1/* 2 * Copyright (c) 2000-2012 Apple Inc. All rights reserved. 3 * 4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ 5 * 6 * This file contains Original Code and/or Modifications of Original Code 7 * as defined in and that are subject to the Apple Public Source License 8 * Version 2.0 (the 'License'). You may not use this file except in 9 * compliance with the License. The rights granted to you under the License 10 * may not be used to create, or enable the creation or redistribution of, 11 * unlawful or unlicensed copies of an Apple operating system, or to 12 * circumvent, violate, or enable the circumvention or violation of, any 13 * terms of an Apple operating system software license agreement. 14 * 15 * Please obtain a copy of the License at 16 * http://www.opensource.apple.com/apsl/ and read it before using this file. 17 * 18 * The Original Code and all software distributed under the License are 19 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER 20 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, 21 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY, 22 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT. 23 * Please see the License for the specific language governing rights and 24 * limitations under the License. 25 * 26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@ 27 */ 28/* 29 * Copyright (c) 1982, 1986, 1989, 1991, 1993 30 * The Regents of the University of California. All rights reserved. 31 * 32 * Redistribution and use in source and binary forms, with or without 33 * modification, are permitted provided that the following conditions 34 * are met: 35 * 1. Redistributions of source code must retain the above copyright 36 * notice, this list of conditions and the following disclaimer. 37 * 2. Redistributions in binary form must reproduce the above copyright 38 * notice, this list of conditions and the following disclaimer in the 39 * documentation and/or other materials provided with the distribution. 40 * 3. All advertising materials mentioning features or use of this software 41 * must display the following acknowledgement: 42 * This product includes software developed by the University of 43 * California, Berkeley and its contributors. 44 * 4. Neither the name of the University nor the names of its contributors 45 * may be used to endorse or promote products derived from this software 46 * without specific prior written permission. 47 * 48 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 49 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 50 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 51 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 52 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 53 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 54 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 55 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 56 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 57 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 58 * SUCH DAMAGE. 59 * 60 * From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94 61 */ 62/* 63 * NOTICE: This file was modified by SPARTA, Inc. in 2005 to introduce 64 * support for mandatory and extensible security protections. This notice 65 * is included in support of clause 2.2 (b) of the Apple Public License, 66 * Version 2.0. 67 */ 68 69#include <sys/param.h> 70#include <sys/systm.h> 71#include <sys/kernel.h> 72#include <sys/domain.h> 73#include <sys/fcntl.h> 74#include <sys/malloc.h> /* XXX must be before <sys/file.h> */ 75#include <sys/file_internal.h> 76#include <sys/guarded.h> 77#include <sys/filedesc.h> 78#include <sys/lock.h> 79#include <sys/mbuf.h> 80#include <sys/namei.h> 81#include <sys/proc_internal.h> 82#include <sys/kauth.h> 83#include <sys/protosw.h> 84#include <sys/socket.h> 85#include <sys/socketvar.h> 86#include <sys/stat.h> 87#include <sys/sysctl.h> 88#include <sys/un.h> 89#include <sys/unpcb.h> 90#include <sys/vnode_internal.h> 91#include <sys/kdebug.h> 92 93#include <kern/zalloc.h> 94#include <kern/locks.h> 95 96#if CONFIG_MACF 97#include <security/mac_framework.h> 98#endif /* CONFIG_MACF */ 99 100#include <mach/vm_param.h> 101 102#define f_msgcount f_fglob->fg_msgcount 103#define f_cred f_fglob->fg_cred 104#define f_ops f_fglob->fg_ops 105#define f_offset f_fglob->fg_offset 106#define f_data f_fglob->fg_data 107struct zone *unp_zone; 108static unp_gen_t unp_gencnt; 109static u_int unp_count; 110 111static lck_attr_t *unp_mtx_attr; 112static lck_grp_t *unp_mtx_grp; 113static lck_grp_attr_t *unp_mtx_grp_attr; 114static lck_rw_t *unp_list_mtx; 115 116static lck_mtx_t *unp_disconnect_lock; 117static lck_mtx_t *unp_connect_lock; 118static u_int disconnect_in_progress; 119 120extern lck_mtx_t *uipc_lock; 121static struct unp_head unp_shead, unp_dhead; 122 123/* 124 * mDNSResponder tracing. When enabled, endpoints connected to 125 * /var/run/mDNSResponder will be traced; during each send on 126 * the traced socket, we log the PID and process name of the 127 * sending process. We also print out a bit of info related 128 * to the data itself; this assumes ipc_msg_hdr in dnssd_ipc.h 129 * of mDNSResponder stays the same. 130 */ 131#define MDNSRESPONDER_PATH "/var/run/mDNSResponder" 132 133static int unpst_tracemdns; /* enable tracing */ 134 135#define MDNS_IPC_MSG_HDR_VERSION_1 1 136 137struct mdns_ipc_msg_hdr { 138 uint32_t version; 139 uint32_t datalen; 140 uint32_t ipc_flags; 141 uint32_t op; 142 union { 143 void *context; 144 uint32_t u32[2]; 145 } __attribute__((packed)); 146 uint32_t reg_index; 147} __attribute__((packed)); 148 149/* 150 * Unix communications domain. 151 * 152 * TODO: 153 * SEQPACKET, RDM 154 * rethink name space problems 155 * need a proper out-of-band 156 * lock pushdown 157 */ 158static struct sockaddr sun_noname = { sizeof (sun_noname), AF_LOCAL, { 0 } }; 159static ino_t unp_ino; /* prototype for fake inode numbers */ 160 161static int unp_attach(struct socket *); 162static void unp_detach(struct unpcb *); 163static int unp_bind(struct unpcb *, struct sockaddr *, proc_t); 164static int unp_connect(struct socket *, struct sockaddr *, proc_t); 165static void unp_disconnect(struct unpcb *); 166static void unp_shutdown(struct unpcb *); 167static void unp_drop(struct unpcb *, int); 168__private_extern__ void unp_gc(void); 169static void unp_scan(struct mbuf *, void (*)(struct fileglob *)); 170static void unp_mark(struct fileglob *); 171static void unp_discard(struct fileglob *); 172static void unp_discard_fdlocked(struct fileglob *, proc_t); 173static int unp_internalize(struct mbuf *, proc_t); 174static int unp_listen(struct unpcb *, proc_t); 175static void unpcb_to_compat(struct unpcb *, struct unpcb_compat *); 176static void unp_get_locks_in_order(struct socket *so, struct socket *conn_so); 177 178static void 179unp_get_locks_in_order(struct socket *so, struct socket *conn_so) 180{ 181 if (so < conn_so) { 182 socket_lock(conn_so, 1); 183 } else { 184 struct unpcb *unp = sotounpcb(so); 185 unp->unp_flags |= UNP_DONTDISCONNECT; 186 unp->rw_thrcount++; 187 socket_unlock(so, 0); 188 189 /* Get the locks in the correct order */ 190 socket_lock(conn_so, 1); 191 socket_lock(so, 0); 192 unp->rw_thrcount--; 193 if (unp->rw_thrcount == 0) { 194 unp->unp_flags &= ~UNP_DONTDISCONNECT; 195 wakeup(unp); 196 } 197 } 198} 199 200static int 201uipc_abort(struct socket *so) 202{ 203 struct unpcb *unp = sotounpcb(so); 204 205 if (unp == 0) 206 return (EINVAL); 207 unp_drop(unp, ECONNABORTED); 208 unp_detach(unp); 209 sofree(so); 210 return (0); 211} 212 213static int 214uipc_accept(struct socket *so, struct sockaddr **nam) 215{ 216 struct unpcb *unp = sotounpcb(so); 217 218 if (unp == 0) 219 return (EINVAL); 220 221 /* 222 * Pass back name of connected socket, 223 * if it was bound and we are still connected 224 * (our peer may have closed already!). 