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