uipc_usrreq.c revision 152283
1/*- 2 * Copyright (c) 1982, 1986, 1989, 1991, 1993 3 * The Regents of the University of California. 4 * Copyright 2004-2005 Robert N. M. Watson 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 4. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94 32 */ 33 34#include <sys/cdefs.h> 35__FBSDID("$FreeBSD: head/sys/kern/uipc_usrreq.c 152283 2005-11-10 16:06:04Z rwatson $"); 36 37#include "opt_mac.h" 38 39#include <sys/param.h> 40#include <sys/domain.h> 41#include <sys/fcntl.h> 42#include <sys/malloc.h> /* XXX must be before <sys/file.h> */ 43#include <sys/file.h> 44#include <sys/filedesc.h> 45#include <sys/jail.h> 46#include <sys/kernel.h> 47#include <sys/lock.h> 48#include <sys/mac.h> 49#include <sys/mbuf.h> 50#include <sys/mutex.h> 51#include <sys/namei.h> 52#include <sys/proc.h> 53#include <sys/protosw.h> 54#include <sys/resourcevar.h> 55#include <sys/socket.h> 56#include <sys/socketvar.h> 57#include <sys/signalvar.h> 58#include <sys/stat.h> 59#include <sys/sx.h> 60#include <sys/sysctl.h> 61#include <sys/systm.h> 62#include <sys/taskqueue.h> 63#include <sys/un.h> 64#include <sys/unpcb.h> 65#include <sys/vnode.h> 66 67#include <vm/uma.h> 68 69static uma_zone_t unp_zone; 70static unp_gen_t unp_gencnt; 71static u_int unp_count; 72 73static struct unp_head unp_shead, unp_dhead; 74 75/* 76 * Unix communications domain. 77 * 78 * TODO: 79 * SEQPACKET, RDM 80 * rethink name space problems 81 * need a proper out-of-band 82 * lock pushdown 83 */ 84static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL }; 85static ino_t unp_ino; /* prototype for fake inode numbers */ 86struct mbuf *unp_addsockcred(struct thread *, struct mbuf *); 87 88/* 89 * Currently, UNIX domain sockets are protected by a single subsystem lock, 90 * which covers global data structures and variables, the contents of each 91 * per-socket unpcb structure, and the so_pcb field in sockets attached to 92 * the UNIX domain. This provides for a moderate degree of paralellism, as 93 * receive operations on UNIX domain sockets do not need to acquire the 94 * subsystem lock. Finer grained locking to permit send() without acquiring 95 * a global lock would be a logical next step. 96 * 97 * The UNIX domain socket lock preceds all socket layer locks, including the 98 * socket lock and socket buffer lock, permitting UNIX domain socket code to 99 * call into socket support routines without releasing its locks. 100 * 101 * Some caution is required in areas where the UNIX domain socket code enters 102 * VFS in order to create or find rendezvous points. This results in 103 * dropping of the UNIX domain socket subsystem lock, acquisition of the 104 * Giant lock, and potential sleeping. This increases the chances of races, 105 * and exposes weaknesses in the socket->protocol API by offering poor 106 * failure modes. 107 */ 108static struct mtx unp_mtx; 109#define UNP_LOCK_INIT() \ 110 mtx_init(&unp_mtx, "unp", NULL, MTX_DEF) 111#define UNP_LOCK() mtx_lock(&unp_mtx) 112#define UNP_UNLOCK() mtx_unlock(&unp_mtx) 113#define UNP_LOCK_ASSERT() mtx_assert(&unp_mtx, MA_OWNED) 114#define UNP_UNLOCK_ASSERT() mtx_assert(&unp_mtx, MA_NOTOWNED) 115 116/* 117 * Garbage collection of cyclic file descriptor/socket references occurs 118 * asynchronously in a taskqueue context in order to avoid recursion and 119 * reentrance in the UNIX domain socket, file descriptor, and socket layer 120 * code. See unp_gc() for a full description. 121 */ 122static struct task unp_gc_task; 123 124static int unp_attach(struct socket *); 125static void unp_detach(struct unpcb *); 126static int unp_bind(struct unpcb *,struct sockaddr *, struct thread *); 127static int unp_connect(struct socket *,struct sockaddr *, struct thread *); 128static int unp_connect2(struct socket *so, struct socket *so2, int); 129static void unp_disconnect(struct unpcb *); 130static void unp_shutdown(struct unpcb *); 131static void unp_drop(struct unpcb *, int); 132static void unp_gc(__unused void *, int); 133static void unp_scan(struct mbuf *, void (*)(struct file *)); 134static void unp_mark(struct file *); 135static void unp_discard(struct file *); 136static void unp_freerights(struct file **, int); 137static int unp_internalize(struct mbuf **, struct thread *); 138static int unp_listen(struct socket *, struct unpcb *, int, 139 struct thread *); 140 141static int 142uipc_abort(struct socket *so) 143{ 144 struct unpcb *unp; 145 146 UNP_LOCK(); 147 unp = sotounpcb(so); 148 if (unp == NULL) { 149 UNP_UNLOCK(); 150 return (EINVAL); 151 } 152 unp_drop(unp, ECONNABORTED); 153 unp_detach(unp); 154 UNP_UNLOCK_ASSERT(); 155 ACCEPT_LOCK(); 156 SOCK_LOCK(so); 157 sotryfree(so); 158 return (0); 159} 160 161static int 162uipc_accept(struct socket *so, struct sockaddr **nam) 163{ 164 struct unpcb *unp; 165 const struct sockaddr *sa; 166 167 /* 168 * Pass back name of connected socket, 169 * if it was bound and we are still connected 170 * (our peer may have closed already!). 171 */ 172 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 173 UNP_LOCK(); 174 unp = sotounpcb(so); 175 if (unp == NULL) { 176 UNP_UNLOCK(); 177 free(*nam, M_SONAME); 178 *nam = NULL; 179 return (EINVAL); 180 } 181 if (unp->unp_conn != NULL && unp->unp_conn->unp_addr != NULL) 182 sa = (struct sockaddr *) unp->unp_conn->unp_addr; 183 else 184 sa = &sun_noname; 185 bcopy(sa, *nam, sa->sa_len); 186 UNP_UNLOCK(); 187 return (0); 188} 189 190static int 191uipc_attach(struct socket *so, int proto, struct thread *td) 192{ 193 struct unpcb *unp = sotounpcb(so); 194 195 if (unp != NULL) 196 return (EISCONN); 197 return (unp_attach(so)); 198} 199 200static int 201uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 202{ 203 struct unpcb *unp; 204 int error; 205 206 UNP_LOCK(); 207 unp = sotounpcb(so); 208 if (unp == NULL) { 209 UNP_UNLOCK(); 210 return (EINVAL); 211 } 212 error = unp_bind(unp, nam, td); 213 UNP_UNLOCK(); 214 return (error); 215} 216 217static int 218uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 219{ 220 struct unpcb *unp; 221 int