uipc_usrreq.c revision 142155
1/*- 2 * Copyright 2004-2005 Robert N. M. Watson 3 * Copyright (c) 1982, 1986, 1989, 1991, 1993 4 * The Regents of the University of California. All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 4. Neither the name of the University nor the names of its contributors 15 * may be used to endorse or promote products derived from this software 16 * without specific prior written permission. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 21 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 28 * SUCH DAMAGE. 29 * 30 * From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94 31 */ 32 33#include <sys/cdefs.h> 34__FBSDID("$FreeBSD: head/sys/kern/uipc_usrreq.c 142155 2005-02-20 23:22:13Z rwatson $"); 35 36#include "opt_mac.h" 37 38#include <sys/param.h> 39#include <sys/domain.h> 40#include <sys/fcntl.h> 41#include <sys/malloc.h> /* XXX must be before <sys/file.h> */ 42#include <sys/file.h> 43#include <sys/filedesc.h> 44#include <sys/jail.h> 45#include <sys/kernel.h> 46#include <sys/lock.h> 47#include <sys/mac.h> 48#include <sys/mbuf.h> 49#include <sys/mutex.h> 50#include <sys/namei.h> 51#include <sys/proc.h> 52#include <sys/protosw.h> 53#include <sys/resourcevar.h> 54#include <sys/socket.h> 55#include <sys/socketvar.h> 56#include <sys/signalvar.h> 57#include <sys/stat.h> 58#include <sys/sx.h> 59#include <sys/sysctl.h> 60#include <sys/systm.h> 61#include <sys/un.h> 62#include <sys/unpcb.h> 63#include <sys/vnode.h> 64 65#include <vm/uma.h> 66 67static uma_zone_t unp_zone; 68static unp_gen_t unp_gencnt; 69static u_int unp_count; 70 71static struct unp_head unp_shead, unp_dhead; 72 73/* 74 * Unix communications domain. 75 * 76 * TODO: 77 * SEQPACKET, RDM 78 * rethink name space problems 79 * need a proper out-of-band 80 * lock pushdown 81 */ 82static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL }; 83static ino_t unp_ino; /* prototype for fake inode numbers */ 84 85/* 86 * Currently, UNIX domain sockets are protected by a single subsystem lock, 87 * which covers global data structures and variables, the contents of each 88 * per-socket unpcb structure, and the so_pcb field in sockets attached to 89 * the UNIX domain. This provides for a moderate degree of paralellism, as 90 * receive operations on UNIX domain sockets do not need to acquire the 91 * subsystem lock. Finer grained locking to permit send() without acquiring 92 * a global lock would be a logical next step. 93 * 94 * The UNIX domain socket lock preceds all socket layer locks, including the 95 * socket lock and socket buffer lock, permitting UNIX domain socket code to 96 * call into socket support routines without releasing its locks. 97 * 98 * Some caution is required in areas where the UNIX domain socket code enters 99 * VFS in order to create or find rendezvous points. This results in 100 * dropping of the UNIX domain socket subsystem lock, acquisition of the 101 * Giant lock, and potential sleeping. This increases the chances of races, 102 * and exposes weaknesses in the socket->protocol API by offering poor 103 * failure modes. 104 */ 105static struct mtx unp_mtx; 106#define UNP_LOCK_INIT() \ 107 mtx_init(&unp_mtx, "unp", NULL, MTX_DEF) 108#define UNP_LOCK() mtx_lock(&unp_mtx) 109#define UNP_UNLOCK() mtx_unlock(&unp_mtx) 110#define UNP_LOCK_ASSERT() mtx_assert(&unp_mtx, MA_OWNED) 111#define UNP_UNLOCK_ASSERT() mtx_assert(&unp_mtx, MA_NOTOWNED) 112 113static int unp_attach(struct socket *); 114static void unp_detach(struct unpcb *); 115static int unp_bind(struct unpcb *,struct sockaddr *, struct thread *); 116static int unp_connect(struct socket *,struct sockaddr *, struct thread *); 117static int unp_connect2(struct socket *so, struct socket *so2); 118static void unp_disconnect(struct unpcb *); 119static void unp_shutdown(struct unpcb *); 120static void unp_drop(struct unpcb *, int); 121static void unp_gc(void); 122static void unp_scan(struct mbuf *, void (*)(struct file *)); 123static void unp_mark(struct file *); 124static void unp_discard(struct file *); 125static void unp_freerights(struct file **, int); 126static int unp_internalize(struct mbuf **, struct thread *); 127static int unp_listen(struct unpcb *, struct thread *); 128 129static int 130uipc_abort(struct socket *so) 131{ 132 struct unpcb *unp; 133 134 UNP_LOCK(); 135 unp = sotounpcb(so); 136 if (unp == NULL) { 137 UNP_UNLOCK(); 138 return (EINVAL); 139 } 140 unp_drop(unp, ECONNABORTED); 141 unp_detach(unp); 142 UNP_UNLOCK_ASSERT(); 143 ACCEPT_LOCK(); 144 SOCK_LOCK(so); 145 sotryfree(so); 146 return (0); 147} 148 149static int 150uipc_accept(struct socket *so, struct sockaddr **nam) 151{ 152 struct unpcb *unp; 153 const struct sockaddr *sa; 154 155 /* 156 * Pass back name of connected socket, 157 * if it was bound and we are still connected 158 * (our peer may have closed already!). 