225 */ 226 if (unp->unp_conn && unp->unp_conn->unp_addr) { 227 *nam = dup_sockaddr((struct sockaddr *) 228 unp->unp_conn->unp_addr, 1); 229 } else { 230 *nam = dup_sockaddr((struct sockaddr *)&sun_noname, 1); 231 } 232 return (0); 233} 234 235/* 236 * Returns: 0 Success 237 * EISCONN 238 * unp_attach: 239 */ 240static int 241uipc_attach(struct socket *so, __unused int proto, __unused proc_t p) 242{ 243 struct unpcb *unp = sotounpcb(so); 244 245 if (unp != 0) 246 return (EISCONN); 247 return (unp_attach(so)); 248} 249 250static int 251uipc_bind(struct socket *so, struct sockaddr *nam, proc_t p) 252{ 253 struct unpcb *unp = sotounpcb(so); 254 255 if (unp == 0) 256 return (EINVAL); 257 258 return (unp_bind(unp, nam, p)); 259} 260 261/* 262 * Returns: 0 Success 263 * EINVAL 264 * unp_connect:??? [See elsewhere in this file] 265 */ 266static int 267uipc_connect(struct socket *so, struct sockaddr *nam, proc_t p) 268{ 269 struct unpcb *unp = sotounpcb(so); 270 271 if (unp == 0) 272 return (EINVAL); 273 return (unp_connect(so, nam, p)); 274} 275 276/* 277 * Returns: 0 Success 278 * EINVAL 279 * unp_connect2:EPROTOTYPE Protocol wrong type for socket 280 * unp_connect2:EINVAL Invalid argument 281 */ 282static int 283uipc_connect2(struct socket *so1, struct socket *so2) 284{ 285 struct unpcb *unp = sotounpcb(so1); 286 287 if (unp == 0) 288 return (EINVAL); 289 290 return (unp_connect2(so1, so2)); 291} 292 293/* control is EOPNOTSUPP */ 294 295static int 296uipc_detach(struct socket *so) 297{ 298 struct unpcb *unp = sotounpcb(so); 299 300 if (unp == 0) 301 return (EINVAL); 302 303 lck_mtx_assert(&unp->unp_mtx, LCK_MTX_ASSERT_OWNED); 304 unp_detach(unp); 305 return (0); 306} 307 308static int 309uipc_disconnect(struct socket *so) 310{ 311 struct unpcb *unp = sotounpcb(so); 312 313 if (unp == 0) 314 return (EINVAL); 315 unp_disconnect(unp); 316 return (0); 317} 318 319/* 320 * Returns: 0 Success 321 * EINVAL 322 */ 323static int 324uipc_listen(struct socket *so, __unused proc_t p) 325{ 326 struct unpcb *unp = sotounpcb(so); 327 328 if (unp == 0 || unp->unp_vnode == 0) 329 return (EINVAL); 330 return (unp_listen(unp, p)); 331} 332 333static int 334uipc_peeraddr(struct socket *so, struct sockaddr **nam) 335{ 336 struct unpcb *unp = sotounpcb(so); 337 338 if (unp == NULL) 339 return (EINVAL); 340 if (unp->unp_conn != NULL && unp->unp_conn->unp_addr != NULL) { 341 *nam = dup_sockaddr((struct sockaddr *) 342 unp->unp_conn->unp_addr, 1); 343 } else { 344 *nam = dup_sockaddr((struct sockaddr *)&sun_noname, 1); 345 } 346 return (0); 347} 348 349static int 350uipc_rcvd(struct socket *so, __unused int flags) 351{ 352 struct unpcb *unp = sotounpcb(so); 353 struct socket *so2; 354 355 if (unp == 0) 356 return (EINVAL); 357 switch (so->so_type) { 358 case SOCK_DGRAM: 359 panic("uipc_rcvd DGRAM?"); 360 /*NOTREACHED*/ 361 362 case SOCK_STREAM: 363#define rcv (&so->so_rcv) 364#define snd (&so2->so_snd) 365 if (unp->unp_conn == 0) 366 break; 367 368 so2 = unp->unp_conn->unp_socket; 369 unp_get_locks_in_order(so, so2); 370 /* 371 * Adjust backpressure on sender 372 * and wakeup any waiting to write. 373 */ 374 snd->sb_mbmax += unp->unp_mbcnt - rcv->sb_mbcnt; 375 unp->unp_mbcnt = rcv->sb_mbcnt; 376 snd->sb_hiwat += unp->unp_cc - rcv->sb_cc; 377 unp->unp_cc = rcv->sb_cc; 378 sowwakeup(so2); 379 380 socket_unlock(so2, 1); 381 382#undef snd 383#undef rcv 384 break; 385 386 default: 387 panic("uipc_rcvd unknown socktype"); 388 } 389 return (0); 390} 391 392/* pru_rcvoob is EOPNOTSUPP */ 393 394/* 395 * Returns: 0 Success 396 * EINVAL 397 * EOPNOTSUPP 398 * EPIPE 399 * ENOTCONN 400 * EISCONN 401 * unp_internalize:EINVAL 402 * unp_internalize:EBADF 403 * unp_connect:EAFNOSUPPORT Address family not supported 404 * unp_connect:EINVAL Invalid argument 405 * unp_connect:ENOTSOCK Not a socket 406 * unp_connect:ECONNREFUSED Connection refused 407 * unp_connect:EISCONN Socket is connected 408 * unp_connect:EPROTOTYPE Protocol wrong type for socket 409 * unp_connect:??? 410 * sbappendaddr:ENOBUFS [5th argument, contents modified] 411 * sbappendaddr:??? [whatever a filter author chooses] 412 */ 413static int 414uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 415 struct mbuf *control, proc_t p) 416{ 417 int error = 0; 418 struct unpcb *unp = sotounpcb(so); 419 struct socket *so2; 420 421 if (unp == 0) { 422 error = EINVAL; 423 goto release; 424 } 425 if (flags & PRUS_OOB) { 426 error = EOPNOTSUPP; 427 goto release; 428 } 429 430 if (control) { 431 /* release lock to avoid deadlock (4436174) */ 432 socket_unlock(so, 0); 433 error = unp_internalize(control, p); 434 socket_lock(so, 0); 435 if (error) 436 goto release; 437 } 438 439 switch (so->so_type) { 440 case SOCK_DGRAM: 441 { 442 struct sockaddr *from; 443 444 if (nam) { 445 if (unp->unp_conn) { 446 error = EISCONN; 447 break; 448 } 449 error = unp_connect(so, nam, p); 450 if (error) 451 break; 452 } else { 453 if (unp->unp_conn == 0) { 454 error = ENOTCONN; 455 break; 456 } 457 } 458 459 so2 = unp->unp_conn->unp_socket; 460 if (so != so2) 461 unp_get_locks_in_order(so, so2); 462 463 if (unp->unp_addr) 464 from = (struct sockaddr *)unp->unp_addr; 465 else 466 from = &sun_noname; 467 /* 468 * sbappendaddr() will fail when the receiver runs out of 469 * space; in contrast to SOCK_STREAM, we will lose messages 470 * for the SOCK_DGRAM case when the receiver's queue overflows. 471 * SB_UNIX on the socket buffer implies that the callee will 472 * not free the control message, if any, because we would need 473 * to call unp_dispose() on it. 474 */ 475 if (sbappendaddr(&so2->so_rcv, from, m, control, &error)) { 476 control = NULL; 477 sorwakeup(so2); 478 } else if (control != NULL && error == 0) { 479 /* A socket filter took control; don't touch it */ 480 control = NULL; 481 } 482 483 if (so != so2) 484 socket_unlock(so2, 1); 485 486 m = NULL; 487 if (nam) 488 unp_disconnect(unp); 489 break; 490 } 491 492 case SOCK_STREAM: { 493 int didreceive = 0; 494#define rcv (&so2->so_rcv) 495#define snd (&so->so_snd) 496 /* Connect if not connected yet. */ 497 /* 498 * Note: A better implementation would complain 499 * if not equal to the peer's address. 500 */ 501 if ((so->so_state & SS_ISCONNECTED) == 0) { 502 if (nam) { 503 error = unp_connect(so, nam, p); 504 if (error) 505 break; /* XXX */ 506 } else { 507 error = ENOTCONN; 508 break; 509 } 510 } 511 512 if (so->so_state & SS_CANTSENDMORE) { 513 error = EPIPE; 514 break; 515 } 516 if (unp->unp_conn == 0) 517 panic("uipc_send connected but no connection?"); 518 519 so2 = unp->unp_conn->unp_socket; 520 unp_get_locks_in_order(so, so2); 521 522 /* Check socket state again as we might have unlocked the socket 523 * while trying to get the locks in order 524 */ 525 526 if ((so->so_state & SS_CANTSENDMORE)) { 527 error = EPIPE; 528 socket_unlock(so2, 1); 529 break; 530 } 531 532 if (unp->unp_flags & UNP_TRACE_MDNS) { 533 struct mdns_ipc_msg_hdr hdr; 534 535 if (mbuf_copydata(m, 0, sizeof (hdr), &hdr) == 0 && 536 hdr.version == ntohl(MDNS_IPC_MSG_HDR_VERSION_1)) { 537 printf("%s[mDNSResponder] pid=%d (%s): op=0x%x\n", 538 __func__, p->p_pid, p->p_comm, ntohl(hdr.op)); 539 } 540 } 541 542 /* 543 * Send to paired receive port, and then reduce send buffer 544 * hiwater marks to maintain backpressure. Wake up readers. 545 * SB_UNIX flag will allow new record to be appended to the 546 * receiver's queue even when it is already full. It is 547 * possible, however, that append might fail. In that case, 548 * we will need to call unp_dispose() on the control message; 549 * the callee will not free it since SB_UNIX is set. 550 */ 551 didreceive = control ? 552 sbappendcontrol(rcv, m, control, &error) : sbappend(rcv, m); 553 554 snd->sb_mbmax -= rcv->sb_mbcnt - unp->unp_conn->unp_mbcnt; 555 unp->unp_conn->unp_mbcnt = rcv->sb_mbcnt; 556 snd->sb_hiwat -= rcv->sb_cc - unp->unp_conn->unp_cc; 557 unp->unp_conn->unp_cc = rcv->sb_cc; 558 if (didreceive) { 559 control = NULL; 560 sorwakeup(so2); 561 } else if (control != NULL && error == 0) { 562 /* A socket filter took control; don't touch it */ 563 control = NULL; 564 } 565 566 socket_unlock(so2, 1); 567 m = NULL; 568#undef snd 569#undef rcv 570 } 571 break; 572 573 default: 574 panic("uipc_send unknown socktype"); 575 } 576 577 /* 578 * SEND_EOF is equivalent to a SEND followed by 579 * a SHUTDOWN. 