error; 222 223 KASSERT(td == curthread, ("uipc_connect: td != curthread")); 224 225 UNP_LOCK(); 226 unp = sotounpcb(so); 227 if (unp == NULL) { 228 UNP_UNLOCK(); 229 return (EINVAL); 230 } 231 error = unp_connect(so, nam, td); 232 UNP_UNLOCK(); 233 return (error); 234} 235 236int 237uipc_connect2(struct socket *so1, struct socket *so2) 238{ 239 struct unpcb *unp; 240 int error; 241 242 UNP_LOCK(); 243 unp = sotounpcb(so1); 244 if (unp == NULL) { 245 UNP_UNLOCK(); 246 return (EINVAL); 247 } 248 error = unp_connect2(so1, so2, PRU_CONNECT2); 249 UNP_UNLOCK(); 250 return (error); 251} 252 253/* control is EOPNOTSUPP */ 254 255static int 256uipc_detach(struct socket *so) 257{ 258 struct unpcb *unp; 259 260 UNP_LOCK(); 261 unp = sotounpcb(so); 262 if (unp == NULL) { 263 UNP_UNLOCK(); 264 return (EINVAL); 265 } 266 unp_detach(unp); 267 UNP_UNLOCK_ASSERT(); 268 return (0); 269} 270 271static int 272uipc_disconnect(struct socket *so) 273{ 274 struct unpcb *unp; 275 276 UNP_LOCK(); 277 unp = sotounpcb(so); 278 if (unp == NULL) { 279 UNP_UNLOCK(); 280 return (EINVAL); 281 } 282 unp_disconnect(unp); 283 UNP_UNLOCK(); 284 return (0); 285} 286 287static int 288uipc_listen(struct socket *so, int backlog, struct thread *td) 289{ 290 struct unpcb *unp; 291 int error; 292 293 UNP_LOCK(); 294 unp = sotounpcb(so); 295 if (unp == NULL || unp->unp_vnode == NULL) { 296 UNP_UNLOCK(); 297 return (EINVAL); 298 } 299 error = unp_listen(so, unp, backlog, td); 300 UNP_UNLOCK(); 301 return (error); 302} 303 304static int 305uipc_peeraddr(struct socket *so, struct sockaddr **nam) 306{ 307 struct unpcb *unp; 308 const struct sockaddr *sa; 309 310 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 311 UNP_LOCK(); 312 unp = sotounpcb(so); 313 if (unp == NULL) { 314 UNP_UNLOCK(); 315 free(*nam, M_SONAME); 316 *nam = NULL; 317 return (EINVAL); 318 } 319 if (unp->unp_conn != NULL && unp->unp_conn->unp_addr!= NULL) 320 sa = (struct sockaddr *) unp->unp_conn->unp_addr; 321 else { 322 /* 323 * XXX: It seems that this test always fails even when 324 * connection is established. So, this else clause is 325 * added as workaround to return PF_LOCAL sockaddr. 326 */ 327 sa = &sun_noname; 328 } 329 bcopy(sa, *nam, sa->sa_len); 330 UNP_UNLOCK(); 331 return (0); 332} 333 334static int 335uipc_rcvd(struct socket *so, int flags) 336{ 337 struct unpcb *unp; 338 struct socket *so2; 339 u_long newhiwat; 340 341 UNP_LOCK(); 342 unp = sotounpcb(so); 343 if (unp == NULL) { 344 UNP_UNLOCK(); 345 return (EINVAL); 346 } 347 switch (so->so_type) { 348 case SOCK_DGRAM: 349 panic("uipc_rcvd DGRAM?"); 350 /*NOTREACHED*/ 351 352 case SOCK_STREAM: 353 if (unp->unp_conn == NULL) 354 break; 355 so2 = unp->unp_conn->unp_socket; 356 SOCKBUF_LOCK(&so2->so_snd); 357 SOCKBUF_LOCK(&so->so_rcv); 358 /* 359 * Adjust backpressure on sender 360 * and wakeup any waiting to write. 361 */ 362 so2->so_snd.sb_mbmax += unp->unp_mbcnt - so->so_rcv.sb_mbcnt; 363 unp->unp_mbcnt = so->so_rcv.sb_mbcnt; 364 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - 365 so->so_rcv.sb_cc; 366 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat, 367 newhiwat, RLIM_INFINITY); 368 unp->unp_cc = so->so_rcv.sb_cc; 369 SOCKBUF_UNLOCK(&so->so_rcv); 370 sowwakeup_locked(so2); 371 break; 372 373 default: 374 panic("uipc_rcvd unknown socktype"); 375 } 376 UNP_UNLOCK(); 377 return (0); 378} 379 380/* pru_rcvoob is EOPNOTSUPP */ 381 382static int 383uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 384 struct mbuf *control, struct thread *td) 385{ 386 int error = 0; 387 struct unpcb *unp; 388 struct socket *so2; 389 u_long newhiwat; 390 391 unp = sotounpcb(so); 392 if (unp == NULL) { 393 error = EINVAL; 394 goto release; 395 } 396 if (flags & PRUS_OOB) { 397 error = EOPNOTSUPP; 398 goto release; 399 } 400 401 if (control != NULL && (error = unp_internalize(&control, td))) 402 goto release; 403 404 UNP_LOCK(); 405 unp = sotounpcb(so); 406 if (unp == NULL) { 407 UNP_UNLOCK(); 408 error = EINVAL; 409 goto dispose_release; 410 } 411 412 switch (so->so_type) { 413 case SOCK_DGRAM: 414 { 415 const struct sockaddr *from; 416 417 if (nam != NULL) { 418 if (unp->unp_conn != NULL) { 419 error = EISCONN; 420 break; 421 } 422 error = unp_connect(so, nam, td); 423 if (error) 424 break; 425 } else { 426 if (unp->unp_conn == NULL) { 427 error = ENOTCONN; 428 break; 429 } 430 } 431 so2 = unp->unp_conn->unp_socket; 432 if (unp->unp_addr != NULL) 433 from = (struct sockaddr *)unp->unp_addr; 434 else 435 from = &sun_noname; 436 if (unp->unp_conn->unp_flags & UNP_WANTCRED) 437 control = unp_addsockcred(td, control); 438 SOCKBUF_LOCK(&so2->so_rcv); 439 if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) { 440 sorwakeup_locked(so2); 441 m = NULL; 442 control = NULL; 443 } else { 444 SOCKBUF_UNLOCK(&so2->so_rcv); 445 error = ENOBUFS; 446 } 447 if (nam != NULL) 448 unp_disconnect(unp); 449 break; 450 } 451 452 case SOCK_STREAM: 453 /* Connect if not connected yet. */ 454 /* 455 * Note: A better implementation would complain 456 * if not equal to the peer's address. 457 */ 458 if ((so->so_state & SS_ISCONNECTED) == 0) { 459 if (nam != NULL) { 460 error = unp_connect(so, nam, td); 461 if (error) 462 break; /* XXX */ 463 } else { 464 error = ENOTCONN; 465 break; 466 } 467 } 468 469 SOCKBUF_LOCK(&so->so_snd); 470 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 471 SOCKBUF_UNLOCK(&so->so_snd); 472 error = EPIPE; 473 break; 474 } 475 if (unp->unp_conn == NULL) 476 panic("uipc_send connected but no connection?"); 477 so2 = unp->unp_conn->unp_socket; 478 SOCKBUF_LOCK(&so2->so_rcv); 479 if (unp->unp_conn->unp_flags & UNP_WANTCRED) { 480 /* 481 * Credentials are passed only once on 482 * SOCK_STREAM. 483 */ 484 unp->unp_conn->unp_flags &= ~UNP_WANTCRED; 485 control = unp_addsockcred(td, control); 486 } 487 /* 488 * Send to paired receive port, and then reduce 489 * send buffer hiwater marks to maintain backpressure. 490 * Wake up readers. 