159 */ 160 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 161 UNP_LOCK(); 162 unp = sotounpcb(so); 163 if (unp == NULL) { 164 UNP_UNLOCK(); 165 free(*nam, M_SONAME); 166 *nam = NULL; 167 return (EINVAL); 168 } 169 if (unp->unp_conn != NULL && unp->unp_conn->unp_addr != NULL) 170 sa = (struct sockaddr *) unp->unp_conn->unp_addr; 171 else 172 sa = &sun_noname; 173 bcopy(sa, *nam, sa->sa_len); 174 UNP_UNLOCK(); 175 return (0); 176} 177 178static int 179uipc_attach(struct socket *so, int proto, struct thread *td) 180{ 181 struct unpcb *unp = sotounpcb(so); 182 183 if (unp != NULL) 184 return (EISCONN); 185 return (unp_attach(so)); 186} 187 188static int 189uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 190{ 191 struct unpcb *unp; 192 int error; 193 194 UNP_LOCK(); 195 unp = sotounpcb(so); 196 if (unp == NULL) { 197 UNP_UNLOCK(); 198 return (EINVAL); 199 } 200 error = unp_bind(unp, nam, td); 201 UNP_UNLOCK(); 202 return (error); 203} 204 205static int 206uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 207{ 208 struct unpcb *unp; 209 int error; 210 211 KASSERT(td == curthread, ("uipc_connect: td != curthread")); 212 213 UNP_LOCK(); 214 unp = sotounpcb(so); 215 if (unp == NULL) { 216 UNP_UNLOCK(); 217 return (EINVAL); 218 } 219 error = unp_connect(so, nam, td); 220 UNP_UNLOCK(); 221 return (error); 222} 223 224int 225uipc_connect2(struct socket *so1, struct socket *so2) 226{ 227 struct unpcb *unp; 228 int error; 229 230 UNP_LOCK(); 231 unp = sotounpcb(so1); 232 if (unp == NULL) { 233 UNP_UNLOCK(); 234 return (EINVAL); 235 } 236 error = unp_connect2(so1, so2); 237 UNP_UNLOCK(); 238 return (error); 239} 240 241/* control is EOPNOTSUPP */ 242 243static int 244uipc_detach(struct socket *so) 245{ 246 struct unpcb *unp; 247 248 UNP_LOCK(); 249 unp = sotounpcb(so); 250 if (unp == NULL) { 251 UNP_UNLOCK(); 252 return (EINVAL); 253 } 254 unp_detach(unp); 255 UNP_UNLOCK_ASSERT(); 256 return (0); 257} 258 259static int 260uipc_disconnect(struct socket *so) 261{ 262 struct unpcb *unp; 263 264 UNP_LOCK(); 265 unp = sotounpcb(so); 266 if (unp == NULL) { 267 UNP_UNLOCK(); 268 return (EINVAL); 269 } 270 unp_disconnect(unp); 271 UNP_UNLOCK(); 272 return (0); 273} 274 275static int 276uipc_listen(struct socket *so, struct thread *td) 277{ 278 struct unpcb *unp; 279 int error; 280 281 UNP_LOCK(); 282 unp = sotounpcb(so); 283 if (unp == NULL || unp->unp_vnode == NULL) { 284 UNP_UNLOCK(); 285 return (EINVAL); 286 } 287 error = unp_listen(unp, td); 288 UNP_UNLOCK(); 289 return (error); 290} 291 292static int 293uipc_peeraddr(struct socket *so, struct sockaddr **nam) 294{ 295 struct unpcb *unp; 296 const struct sockaddr *sa; 297 298 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 299 UNP_LOCK(); 300 unp = sotounpcb(so); 301 if (unp == NULL) { 302 UNP_UNLOCK(); 303 free(*nam, M_SONAME); 304 *nam = NULL; 305 return (EINVAL); 306 } 307 if (unp->unp_conn != NULL && unp->unp_conn->unp_addr!= NULL) 308 sa = (struct sockaddr *) unp->unp_conn->unp_addr; 309 else { 310 /* 311 * XXX: It seems that this test always fails even when 312 * connection is established. So, this else clause is 313 * added as workaround to return PF_LOCAL sockaddr. 314 */ 315 sa = &sun_noname; 316 } 317 bcopy(sa, *nam, sa->sa_len); 318 UNP_UNLOCK(); 319 return (0); 320} 321 322static int 323uipc_rcvd(struct socket *so, int flags) 324{ 325 struct unpcb *unp; 326 struct socket *so2; 327 u_long newhiwat; 328 329 UNP_LOCK(); 330 unp = sotounpcb(so); 331 if (unp == NULL) { 332 UNP_UNLOCK(); 333 return (EINVAL); 334 } 335 switch (so->so_type) { 336 case SOCK_DGRAM: 337 panic("uipc_rcvd DGRAM?"); 338 /*NOTREACHED*/ 339 340 case SOCK_STREAM: 341 if (unp->unp_conn == NULL) 342 break; 343 so2 = unp->unp_conn->unp_socket; 344 SOCKBUF_LOCK(&so2->so_snd); 345 SOCKBUF_LOCK(&so->so_rcv); 346 /* 347 * Adjust backpressure on sender 348 * and wakeup any waiting to write. 349 */ 350 so2->so_snd.sb_mbmax += unp->unp_mbcnt - so->so_rcv.sb_mbcnt; 351 unp->unp_mbcnt = so->so_rcv.sb_mbcnt; 352 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - 353 so->so_rcv.sb_cc; 354 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat, 355 newhiwat, RLIM_INFINITY); 356 unp->unp_cc = so->so_rcv.sb_cc; 357 SOCKBUF_UNLOCK(&so->so_rcv); 358 sowwakeup_locked(so2); 359 break; 360 361 default: 362 panic("uipc_rcvd unknown socktype"); 363 } 364 UNP_UNLOCK(); 365 return (0); 366} 367 368/* pru_rcvoob is EOPNOTSUPP */ 369 370static int 371uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 372 struct mbuf *control, struct thread *td) 373{ 374 int error = 0; 375 struct unpcb *unp; 376 struct socket *so2; 377 u_long newhiwat; 378 379 unp = sotounpcb(so); 380 if (unp == NULL) { 381 error = EINVAL; 382 goto release; 383 } 384 if (flags & PRUS_OOB) { 385 error = EOPNOTSUPP; 386 goto release; 387 } 388 389 if (control != NULL && (error = unp_internalize(&control, td))) 390 goto release; 391 392 UNP_LOCK(); 393 unp = sotounpcb(so); 394 if (unp == NULL) { 395 UNP_UNLOCK(); 396 error = EINVAL; 397 goto dispose_release; 398 } 399 400 switch (so->so_type) { 401 case SOCK_DGRAM: 402 { 403 const struct sockaddr *from; 404 405 if (nam != NULL) { 406 if (unp->unp_conn != NULL) { 407 error = EISCONN; 408 break; 409 } 410 error = unp_connect(so, nam, td); 411 if (error) 412 break; 413 } else { 414 if (unp->unp_conn == NULL) { 415 error = ENOTCONN; 416 break; 417 } 418 } 419 so2 = unp->unp_conn->unp_socket; 420 if (unp->unp_addr != NULL) 421 from = (struct sockaddr *)unp->unp_addr; 422 else 423 from = &sun_noname; 424 SOCKBUF_LOCK(&so2->so_rcv); 425 if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) { 426 sorwakeup_locked(so2); 427 m = NULL; 428 control = NULL; 429 } else { 430 SOCKBUF_UNLOCK(&so2->so_rcv); 431 error = ENOBUFS; 432 } 433 if (nam != NULL) 434 unp_disconnect(unp); 435 break; 436 } 437 438 case SOCK_STREAM: 439 /* Connect if not connected yet. */ 440 /* 441 * Note: A better implementation would complain 442 * if not equal to the peer's address. 443 */ 444 if ((so->so_state & SS_ISCONNECTED) == 0) { 445 if (nam != NULL) { 446 error = unp_connect(so, nam, td); 447 if (error) 448 break; /* XXX */ 449 } else { 450 error = ENOTCONN; 451 break; 452 } 453 } 454 455 SOCKBUF_LOCK(&so->so_snd); 456 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 457 SOCKBUF_UNLOCK(&so->so_snd); 458 error = EPIPE; 459 break; 460 } 461 if (unp->unp_conn == NULL) 462 panic("uipc_send connected but no connection?"); 463 so2 = unp->unp_conn->unp_socket; 464 SOCKBUF_LOCK(&so2->so_rcv); 465 /* 466 * Send to paired receive port, and then reduce 467 * send buffer hiwater marks to maintain backpressure. 468 * Wake up readers. 469 */ 470 if (control != NULL) { 471 if (sbappendcontrol_locked(&so2->so_rcv, m, control)) 472 control = NULL; 473 } else { 474 sbappend_locked(&so2->so_rcv, m); 475 } 476 so->so_snd.sb_mbmax -= 477 so2->so_rcv.sb_mbcnt - unp->unp_conn->unp_mbcnt; 478 unp->unp_conn->unp_mbcnt = so2->so_rcv.sb_mbcnt; 479 newhiwat = so->so_snd.sb_hiwat - 480 (so2->so_rcv.sb_cc - unp->unp_conn->unp_cc); 481 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat, 482 newhiwat, RLIM_INFINITY); 483 SOCKBUF_UNLOCK(&so->so_snd); 484 unp->unp_conn->unp_cc = so2->so_rcv.sb_cc; 485 sorwakeup_locked(so2); 486 m = NULL; 487 break; 488 489 default: 490 panic("uipc_send unknown socktype"); 491 } 492 493 /* 494 * SEND_EOF is equivalent to a SEND followed by 495 * a SHUTDOWN. 496 */ 497 if (flags & PRUS_EOF) { 498 socantsendmore(so); 499 unp_shutdown(unp); 500 } 501 UNP_UNLOCK(); 502 503dispose_release: 504 if (control != NULL && error != 0) 505 unp_dispose(control); 506 507release: 508 if (control != NULL) 509 m_freem(control); 510 if (m != NULL) 511 m_freem(m); 512 return (error); 513} 514 515static int 516uipc_sense(struct socket *so, struct stat *sb) 517{ 518 struct unpcb *unp; 519 struct socket *so2; 520 521 UNP_LOCK(); 522 unp = sotounpcb(so); 523 if (unp == NULL) { 524 UNP_UNLOCK(); 525 return (EINVAL); 526 } 527 sb->st_blksize = so->so_snd.sb_hiwat; 528 if (so->so_type == SOCK_STREAM && unp->unp_conn != NULL) { 529 so2 = unp->unp_conn->unp_socket; 530 sb->st_blksize += so2->so_rcv.sb_cc; 531 } 532 sb->st_dev = NODEV; 533 if (unp->unp_ino == 0) 534 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino; 535 sb->st_ino = unp->unp_ino; 536 UNP_UNLOCK(); 537 return (0); 538} 539 540static int 541uipc_shutdown(struct socket *so) 542{ 543 struct unpcb *unp; 544 545 UNP_LOCK(); 546 unp = sotounpcb(so); 547 if (unp == NULL) { 548 UNP_UNLOCK(); 549 return (EINVAL); 550 } 551 socantsendmore(so); 552 unp_shutdown(unp); 553 UNP_UNLOCK(); 554 return (0); 555} 556 557static int 558uipc_sockaddr(struct socket *so, struct sockaddr **nam) 559{ 560 struct unpcb *unp; 561 const struct sockaddr *sa; 562 563 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 564 UNP_LOCK(); 565 unp = sotounpcb(so); 566 if (unp == NULL) { 567 UNP_UNLOCK(); 568 free(*nam, M_SONAME); 569 *nam = NULL; 570 return (EINVAL); 571 } 572 if (unp->unp_addr != NULL) 573 sa = (struct sockaddr *) unp->unp_addr; 574 else 575 sa = &sun_noname; 576 bcopy(sa, *nam, sa->sa_len); 577 UNP_UNLOCK(); 578 return (0); 579} 580 581struct pr_usrreqs uipc_usrreqs = { 582 .pru_abort = uipc_abort, 583 .pru_accept = uipc_accept, 584 .pru_attach = uipc_attach, 585 .pru_bind = uipc_bind, 586 .pru_connect = uipc_connect, 587 .pru_connect2 = uipc_connect2, 588 .pru_detach = uipc_detach, 589 .pru_disconnect = uipc_disconnect, 590 .pru_listen = uipc_listen, 591 .pru_peeraddr = uipc_peeraddr, 592 .pru_rcvd = uipc_rcvd, 593 .pru_send = uipc_send, 594 .pru_sense = uipc_sense, 