580 */ 581 if (flags & PRUS_EOF) { 582 socantsendmore(so); 583 unp_shutdown(unp); 584 } 585 586 if (control && error != 0) { 587 socket_unlock(so, 0); 588 unp_dispose(control); 589 socket_lock(so, 0); 590 } 591 592release: 593 if (control) 594 m_freem(control); 595 if (m) 596 m_freem(m); 597 return (error); 598} 599 600static int 601uipc_sense(struct socket *so, void *ub, int isstat64) 602{ 603 struct unpcb *unp = sotounpcb(so); 604 struct socket *so2; 605 blksize_t blksize; 606 607 if (unp == 0) 608 return (EINVAL); 609 610 blksize = so->so_snd.sb_hiwat; 611 if (so->so_type == SOCK_STREAM && unp->unp_conn != 0) { 612 so2 = unp->unp_conn->unp_socket; 613 blksize += so2->so_rcv.sb_cc; 614 } 615 if (unp->unp_ino == 0) 616 unp->unp_ino = unp_ino++; 617 618 if (isstat64 != 0) { 619 struct stat64 *sb64; 620 621 sb64 = (struct stat64 *)ub; 622 sb64->st_blksize = blksize; 623 sb64->st_dev = NODEV; 624 sb64->st_ino = (ino64_t)unp->unp_ino; 625 } else { 626 struct stat *sb; 627 628 sb = (struct stat *)ub; 629 sb->st_blksize = blksize; 630 sb->st_dev = NODEV; 631 sb->st_ino = (ino_t)(uintptr_t)unp->unp_ino; 632 } 633 634 return (0); 635} 636 637/* 638 * Returns: 0 Success 639 * EINVAL 640 * 641 * Notes: This is not strictly correct, as unp_shutdown() also calls 642 * socantrcvmore(). These should maybe both be conditionalized 643 * on the 'how' argument in soshutdown() as called from the 644 * shutdown() system call. 645 */ 646static int 647uipc_shutdown(struct socket *so) 648{ 649 struct unpcb *unp = sotounpcb(so); 650 651 if (unp == 0) 652 return (EINVAL); 653 socantsendmore(so); 654 unp_shutdown(unp); 655 return (0); 656} 657 658/* 659 * Returns: 0 Success 660 * EINVAL Invalid argument 661 */ 662static int 663uipc_sockaddr(struct socket *so, struct sockaddr **nam) 664{ 665 struct unpcb *unp = sotounpcb(so); 666 667 if (unp == NULL) 668 return (EINVAL); 669 if (unp->unp_addr != NULL) { 670 *nam = dup_sockaddr((struct sockaddr *)unp->unp_addr, 1); 671 } else { 672 *nam = dup_sockaddr((struct sockaddr *)&sun_noname, 1); 673 } 674 return (0); 675} 676 677struct pr_usrreqs uipc_usrreqs = { 678 .pru_abort = uipc_abort, 679 .pru_accept = uipc_accept, 680 .pru_attach = uipc_attach, 681 .pru_bind = uipc_bind, 682 .pru_connect = uipc_connect, 683 .pru_connect2 = uipc_connect2, 684 .pru_detach = uipc_detach, 685 .pru_disconnect = uipc_disconnect, 686 .pru_listen = uipc_listen, 687 .pru_peeraddr = uipc_peeraddr, 688 .pru_rcvd = uipc_rcvd, 689 .pru_send = uipc_send, 690 .pru_sense = uipc_sense, 691 .pru_shutdown = uipc_shutdown, 692 .pru_sockaddr = uipc_sockaddr, 693 .pru_sosend = sosend, 694 .pru_soreceive = soreceive, 695}; 696 697int 698uipc_ctloutput(struct socket *so, struct sockopt *sopt) 699{ 700 struct unpcb *unp = sotounpcb(so); 701 int error = 0; 702 pid_t peerpid; 703 struct socket *peerso; 704 705 switch (sopt->sopt_dir) { 706 case SOPT_GET: 707 switch (sopt->sopt_name) { 708 case LOCAL_PEERCRED: 709 if (unp->unp_flags & UNP_HAVEPC) { 710 error = sooptcopyout(sopt, &unp->unp_peercred, 711 sizeof (unp->unp_peercred)); 712 } else { 713 if (so->so_type == SOCK_STREAM) 714 error = ENOTCONN; 715 else 716 error = EINVAL; 717 } 718 break; 719 case LOCAL_PEERPID: 720 case LOCAL_PEEREPID: 721 if (unp->unp_conn == NULL) { 722 error = ENOTCONN; 723 break; 724 } 725 peerso = unp->unp_conn->unp_socket; 726 if (peerso == NULL) 727 panic("peer is connected but has no socket?"); 728 unp_get_locks_in_order(so, peerso); 729 if (sopt->sopt_name == LOCAL_PEEREPID && 730 peerso->so_flags & SOF_DELEGATED) 731 peerpid = peerso->e_pid; 732 else 733 peerpid = peerso->last_pid; 734 socket_unlock(peerso, 1); 735 error = sooptcopyout(sopt, &peerpid, sizeof (peerpid)); 736 break; 737 case LOCAL_PEERUUID: 738 case LOCAL_PEEREUUID: 739 if (unp->unp_conn == NULL) { 740 error = ENOTCONN; 741 break; 742 } 743 peerso = unp->unp_conn->unp_socket; 744 if (peerso == NULL) 745 panic("peer is connected but has no socket?"); 746 unp_get_locks_in_order(so, peerso); 747 if (sopt->sopt_name == LOCAL_PEEREUUID && 748 peerso->so_flags & SOF_DELEGATED) 749 error = sooptcopyout(sopt, &peerso->e_uuid, 750 sizeof (peerso->e_uuid)); 751 else 752 error = sooptcopyout(sopt, &peerso->last_uuid, 753 sizeof (peerso->last_uuid)); 754 socket_unlock(peerso, 1); 755 break; 756 default: 757 error = EOPNOTSUPP; 758 break; 759 } 760 break; 761 case SOPT_SET: 762 default: 763 error = EOPNOTSUPP; 764 break; 765 } 766 767 return (error); 768} 769 770/* 771 * Both send and receive buffers are allocated PIPSIZ bytes of buffering 772 * for stream sockets, although the total for sender and receiver is 773 * actually only PIPSIZ. 774 * Datagram sockets really use the sendspace as the maximum datagram size, 775 * and don't really want to reserve the sendspace. Their recvspace should 776 * be large enough for at least one max-size datagram plus address. 777 */ 778#ifndef PIPSIZ 779#define PIPSIZ 8192 780#endif 781static u_int32_t unpst_sendspace = PIPSIZ; 782static u_int32_t unpst_recvspace = PIPSIZ; 783static u_int32_t unpdg_sendspace = 2*1024; /* really max datagram size */ 784static u_int32_t unpdg_recvspace = 4*1024; 785 786static int unp_rights; /* file descriptors in flight */ 787static int unp_disposed; /* discarded file descriptors */ 788 789SYSCTL_DECL(_net_local_stream); 790SYSCTL_INT(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW | CTLFLAG_LOCKED, 791 &unpst_sendspace, 0, ""); 792SYSCTL_INT(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW | CTLFLAG_LOCKED, 793 &unpst_recvspace, 0, ""); 794SYSCTL_INT(_net_local_stream, OID_AUTO, tracemdns, CTLFLAG_RW | CTLFLAG_LOCKED, 795 &unpst_tracemdns, 0, ""); 796SYSCTL_DECL(_net_local_dgram); 797SYSCTL_INT(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW | CTLFLAG_LOCKED, 798 &unpdg_sendspace, 0, ""); 799SYSCTL_INT(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW | CTLFLAG_LOCKED, 800 &unpdg_recvspace, 0, ""); 801SYSCTL_DECL(_net_local); 802SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD | CTLFLAG_LOCKED, &unp_rights, 0, ""); 803 804/* 805 * Returns: 0 Success 806 * ENOBUFS 807 * soreserve:ENOBUFS 808 */ 809static int 810unp_attach(struct socket *so) 811{ 812 struct unpcb *unp; 813 int error = 0; 814 815 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 816 switch (so->so_type) { 817 818 case SOCK_STREAM: 819 error = soreserve(so, unpst_sendspace, unpst_recvspace); 820 break; 821 822 case SOCK_DGRAM: 823 error = soreserve(so, unpdg_sendspace, unpdg_recvspace); 824 break; 825 826 default: 827 panic("unp_attach"); 828 } 829 if (error) 830 return (error); 831 } 832 unp = (struct unpcb *)zalloc(unp_zone); 833 if (unp == NULL) 834 return (ENOBUFS); 835 bzero(unp, sizeof (*unp)); 836 837 lck_mtx_init(&unp->unp_mtx, 838 unp_mtx_grp, unp_mtx_attr); 839 840 lck_rw_lock_exclusive(unp_list_mtx); 841 LIST_INIT(&unp->unp_refs); 842 unp->unp_socket = so; 843 unp->unp_gencnt = ++unp_gencnt; 844 unp_count++; 845 LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? 846 &unp_dhead : &unp_shead, unp, unp_link); 847 lck_rw_done(unp_list_mtx); 848 so->so_pcb = (caddr_t)unp; 849 /* 850 * Mark AF_UNIX socket buffers accordingly so that: 851 * 852 * a. In the SOCK_STREAM case, socket buffer append won't fail due to 853 * the lack of space; this essentially loosens the sbspace() check, 854 * since there is disconnect between sosend() and uipc_send() with 855 * respect to flow control that might result in our dropping the 856 * data in uipc_send(). By setting this, we allow for slightly 857 * more records to be appended to the receiving socket to avoid 858 * losing data (which we can't afford in the SOCK_STREAM case). 859 * Flow control still takes place since we adjust the sender's 860 * hiwat during each send. This doesn't affect the SOCK_DGRAM 861 * case and append would still fail when the queue overflows. 862 * 863 * b. In the presence of control messages containing internalized 864 * file descriptors, the append routines will not free them since 865 * we'd need to undo the work first via unp_dispose(). 866 */ 867 so->so_rcv.sb_flags |= SB_UNIX; 868 so->so_snd.sb_flags |= SB_UNIX; 869 return (0); 870} 871 872static void 873unp_detach(struct unpcb *unp) 874{ 875 int so_locked = 1; 876 877 lck_rw_lock_exclusive(unp_list_mtx); 878 LIST_REMOVE(unp, unp_link); 879 --unp_count; 880 ++unp_gencnt; 881 lck_rw_done(unp_list_mtx); 882 if (unp->unp_vnode) { 883 struct vnode *tvp = NULL; 884 socket_unlock(unp->unp_socket, 0); 885 886 /* Holding unp_connect_lock will avoid a race between 887 * a thread closing the listening socket and a thread 888 * connecting to it. 889 */ 890 lck_mtx_lock(unp_connect_lock); 891 socket_lock(unp->unp_socket, 0); 892 if (unp->unp_vnode) { 893 tvp = unp->unp_vnode; 894 unp->unp_vnode->v_socket = NULL; 895 unp->unp_vnode = NULL; 896 } 897 lck_mtx_unlock(unp_connect_lock); 898 if (tvp != NULL) 899 vnode_rele(tvp); /* drop the usecount */ 900 } 901 if (unp->unp_conn) 902 unp_disconnect(unp); 903 while (unp->unp_refs.lh_first) { 904 struct unpcb *unp2 = NULL; 905 906 /* This datagram socket is connected to one or more 907 * sockets. In order to avoid a race condition between removing 908 * this reference and closing the connected socket, we need 909 * to check disconnect_in_progress 910 */ 911 if (so_locked == 1) { 912 socket_unlock(unp->unp_socket, 0); 913 