491 */ 492 if (control != NULL) { 493 if (sbappendcontrol_locked(&so2->so_rcv, m, control)) 494 control = NULL; 495 } else { 496 sbappend_locked(&so2->so_rcv, m); 497 } 498 so->so_snd.sb_mbmax -= 499 so2->so_rcv.sb_mbcnt - unp->unp_conn->unp_mbcnt; 500 unp->unp_conn->unp_mbcnt = so2->so_rcv.sb_mbcnt; 501 newhiwat = so->so_snd.sb_hiwat - 502 (so2->so_rcv.sb_cc - unp->unp_conn->unp_cc); 503 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat, 504 newhiwat, RLIM_INFINITY); 505 SOCKBUF_UNLOCK(&so->so_snd); 506 unp->unp_conn->unp_cc = so2->so_rcv.sb_cc; 507 sorwakeup_locked(so2); 508 m = NULL; 509 break; 510 511 default: 512 panic("uipc_send unknown socktype"); 513 } 514 515 /* 516 * SEND_EOF is equivalent to a SEND followed by 517 * a SHUTDOWN. 518 */ 519 if (flags & PRUS_EOF) { 520 socantsendmore(so); 521 unp_shutdown(unp); 522 } 523 UNP_UNLOCK(); 524 525dispose_release: 526 if (control != NULL && error != 0) 527 unp_dispose(control); 528 529release: 530 if (control != NULL) 531 m_freem(control); 532 if (m != NULL) 533 m_freem(m); 534 return (error); 535} 536 537static int 538uipc_sense(struct socket *so, struct stat *sb) 539{ 540 struct unpcb *unp; 541 struct socket *so2; 542 543 UNP_LOCK(); 544 unp = sotounpcb(so); 545 if (unp == NULL) { 546 UNP_UNLOCK(); 547 return (EINVAL); 548 } 549 sb->st_blksize = so->so_snd.sb_hiwat; 550 if (so->so_type == SOCK_STREAM && unp->unp_conn != NULL) { 551 so2 = unp->unp_conn->unp_socket; 552 sb->st_blksize += so2->so_rcv.sb_cc; 553 } 554 sb->st_dev = NODEV; 555 if (unp->unp_ino == 0) 556 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino; 557 sb->st_ino = unp->unp_ino; 558 UNP_UNLOCK(); 559 return (0); 560} 561 562static int 563uipc_shutdown(struct socket *so) 564{ 565 struct unpcb *unp; 566 567 UNP_LOCK(); 568 unp = sotounpcb(so); 569 if (unp == NULL) { 570 UNP_UNLOCK(); 571 return (EINVAL); 572 } 573 socantsendmore(so); 574 unp_shutdown(unp); 575 UNP_UNLOCK(); 576 return (0); 577} 578 579static int 580uipc_sockaddr(struct socket *so, struct sockaddr **nam) 581{ 582 struct unpcb *unp; 583 const struct sockaddr *sa; 584 585 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 586 UNP_LOCK(); 587 unp = sotounpcb(so); 588 if (unp == NULL) { 589 UNP_UNLOCK(); 590 free(*nam, M_SONAME); 591 *nam = NULL; 592 return (EINVAL); 593 } 594 if (unp->unp_addr != NULL) 595 sa = (struct sockaddr *) unp->unp_addr; 596 else 597 sa = &sun_noname; 598 bcopy(sa, *nam, sa->sa_len); 599 UNP_UNLOCK(); 600 return (0); 601} 602 603struct pr_usrreqs uipc_usrreqs = { 604 .pru_abort = uipc_abort, 605 .pru_accept = uipc_accept, 606 .pru_attach = uipc_attach, 607 .pru_bind = uipc_bind, 608 .pru_connect = uipc_connect, 609 .pru_connect2 = uipc_connect2, 610 .pru_detach = uipc_detach, 611 .pru_disconnect = uipc_disconnect, 612 .pru_listen = uipc_listen, 613 .pru_peeraddr = uipc_peeraddr, 614 .pru_rcvd = uipc_rcvd, 615 .pru_send = uipc_send, 616 .pru_sense = uipc_sense, 617 .pru_shutdown = uipc_shutdown, 618 .pru_sockaddr = uipc_sockaddr, 619 .pru_sosend = sosend, 620 .pru_soreceive = soreceive, 621 .pru_sopoll = sopoll, 622}; 623 624int 625uipc_ctloutput(struct socket *so, struct sockopt *sopt) 626{ 627 struct unpcb *unp; 628 struct xucred xu; 629 int error, optval; 630 631 if (sopt->sopt_level != 0) 632 return (EINVAL); 633 634 UNP_LOCK(); 635 unp = sotounpcb(so); 636 if (unp == NULL) { 637 UNP_UNLOCK(); 638 return (EINVAL); 639 } 640 error = 0; 641 642 switch (sopt->sopt_dir) { 643 case SOPT_GET: 644 switch (sopt->sopt_name) { 645 case LOCAL_PEERCRED: 646 if (unp->unp_flags & UNP_HAVEPC) 647 xu = unp->unp_peercred; 648 else { 649 if (so->so_type == SOCK_STREAM) 650 error = ENOTCONN; 651 else 652 error = EINVAL; 653 } 654 if (error == 0) 655 error = sooptcopyout(sopt, &xu, sizeof(xu)); 656 break; 657 case LOCAL_CREDS: 658 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0; 659 error = sooptcopyout(sopt, &optval, sizeof(optval)); 660 break; 661 case LOCAL_CONNWAIT: 662 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0; 663 error = sooptcopyout(sopt, &optval, sizeof(optval)); 664 break; 665 default: 666 error = EOPNOTSUPP; 667 break; 668 } 669 break; 670 case SOPT_SET: 671 switch (sopt->sopt_name) { 672 case LOCAL_CREDS: 673 case LOCAL_CONNWAIT: 674 error = sooptcopyin(sopt, &optval, sizeof(optval), 675 sizeof(optval)); 676 if (error) 677 break; 678 679#define OPTSET(bit) \ 680 if (optval) \ 681 unp->unp_flags |= bit; \ 682 else \ 683 unp->unp_flags &= ~bit; 684 685 switch (sopt->sopt_name) { 686 case LOCAL_CREDS: 687 OPTSET(UNP_WANTCRED); 688 break; 689 case LOCAL_CONNWAIT: 690 OPTSET(UNP_CONNWAIT); 691 break; 692 default: 693 break; 694 } 695 break; 696#undef OPTSET 697 default: 698 error = ENOPROTOOPT; 699 break; 700 } 701 break; 702 default: 703 error = EOPNOTSUPP; 704 break; 705 } 706 UNP_UNLOCK(); 707 return (error); 708} 709 710/* 711 * Both send and receive buffers are allocated PIPSIZ bytes of buffering 712 * for stream sockets, although the total for sender and receiver is 713 * actually only PIPSIZ. 714 * Datagram sockets really use the sendspace as the maximum datagram size, 715 * and don't really want to reserve the sendspace. Their recvspace should 716 * be large enough for at least one max-size datagram plus address. 717 */ 718#ifndef PIPSIZ 719#define PIPSIZ 8192 720#endif 721static u_long unpst_sendspace = PIPSIZ; 722static u_long unpst_recvspace = PIPSIZ; 723static u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 724static u_long unpdg_recvspace = 4*1024; 725 726static int unp_rights; /* file descriptors in flight */ 727 728SYSCTL_DECL(_net_local_stream); 729SYSCTL_INT(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 730 &unpst_sendspace, 0, ""); 731SYSCTL_INT(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 732 &unpst_recvspace, 0, ""); 733SYSCTL_DECL(_net_local_dgram); 734SYSCTL_INT(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 735 &unpdg_sendspace, 0, ""); 736SYSCTL_INT(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 737 &unpdg_recvspace, 0, ""); 738SYSCTL_DECL(_net_local); 739SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, ""); 740 741static int 742unp_attach(struct