595 .pru_shutdown = uipc_shutdown, 596 .pru_sockaddr = uipc_sockaddr, 597 .pru_sosend = sosend, 598 .pru_soreceive = soreceive, 599 .pru_sopoll = sopoll, 600}; 601 602int 603uipc_ctloutput(struct socket *so, struct sockopt *sopt) 604{ 605 struct unpcb *unp; 606 struct xucred xu; 607 int error; 608 609 switch (sopt->sopt_dir) { 610 case SOPT_GET: 611 switch (sopt->sopt_name) { 612 case LOCAL_PEERCRED: 613 error = 0; 614 UNP_LOCK(); 615 unp = sotounpcb(so); 616 if (unp == NULL) { 617 UNP_UNLOCK(); 618 error = EINVAL; 619 break; 620 } 621 if (unp->unp_flags & UNP_HAVEPC) 622 xu = unp->unp_peercred; 623 else { 624 if (so->so_type == SOCK_STREAM) 625 error = ENOTCONN; 626 else 627 error = EINVAL; 628 } 629 UNP_UNLOCK(); 630 if (error == 0) 631 error = sooptcopyout(sopt, &xu, sizeof(xu)); 632 break; 633 default: 634 error = EOPNOTSUPP; 635 break; 636 } 637 break; 638 case SOPT_SET: 639 default: 640 error = EOPNOTSUPP; 641 break; 642 } 643 return (error); 644} 645 646/* 647 * Both send and receive buffers are allocated PIPSIZ bytes of buffering 648 * for stream sockets, although the total for sender and receiver is 649 * actually only PIPSIZ. 650 * Datagram sockets really use the sendspace as the maximum datagram size, 651 * and don't really want to reserve the sendspace. Their recvspace should 652 * be large enough for at least one max-size datagram plus address. 653 */ 654#ifndef PIPSIZ 655#define PIPSIZ 8192 656#endif 657static u_long unpst_sendspace = PIPSIZ; 658static u_long unpst_recvspace = PIPSIZ; 659static u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 660static u_long unpdg_recvspace = 4*1024; 661 662static int unp_rights; /* file descriptors in flight */ 663 664SYSCTL_DECL(_net_local_stream); 665SYSCTL_INT(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 666 &unpst_sendspace, 0, ""); 667SYSCTL_INT(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 668 &unpst_recvspace, 0, ""); 669SYSCTL_DECL(_net_local_dgram); 670SYSCTL_INT(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 671 &unpdg_sendspace, 0, ""); 672SYSCTL_INT(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 673 &unpdg_recvspace, 0, ""); 674SYSCTL_DECL(_net_local); 675SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, ""); 676 677static int 678unp_attach(struct socket *so) 679{ 680 struct unpcb *unp; 681 int error; 682 683 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 684 switch (so->so_type) { 685 686 case SOCK_STREAM: 687 error = soreserve(so, unpst_sendspace, unpst_recvspace); 688 break; 689 690 case SOCK_DGRAM: 691 error = soreserve(so, unpdg_sendspace, unpdg_recvspace); 692 break; 693 694 default: 695 panic("unp_attach"); 696 } 697 if (error) 698 return (error); 699 } 700 unp = uma_zalloc(unp_zone, M_WAITOK | M_ZERO); 701 if (unp == NULL) 702 return (ENOBUFS); 703 LIST_INIT(&unp->unp_refs); 704 unp->unp_socket = so; 705 so->so_pcb = unp; 706 707 UNP_LOCK(); 708 unp->unp_gencnt = ++unp_gencnt; 709 unp_count++; 710 LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead 711 : &unp_shead, unp, unp_link); 712 UNP_UNLOCK(); 713 714 return (0); 715} 716 717static void 718unp_detach(struct unpcb *unp) 719{ 720 struct vnode *vp; 721 722 UNP_LOCK_ASSERT(); 723 724 LIST_REMOVE(unp, unp_link); 725 unp->unp_gencnt = ++unp_gencnt; 726 --unp_count; 727 if ((vp = unp->unp_vnode) != NULL) { 728 /* 729 * XXXRW: should v_socket be frobbed only while holding 730 * Giant? 731 */ 732 unp->unp_vnode->v_socket = NULL; 733 unp->unp_vnode = NULL; 734 } 735 if (unp->unp_conn != NULL) 736 unp_disconnect(unp); 737 while (!LIST_EMPTY(&unp->unp_refs)) { 738 struct unpcb *ref = LIST_FIRST(&unp->unp_refs); 739 unp_drop(ref, ECONNRESET); 740 } 741 soisdisconnected(unp->unp_socket); 742 unp->unp_socket->so_pcb = NULL; 743 if (unp_rights) { 744 /* 745 * Normally the receive buffer is flushed later, 746 * in sofree, but if our receive buffer holds references 747 * to descriptors that are now garbage, we will dispose 748 * of those descriptor references after the garbage collector 749 * gets them (resulting in a "panic: closef: count < 0"). 750 */ 751 sorflush(unp->unp_socket); 752 unp_gc(); /* Will unlock UNP. */ 753 } else 754 UNP_UNLOCK(); 755 UNP_UNLOCK_ASSERT(); 756 if (unp->unp_addr != NULL) 757 FREE(unp->unp_addr, M_SONAME); 758 uma_zfree(unp_zone, unp); 759 if (vp) { 760 mtx_lock(&Giant); 761 vrele(vp); 762 mtx_unlock(&Giant); 763 } 764} 765 766static int 767unp_bind(struct unpcb *unp, struct sockaddr *nam, struct thread *td) 768{ 769 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 770 struct vnode *vp; 771 struct mount *mp; 772 struct vattr vattr; 773 int error, namelen; 774 struct nameidata nd; 775 char *buf; 776 777 UNP_LOCK_ASSERT(); 778 779 /* 780 * XXXRW: This test-and-set of unp_vnode is non-atomic; the 781 * unlocked read here is fine, but the value of unp_vnode needs 782 * to be tested again after we do all the lookups to see if the 783 * pcb is still unbound? 