so_locked = 0; 914 } 915 lck_mtx_lock(unp_disconnect_lock); 916 while (disconnect_in_progress != 0) { 917 (void)msleep((caddr_t)&disconnect_in_progress, unp_disconnect_lock, 918 PSOCK, "disconnect", NULL); 919 } 920 disconnect_in_progress = 1; 921 lck_mtx_unlock(unp_disconnect_lock); 922 923 /* Now we are sure that any unpcb socket disconnect is not happening */ 924 if (unp->unp_refs.lh_first != NULL) { 925 unp2 = unp->unp_refs.lh_first; 926 socket_lock(unp2->unp_socket, 1); 927 } 928 929 lck_mtx_lock(unp_disconnect_lock); 930 disconnect_in_progress = 0; 931 wakeup(&disconnect_in_progress); 932 lck_mtx_unlock(unp_disconnect_lock); 933 934 if (unp2 != NULL) { 935 /* We already locked this socket and have a reference on it */ 936 unp_drop(unp2, ECONNRESET); 937 socket_unlock(unp2->unp_socket, 1); 938 } 939 } 940 941 if (so_locked == 0) { 942 socket_lock(unp->unp_socket, 0); 943 so_locked = 1; 944 } 945 soisdisconnected(unp->unp_socket); 946 /* makes sure we're getting dealloced */ 947 unp->unp_socket->so_flags |= SOF_PCBCLEARING; 948} 949 950/* 951 * Returns: 0 Success 952 * EAFNOSUPPORT 953 * EINVAL 954 * EADDRINUSE 955 * namei:??? [anything namei can return] 956 * vnode_authorize:??? [anything vnode_authorize can return] 957 * 958 * Notes: p at this point is the current process, as this function is 959 * only called by sobind(). 960 */ 961static int 962unp_bind( 963 struct unpcb *unp, 964 struct sockaddr *nam, 965 proc_t p) 966{ 967 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 968 struct vnode *vp, *dvp; 969 struct vnode_attr va; 970 vfs_context_t ctx = vfs_context_current(); 971 int error, namelen; 972 struct nameidata nd; 973 struct socket *so = unp->unp_socket; 974 char buf[SOCK_MAXADDRLEN]; 975 976 if (nam->sa_family != 0 && nam->sa_family != AF_UNIX) { 977 return (EAFNOSUPPORT); 978 } 979 980 if (unp->unp_vnode != NULL) 981 return (EINVAL); 982 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 983 if (namelen <= 0) 984 return (EINVAL); 985 986 socket_unlock(so, 0); 987 988 strlcpy(buf, soun->sun_path, namelen+1); 989 NDINIT(&nd, CREATE, OP_MKFIFO, FOLLOW | LOCKPARENT, UIO_SYSSPACE, 990 CAST_USER_ADDR_T(buf), ctx); 991 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 992 error = namei(&nd); 993 if (error) { 994 socket_lock(so, 0); 995 return (error); 996 } 997 dvp = nd.ni_dvp; 998 vp = nd.ni_vp; 999 1000 if (vp != NULL) { 1001 /* 1002 * need to do this before the vnode_put of dvp 1003 * since we may have to release an fs_nodelock 1004 */ 1005 nameidone(&nd); 1006 1007 vnode_put(dvp); 1008 vnode_put(vp); 1009 1010 socket_lock(so, 0); 1011 return (EADDRINUSE); 1012 } 1013 1014 VATTR_INIT(&va); 1015 VATTR_SET(&va, va_type, VSOCK); 1016 VATTR_SET(&va, va_mode, (ACCESSPERMS & ~p->p_fd->fd_cmask)); 1017 1018#if CONFIG_MACF 1019 error = mac_vnode_check_create(ctx, 1020 nd.ni_dvp, &nd.ni_cnd, &va); 1021 1022 if (error == 0) 1023#endif /* CONFIG_MACF */ 1024#if CONFIG_MACF_SOCKET_SUBSET 1025 error = mac_vnode_check_uipc_bind(ctx, 1026 nd.ni_dvp, &nd.ni_cnd, &va); 1027 1028 if (error == 0) 1029#endif /* MAC_SOCKET_SUBSET */ 1030 /* authorize before creating */ 1031 error = vnode_authorize(dvp, NULL, KAUTH_VNODE_ADD_FILE, ctx); 1032 1033 if (!error) { 1034 /* create the socket */ 1035 error = vn_create(dvp, &vp, &nd, &va, 0, 0, NULL, ctx); 1036 } 1037 1038 nameidone(&nd); 1039 vnode_put(dvp); 1040 1041 if (error) { 1042 socket_lock(so, 0); 1043 return (error); 1044 } 1045 vnode_ref(vp); /* gain a longterm reference */ 1046 socket_lock(so, 0); 1047 vp->v_socket = unp->unp_socket; 1048 unp->unp_vnode = vp; 1049 unp->unp_addr = (struct sockaddr_un *)dup_sockaddr(nam, 1); 1050 vnode_put(vp); /* drop the iocount */ 1051 1052 return (0); 1053} 1054 1055 1056/* 1057 * Returns: 0 Success 1058 * EAFNOSUPPORT Address family not supported 1059 * EINVAL Invalid argument 1060 * ENOTSOCK Not a socket 1061 * ECONNREFUSED Connection refused 1062 * EPROTOTYPE Protocol wrong type for socket 1063 * EISCONN Socket is connected 1064 * unp_connect2:EPROTOTYPE Protocol wrong type for socket 1065 * unp_connect2:EINVAL Invalid argument 1066 * namei:??? [anything namei can return] 1067 * vnode_authorize:???? [anything vnode_authorize can return] 1068 * 1069 * Notes: p at this point is the current process, as this function is 1070 * only called by sosend(), sendfile(), and soconnectlock(). 1071 */ 1072static int 1073unp_connect(struct socket *so, struct sockaddr *nam, __unused proc_t p) 1074{ 1075 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 1076 struct vnode *vp; 1077 struct socket *so2, *so3, *list_so=NULL; 1078 struct unpcb *unp, *unp2, *unp3; 1079 vfs_context_t ctx = vfs_context_current(); 1080 int error, len; 1081 struct nameidata nd; 1082 char buf[SOCK_MAXADDRLEN]; 1083 1084 if (nam->sa_family != 0 && nam->sa_family != AF_UNIX) { 1085 return (EAFNOSUPPORT); 1086 } 1087 1088 unp = sotounpcb(so); 1089 so2 = so3 = NULL; 1090 1091 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 1092 if (len <= 0) 1093 return (EINVAL); 1094 1095 strlcpy(buf, soun->sun_path, len+1); 1096 socket_unlock(so, 0); 1097 1098 NDINIT(&nd, LOOKUP, OP_LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, 1099 CAST_USER_ADDR_T(buf), ctx); 1100 error = namei(&nd); 1101 if (error) { 1102 socket_lock(so, 0); 1103 return (error); 1104 } 1105 nameidone(&nd); 1106 vp = nd.ni_vp; 1107 if (vp->v_type != VSOCK) { 1108 error = ENOTSOCK; 1109 socket_lock(so, 0); 1110 goto out; 1111 } 1112 1113#if CONFIG_MACF_SOCKET_SUBSET 1114 error = mac_vnode_check_uipc_connect(ctx, vp); 1115 if (error) { 1116 socket_lock(so, 0); 1117 goto out; 1118 } 1119#endif /* MAC_SOCKET_SUBSET */ 1120 1121 error = vnode_authorize(vp, NULL, KAUTH_VNODE_WRITE_DATA, ctx); 1122 if (error) { 1123 socket_lock(so, 0); 1124 goto out; 1125 } 1126 1127 lck_mtx_lock(unp_connect_lock); 1128 1129 if (vp->v_socket == 0) { 1130 lck_mtx_unlock(unp_connect_lock); 1131 error = ECONNREFUSED; 1132 socket_lock(so, 0); 1133 goto out; 1134 } 1135 1136 socket_lock(vp->v_socket, 1); /* Get a reference on the listening socket */ 1137 so2 = vp->v_socket; 1138 lck_mtx_unlock(unp_connect_lock); 1139 1140 1141 if (so2->so_pcb == NULL) { 1142 error = ECONNREFUSED; 1143 if (so != so2) { 1144 socket_unlock(so2, 1); 1145 socket_lock(so, 0); 1146 } else { 1147 /* Release the reference held for the listen socket */ 1148 so2->so_usecount--; 1149 } 1150 goto out; 1151 } 1152 1153 if (so < so2) { 1154 socket_unlock(so2, 0); 1155 socket_lock(so, 0); 1156 socket_lock(so2, 0); 1157 } else if (so > so2) { 1158 socket_lock(so, 0); 1159 } 1160 /* 1161 * Check if socket was connected while we were trying to 1162 * get the socket locks in order. 1163 * XXX - probably shouldn't return an error for SOCK_DGRAM 1164 */ 1165 if ((so->so_state & SS_ISCONNECTED) != 0) { 1166 error = EISCONN; 1167 goto decref_out; 1168 } 1169 1170 if (so->so_type != so2->so_type) { 1171 error = EPROTOTYPE; 1172 goto decref_out; 1173 } 1174 1175 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 1176 /* Release the incoming socket but keep a reference */ 1177 socket_unlock(so, 0); 1178 1179 if ((so2->so_options & SO_ACCEPTCONN) == 0 || 1180 (so3 = sonewconn(so2, 0, nam)) == 0) { 1181 error = ECONNREFUSED; 1182 if (so != so2) { 1183 socket_unlock(so2, 1); 1184 socket_lock(so, 0); 1185 } else { 1186 socket_lock(so, 0); 1187 /* Release the reference held for 1188 * listen socket. 1189 */ 1190 so2->so_usecount--; 1191 } 1192 goto out; 1193 } 1194 unp2 = sotounpcb(so2); 1195 unp3 = sotounpcb(so3); 1196 if (unp2->unp_addr) 1197 unp3->unp_addr = (struct sockaddr_un *) 1198 dup_sockaddr((struct sockaddr *)unp2->unp_addr, 1); 1199 1200 /* 1201 * unp_peercred management: 1202 * 1203 * The connecter's (client's) credentials are copied 1204 * from its process structure at the time of connect() 1205 * (which is now). 1206 */ 1207 cru2x(vfs_context_ucred(ctx), &unp3->unp_peercred); 1208 unp3->unp_flags |= UNP_HAVEPC; 1209 /* 1210 * The receiver's (server's) credentials are copied 1211 * from the unp_peercred member of socket on which the 1212 * former called listen(); unp_listen() cached that 1213 * process's credentials at that time so we can use 1214 * them now. 1215 */ 1216 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED, 1217 ("unp_connect: listener without cached peercred")); 1218 1219 /* Here we need to have both so and so2 locks and so2 1220 * is already locked. Lock ordering is required. 