socket *so) 743{ 744 struct unpcb *unp; 745 int error; 746 747 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 748 switch (so->so_type) { 749 750 case SOCK_STREAM: 751 error = soreserve(so, unpst_sendspace, unpst_recvspace); 752 break; 753 754 case SOCK_DGRAM: 755 error = soreserve(so, unpdg_sendspace, unpdg_recvspace); 756 break; 757 758 default: 759 panic("unp_attach"); 760 } 761 if (error) 762 return (error); 763 } 764 unp = uma_zalloc(unp_zone, M_WAITOK | M_ZERO); 765 if (unp == NULL) 766 return (ENOBUFS); 767 LIST_INIT(&unp->unp_refs); 768 unp->unp_socket = so; 769 so->so_pcb = unp; 770 771 UNP_LOCK(); 772 unp->unp_gencnt = ++unp_gencnt; 773 unp_count++; 774 LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead 775 : &unp_shead, unp, unp_link); 776 UNP_UNLOCK(); 777 778 return (0); 779} 780 781static void 782unp_detach(struct unpcb *unp) 783{ 784 struct vnode *vp; 785 int local_unp_rights; 786 787 UNP_LOCK_ASSERT(); 788 789 LIST_REMOVE(unp, unp_link); 790 unp->unp_gencnt = ++unp_gencnt; 791 --unp_count; 792 if ((vp = unp->unp_vnode) != NULL) { 793 /* 794 * XXXRW: should v_socket be frobbed only while holding 795 * Giant? 796 */ 797 unp->unp_vnode->v_socket = NULL; 798 unp->unp_vnode = NULL; 799 } 800 if (unp->unp_conn != NULL) 801 unp_disconnect(unp); 802 while (!LIST_EMPTY(&unp->unp_refs)) { 803 struct unpcb *ref = LIST_FIRST(&unp->unp_refs); 804 unp_drop(ref, ECONNRESET); 805 } 806 soisdisconnected(unp->unp_socket); 807 unp->unp_socket->so_pcb = NULL; 808 local_unp_rights = unp_rights; 809 UNP_UNLOCK(); 810 if (unp->unp_addr != NULL) 811 FREE(unp->unp_addr, M_SONAME); 812 uma_zfree(unp_zone, unp); 813 if (vp) { 814 mtx_lock(&Giant); 815 vrele(vp); 816 mtx_unlock(&Giant); 817 } 818 if (local_unp_rights) 819 taskqueue_enqueue(taskqueue_thread, &unp_gc_task); 820} 821 822static int 823unp_bind(struct unpcb *unp, struct sockaddr *nam, struct thread *td) 824{ 825 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 826 struct vnode *vp; 827 struct mount *mp; 828 struct vattr vattr; 829 int error, namelen; 830 struct nameidata nd; 831 char *buf; 832 833 UNP_LOCK_ASSERT(); 834 835 /* 836 * XXXRW: This test-and-set of unp_vnode is non-atomic; the 837 * unlocked read here is fine, but the value of unp_vnode needs 838 * to be tested again after we do all the lookups to see if the 839 * pcb is still unbound? 840 */ 841 if (unp->unp_vnode != NULL) 842 return (EINVAL); 843 844 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 845 if (namelen <= 0) 846 return (EINVAL); 847 848 UNP_UNLOCK(); 849 850 buf = malloc(namelen + 1, M_TEMP, M_WAITOK); 851 strlcpy(buf, soun->sun_path, namelen + 1); 852 853 mtx_lock(&Giant); 854restart: 855 mtx_assert(&Giant, MA_OWNED); 856 NDINIT(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME, UIO_SYSSPACE, 857 buf, td); 858/* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 859 error = namei(&nd); 860 if (error) 861 goto done; 862 vp = nd.ni_vp; 863 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { 864 NDFREE(&nd, NDF_ONLY_PNBUF); 865 if (nd.ni_dvp == vp) 866 vrele(nd.ni_dvp); 867 else 868 vput(nd.ni_dvp); 869 if (vp != NULL) { 870 vrele(vp); 871 error = EADDRINUSE; 872 goto done; 873 } 874 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH); 875 if (error) 876 goto done; 877 goto restart; 878 } 879 VATTR_NULL(&vattr); 880 vattr.va_type = VSOCK; 881 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask); 882#ifdef MAC 883 error = mac_check_vnode_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, 884 &vattr); 885#endif 886 if (error == 0) { 887 VOP_LEASE(nd.ni_dvp, td, td->td_ucred, LEASE_WRITE); 888 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 889 } 890 NDFREE(&nd, NDF_ONLY_PNBUF); 891 vput(nd.ni_dvp); 892 if (error) { 893 vn_finished_write(mp); 894 goto done; 895 } 896 vp = nd.ni_vp; 897 ASSERT_VOP_LOCKED(vp, "unp_bind"); 898 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK); 899 UNP_LOCK(); 900 vp->v_socket = unp->unp_socket; 901 unp->unp_vnode = vp; 902 unp->unp_addr = soun; 903 UNP_UNLOCK(); 904 VOP_UNLOCK(vp, 0, td); 905 vn_finished_write(mp); 906done: 907 mtx_unlock(&Giant); 908 free(buf, M_TEMP); 909 UNP_LOCK(); 910 return (error); 911} 912 913static int 914unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 915{ 916 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 917 struct vnode *vp; 918 struct socket *so2, *so3; 919 struct unpcb *unp, *unp2, *unp3; 920 int error, len; 921 struct nameidata nd; 922 char buf[SOCK_MAXADDRLEN]; 923 struct sockaddr *sa; 924 925 UNP_LOCK_ASSERT(); 926 unp = sotounpcb(so); 927 928 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 929 if (len <= 0) 930 return (EINVAL); 931 strlcpy(buf, soun->sun_path, len + 1); 932 UNP_UNLOCK(); 933 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 934 mtx_lock(&Giant); 935 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, td); 936 error = namei(&nd); 937 if (error) 938 vp = NULL; 939 else 940 vp = nd.ni_vp; 941 ASSERT_VOP_LOCKED(vp, "unp_connect"); 942 NDFREE(&nd, NDF_ONLY_PNBUF); 943 if (error) 944 goto bad; 945 946 if (vp->v_type != VSOCK) { 947 error = ENOTSOCK; 948 goto bad; 949 } 950 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); 951 if (error) 952 goto bad; 953 mtx_unlock(&Giant); 954 UNP_LOCK(); 955 unp = sotounpcb(so); 956 if (unp == NULL) { 957 error = EINVAL; 958 goto bad2; 959 } 960 so2 = vp->v_socket; 961 if (so2 == NULL) { 962 error = ECONNREFUSED; 963 goto bad2; 964 } 965 if (so->so_type != so2->so_type) { 966 error = EPROTOTYPE; 967 goto bad2; 968 } 969 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 970 if (so2->so_options & SO_ACCEPTCONN) { 971 /* 972 * NB: drop locks here so unp_attach is entered 973 * w/o locks; this avoids a recursive lock 974 * of the head and holding sleep locks across 975 * a (potentially) blocking malloc. 