784 */ 785 if (unp->unp_vnode != NULL) 786 return (EINVAL); 787 788 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 789 if (namelen <= 0) 790 return (EINVAL); 791 792 UNP_UNLOCK(); 793 794 buf = malloc(namelen + 1, M_TEMP, M_WAITOK); 795 strlcpy(buf, soun->sun_path, namelen + 1); 796 797 mtx_lock(&Giant); 798restart: 799 mtx_assert(&Giant, MA_OWNED); 800 NDINIT(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME, UIO_SYSSPACE, 801 buf, td); 802/* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 803 error = namei(&nd); 804 if (error) 805 goto done; 806 vp = nd.ni_vp; 807 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { 808 NDFREE(&nd, NDF_ONLY_PNBUF); 809 if (nd.ni_dvp == vp) 810 vrele(nd.ni_dvp); 811 else 812 vput(nd.ni_dvp); 813 if (vp != NULL) { 814 vrele(vp); 815 error = EADDRINUSE; 816 goto done; 817 } 818 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH); 819 if (error) 820 goto done; 821 goto restart; 822 } 823 VATTR_NULL(&vattr); 824 vattr.va_type = VSOCK; 825 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask); 826#ifdef MAC 827 error = mac_check_vnode_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, 828 &vattr); 829#endif 830 if (error == 0) { 831 VOP_LEASE(nd.ni_dvp, td, td->td_ucred, LEASE_WRITE); 832 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 833 } 834 NDFREE(&nd, NDF_ONLY_PNBUF); 835 vput(nd.ni_dvp); 836 if (error) 837 goto done; 838 vp = nd.ni_vp; 839 ASSERT_VOP_LOCKED(vp, "unp_bind"); 840 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK); 841 UNP_LOCK(); 842 vp->v_socket = unp->unp_socket; 843 unp->unp_vnode = vp; 844 unp->unp_addr = soun; 845 UNP_UNLOCK(); 846 VOP_UNLOCK(vp, 0, td); 847 vn_finished_write(mp); 848done: 849 mtx_unlock(&Giant); 850 free(buf, M_TEMP); 851 UNP_LOCK(); 852 return (error); 853} 854 855static int 856unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 857{ 858 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 859 struct vnode *vp; 860 struct socket *so2, *so3; 861 struct unpcb *unp, *unp2, *unp3; 862 int error, len; 863 struct nameidata nd; 864 char buf[SOCK_MAXADDRLEN]; 865 struct sockaddr *sa; 866 867 UNP_LOCK_ASSERT(); 868 unp = sotounpcb(so); 869 870 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 871 if (len <= 0) 872 return (EINVAL); 873 strlcpy(buf, soun->sun_path, len + 1); 874 UNP_UNLOCK(); 875 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 876 mtx_lock(&Giant); 877 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, td); 878 error = namei(&nd); 879 if (error) 880 vp = NULL; 881 else 882 vp = nd.ni_vp; 883 ASSERT_VOP_LOCKED(vp, "unp_connect"); 884 NDFREE(&nd, NDF_ONLY_PNBUF); 885 if (error) 886 goto bad; 887 888 if (vp->v_type != VSOCK) { 889 error = ENOTSOCK; 890 goto bad; 891 } 892 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); 893 if (error) 894 goto bad; 895 mtx_unlock(&Giant); 896 UNP_LOCK(); 897 unp = sotounpcb(so); 898 if (unp == NULL) { 899 error = EINVAL; 900 goto bad2; 901 } 902 so2 = vp->v_socket; 903 if (so2 == NULL) { 904 error = ECONNREFUSED; 905 goto bad2; 906 } 907 if (so->so_type != so2->so_type) { 908 error = EPROTOTYPE; 909 goto bad2; 910 } 911 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 912 if (so2->so_options & SO_ACCEPTCONN) { 913 /* 914 * NB: drop locks here so unp_attach is entered 915 * w/o locks; this avoids a recursive lock 916 * of the head and holding sleep locks across 917 * a (potentially) blocking malloc. 918 */ 919 UNP_UNLOCK(); 920 so3 = sonewconn(so2, 0); 921 UNP_LOCK(); 922 } else 923 so3 = NULL; 924 if (so3 == NULL) { 925 error = ECONNREFUSED; 926 goto bad2; 927 } 928 unp = sotounpcb(so); 929 unp2 = sotounpcb(so2); 930 unp3 = sotounpcb(so3); 931 if (unp2->unp_addr != NULL) { 932 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); 933 unp3->unp_addr = (struct sockaddr_un *) sa; 934 sa = NULL; 935 } 936 /* 937 * unp_peercred management: 938 * 939 * The connecter's (client's) credentials are copied 940 * from its process structure at the time of connect() 941 * (which is now). 