1221 */ 1222 if (so < so2) { 1223 socket_unlock(so2, 0); 1224 socket_lock(so, 0); 1225 socket_lock(so2, 0); 1226 } else { 1227 socket_lock(so, 0); 1228 } 1229 1230 /* Check again if the socket state changed when its lock was released */ 1231 if ((so->so_state & SS_ISCONNECTED) != 0) { 1232 error = EISCONN; 1233 socket_unlock(so2, 1); 1234 socket_lock(so3, 0); 1235 sofreelastref(so3, 1); 1236 goto out; 1237 } 1238 memcpy(&unp->unp_peercred, &unp2->unp_peercred, 1239 sizeof (unp->unp_peercred)); 1240 unp->unp_flags |= UNP_HAVEPC; 1241 1242#if CONFIG_MACF_SOCKET 1243 /* XXXMAC: recursive lock: SOCK_LOCK(so); */ 1244 mac_socketpeer_label_associate_socket(so, so3); 1245 mac_socketpeer_label_associate_socket(so3, so); 1246 /* XXXMAC: SOCK_UNLOCK(so); */ 1247#endif /* MAC_SOCKET */ 1248 1249 /* Hold the reference on listening socket until the end */ 1250 socket_unlock(so2, 0); 1251 list_so = so2; 1252 1253 /* Lock ordering doesn't matter because so3 was just created */ 1254 socket_lock(so3, 1); 1255 so2 = so3; 1256 1257 /* 1258 * Enable tracing for mDNSResponder endpoints. (The use 1259 * of sizeof instead of strlen below takes the null 1260 * terminating character into account.) 1261 */ 1262 if (unpst_tracemdns && 1263 !strncmp(soun->sun_path, MDNSRESPONDER_PATH, 1264 sizeof (MDNSRESPONDER_PATH))) { 1265 unp->unp_flags |= UNP_TRACE_MDNS; 1266 unp2->unp_flags |= UNP_TRACE_MDNS; 1267 } 1268 } 1269 1270 error = unp_connect2(so, so2); 1271 1272decref_out: 1273 if (so2 != NULL) { 1274 if (so != so2) { 1275 socket_unlock(so2, 1); 1276 } else { 1277 /* Release the extra reference held for the listen socket. 1278 * This is possible only for SOCK_DGRAM sockets. We refuse 1279 * connecting to the same socket for SOCK_STREAM sockets. 1280 */ 1281 so2->so_usecount--; 1282 } 1283 } 1284 1285 if (list_so != NULL) { 1286 socket_lock(list_so, 0); 1287 socket_unlock(list_so, 1); 1288 } 1289 1290out: 1291 lck_mtx_assert(&unp->unp_mtx, LCK_MTX_ASSERT_OWNED); 1292 vnode_put(vp); 1293 return (error); 1294} 1295 1296/* 1297 * Returns: 0 Success 1298 * EPROTOTYPE Protocol wrong type for socket 1299 * EINVAL Invalid argument 1300 */ 1301int 1302unp_connect2(struct socket *so, struct socket *so2) 1303{ 1304 struct unpcb *unp = sotounpcb(so); 1305 struct unpcb *unp2; 1306 1307 if (so2->so_type != so->so_type) 1308 return (EPROTOTYPE); 1309 1310 unp2 = sotounpcb(so2); 1311 1312 lck_mtx_assert(&unp->unp_mtx, LCK_MTX_ASSERT_OWNED); 1313 lck_mtx_assert(&unp2->unp_mtx, LCK_MTX_ASSERT_OWNED); 1314 1315 /* Verify both sockets are still opened */ 1316 if (unp == 0 || unp2 == 0) 1317 return (EINVAL); 1318 1319 unp->unp_conn = unp2; 1320 so2->so_usecount++; 1321 1322 switch (so->so_type) { 1323 1324 case SOCK_DGRAM: 1325 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 1326 1327 if (so != so2) { 1328 /* Avoid lock order reversals due to drop/acquire in soisconnected. */ 1329 /* Keep an extra reference on so2 that will be dropped 1330 * soon after getting the locks in order 1331 */ 1332 socket_unlock(so2, 0); 1333 soisconnected(so); 1334 unp_get_locks_in_order(so, so2); 1335 so2->so_usecount--; 1336 } else { 1337 soisconnected(so); 1338 } 1339 1340 break; 1341 1342 case SOCK_STREAM: 1343 /* This takes care of socketpair */ 1344 if (!(unp->unp_flags & UNP_HAVEPC) && 1345 !(unp2->unp_flags & UNP_HAVEPC)) { 1346 cru2x(kauth_cred_get(), &unp->unp_peercred); 1347 unp->unp_flags |= UNP_HAVEPC; 1348 1349 cru2x(kauth_cred_get(), &unp2->unp_peercred); 1350 unp2->unp_flags |= UNP_HAVEPC; 1351 } 1352 unp2->unp_conn = unp; 1353 so->so_usecount++; 1354 1355 /* Avoid lock order reversals due to drop/acquire in soisconnected. */ 1356 socket_unlock(so, 0); 1357 soisconnected(so2); 1358 1359 /* Keep an extra reference on so2, that will be dropped soon after 1360 * getting the locks in order again. 1361 */ 1362 socket_unlock(so2, 0); 1363 1364 socket_lock(so, 0); 1365 soisconnected(so); 1366 1367 unp_get_locks_in_order(so, so2); 1368 /* Decrement the extra reference left before */ 1369 so2->so_usecount--; 1370 break; 1371 1372 default: 1373 panic("unknown socket type %d in unp_connect2", so->so_type); 1374 } 1375 lck_mtx_assert(&unp->unp_mtx, LCK_MTX_ASSERT_OWNED); 1376 lck_mtx_assert(&unp2->unp_mtx, LCK_MTX_ASSERT_OWNED); 1377 return (0); 1378} 1379 1380static void 1381unp_disconnect(struct unpcb *unp) 1382{ 1383 struct unpcb *unp2 = NULL; 1384 struct socket *so2 = NULL, *so; 1385 struct socket *waitso; 1386 int so_locked = 1, strdisconn = 0; 1387 1388 so = unp->unp_socket; 1389 if (unp->unp_conn == NULL) { 1390 return; 1391 } 1392 lck_mtx_lock(unp_disconnect_lock); 1393 while (disconnect_in_progress != 0) { 1394 if (so_locked == 1) { 1395 socket_unlock(so, 0); 1396 so_locked = 0; 1397 } 1398 (void)msleep((caddr_t)&disconnect_in_progress, unp_disconnect_lock, 1399 PSOCK, "disconnect", NULL); 1400 } 1401 disconnect_in_progress = 1; 1402 lck_mtx_unlock(unp_disconnect_lock); 1403 1404 if (so_locked == 0) { 1405 socket_lock(so, 0); 1406 so_locked = 1; 1407 } 1408 1409 unp2 = unp->unp_conn; 1410 1411 if (unp2 == 0 || unp2->unp_socket == NULL) { 1412 goto out; 1413 } 1414 so2 = unp2->unp_socket; 1415 1416try_again: 1417 if (so == so2) { 1418 if (so_locked == 0) { 1419 socket_lock(so, 0); 1420 } 1421 waitso = so; 1422 } else if (so < so2) { 1423 if (so_locked == 0) { 1424 socket_lock(so, 0); 1425 } 1426 socket_lock(so2, 1); 1427 waitso = so2; 1428 } else { 1429 if (so_locked == 1) { 1430 socket_unlock(so, 0); 1431 } 1432 socket_lock(so2, 1); 1433 socket_lock(so, 0); 1434 waitso = so; 1435 } 1436 so_locked = 1; 1437 1438 lck_mtx_assert(&unp->unp_mtx, LCK_MTX_ASSERT_OWNED); 1439 lck_mtx_assert(&unp2->unp_mtx, LCK_MTX_ASSERT_OWNED); 1440 1441 /* Check for the UNP_DONTDISCONNECT flag, if it 1442 * is set, release both sockets and go to sleep 1443 */ 1444 1445 if ((((struct unpcb *)waitso->so_pcb)->unp_flags & UNP_DONTDISCONNECT) != 0) { 1446 if (so != so2) { 1447 socket_unlock(so2, 1); 1448 } 1449 so_locked = 0; 1450 1451 (void)msleep(waitso->so_pcb, &unp->unp_mtx, 1452 PSOCK | PDROP, "unpdisconnect", NULL); 1453 goto try_again; 1454 } 1455 1456 if (unp->unp_conn == NULL) { 1457 panic("unp_conn became NULL after sleep"); 1458 } 1459 1460 unp->unp_conn = NULL; 1461 so2->so_usecount--; 1462 1463 if (unp->unp_flags & UNP_TRACE_MDNS) 1464 unp->unp_flags &= ~UNP_TRACE_MDNS; 1465 1466 switch (unp->unp_socket->so_type) { 1467 1468 case SOCK_DGRAM: 1469 LIST_REMOVE(unp, unp_reflink); 1470 unp->unp_socket->so_state &= ~SS_ISCONNECTED; 1471 if (so != so2) 1472 socket_unlock(so2, 1); 1473 break; 1474 1475 case SOCK_STREAM: 1476 unp2->unp_conn = NULL; 1477 so->so_usecount--; 1478 1479 /* Set the socket state correctly but do a wakeup later when 1480 * we release all locks except the socket lock, this will avoid 1481 * a deadlock. 1482 */ 1483 unp->unp_socket->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING); 1484 unp->unp_socket->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE|SS_ISDISCONNECTED); 1485 1486 unp2->unp_socket->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING); 1487 unp->unp_socket->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE|SS_ISDISCONNECTED); 1488 1489 if (unp2->unp_flags & UNP_TRACE_MDNS) 1490 unp2->unp_flags &= ~UNP_TRACE_MDNS; 1491 1492 strdisconn = 1; 1493 break; 1494 default: 1495 panic("unknown socket type %d", so->so_type); 1496 } 1497out: 1498 lck_mtx_lock(unp_disconnect_lock); 1499 disconnect_in_progress = 0; 1500 wakeup(&disconnect_in_progress); 1501 lck_mtx_unlock(unp_disconnect_lock); 1502 1503 if (strdisconn) { 1504 socket_unlock(so, 0); 1505 soisdisconnected(so2); 1506 socket_unlock(so2, 1); 1507 1508 socket_lock(so,0); 1509 soisdisconnected(so); 1510 } 1511 lck_mtx_assert(&unp->unp_mtx, LCK_MTX_ASSERT_OWNED); 1512 return; 1513} 1514 1515/* 1516 * unpcb_to_compat copies specific bits of a unpcb to a unpcb_compat format. 1517 * The unpcb_compat data structure is passed to user space and must not change. 