976 */ 977 UNP_UNLOCK(); 978 so3 = sonewconn(so2, 0); 979 UNP_LOCK(); 980 } else 981 so3 = NULL; 982 if (so3 == NULL) { 983 error = ECONNREFUSED; 984 goto bad2; 985 } 986 unp = sotounpcb(so); 987 unp2 = sotounpcb(so2); 988 unp3 = sotounpcb(so3); 989 if (unp2->unp_addr != NULL) { 990 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); 991 unp3->unp_addr = (struct sockaddr_un *) sa; 992 sa = NULL; 993 } 994 /* 995 * unp_peercred management: 996 * 997 * The connecter's (client's) credentials are copied 998 * from its process structure at the time of connect() 999 * (which is now). 1000 */ 1001 cru2x(td->td_ucred, &unp3->unp_peercred); 1002 unp3->unp_flags |= UNP_HAVEPC; 1003 /* 1004 * The receiver's (server's) credentials are copied 1005 * from the unp_peercred member of socket on which the 1006 * former called listen(); unp_listen() cached that 1007 * process's credentials at that time so we can use 1008 * them now. 1009 */ 1010 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED, 1011 ("unp_connect: listener without cached peercred")); 1012 memcpy(&unp->unp_peercred, &unp2->unp_peercred, 1013 sizeof(unp->unp_peercred)); 1014 unp->unp_flags |= UNP_HAVEPC; 1015#ifdef MAC 1016 SOCK_LOCK(so); 1017 mac_set_socket_peer_from_socket(so, so3); 1018 mac_set_socket_peer_from_socket(so3, so); 1019 SOCK_UNLOCK(so); 1020#endif 1021 1022 so2 = so3; 1023 } 1024 error = unp_connect2(so, so2, PRU_CONNECT); 1025bad2: 1026 UNP_UNLOCK(); 1027 mtx_lock(&Giant); 1028bad: 1029 mtx_assert(&Giant, MA_OWNED); 1030 if (vp != NULL) 1031 vput(vp); 1032 mtx_unlock(&Giant); 1033 free(sa, M_SONAME); 1034 UNP_LOCK(); 1035 return (error); 1036} 1037 1038static int 1039unp_connect2(struct socket *so, struct socket *so2, int req) 1040{ 1041 struct unpcb *unp = sotounpcb(so); 1042 struct unpcb *unp2; 1043 1044 UNP_LOCK_ASSERT(); 1045 1046 if (so2->so_type != so->so_type) 1047 return (EPROTOTYPE); 1048 unp2 = sotounpcb(so2); 1049 unp->unp_conn = unp2; 1050 switch (so->so_type) { 1051 1052 case SOCK_DGRAM: 1053 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 1054 soisconnected(so); 1055 break; 1056 1057 case SOCK_STREAM: 1058 unp2->unp_conn = unp; 1059 if (req == PRU_CONNECT && 1060 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 1061 soisconnecting(so); 1062 else 1063 soisconnected(so); 1064 soisconnected(so2); 1065 break; 1066 1067 default: 1068 panic("unp_connect2"); 1069 } 1070 return (0); 1071} 1072 1073static void 1074unp_disconnect(struct unpcb *unp) 1075{ 1076 struct unpcb *unp2 = unp->unp_conn; 1077 struct socket *so; 1078 1079 UNP_LOCK_ASSERT(); 1080 1081 if (unp2 == NULL) 1082 return; 1083 unp->unp_conn = NULL; 1084 switch (unp->unp_socket->so_type) { 1085 1086 case SOCK_DGRAM: 1087 LIST_REMOVE(unp, unp_reflink); 1088 so = unp->unp_socket; 1089 SOCK_LOCK(so); 1090 so->so_state &= ~SS_ISCONNECTED; 1091 SOCK_UNLOCK(so); 1092 break; 1093 1094 case SOCK_STREAM: 1095 soisdisconnected(unp->unp_socket); 1096 unp2->unp_conn = NULL; 1097 soisdisconnected(unp2->unp_socket); 1098 break; 1099 } 1100} 1101 1102#ifdef notdef 1103void 1104unp_abort(struct unpcb *unp) 1105{ 1106 1107 unp_detach(unp); 1108 UNP_UNLOCK_ASSERT(); 1109} 1110#endif 1111 1112/* 1113 * unp_pcblist() assumes that UNIX domain socket memory is never reclaimed 1114 * by the zone (UMA_ZONE_NOFREE), and as such potentially stale pointers 1115 * are safe to reference. It first scans the list of struct unpcb's to 1116 * generate a pointer list, then it rescans its list one entry at a time to 1117 * externalize and copyout. It checks the generation number to see if a 1118 * struct unpcb has been reused, and will skip it if so. 1119 */ 1120static int 1121unp_pcblist(SYSCTL_HANDLER_ARGS) 1122{ 1123 int error, i, n; 1124 struct unpcb *unp, **unp_list; 1125 unp_gen_t gencnt; 1126 struct xunpgen *xug; 1127 struct unp_head *head; 1128 struct xunpcb *xu; 1129 1130 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead); 1131 1132 /* 1133 * The process of preparing the PCB list is too time-consuming and 1134 * resource-intensive to repeat twice on every request. 1135 */ 1136 if (req->oldptr == NULL) { 1137 n = unp_count; 1138 req->oldidx = 2 * (sizeof *xug) 1139 + (n + n/8) * sizeof(struct xunpcb); 1140 return (0); 1141 } 1142 1143 if (req->newptr != NULL) 1144 return (EPERM); 1145 1146 /* 1147 * OK, now we're committed to doing something. 1148 */ 1149 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK); 1150 UNP_LOCK(); 1151 gencnt = unp_gencnt; 1152 n = unp_count; 1153 UNP_UNLOCK(); 1154 1155 xug->xug_len = sizeof *xug; 1156 xug->xug_count = n; 1157 xug->xug_gen = gencnt; 1158 xug->xug_sogen = so_gencnt; 1159 error = SYSCTL_OUT(req, xug, sizeof *xug); 1160 if (error) { 1161 free(xug, M_TEMP); 1162 return (error); 1163 } 1164 1165 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 1166 1167 UNP_LOCK(); 1168 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 1169 unp = LIST_NEXT(unp, unp_link)) { 1170 if (unp->unp_gencnt <= gencnt) { 1171 if (cr_cansee(req->td->td_ucred, 1172 unp->unp_socket->so_cred)) 1173 continue; 1174 unp_list[i++] = unp; 1175 } 1176 } 1177 UNP_UNLOCK(); 1178 n = i; /* in case we lost some during malloc */ 1179 1180 error = 0; 1181 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO); 1182 for (i = 0; i < n; i++) { 1183 unp = unp_list[i]; 1184 if (unp->unp_gencnt <= gencnt) { 1185 xu->xu_len = sizeof *xu; 1186 xu->xu_unpp = unp; 1187 /* 1188 * XXX - need more locking here to protect against 1189 * connect/disconnect races for SMP. 1190 */ 1191 if (unp->unp_addr != NULL) 1192 bcopy(unp->unp_addr, &xu->xu_addr, 1193 unp->unp_addr->sun_len); 1194 if (unp->unp_conn != NULL && 1195 unp->unp_conn->unp_addr != NULL) 1196 bcopy(unp->unp_conn->unp_addr, 1197 &xu->xu_caddr, 1198 unp->unp_conn->unp_addr->sun_len); 1199 bcopy(unp, &xu->xu_unp, sizeof *unp); 1200 sotoxsocket(unp->unp_socket, &xu->xu_socket); 1201 error = SYSCTL_OUT(req, xu, sizeof *xu); 1202 } 1203 } 1204 free(xu, M_TEMP); 1205 if (!error) { 1206 /* 1207 * Give the user an updated idea of our state. 1208 * If the generation differs from what we told 1209 * her before, she knows that something happened 1210 * while we were processing this request, and it 1211 * might be necessary to retry. 