942 */ 943 cru2x(td->td_ucred, &unp3->unp_peercred); 944 unp3->unp_flags |= UNP_HAVEPC; 945 /* 946 * The receiver's (server's) credentials are copied 947 * from the unp_peercred member of socket on which the 948 * former called listen(); unp_listen() cached that 949 * process's credentials at that time so we can use 950 * them now. 951 */ 952 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED, 953 ("unp_connect: listener without cached peercred")); 954 memcpy(&unp->unp_peercred, &unp2->unp_peercred, 955 sizeof(unp->unp_peercred)); 956 unp->unp_flags |= UNP_HAVEPC; 957#ifdef MAC 958 SOCK_LOCK(so); 959 mac_set_socket_peer_from_socket(so, so3); 960 mac_set_socket_peer_from_socket(so3, so); 961 SOCK_UNLOCK(so); 962#endif 963 964 so2 = so3; 965 } 966 error = unp_connect2(so, so2); 967bad2: 968 UNP_UNLOCK(); 969 mtx_lock(&Giant); 970bad: 971 mtx_assert(&Giant, MA_OWNED); 972 if (vp != NULL) 973 vput(vp); 974 mtx_unlock(&Giant); 975 free(sa, M_SONAME); 976 UNP_LOCK(); 977 return (error); 978} 979 980static int 981unp_connect2(struct socket *so, struct socket *so2) 982{ 983 struct unpcb *unp = sotounpcb(so); 984 struct unpcb *unp2; 985 986 UNP_LOCK_ASSERT(); 987 988 if (so2->so_type != so->so_type) 989 return (EPROTOTYPE); 990 unp2 = sotounpcb(so2); 991 unp->unp_conn = unp2; 992 switch (so->so_type) { 993 994 case SOCK_DGRAM: 995 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 996 soisconnected(so); 997 break; 998 999 case SOCK_STREAM: 1000 unp2->unp_conn = unp; 1001 soisconnected(so); 1002 soisconnected(so2); 1003 break; 1004 1005 default: 1006 panic("unp_connect2"); 1007 } 1008 return (0); 1009} 1010 1011static void 1012unp_disconnect(struct unpcb *unp) 1013{ 1014 struct unpcb *unp2 = unp->unp_conn; 1015 struct socket *so; 1016 1017 UNP_LOCK_ASSERT(); 1018 1019 if (unp2 == NULL) 1020 return; 1021 unp->unp_conn = NULL; 1022 switch (unp->unp_socket->so_type) { 1023 1024 case SOCK_DGRAM: 1025 LIST_REMOVE(unp, unp_reflink); 1026 so = unp->unp_socket; 1027 SOCK_LOCK(so); 1028 so->so_state &= ~SS_ISCONNECTED; 1029 SOCK_UNLOCK(so); 1030 break; 1031 1032 case SOCK_STREAM: 1033 soisdisconnected(unp->unp_socket); 1034 unp2->unp_conn = NULL; 1035 soisdisconnected(unp2->unp_socket); 1036 break; 1037 } 1038} 1039 1040#ifdef notdef 1041void 1042unp_abort(struct unpcb *unp) 1043{ 1044 1045 unp_detach(unp); 1046 UNP_UNLOCK_ASSERT(); 1047} 1048#endif 1049 1050/* 1051 * unp_pcblist() assumes that UNIX domain socket memory is never reclaimed 1052 * by the zone (UMA_ZONE_NOFREE), and as such potentially stale pointers 1053 * are safe to reference. It first scans the list of struct unpcb's to 1054 * generate a pointer list, then it rescans its list one entry at a time to 1055 * externalize and copyout. It checks the generation number to see if a 1056 * struct unpcb has been reused, and will skip it if so. 1057 */ 1058static int 1059unp_pcblist(SYSCTL_HANDLER_ARGS) 1060{ 1061 int error, i, n; 1062 struct unpcb *unp, **unp_list; 1063 unp_gen_t gencnt; 1064 struct xunpgen *xug; 1065 struct unp_head *head; 1066 struct xunpcb *xu; 1067 1068 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead); 1069 1070 /* 1071 * The process of preparing the PCB list is too time-consuming and 1072 * resource-intensive to repeat twice on every request. 1073 */ 1074 if (req->oldptr == NULL) { 1075 n = unp_count; 1076 req->oldidx = 2 * (sizeof *xug) 1077 + (n + n/8) * sizeof(struct xunpcb); 1078 return (0); 1079 } 1080 1081 if (req->newptr != NULL) 1082 return (EPERM); 1083 1084 /* 1085 * OK, now we're committed to doing something. 1086 */ 1087 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK); 1088 UNP_LOCK(); 1089 gencnt = unp_gencnt; 1090 n = unp_count; 1091 UNP_UNLOCK(); 1092 1093 xug->xug_len = sizeof *xug; 1094 xug->xug_count = n; 1095 xug->xug_gen = gencnt; 1096 xug->xug_sogen = so_gencnt; 1097 error = SYSCTL_OUT(req, xug, sizeof *xug); 1098 if (error) { 1099 free(xug, M_TEMP); 1100 return (error); 1101 } 1102 1103 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 1104 1105 UNP_LOCK(); 1106 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 1107 unp = LIST_NEXT(unp, unp_link)) { 1108 if (unp->unp_gencnt <= gencnt) { 1109 if (cr_cansee(req->td->td_ucred, 1110 unp->unp_socket->so_cred)) 1111 continue; 1112 unp_list[i++] = unp; 1113 } 1114 } 1115 UNP_UNLOCK(); 1116 n = i; /* in case we lost some during malloc */ 1117 1118 error = 0; 1119 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK); 1120 for (i = 0; i < n; i++) { 1121 unp = unp_list[i]; 1122 if (unp->unp_gencnt <= gencnt) { 1123 xu->xu_len = sizeof *xu; 1124 xu->xu_unpp = unp; 1125 /* 1126 * XXX - need more locking here to protect against 1127 * connect/disconnect races for SMP. 1128 */ 1129 if (unp->unp_addr != NULL) 1130 bcopy(unp->unp_addr, &xu->xu_addr, 1131 unp->unp_addr->sun_len); 1132 if (unp->unp_conn != NULL && 1133 unp->unp_conn->unp_addr != NULL) 1134 bcopy(unp->unp_conn->unp_addr, 1135 &xu->xu_caddr, 1136 unp->unp_conn->unp_addr->sun_len); 1137 bcopy(unp, &xu->xu_unp, sizeof *unp); 1138 sotoxsocket(unp->unp_socket, &xu->xu_socket); 1139 error = SYSCTL_OUT(req, xu, sizeof *xu); 1140 } 1141 } 1142 free(xu, M_TEMP); 1143 if (!error) { 1144 /* 1145 * Give the user an updated idea of our state. 1146 * If the generation differs from what we told 1147 * her before, she knows that something happened 1148 * while we were processing this request, and it 1149 * might be necessary to retry. 