->mux_list));
2035 }
2036
2037 static struct pernet_operations kcm_net_ops = {
2038         .init = kcm_init_net,
2039         .exit = kcm_exit_net,
2040         .id   = &kcm_net_id,
2041         .size = sizeof(struct kcm_net),
2042 };
2043
2044 static int __init kcm_init(void)
2045 {
2046         int err = -ENOMEM;
2047
2048         kcm_muxp = kmem_cache_create("kcm_mux_cache",
2049                                      sizeof(struct kcm_mux), 0,
2050                                      SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2051         if (!kcm_muxp)
2052                 goto fail;
2053
2054         kcm_psockp = kmem_cache_create("kcm_psock_cache",
2055                                        sizeof(struct kcm_psock), 0,
2056                                         SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2057         if (!kcm_psockp)
2058                 goto fail;
2059
2060         kcm_wq = create_singlethread_workqueue("kkcmd");
2061         if (!kcm_wq)
2062                 goto fail;
2063
2064         err = proto_register(&kcm_proto, 1);
2065         if (err)
2066                 goto fail;
2067
2068         err = register_pernet_device(&kcm_net_ops);
2069         if (err)
2070                 goto net_ops_fail;
2071
2072         err = sock_register(&kcm_family_ops);
2073         if (err)
2074                 goto sock_register_fail;
2075
2076         err = kcm_proc_init();
2077         if (err)
2078                 goto proc_init_fail;
2079
2080         return 0;
2081
2082 proc_init_fail:
2083         sock_unregister(PF_KCM);
2084
2085 sock_register_fail:
2086         unregister_pernet_device(&kcm_net_ops);
2087
2088 net_ops_fail:
2089         proto_unregister(&kcm_proto);
2090
2091 fail:
2092         kmem_cache_destroy(kcm_muxp);
2093         kmem_cache_destroy(kcm_psockp);
2094
2095         if (kcm_wq)
2096                 destroy_workqueue(kcm_wq);
2097
2098         return err;
2099 }
2100
2101 static void __exit kcm_exit(void)
2102 {
2103         kcm_proc_exit();
2104         sock_unregister(PF_KCM);
2105         unregister_pernet_device(&kcm_net_ops);
2106         proto_unregister(&kcm_proto);
2107         destroy_workqueue(kcm_wq);
2108
2109         kmem_cache_destroy(kcm_muxp);
2110         kmem_cache_destroy(kcm_psockp);
2111 }
2112
2113 module_init(kcm_init);
2114 module_exit(kcm_exit);
2115
2116 MODULE_LICENSE("GPL");
2117 MODULE_ALIAS_NETPROTO(PF_KCM);
2118