1212 */ 1213 xug->xug_gen = unp_gencnt; 1214 xug->xug_sogen = so_gencnt; 1215 xug->xug_count = unp_count; 1216 error = SYSCTL_OUT(req, xug, sizeof *xug); 1217 } 1218 free(unp_list, M_TEMP); 1219 free(xug, M_TEMP); 1220 return (error); 1221} 1222 1223SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD, 1224 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 1225 "List of active local datagram sockets"); 1226SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD, 1227 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 1228 "List of active local stream sockets"); 1229 1230static void 1231unp_shutdown(struct unpcb *unp) 1232{ 1233 struct socket *so; 1234 1235 UNP_LOCK_ASSERT(); 1236 1237 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn && 1238 (so = unp->unp_conn->unp_socket)) 1239 socantrcvmore(so); 1240} 1241 1242static void 1243unp_drop(struct unpcb *unp, int errno) 1244{ 1245 struct socket *so = unp->unp_socket; 1246 1247 UNP_LOCK_ASSERT(); 1248 1249 so->so_error = errno; 1250 unp_disconnect(unp); 1251} 1252 1253#ifdef notdef 1254void 1255unp_drain(void) 1256{ 1257 1258} 1259#endif 1260 1261static void 1262unp_freerights(struct file **rp, int fdcount) 1263{ 1264 int i; 1265 struct file *fp; 1266 1267 for (i = 0; i < fdcount; i++) { 1268 fp = *rp; 1269 /* 1270 * zero the pointer before calling 1271 * unp_discard since it may end up 1272 * in unp_gc().. 1273 */ 1274 *rp++ = 0; 1275 unp_discard(fp); 1276 } 1277} 1278 1279int 1280unp_externalize(struct mbuf *control, struct mbuf **controlp) 1281{ 1282 struct thread *td = curthread; /* XXX */ 1283 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1284 int i; 1285 int *fdp; 1286 struct file **rp; 1287 struct file *fp; 1288 void *data; 1289 socklen_t clen = control->m_len, datalen; 1290 int error, newfds; 1291 int f; 1292 u_int newlen; 1293 1294 UNP_UNLOCK_ASSERT(); 1295 1296 error = 0; 1297 if (controlp != NULL) /* controlp == NULL => free control messages */ 1298 *controlp = NULL; 1299 1300 while (cm != NULL) { 1301 if (sizeof(*cm) > clen || cm->cmsg_len > clen) { 1302 error = EINVAL; 1303 break; 1304 } 1305 1306 data = CMSG_DATA(cm); 1307 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1308 1309 if (cm->cmsg_level == SOL_SOCKET 1310 && cm->cmsg_type == SCM_RIGHTS) { 1311 newfds = datalen / sizeof(struct file *); 1312 rp = data; 1313 1314 /* If we're not outputting the descriptors free them. */ 1315 if (error || controlp == NULL) { 1316 unp_freerights(rp, newfds); 1317 goto next; 1318 } 1319 FILEDESC_LOCK(td->td_proc->p_fd); 1320 /* if the new FD's will not fit free them. */ 1321 if (!fdavail(td, newfds)) { 1322 FILEDESC_UNLOCK(td->td_proc->p_fd); 1323 error = EMSGSIZE; 1324 unp_freerights(rp, newfds); 1325 goto next; 1326 } 1327 /* 1328 * now change each pointer to an fd in the global 1329 * table to an integer that is the index to the 1330 * local fd table entry that we set up to point 1331 * to the global one we are transferring. 1332 */ 1333 newlen = newfds * sizeof(int); 1334 *controlp = sbcreatecontrol(NULL, newlen, 1335 SCM_RIGHTS, SOL_SOCKET); 1336 if (*controlp == NULL) { 1337 FILEDESC_UNLOCK(td->td_proc->p_fd); 1338 error = E2BIG; 1339 unp_freerights(rp, newfds); 1340 goto next; 1341 } 1342 1343 fdp = (int *) 1344 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1345 for (i = 0; i < newfds; i++) { 1346 if (fdalloc(td, 0, &f)) 1347 panic("unp_externalize fdalloc failed"); 1348 fp = *rp++; 1349 td->td_proc->p_fd->fd_ofiles[f] = fp; 1350 FILE_LOCK(fp); 1351 fp->f_msgcount--; 1352 FILE_UNLOCK(fp); 1353 unp_rights--; 1354 *fdp++ = f; 1355 } 1356 FILEDESC_UNLOCK(td->td_proc->p_fd); 1357 } else { /* We can just copy anything else across */ 1358 if (error || controlp == NULL) 1359 goto next; 1360 *controlp = sbcreatecontrol(NULL, datalen, 1361 cm->cmsg_type, cm->cmsg_level); 1362 if (*controlp == NULL) { 1363 error = ENOBUFS; 1364 goto next; 1365 } 1366 bcopy(data, 1367 CMSG_DATA(mtod(*controlp, struct cmsghdr *)), 1368 datalen); 1369 } 1370 1371 controlp = &(*controlp)->m_next; 1372 1373next: 1374 if (CMSG_SPACE(datalen) < clen) { 1375 clen -= CMSG_SPACE(datalen); 1376 cm = (struct cmsghdr *) 1377 ((caddr_t)cm + CMSG_SPACE(datalen)); 1378 } else { 1379 clen = 0; 1380 cm = NULL; 1381 } 1382 } 1383 1384 m_freem(control); 1385 1386 return (error); 1387} 1388 1389void 1390unp_init(void) 1391{ 1392 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL, 1393 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 1394 if (unp_zone == NULL) 1395 panic("unp_init"); 1396 uma_zone_set_max(unp_zone, nmbclusters); 1397 LIST_INIT(&unp_dhead); 1398 LIST_INIT(&unp_shead); 1399 TASK_INIT(&unp_gc_task, 0, unp_gc, NULL); 1400 UNP_LOCK_INIT(); 1401} 1402 1403static int 1404unp_internalize(struct mbuf **controlp, struct thread *td) 1405{ 1406 struct mbuf *control = *controlp; 1407 struct proc *p = td->td_proc; 1408 struct filedesc *fdescp = p->p_fd; 1409 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1410 struct cmsgcred *cmcred; 1411 struct file **rp; 1412 struct file *fp; 1413 struct timeval *tv; 1414 int i, fd, *fdp; 1415 void *data; 1416 socklen_t clen = control->m_len, datalen; 1417 int error, oldfds; 1418 u_int newlen; 1419 1420 UNP_UNLOCK_ASSERT(); 1421 1422 error = 0; 1423 *controlp = NULL; 1424 1425 while (cm != NULL) { 1426 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET 1427 || cm->cmsg_len > clen) { 1428 error = EINVAL; 1429 goto out; 1430 } 1431 1432 data = CMSG_DATA(cm); 1433 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1434 1435 switch (cm->cmsg_type) { 1436 /* 1437 * Fill in credential information. 1438 */ 1439 case SCM_CREDS: 1440 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred), 1441 SCM_CREDS, SOL_SOCKET); 1442 if (*controlp == NULL) { 1443 error = ENOBUFS; 1444 goto out; 1445 } 1446 1447 cmcred = (struct cmsgcred *) 1448 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1449 cmcred->cmcred_pid = p->p_pid; 1450 cmcred->cmcred_uid = td->td_ucred->cr_ruid; 1451 cmcred->cmcred_gid = td->td_ucred->cr_rgid; 1452 cmcred->cmcred_euid = td->td_ucred->cr_uid; 1453 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups, 1454 CMGROUP_MAX); 1455 for (i = 0; i < cmcred->cmcred_ngroups; i++) 1456 cmcred->cmcred_groups[i] = 1457 td->td_ucred->cr_groups[i]; 1458 break; 1459 1460 case SCM_RIGHTS: 1461 oldfds = datalen / sizeof (int); 1462 /* 1463 * check that all the FDs passed in refer to legal files 1464 * If not, reject the entire operation. 