1150 */ 1151 xug->xug_gen = unp_gencnt; 1152 xug->xug_sogen = so_gencnt; 1153 xug->xug_count = unp_count; 1154 error = SYSCTL_OUT(req, xug, sizeof *xug); 1155 } 1156 free(unp_list, M_TEMP); 1157 free(xug, M_TEMP); 1158 return (error); 1159} 1160 1161SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD, 1162 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 1163 "List of active local datagram sockets"); 1164SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD, 1165 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 1166 "List of active local stream sockets"); 1167 1168static void 1169unp_shutdown(struct unpcb *unp) 1170{ 1171 struct socket *so; 1172 1173 UNP_LOCK_ASSERT(); 1174 1175 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn && 1176 (so = unp->unp_conn->unp_socket)) 1177 socantrcvmore(so); 1178} 1179 1180static void 1181unp_drop(struct unpcb *unp, int errno) 1182{ 1183 struct socket *so = unp->unp_socket; 1184 1185 UNP_LOCK_ASSERT(); 1186 1187 so->so_error = errno; 1188 unp_disconnect(unp); 1189} 1190 1191#ifdef notdef 1192void 1193unp_drain(void) 1194{ 1195 1196} 1197#endif 1198 1199static void 1200unp_freerights(struct file **rp, int fdcount) 1201{ 1202 int i; 1203 struct file *fp; 1204 1205 for (i = 0; i < fdcount; i++) { 1206 fp = *rp; 1207 /* 1208 * zero the pointer before calling 1209 * unp_discard since it may end up 1210 * in unp_gc().. 1211 */ 1212 *rp++ = 0; 1213 unp_discard(fp); 1214 } 1215} 1216 1217int 1218unp_externalize(struct mbuf *control, struct mbuf **controlp) 1219{ 1220 struct thread *td = curthread; /* XXX */ 1221 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1222 int i; 1223 int *fdp; 1224 struct file **rp; 1225 struct file *fp; 1226 void *data; 1227 socklen_t clen = control->m_len, datalen; 1228 int error, newfds; 1229 int f; 1230 u_int newlen; 1231 1232 UNP_UNLOCK_ASSERT(); 1233 1234 error = 0; 1235 if (controlp != NULL) /* controlp == NULL => free control messages */ 1236 *controlp = NULL; 1237 1238 while (cm != NULL) { 1239 if (sizeof(*cm) > clen || cm->cmsg_len > clen) { 1240 error = EINVAL; 1241 break; 1242 } 1243 1244 data = CMSG_DATA(cm); 1245 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1246 1247 if (cm->cmsg_level == SOL_SOCKET 1248 && cm->cmsg_type == SCM_RIGHTS) { 1249 newfds = datalen / sizeof(struct file *); 1250 rp = data; 1251 1252 /* If we're not outputting the descriptors free them. */ 1253 if (error || controlp == NULL) { 1254 unp_freerights(rp, newfds); 1255 goto next; 1256 } 1257 FILEDESC_LOCK(td->td_proc->p_fd); 1258 /* if the new FD's will not fit free them. */ 1259 if (!fdavail(td, newfds)) { 1260 FILEDESC_UNLOCK(td->td_proc->p_fd); 1261 error = EMSGSIZE; 1262 unp_freerights(rp, newfds); 1263 goto next; 1264 } 1265 /* 1266 * now change each pointer to an fd in the global 1267 * table to an integer that is the index to the 1268 * local fd table entry that we set up to point 1269 * to the global one we are transferring. 1270 */ 1271 newlen = newfds * sizeof(int); 1272 *controlp = sbcreatecontrol(NULL, newlen, 1273 SCM_RIGHTS, SOL_SOCKET); 1274 if (*controlp == NULL) { 1275 FILEDESC_UNLOCK(td->td_proc->p_fd); 1276 error = E2BIG; 1277 unp_freerights(rp, newfds); 1278 goto next; 1279 } 1280 1281 fdp = (int *) 1282 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1283 for (i = 0; i < newfds; i++) { 1284 if (fdalloc(td, 0, &f)) 1285 panic("unp_externalize fdalloc failed"); 1286 fp = *rp++; 1287 td->td_proc->p_fd->fd_ofiles[f] = fp; 1288 FILE_LOCK(fp); 1289 fp->f_msgcount--; 1290 FILE_UNLOCK(fp); 1291 unp_rights--; 1292 *fdp++ = f; 1293 } 1294 FILEDESC_UNLOCK(td->td_proc->p_fd); 1295 } else { /* We can just copy anything else across */ 1296 if (error || controlp == NULL) 1297 goto next; 1298 *controlp = sbcreatecontrol(NULL, datalen, 1299 cm->cmsg_type, cm->cmsg_level); 1300 if (*controlp == NULL) { 1301 error = ENOBUFS; 1302 goto next; 1303 } 1304 bcopy(data, 1305 CMSG_DATA(mtod(*controlp, struct cmsghdr *)), 1306 datalen); 1307 } 1308 1309 controlp = &(*controlp)->m_next; 1310 1311next: 1312 if (CMSG_SPACE(datalen) < clen) { 1313 clen -= CMSG_SPACE(datalen); 1314 cm = (struct cmsghdr *) 1315 ((caddr_t)cm + CMSG_SPACE(datalen)); 1316 } else { 1317 clen = 0; 1318 cm = NULL; 1319 } 1320 } 1321 1322 m_freem(control); 1323 1324 return (error); 1325} 1326 1327void 1328unp_init(void) 1329{ 1330 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL, 1331 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 1332 if (unp_zone == NULL) 1333 panic("unp_init"); 1334 uma_zone_set_max(unp_zone, nmbclusters); 1335 LIST_INIT(&unp_dhead); 1336 LIST_INIT(&unp_shead); 1337 1338 UNP_LOCK_INIT(); 1339} 1340 1341static int 1342unp_internalize(struct mbuf **controlp, struct thread *td) 1343{ 1344 struct mbuf *control = *controlp; 1345 struct proc *p = td->td_proc; 