1518 */ 1519static void 1520unpcb_to_compat(struct unpcb *up, struct unpcb_compat *cp) 1521{ 1522#if defined(__LP64__) 1523 cp->unp_link.le_next = (u_int32_t) 1524 VM_KERNEL_ADDRPERM(up->unp_link.le_next); 1525 cp->unp_link.le_prev = (u_int32_t) 1526 VM_KERNEL_ADDRPERM(up->unp_link.le_prev); 1527#else 1528 cp->unp_link.le_next = (struct unpcb_compat *) 1529 VM_KERNEL_ADDRPERM(up->unp_link.le_next); 1530 cp->unp_link.le_prev = (struct unpcb_compat **) 1531 VM_KERNEL_ADDRPERM(up->unp_link.le_prev); 1532#endif 1533 cp->unp_socket = (_UNPCB_PTR(struct socket *)) 1534 VM_KERNEL_ADDRPERM(up->unp_socket); 1535 cp->unp_vnode = (_UNPCB_PTR(struct vnode *)) 1536 VM_KERNEL_ADDRPERM(up->unp_vnode); 1537 cp->unp_ino = up->unp_ino; 1538 cp->unp_conn = (_UNPCB_PTR(struct unpcb_compat *)) 1539 VM_KERNEL_ADDRPERM(up->unp_conn); 1540 cp->unp_refs = (u_int32_t)VM_KERNEL_ADDRPERM(up->unp_refs.lh_first); 1541#if defined(__LP64__) 1542 cp->unp_reflink.le_next = 1543 (u_int32_t)VM_KERNEL_ADDRPERM(up->unp_reflink.le_next); 1544 cp->unp_reflink.le_prev = 1545 (u_int32_t)VM_KERNEL_ADDRPERM(up->unp_reflink.le_prev); 1546#else 1547 cp->unp_reflink.le_next = 1548 (struct unpcb_compat *)VM_KERNEL_ADDRPERM(up->unp_reflink.le_next); 1549 cp->unp_reflink.le_prev = 1550 (struct unpcb_compat **)VM_KERNEL_ADDRPERM(up->unp_reflink.le_prev); 1551#endif 1552 cp->unp_addr = (_UNPCB_PTR(struct sockaddr_un *)) 1553 VM_KERNEL_ADDRPERM(up->unp_addr); 1554 cp->unp_cc = up->unp_cc; 1555 cp->unp_mbcnt = up->unp_mbcnt; 1556 cp->unp_gencnt = up->unp_gencnt; 1557} 1558 1559static int 1560unp_pcblist SYSCTL_HANDLER_ARGS 1561{ 1562#pragma unused(oidp,arg2) 1563 int error, i, n; 1564 struct unpcb *unp, **unp_list; 1565 unp_gen_t gencnt; 1566 struct xunpgen xug; 1567 struct unp_head *head; 1568 1569 lck_rw_lock_shared(unp_list_mtx); 1570 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead); 1571 1572 /* 1573 * The process of preparing the PCB list is too time-consuming and 1574 * resource-intensive to repeat twice on every request. 1575 */ 1576 if (req->oldptr == USER_ADDR_NULL) { 1577 n = unp_count; 1578 req->oldidx = 2 * sizeof (xug) + (n + n / 8) * 1579 sizeof (struct xunpcb); 1580 lck_rw_done(unp_list_mtx); 1581 return (0); 1582 } 1583 1584 if (req->newptr != USER_ADDR_NULL) { 1585 lck_rw_done(unp_list_mtx); 1586 return (EPERM); 1587 } 1588 1589 /* 1590 * OK, now we're committed to doing something. 1591 */ 1592 gencnt = unp_gencnt; 1593 n = unp_count; 1594 1595 bzero(&xug, sizeof (xug)); 1596 xug.xug_len = sizeof (xug); 1597 xug.xug_count = n; 1598 xug.xug_gen = gencnt; 1599 xug.xug_sogen = so_gencnt; 1600 error = SYSCTL_OUT(req, &xug, sizeof (xug)); 1601 if (error) { 1602 lck_rw_done(unp_list_mtx); 1603 return (error); 1604 } 1605 1606 /* 1607 * We are done if there is no pcb 1608 */ 1609 if (n == 0) { 1610 lck_rw_done(unp_list_mtx); 1611 return (0); 1612 } 1613 1614 MALLOC(unp_list, struct unpcb **, n * sizeof (*unp_list), 1615 M_TEMP, M_WAITOK); 1616 if (unp_list == 0) { 1617 lck_rw_done(unp_list_mtx); 1618 return (ENOMEM); 1619 } 1620 1621 for (unp = head->lh_first, i = 0; unp && i < n; 1622 unp = unp->unp_link.le_next) { 1623 if (unp->unp_gencnt <= gencnt) 1624 unp_list[i++] = unp; 1625 } 1626 n = i; /* in case we lost some during malloc */ 1627 1628 error = 0; 1629 for (i = 0; i < n; i++) { 1630 unp = unp_list[i]; 1631 if (unp->unp_gencnt <= gencnt) { 1632 struct xunpcb xu; 1633 1634 bzero(&xu, sizeof (xu)); 1635 xu.xu_len = sizeof (xu); 1636 xu.xu_unpp = (_UNPCB_PTR(struct unpcb_compat *)) 1637 VM_KERNEL_ADDRPERM(unp); 1638 /* 1639 * XXX - need more locking here to protect against 1640 * connect/disconnect races for SMP. 1641 */ 1642 if (unp->unp_addr) 1643 bcopy(unp->unp_addr, &xu.xu_addr, 1644 unp->unp_addr->sun_len); 1645 if (unp->unp_conn && unp->unp_conn->unp_addr) 1646 bcopy(unp->unp_conn->unp_addr, 1647 &xu.xu_caddr, 1648 unp->unp_conn->unp_addr->sun_len); 1649 unpcb_to_compat(unp, &xu.xu_unp); 1650 sotoxsocket(unp->unp_socket, &xu.xu_socket); 1651 error = SYSCTL_OUT(req, &xu, sizeof (xu)); 1652 } 1653 } 1654 if (!error) { 1655 /* 1656 * Give the user an updated idea of our state. 1657 * If the generation differs from what we told 1658 * her before, she knows that something happened 1659 * while we were processing this request, and it 1660 * might be necessary to retry. 1661 */ 1662 bzero(&xug, sizeof (xug)); 1663 xug.xug_len = sizeof (xug); 1664 xug.xug_gen = unp_gencnt; 1665 xug.xug_sogen = so_gencnt; 1666 xug.xug_count = unp_count; 1667 error = SYSCTL_OUT(req, &xug, sizeof (xug)); 1668 } 1669 FREE(unp_list, M_TEMP); 1670 lck_rw_done(unp_list_mtx); 1671 return (error); 1672} 1673 1674SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD | CTLFLAG_LOCKED, 1675 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 1676 "List of active local datagram sockets"); 1677SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD | CTLFLAG_LOCKED, 1678 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 1679 "List of active local stream sockets"); 1680 1681 1682static int 1683unp_pcblist64 SYSCTL_HANDLER_ARGS 1684{ 1685#pragma unused(oidp,arg2) 1686 int error, i, n; 1687 struct unpcb *unp, **unp_list; 1688 unp_gen_t gencnt; 1689 struct xunpgen xug; 1690 struct unp_head *head; 1691 1692 lck_rw_lock_shared(unp_list_mtx); 1693 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead); 1694 1695 /* 1696 * The process of preparing the PCB list is too time-consuming and 1697 * resource-intensive to repeat twice on every request. 1698 */ 1699 if (req->oldptr == USER_ADDR_NULL) { 1700 n = unp_count; 1701 req->oldidx = 2 * sizeof (xug) + (n + n / 8) * 1702 (sizeof (struct xunpcb64)); 1703 lck_rw_done(unp_list_mtx); 1704 return (0); 1705 } 1706 1707 if (req->newptr != USER_ADDR_NULL) { 1708 lck_rw_done(unp_list_mtx); 1709 return (EPERM); 1710 } 1711 1712 /* 1713 * OK, now we're committed to doing something. 1714 */ 1715 gencnt = unp_gencnt; 1716 n = unp_count; 1717 1718 bzero(&xug, sizeof (xug)); 1719 xug.xug_len = sizeof (xug); 1720 xug.xug_count = n; 1721 xug.xug_gen = gencnt; 1722 xug.xug_sogen = so_gencnt; 1723 error = SYSCTL_OUT(req, &xug, sizeof (xug)); 1724 if (error) { 1725 lck_rw_done(unp_list_mtx); 1726 return (error); 1727 } 1728 1729 /* 1730 * We are done if there is no pcb 1731 */ 1732 if (n == 0) { 1733 lck_rw_done(unp_list_mtx); 1734 return (0); 1735 } 1736 1737 MALLOC(unp_list, struct unpcb **, n * sizeof (*unp_list), 1738 M_TEMP, M_WAITOK); 1739 if (unp_list == 0) { 1740 lck_rw_done(unp_list_mtx); 1741 return (ENOMEM); 1742 } 1743 1744 for (unp = head->lh_first, i = 0; unp && i < n; 1745 unp = unp->unp_link.le_next) { 1746 if (unp->unp_gencnt <= gencnt) 1747 unp_list[i++] = unp; 1748 } 1749 n = i; /* in case we lost some during malloc */ 1750 1751 error = 0; 1752 for (i = 0; i < n; i++) { 1753 unp = unp_list[i]; 1754 if (unp->unp_gencnt <= gencnt) { 1755 struct xunpcb64 xu; 1756 size_t xu_len = sizeof(struct xunpcb64); 1757 1758 bzero(&xu, xu_len); 1759 xu.xu_len = xu_len; 1760 xu.xu_unpp = (u_int64_t)VM_KERNEL_ADDRPERM(unp); 1761 xu.xunp_link.le_next = (u_int64_t) 1762 VM_KERNEL_ADDRPERM(unp->unp_link.le_next); 1763 xu.xunp_link.le_prev = (u_int64_t) 1764 VM_KERNEL_ADDRPERM(unp->unp_link.le_prev); 1765 xu.xunp_socket = (u_int64_t) 1766 VM_KERNEL_ADDRPERM(unp->unp_socket); 1767 xu.xunp_vnode = (u_int64_t) 1768 VM_KERNEL_ADDRPERM(unp->unp_vnode); 1769 xu.xunp_ino = unp->unp_ino; 1770 xu.xunp_conn = (u_int64_t) 1771 VM_KERNEL_ADDRPERM(unp->unp_conn); 1772 xu.xunp_refs = (u_int64_t) 1773 VM_KERNEL_ADDRPERM(unp->unp_refs.lh_first); 1774 xu.xunp_reflink.le_next = (u_int64_t) 1775 VM_KERNEL_ADDRPERM(unp->unp_reflink.le_next); 1776 xu.xunp_reflink.le_prev = (u_int64_t) 1777 VM_KERNEL_ADDRPERM(unp->unp_reflink.le_prev); 1778 xu.xunp_cc = unp->unp_cc; 1779 xu.xunp_mbcnt = unp->unp_mbcnt; 1780 xu.xunp_gencnt = unp->unp_gencnt; 1781 1782 if (unp->unp_socket) 1783 sotoxsocket64(unp->unp_socket, &xu.xu_socket); 1784 1785 /* 1786 * XXX - need more locking here to protect against 1787 * connect/disconnect races for SMP. 1788 */ 1789 if (unp->unp_addr) 1790 bcopy(unp->unp_addr, &xu.xunp_addr, 1791 unp->unp_addr->sun_len); 1792 if (unp->unp_conn && unp->unp_conn->unp_addr) 1793 bcopy(unp->unp_conn->unp_addr, 1794 &xu.xunp_caddr, 1795 unp->unp_conn->unp_addr->sun_len); 1796 1797 error = SYSCTL_OUT(req, &xu, xu_len); 1798 } 1799 } 1800 if (!error) { 1801 /* 1802 * Give the user an updated idea of our state. 1803 * If the generation differs from what we told 1804 * her before, she knows that something happened 1805 * while we were processing this request, and it 1806 * might be necessary to retry. 