1465 */ 1466 fdp = data; 1467 FILEDESC_LOCK(fdescp); 1468 for (i = 0; i < oldfds; i++) { 1469 fd = *fdp++; 1470 if ((unsigned)fd >= fdescp->fd_nfiles || 1471 fdescp->fd_ofiles[fd] == NULL) { 1472 FILEDESC_UNLOCK(fdescp); 1473 error = EBADF; 1474 goto out; 1475 } 1476 fp = fdescp->fd_ofiles[fd]; 1477 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { 1478 FILEDESC_UNLOCK(fdescp); 1479 error = EOPNOTSUPP; 1480 goto out; 1481 } 1482 1483 } 1484 /* 1485 * Now replace the integer FDs with pointers to 1486 * the associated global file table entry.. 1487 */ 1488 newlen = oldfds * sizeof(struct file *); 1489 *controlp = sbcreatecontrol(NULL, newlen, 1490 SCM_RIGHTS, SOL_SOCKET); 1491 if (*controlp == NULL) { 1492 FILEDESC_UNLOCK(fdescp); 1493 error = E2BIG; 1494 goto out; 1495 } 1496 1497 fdp = data; 1498 rp = (struct file **) 1499 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1500 for (i = 0; i < oldfds; i++) { 1501 fp = fdescp->fd_ofiles[*fdp++]; 1502 *rp++ = fp; 1503 FILE_LOCK(fp); 1504 fp->f_count++; 1505 fp->f_msgcount++; 1506 FILE_UNLOCK(fp); 1507 unp_rights++; 1508 } 1509 FILEDESC_UNLOCK(fdescp); 1510 break; 1511 1512 case SCM_TIMESTAMP: 1513 *controlp = sbcreatecontrol(NULL, sizeof(*tv), 1514 SCM_TIMESTAMP, SOL_SOCKET); 1515 if (*controlp == NULL) { 1516 error = ENOBUFS; 1517 goto out; 1518 } 1519 tv = (struct timeval *) 1520 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1521 microtime(tv); 1522 break; 1523 1524 default: 1525 error = EINVAL; 1526 goto out; 1527 } 1528 1529 controlp = &(*controlp)->m_next; 1530 1531 if (CMSG_SPACE(datalen) < clen) { 1532 clen -= CMSG_SPACE(datalen); 1533 cm = (struct cmsghdr *) 1534 ((caddr_t)cm + CMSG_SPACE(datalen)); 1535 } else { 1536 clen = 0; 1537 cm = NULL; 1538 } 1539 } 1540 1541out: 1542 m_freem(control); 1543 1544 return (error); 1545} 1546 1547struct mbuf * 1548unp_addsockcred(struct thread *td, struct mbuf *control) 1549{ 1550 struct mbuf *m, *n; 1551 struct sockcred *sc; 1552 int ngroups; 1553 int i; 1554 1555 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); 1556 1557 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET); 1558 if (m == NULL) 1559 return (control); 1560 m->m_next = NULL; 1561 1562 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *)); 1563 sc->sc_uid = td->td_ucred->cr_ruid; 1564 sc->sc_euid = td->td_ucred->cr_uid; 1565 sc->sc_gid = td->td_ucred->cr_rgid; 1566 sc->sc_egid = td->td_ucred->cr_gid; 1567 sc->sc_ngroups = ngroups; 1568 for (i = 0; i < sc->sc_ngroups; i++) 1569 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 1570 1571 /* 1572 * If a control message already exists, append us to the end. 1573 */ 1574 if (control != NULL) { 1575 for (n = control; n->m_next != NULL; n = n->m_next) 1576 ; 1577 n->m_next = m; 1578 } else 1579 control = m; 1580 1581 return (control); 1582} 1583 1584/* 1585 * unp_defer indicates whether additional work has been defered for a future 1586 * pass through unp_gc(). It is thread local and does not require explicit 1587 * synchronization. 1588 */ 1589static int unp_defer; 1590 1591static int unp_taskcount; 1592SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, ""); 1593 1594static int unp_recycled; 1595SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, ""); 1596 1597static void 1598unp_gc(__unused void *arg, int pending) 1599{ 1600 struct file *fp, *nextfp; 1601 struct socket *so; 1602 struct file **extra_ref, **fpp; 1603 int nunref, i; 1604 int nfiles_snap; 1605 int nfiles_slack = 20; 1606 1607 unp_taskcount++; 1608 unp_defer = 0; 1609 /* 1610 * before going through all this, set all FDs to 1611 * be NOT defered and NOT externally accessible 1612 */ 1613 sx_slock(&filelist_lock); 1614 LIST_FOREACH(fp, &filehead, f_list) 1615 fp->f_gcflag &= ~(FMARK|FDEFER); 1616 do { 1617 KASSERT(unp_defer >= 0, ("unp_gc: unp_defer %d", unp_defer)); 1618 LIST_FOREACH(fp, &filehead, f_list) { 1619 FILE_LOCK(fp); 1620 /* 1621 * If the file is not open, skip it -- could be a 1622 * file in the process of being opened, or in the 1623 * process of being closed. If the file is 1624 * "closing", it may have been marked for deferred 1625 * consideration. Clear the flag now if so. 1626 */ 1627 if (fp->f_count == 0) { 1628 if (fp->f_gcflag & FDEFER) 1629 unp_defer--; 1630 fp->f_gcflag &= ~(FMARK|FDEFER); 1631 FILE_UNLOCK(fp); 1632 continue; 1633 } 1634 /* 1635 * If we already marked it as 'defer' in a 1636 * previous pass, then try process it this time 1637 * and un-mark it 1638 */ 1639 if (fp->f_gcflag & FDEFER) { 1640 fp->f_gcflag &= ~FDEFER; 1641 unp_defer--; 1642 } else { 1643 /* 1644 * if it's not defered, then check if it's 1645 * already marked.. if so skip it 1646 */ 1647 if (fp->f_gcflag & FMARK) { 1648 FILE_UNLOCK(fp); 1649 continue; 1650 } 1651 /* 1652 * If all references are from messages 1653 * in transit, then skip it. it's not 1654 * externally accessible. 1655 */ 1656 if (fp->f_count == fp->f_msgcount) { 1657 FILE_UNLOCK(fp); 1658 continue; 1659 } 1660 /* 1661 * If it got this far then it must be 1662 * externally accessible. 1663 */ 1664 fp->f_gcflag |= FMARK; 1665 } 1666 /* 1667 * either it was defered, or it is externally 1668 * accessible and not already marked so. 1669 * Now check if it is possibly one of OUR sockets. 1670 */ 1671 if (fp->f_type != DTYPE_SOCKET || 1672 (so = fp->f_data) == NULL) { 1673 FILE_UNLOCK(fp); 1674 continue; 1675 } 1676 FILE_UNLOCK(fp); 1677 if (so->so_proto->pr_domain != &localdomain || 1678 (so->so_proto->pr_flags&PR_RIGHTS) == 0) 1679 continue; 1680 /* 1681 * So, Ok, it's one of our sockets and it IS externally 1682 * accessible (or was defered). Now we look 1683 * to see if we hold any file descriptors in its 1684 * message buffers. Follow those links and mark them 1685 * as accessible too. 1686 */ 1687 SOCKBUF_LOCK(&so->so_rcv); 1688 unp_scan(so->so_rcv.sb_mb, unp_mark); 1689 SOCKBUF_UNLOCK(&so->so_rcv); 1690 } 1691 } while (unp_defer); 1692 sx_sunlock(&filelist_lock); 1693 /* 1694 * XXXRW: The following comments need updating for a post-SMPng and 1695 * deferred unp_gc() world, but are still generally accurate. 