1807 */ 1808 bzero(&xug, sizeof (xug)); 1809 xug.xug_len = sizeof (xug); 1810 xug.xug_gen = unp_gencnt; 1811 xug.xug_sogen = so_gencnt; 1812 xug.xug_count = unp_count; 1813 error = SYSCTL_OUT(req, &xug, sizeof (xug)); 1814 } 1815 FREE(unp_list, M_TEMP); 1816 lck_rw_done(unp_list_mtx); 1817 return (error); 1818} 1819 1820SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist64, CTLFLAG_RD | CTLFLAG_LOCKED, 1821 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist64, "S,xunpcb64", 1822 "List of active local datagram sockets 64 bit"); 1823SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist64, CTLFLAG_RD | CTLFLAG_LOCKED, 1824 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist64, "S,xunpcb64", 1825 "List of active local stream sockets 64 bit"); 1826 1827 1828static void 1829unp_shutdown(struct unpcb *unp) 1830{ 1831 struct socket *so = unp->unp_socket; 1832 struct socket *so2; 1833 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn) { 1834 so2 = unp->unp_conn->unp_socket; 1835 unp_get_locks_in_order(so, so2); 1836 socantrcvmore(so2); 1837 socket_unlock(so2, 1); 1838 } 1839} 1840 1841static void 1842unp_drop(struct unpcb *unp, int errno) 1843{ 1844 struct socket *so = unp->unp_socket; 1845 1846 so->so_error = errno; 1847 unp_disconnect(unp); 1848} 1849 1850/* 1851 * Returns: 0 Success 1852 * EMSGSIZE The new fd's will not fit 1853 * ENOBUFS Cannot alloc struct fileproc 1854 */ 1855int 1856unp_externalize(struct mbuf *rights) 1857{ 1858 proc_t p = current_proc(); /* XXX */ 1859 int i; 1860 struct cmsghdr *cm = mtod(rights, struct cmsghdr *); 1861 struct fileglob **rp = (struct fileglob **)(cm + 1); 1862 int *fds = (int *)(cm + 1); 1863 struct fileproc *fp; 1864 struct fileglob *fg; 1865 int newfds = (cm->cmsg_len - sizeof (*cm)) / sizeof (int); 1866 int f; 1867 1868 proc_fdlock(p); 1869 1870 /* 1871 * if the new FD's will not fit, then we free them all 1872 */ 1873 if (!fdavail(p, newfds)) { 1874 for (i = 0; i < newfds; i++) { 1875 fg = *rp; 1876 unp_discard_fdlocked(fg, p); 1877 *rp++ = NULL; 1878 } 1879 proc_fdunlock(p); 1880 1881 return (EMSGSIZE); 1882 } 1883 /* 1884 * now change each pointer to an fd in the global table to 1885 * an integer that is the index to the local fd table entry 1886 * that we set up to point to the global one we are transferring. 1887 * XXX (1) this assumes a pointer and int are the same size, 1888 * XXX or the mbuf can hold the expansion 1889 * XXX (2) allocation failures should be non-fatal 1890 */ 1891 for (i = 0; i < newfds; i++) { 1892#if CONFIG_MACF_SOCKET 1893 /* 1894 * If receive access is denied, don't pass along 1895 * and error message, just discard the descriptor. 1896 */ 1897 if (mac_file_check_receive(kauth_cred_get(), *rp)) { 1898 fg = *rp; 1899 *rp++ = 0; 1900 unp_discard_fdlocked(fg, p); 1901 continue; 1902 } 1903#endif 1904 if (fdalloc(p, 0, &f)) 1905 panic("unp_externalize:fdalloc"); 1906 fg = rp[i]; 1907 fp = fileproc_alloc_init(NULL); 1908 if (fp == NULL) 1909 panic("unp_externalize: MALLOC_ZONE"); 1910 fp->f_iocount = 0; 1911 fp->f_fglob = fg; 1912 fg_removeuipc(fg); 1913 procfdtbl_releasefd(p, f, fp); 1914 (void) OSAddAtomic(-1, &unp_rights); 1915 fds[i] = f; 1916 } 1917 proc_fdunlock(p); 1918 1919 return (0); 1920} 1921 1922void 1923unp_init(void) 1924{ 1925 unp_zone = zinit(sizeof (struct unpcb), 1926 (nmbclusters * sizeof (struct unpcb)), 4096, "unpzone"); 1927 1928 if (unp_zone == 0) 1929 panic("unp_init"); 1930 LIST_INIT(&unp_dhead); 1931 LIST_INIT(&unp_shead); 1932 1933 /* 1934 * allocate lock group attribute and group for udp pcb mutexes 1935 */ 1936 unp_mtx_grp_attr = lck_grp_attr_alloc_init(); 1937 1938 unp_mtx_grp = lck_grp_alloc_init("unp_list", unp_mtx_grp_attr); 1939 1940 unp_mtx_attr = lck_attr_alloc_init(); 1941 1942 if ((unp_list_mtx = lck_rw_alloc_init(unp_mtx_grp, 1943 unp_mtx_attr)) == NULL) 1944 return; /* pretty much dead if this fails... */ 1945 1946 if ((unp_disconnect_lock = lck_mtx_alloc_init(unp_mtx_grp, 1947 unp_mtx_attr)) == NULL) 1948 return; 1949 1950 if ((unp_connect_lock = lck_mtx_alloc_init(unp_mtx_grp, 1951 unp_mtx_attr)) == NULL) 1952 return; 1953} 1954 1955#ifndef MIN 1956#define MIN(a, b) (((a) < (b)) ? (a) : (b)) 1957#endif 1958 1959/* 1960 * Returns: 0 Success 1961 * EINVAL 1962 * fdgetf_noref:EBADF 1963 */ 1964static int 1965unp_internalize(struct mbuf *control, proc_t p) 1966{ 1967 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1968 int *fds; 1969 struct fileglob **rp; 1970 struct fileproc *fp; 1971 int i, error; 1972 int oldfds; 1973 1974 /* 64bit: cmsg_len is 'uint32_t', m_len is 'long' */ 1975 if (cm->cmsg_type != SCM_RIGHTS || cm->cmsg_level != SOL_SOCKET || 1976 (socklen_t)cm->cmsg_len != (socklen_t)control->m_len) { 1977 return (EINVAL); 1978 } 1979 oldfds = (cm->cmsg_len - sizeof (*cm)) / sizeof (int); 1980 1981 proc_fdlock(p); 1982 fds = (int *)(cm + 1); 1983 1984 for (i = 0; i < oldfds; i++) { 1985 struct fileproc *tmpfp; 1986 if (((error = fdgetf_noref(p, fds[i], &tmpfp)) != 0)) { 1987 proc_fdunlock(p); 1988 return (error); 1989 } else if (!filetype_issendable(FILEGLOB_DTYPE(tmpfp->f_fglob))) { 1990 proc_fdunlock(p); 1991 return (EINVAL); 1992 } else if (FP_ISGUARDED(tmpfp, GUARD_SOCKET_IPC)) { 1993 error = fp_guard_exception(p, 1994 fds[i], tmpfp, kGUARD_EXC_SOCKET_IPC); 1995 proc_fdunlock(p); 1996 return (error); 1997 } 1998 } 1999 rp = (struct fileglob **)(cm + 1); 2000 2001 /* On K64 we need to walk backwards because a fileglob * is twice the size of an fd 2002 * and doing them in-order would result in stomping over unprocessed fd's 2003 */ 2004 for (i = (oldfds - 1); i >= 0; i--) { 2005 (void) fdgetf_noref(p, fds[i], &fp); 2006 fg_insertuipc(fp->f_fglob); 2007 rp[i] = fp->f_fglob; 2008 (void) OSAddAtomic(1, &unp_rights); 2009 } 2010 proc_fdunlock(p); 2011 2012 return (0); 2013} 2014 2015static int unp_defer, unp_gcing, unp_gcwait; 2016static thread_t unp_gcthread = NULL; 2017 2018/* always called under uipc_lock */ 2019void 2020unp_gc_wait(void) 2021{ 2022 if (unp_gcthread == current_thread()) 2023 return; 2024 2025 while (unp_gcing != 0) { 2026 unp_gcwait = 1; 2027 msleep(&unp_gcing, uipc_lock, 0 , "unp_gc_wait", NULL); 2028 } 2029} 2030 2031 2032__private_extern__ void 2033unp_gc(void) 2034{ 2035 struct fileglob *fg, *nextfg; 2036 struct socket *so; 2037 static struct fileglob **extra_ref; 2038 struct fileglob **fpp; 2039 int nunref, i; 2040 int need_gcwakeup = 0; 2041 2042 lck_mtx_lock(uipc_lock); 2043 if (unp_gcing) { 2044 lck_mtx_unlock(uipc_lock); 2045 return; 2046 } 2047 unp_gcing = 1; 2048 unp_defer = 0; 2049 unp_gcthread = current_thread(); 2050 lck_mtx_unlock(uipc_lock); 2051 /* 2052 * before going through all this, set all FDs to 2053 * be NOT defered and NOT externally accessible 2054 */ 2055 for (fg = fmsghead.lh_first; fg != 0; fg = fg->f_msglist.le_next) { 2056 lck_mtx_lock(&fg->fg_lock); 2057 fg->fg_flag &= ~(FMARK|FDEFER); 2058 lck_mtx_unlock(&fg->fg_lock); 2059 } 2060 do { 2061 for (fg = fmsghead.lh_first; fg != 0; 2062 fg = fg->f_msglist.le_next) { 2063 lck_mtx_lock(&fg->fg_lock); 2064 /* 2065 * If the file is not open, skip it 2066 */ 2067 if (fg->fg_count == 0) { 2068 lck_mtx_unlock(&fg->fg_lock); 2069 continue; 2070 } 2071 /* 2072 * If we already marked it as 'defer' in a 2073 * previous pass, then try process it this time 2074 * and un-mark it 2075 */ 2076 if (fg->fg_flag & FDEFER) { 2077 fg->fg_flag &= ~FDEFER; 2078 unp_defer--; 2079 } else { 2080 /* 2081 * if it's not defered, then check if it's 2082 * already marked.. if so skip it 2083 */ 2084 if (fg->fg_flag & FMARK) { 2085 lck_mtx_unlock(&fg->fg_lock); 2086 continue; 2087 } 2088 /* 2089 * If all references are from messages 2090 * in transit, then skip it. it's not 2091 * externally accessible. 2092 */ 2093 if (fg->fg_count == fg->fg_msgcount) { 2094 lck_mtx_unlock(&fg->fg_lock); 2095 continue; 2096 } 2097 /* 2098 * If it got this far then it must be 2099 * externally accessible. 2100 */ 2101 fg->fg_flag |= FMARK; 2102 } 2103 /* 2104 * either it was defered, or it is externally 2105 * accessible and not already marked so. 2106 * Now check if it is possibly one of OUR sockets. 2107 */ 2108 if (FILEGLOB_DTYPE(fg) != DTYPE_SOCKET || 2109 (so = (struct socket *)fg->fg_data) == 0) { 2110 lck_mtx_unlock(&fg->fg_lock); 2111 continue; 2112 } 2113 if (so->so_proto->pr_domain != localdomain || 2114 (so->so_proto->pr_flags&PR_RIGHTS) == 0) { 2115 lck_mtx_unlock(&fg->fg_lock); 2116 continue; 2117 } 2118#ifdef notdef 2119 /* 2120 * if this code is enabled need to run 2121 * under network funnel 2122 */ 2123 if (so->so_rcv.sb_flags & SB_LOCK) { 2124 /* 2125 * This is problematical; it's not clear 2126 * we need to wait for the sockbuf to be 2127 * unlocked (on a uniprocessor, at least), 2128 * and it's also not clear what to do 2129 * if sbwait returns an error due to receipt 2130 * of a signal. If sbwait does return 2131 * an error, we'll go into an infinite 2132 * loop. Delete all of this for now. 2133 */ 2134 (void) sbwait(&so->so_rcv); 2135 goto restart; 2136 } 2137#endif 2138 /* 2139 * So, Ok, it's one of our sockets and it IS externally 2140 * accessible (or was defered). Now we look 2141 * to see if we hold any file descriptors in its 2142 * message buffers. Follow those links and mark them 2143 * as accessible too. 2144 * 2145 * In case a file is passed onto itself we need to 2146 * release the file lock. 2147 */ 2148 lck_mtx_unlock(&fg->fg_lock); 2149 2150 unp_scan(so->so_rcv.sb_mb, unp_mark); 2151 } 2152 } while (unp_defer); 2153 /* 2154 * We grab an extra reference to each of the file table entries 2155 * that are not otherwise accessible and then free the rights 2156 * that are stored in messages on them. 2157 * 2158 * The bug in the orginal code is a little tricky, so I'll describe 2159 * what's wrong with it here. 2160 * 2161 * It is incorrect to simply unp_discard each entry for f_msgcount 2162 * times -- consider the case of sockets A and B that contain 2163 * references to each other. On a last close of some other socket, 2164 * we trigger a gc since the number of outstanding rights (unp_rights) 2165 * is non-zero. If during the sweep phase the gc code un_discards, 2166 * we end up doing a (full) closef on the descriptor. A closef on A 2167 * results in the following chain. Closef calls soo_close, which 2168 * calls soclose. Soclose calls first (through the switch 2169 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply 2170 * returns because the previous instance had set unp_gcing, and 2171 * we return all the way back to soclose, which marks the socket 2172 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush 2173 * to free up the rights that are queued in messages on the socket A, 2174 * i.e., the reference on B. The sorflush calls via the dom_dispose 2175 * switch unp_dispose, which unp_scans with unp_discard. This second 2176 * instance of unp_discard just calls closef on B. 2177 * 2178 * Well, a similar chain occurs on B, resulting in a sorflush on B, 2179 * which results in another closef on A. Unfortunately, A is already 2180 * being closed, and the descriptor has already been marked with 2181 * SS_NOFDREF, and soclose panics at this point. 2182 * 2183 * Here, we first take an extra reference to each inaccessible 2184 * descriptor. Then, we call sorflush ourself, since we know 2185 * it is a Unix domain socket anyhow. After we destroy all the 2186 * rights carried in messages, we do a last closef to get rid 2187 * of our extra reference. This is the last close, and the 2188 * unp_detach etc will shut down the socket. 2189 * 2190 * 91/09/19, bsy@cs.cmu.edu 2191 */ 2192 extra_ref = _MALLOC(nfiles * sizeof (struct fileglob *), 2193 M_FILEGLOB, M_WAITOK); 2194 if (extra_ref == NULL) 2195 goto bail; 2196 for (nunref = 0, fg = fmsghead.lh_first, fpp = extra_ref; fg != 0; 2197 fg = nextfg) { 2198 lck_mtx_lock(&fg->fg_lock); 2199 2200 nextfg = fg->f_msglist.le_next; 2201 /* 2202 * If it's not open, skip it 2203 */ 2204 if (fg->fg_count == 0) { 2205 lck_mtx_unlock(&fg->fg_lock); 2206 continue; 2207 } 2208 /* 2209 * If all refs are from msgs, and it's not marked accessible 2210 * then it must be referenced from some unreachable cycle 2211 * of (shut-down) FDs, so include it in our 2212 * list of FDs to remove 2213 */ 2214 if (fg->fg_count == fg->fg_msgcount && !(fg->fg_flag & FMARK)) { 2215 fg->fg_count++; 2216 *fpp++ = fg; 2217 nunref++; 2218 } 2219 lck_mtx_unlock(&fg->fg_lock); 2220 } 2221 /* 2222 * for each FD on our hit list, do the following two things 2223 */ 2224 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 2225 struct fileglob *tfg; 2226 2227 tfg = *fpp; 2228 2229 if (FILEGLOB_DTYPE(tfg) == DTYPE_SOCKET && 2230 tfg->fg_data != NULL) { 2231 so = (struct socket *)(tfg->fg_data); 2232 2233 socket_lock(so, 0); 2234 2235 sorflush(so); 2236 2237 socket_unlock(so, 0); 2238 } 2239 } 2240 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) 2241 closef_locked((struct fileproc *)0, *fpp, (proc_t)NULL); 2242 2243 FREE((caddr_t)extra_ref, M_FILEGLOB); 2244bail: 2245 lck_mtx_lock(uipc_lock); 2246 unp_gcing = 0; 2247 unp_gcthread = NULL; 2248 2249 if (unp_gcwait != 0) { 2250 unp_gcwait = 0; 2251 need_gcwakeup = 1; 2252 } 2253 lck_mtx_unlock(uipc_lock); 2254 2255 if (need_gcwakeup != 0) 2256 wakeup(&unp_gcing); 2257} 2258 2259void 2260unp_dispose(struct mbuf *m) 2261{ 2262 if (m) { 2263 unp_scan(m, unp_discard); 2264 } 2265} 2266 2267/* 2268 * Returns: 0 Success 2269 */ 2270static int 2271unp_listen(struct unpcb *unp, proc_t p) 2272{ 2273 kauth_cred_t safecred = kauth_cred_proc_ref(p); 2274 cru2x(safecred, &unp->unp_peercred); 2275 kauth_cred_unref(&safecred); 2276 unp->unp_flags |= UNP_HAVEPCCACHED; 2277 return (0); 2278} 2279 2280/* should run under kernel funnel */ 2281static void 2282unp_scan(struct mbuf *m0, void (*op)(struct fileglob *)) 2283{ 2284 struct mbuf *m; 2285 struct fileglob **rp; 2286 struct cmsghdr *cm; 2287 int i; 2288 int qfds; 2289 2290 while (m0) { 2291 for (m = m0; m; m = m->m_next) 2292 if (m->m_type == MT_CONTROL && 2293 (size_t)m->m_len >= sizeof (*cm)) { 2294 cm = mtod(m, struct cmsghdr *); 2295 if (cm->cmsg_level != SOL_SOCKET || 2296 cm->cmsg_type != SCM_RIGHTS) 2297 continue; 2298 qfds = (cm->cmsg_len - sizeof (*cm)) / 2299 sizeof (int); 2300 rp = (struct fileglob **)(cm + 1); 2301 for (i = 0; i < qfds; i++) 2302 (*op)(*rp++); 2303 break; /* XXX, but saves time */ 2304 } 2305 m0 = m0->m_act; 2306 } 2307} 2308 2309/* should run under kernel funnel */ 2310static void 2311unp_mark(struct fileglob *fg) 2312{ 2313 lck_mtx_lock(&fg->fg_lock); 2314 2315 if (fg->fg_flag & FMARK) { 2316 lck_mtx_unlock(&fg->fg_lock); 2317 return; 2318 } 2319 fg->fg_flag |= (FMARK|FDEFER); 2320 2321 lck_mtx_unlock(&fg->fg_lock); 2322 2323 unp_defer++; 2324} 2325 2326/* should run under kernel funnel */ 2327static void 2328unp_discard(struct fileglob *fg) 2329{ 2330 proc_t p = current_proc(); /* XXX */ 2331 2332 (void) OSAddAtomic(1, &unp_disposed); 2333 2334 proc_fdlock(p); 2335 unp_discard_fdlocked(fg, p); 2336 proc_fdunlock(p); 2337} 2338static void 2339unp_discard_fdlocked(struct fileglob *fg, proc_t p) 2340{ 2341 fg_removeuipc(fg); 2342 2343 (void) OSAddAtomic(-1, &unp_rights); 2344 (void) closef_locked((struct fileproc *)0, fg, p); 2345} 2346 2347int 2348unp_lock(struct socket *so, int refcount, void * lr) 2349 { 2350 void * lr_saved; 2351 if (lr == 0) 2352 lr_saved = (void *) __builtin_return_address(0); 2353 else lr_saved = lr; 2354 2355 if (so->so_pcb) { 2356 lck_mtx_lock(&((struct unpcb *)so->so_pcb)->unp_mtx); 2357 } else { 2358 panic("unp_lock: so=%p NO PCB! lr=%p ref=0x%x\n", 2359 so, lr_saved, so->so_usecount); 2360 } 2361 2362 if (so->so_usecount < 0) 2363 panic("unp_lock: so=%p so_pcb=%p lr=%p ref=0x%x\n", 2364 so, so->so_pcb, lr_saved, so->so_usecount); 2365 2366 if (refcount) 2367 so->so_usecount++; 2368 2369 so->lock_lr[so->next_lock_lr] = lr_saved; 2370 so->next_lock_lr = (so->next_lock_lr+1) % SO_LCKDBG_MAX; 2371 return (0); 2372} 2373 2374int 2375unp_unlock(struct socket *so, int refcount, void * lr) 2376{ 2377 void * lr_saved; 2378 lck_mtx_t * mutex_held = NULL; 2379 struct unpcb *unp = sotounpcb(so); 2380 2381 if (lr == 0) 2382 lr_saved = (void *) __builtin_return_address(0); 2383 else lr_saved = lr; 2384 2385 if (refcount) 2386 so->so_usecount--; 2387 2388 if (so->so_usecount < 0) 2389 panic("unp_unlock: so=%p usecount=%x\n", so, so->so_usecount); 2390 if (so->so_pcb == NULL) { 2391 panic("unp_unlock: so=%p NO PCB usecount=%x\n", so, so->so_usecount); 2392 } else { 2393 mutex_held = &((struct unpcb *)so->so_pcb)->unp_mtx; 2394 } 2395 lck_mtx_assert(mutex_held, LCK_MTX_ASSERT_OWNED); 2396 so->unlock_lr[so->next_unlock_lr] = lr_saved; 2397 so->next_unlock_lr = (so->next_unlock_lr+1) % SO_LCKDBG_MAX; 2398 2399 if (so->so_usecount == 0 && (so->so_flags & SOF_PCBCLEARING)) { 2400 sofreelastref(so, 1); 2401 2402 if (unp->unp_addr) 2403 FREE(unp->unp_addr, M_SONAME); 2404 2405 lck_mtx_unlock(mutex_held); 2406 2407 lck_mtx_destroy(&unp->unp_mtx, unp_mtx_grp); 2408 zfree(unp_zone, unp); 2409 2410 unp_gc(); 2411 } else { 2412 lck_mtx_unlock(mutex_held); 2413 } 2414 2415 return (0); 2416} 2417 2418lck_mtx_t * 2419unp_getlock(struct socket *so, __unused int locktype) 2420{ 2421 struct unpcb *unp = (struct unpcb *)so->so_pcb; 2422 2423 2424 if (so->so_pcb) { 2425 if (so->so_usecount < 0) 2426 panic("unp_getlock: so=%p usecount=%x\n", so, so->so_usecount); 2427 return(&unp->unp_mtx); 2428 } else { 2429 panic("unp_getlock: so=%p NULL so_pcb\n", so); 2430 return (so->so_proto->pr_domain->dom_mtx); 2431 } 2432} 2433 2434