1696 * 1697 * We grab an extra reference to each of the file table entries 1698 * that are not otherwise accessible and then free the rights 1699 * that are stored in messages on them. 1700 * 1701 * The bug in the orginal code is a little tricky, so I'll describe 1702 * what's wrong with it here. 1703 * 1704 * It is incorrect to simply unp_discard each entry for f_msgcount 1705 * times -- consider the case of sockets A and B that contain 1706 * references to each other. On a last close of some other socket, 1707 * we trigger a gc since the number of outstanding rights (unp_rights) 1708 * is non-zero. If during the sweep phase the gc code unp_discards, 1709 * we end up doing a (full) closef on the descriptor. A closef on A 1710 * results in the following chain. Closef calls soo_close, which 1711 * calls soclose. Soclose calls first (through the switch 1712 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply 1713 * returns because the previous instance had set unp_gcing, and 1714 * we return all the way back to soclose, which marks the socket 1715 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush 1716 * to free up the rights that are queued in messages on the socket A, 1717 * i.e., the reference on B. The sorflush calls via the dom_dispose 1718 * switch unp_dispose, which unp_scans with unp_discard. This second 1719 * instance of unp_discard just calls closef on B. 1720 * 1721 * Well, a similar chain occurs on B, resulting in a sorflush on B, 1722 * which results in another closef on A. Unfortunately, A is already 1723 * being closed, and the descriptor has already been marked with 1724 * SS_NOFDREF, and soclose panics at this point. 1725 * 1726 * Here, we first take an extra reference to each inaccessible 1727 * descriptor. Then, we call sorflush ourself, since we know 1728 * it is a Unix domain socket anyhow. After we destroy all the 1729 * rights carried in messages, we do a last closef to get rid 1730 * of our extra reference. This is the last close, and the 1731 * unp_detach etc will shut down the socket. 1732 * 1733 * 91/09/19, bsy@cs.cmu.edu 1734 */ 1735again: 1736 nfiles_snap = openfiles + nfiles_slack; /* some slack */ 1737 extra_ref = malloc(nfiles_snap * sizeof(struct file *), M_TEMP, 1738 M_WAITOK); 1739 sx_slock(&filelist_lock); 1740 if (nfiles_snap < openfiles) { 1741 sx_sunlock(&filelist_lock); 1742 free(extra_ref, M_TEMP); 1743 nfiles_slack += 20; 1744 goto again; 1745 } 1746 for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref; 1747 fp != NULL; fp = nextfp) { 1748 nextfp = LIST_NEXT(fp, f_list); 1749 FILE_LOCK(fp); 1750 /* 1751 * If it's not open, skip it 1752 */ 1753 if (fp->f_count == 0) { 1754 FILE_UNLOCK(fp); 1755 continue; 1756 } 1757 /* 1758 * If all refs are from msgs, and it's not marked accessible 1759 * then it must be referenced from some unreachable cycle 1760 * of (shut-down) FDs, so include it in our 1761 * list of FDs to remove 1762 */ 1763 if (fp->f_count == fp->f_msgcount && !(fp->f_gcflag & FMARK)) { 1764 *fpp++ = fp; 1765 nunref++; 1766 fp->f_count++; 1767 } 1768 FILE_UNLOCK(fp); 1769 } 1770 sx_sunlock(&filelist_lock); 1771 /* 1772 * for each FD on our hit list, do the following two things 1773 */ 1774 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1775 struct file *tfp = *fpp; 1776 FILE_LOCK(tfp); 1777 if (tfp->f_type == DTYPE_SOCKET && 1778 tfp->f_data != NULL) { 1779 FILE_UNLOCK(tfp); 1780 sorflush(tfp->f_data); 1781 } else { 1782 FILE_UNLOCK(tfp); 1783 } 1784 } 1785 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1786 closef(*fpp, (struct thread *) NULL); 1787 unp_recycled++; 1788 } 1789 free(extra_ref, M_TEMP); 1790} 1791 1792void 1793unp_dispose(struct mbuf *m) 1794{ 1795 1796 if (m) 1797 unp_scan(m, unp_discard); 1798} 1799 1800static int 1801unp_listen(struct socket *so, struct unpcb *unp, int backlog, 1802 struct thread *td) 1803{ 1804 int error; 1805 1806 UNP_LOCK_ASSERT(); 1807 1808 SOCK_LOCK(so); 1809 error = solisten_proto_check(so); 1810 if (error == 0) { 1811 cru2x(td->td_ucred, &unp->unp_peercred); 1812 unp->unp_flags |= UNP_HAVEPCCACHED; 1813 solisten_proto(so, backlog); 1814 } 1815 SOCK_UNLOCK(so); 1816 return (error); 1817} 1818 1819static void 1820unp_scan(struct mbuf *m0, void (*op)(struct file *)) 1821{ 1822 struct mbuf *m; 1823 struct file **rp; 1824 struct cmsghdr *cm; 1825 void *data; 1826 int i; 1827 socklen_t clen, datalen; 1828 int qfds; 1829 1830 while (m0 != NULL) { 1831 for (m = m0; m; m = m->m_next) { 1832 if (m->m_type != MT_CONTROL) 1833 continue; 1834 1835 cm = mtod(m, struct cmsghdr *); 1836 clen = m->m_len; 1837 1838 while (cm != NULL) { 1839 if (sizeof(*cm) > clen || cm->cmsg_len > clen) 1840 break; 1841 1842 data = CMSG_DATA(cm); 1843 datalen = (caddr_t)cm + cm->cmsg_len 1844 - (caddr_t)data; 1845 1846 if (cm->cmsg_level == SOL_SOCKET && 1847 cm->cmsg_type == SCM_RIGHTS) { 1848 qfds = datalen / sizeof (struct file *); 1849 rp = data; 1850 for (i = 0; i < qfds; i++) 1851 (*op)(*rp++); 1852 } 1853 1854 if (CMSG_SPACE(datalen) < clen) { 1855 clen -= CMSG_SPACE(datalen); 1856 cm = (struct cmsghdr *) 1857 ((caddr_t)cm + CMSG_SPACE(datalen)); 1858 } else { 1859 clen = 0; 1860 cm = NULL; 1861 } 1862 } 1863 } 1864 m0 = m0->m_act; 1865 } 1866} 1867 1868static void 1869unp_mark(struct file *fp) 1870{ 1871 if (fp->f_gcflag & FMARK) 1872 return; 1873 unp_defer++; 1874 fp->f_gcflag |= (FMARK|FDEFER); 1875} 1876 1877static void 1878unp_discard(struct file *fp) 1879{ 1880 UNP_LOCK(); 1881 FILE_LOCK(fp); 1882 fp->f_msgcount--; 1883 unp_rights--; 1884 FILE_UNLOCK(fp); 1885 UNP_UNLOCK(); 1886 (void) closef(fp, (struct thread *)NULL); 1887} 1888