1346 struct filedesc *fdescp = p->p_fd; 1347 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1348 struct cmsgcred *cmcred; 1349 struct file **rp; 1350 struct file *fp; 1351 struct timeval *tv; 1352 int i, fd, *fdp; 1353 void *data; 1354 socklen_t clen = control->m_len, datalen; 1355 int error, oldfds; 1356 u_int newlen; 1357 1358 UNP_UNLOCK_ASSERT(); 1359 1360 error = 0; 1361 *controlp = NULL; 1362 1363 while (cm != NULL) { 1364 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET 1365 || cm->cmsg_len > clen) { 1366 error = EINVAL; 1367 goto out; 1368 } 1369 1370 data = CMSG_DATA(cm); 1371 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1372 1373 switch (cm->cmsg_type) { 1374 /* 1375 * Fill in credential information. 1376 */ 1377 case SCM_CREDS: 1378 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred), 1379 SCM_CREDS, SOL_SOCKET); 1380 if (*controlp == NULL) { 1381 error = ENOBUFS; 1382 goto out; 1383 } 1384 1385 cmcred = (struct cmsgcred *) 1386 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1387 cmcred->cmcred_pid = p->p_pid; 1388 cmcred->cmcred_uid = td->td_ucred->cr_ruid; 1389 cmcred->cmcred_gid = td->td_ucred->cr_rgid; 1390 cmcred->cmcred_euid = td->td_ucred->cr_uid; 1391 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups, 1392 CMGROUP_MAX); 1393 for (i = 0; i < cmcred->cmcred_ngroups; i++) 1394 cmcred->cmcred_groups[i] = 1395 td->td_ucred->cr_groups[i]; 1396 break; 1397 1398 case SCM_RIGHTS: 1399 oldfds = datalen / sizeof (int); 1400 /* 1401 * check that all the FDs passed in refer to legal files 1402 * If not, reject the entire operation. 1403 */ 1404 fdp = data; 1405 FILEDESC_LOCK(fdescp); 1406 for (i = 0; i < oldfds; i++) { 1407 fd = *fdp++; 1408 if ((unsigned)fd >= fdescp->fd_nfiles || 1409 fdescp->fd_ofiles[fd] == NULL) { 1410 FILEDESC_UNLOCK(fdescp); 1411 error = EBADF; 1412 goto out; 1413 } 1414 fp = fdescp->fd_ofiles[fd]; 1415 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { 1416 FILEDESC_UNLOCK(fdescp); 1417 error = EOPNOTSUPP; 1418 goto out; 1419 } 1420 1421 } 1422 /* 1423 * Now replace the integer FDs with pointers to 1424 * the associated global file table entry.. 1425 */ 1426 newlen = oldfds * sizeof(struct file *); 1427 *controlp = sbcreatecontrol(NULL, newlen, 1428 SCM_RIGHTS, SOL_SOCKET); 1429 if (*controlp == NULL) { 1430 FILEDESC_UNLOCK(fdescp); 1431 error = E2BIG; 1432 goto out; 1433 } 1434 1435 fdp = data; 1436 rp = (struct file **) 1437 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1438 for (i = 0; i < oldfds; i++) { 1439 fp = fdescp->fd_ofiles[*fdp++]; 1440 *rp++ = fp; 1441 FILE_LOCK(fp); 1442 fp->f_count++; 1443 fp->f_msgcount++; 1444 FILE_UNLOCK(fp); 1445 unp_rights++; 1446 } 1447 FILEDESC_UNLOCK(fdescp); 1448 break; 1449 1450 case SCM_TIMESTAMP: 1451 *controlp = sbcreatecontrol(NULL, sizeof(*tv), 1452 SCM_TIMESTAMP, SOL_SOCKET); 1453 if (*controlp == NULL) { 1454 error = ENOBUFS; 1455 goto out; 1456 } 1457 tv = (struct timeval *) 1458 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1459 microtime(tv); 1460 break; 1461 1462 default: 1463 error = EINVAL; 1464 goto out; 1465 } 1466 1467 controlp = &(*controlp)->m_next; 1468 1469 if (CMSG_SPACE(datalen) < clen) { 1470 clen -= CMSG_SPACE(datalen); 1471 cm = (struct cmsghdr *) 1472 ((caddr_t)cm + CMSG_SPACE(datalen)); 1473 } else { 1474 clen = 0; 1475 cm = NULL; 1476 } 1477 } 1478 1479out: 1480 m_freem(control); 1481 1482 return (error); 1483} 1484 1485/* 1486 * unp_defer is thread-local during garbage collection, and does not require 1487 * explicit synchronization. unp_gcing prevents other threads from entering 1488 * garbage collection, and perhaps should be an sx lock instead. 1489 */ 1490static int unp_defer, unp_gcing; 1491 1492static void 1493unp_gc(void) 1494{ 1495 struct file *fp, *nextfp; 1496 struct socket *so; 1497 struct file **extra_ref, **fpp; 1498 int nunref, i; 1499 int nfiles_snap; 1500 int nfiles_slack = 20; 1501 1502 UNP_LOCK_ASSERT(); 1503 1504 if (unp_gcing) { 1505 UNP_UNLOCK(); 1506 return; 1507 } 1508 unp_gcing = 1; 1509 unp_defer = 0; 1510 UNP_UNLOCK(); 1511 /* 1512 * before going through all this, set all FDs to 1513 * be NOT defered and NOT externally accessible 1514 */ 1515 sx_slock(&filelist_lock); 1516 LIST_FOREACH(fp, &filehead, f_list) 1517 fp->f_gcflag &= ~(FMARK|FDEFER); 1518 do { 1519 LIST_FOREACH(fp, &filehead, f_list) { 1520 FILE_LOCK(fp); 1521 /* 1522 * If the file is not open, skip it 1523 */ 1524 if (fp->f_count == 0) { 1525 FILE_UNLOCK(fp); 1526 continue; 1527 } 1528 /* 1529 * If we already marked it as 'defer' in a 1530 * previous pass, then try process it this time 1531 * and un-mark it 1532 */ 1533 if (fp->f_gcflag & FDEFER) { 1534 fp->f_gcflag &= ~FDEFER; 1535 unp_defer--; 1536 } else { 1537 /* 1538 * if it's not defered, then check if it's 1539 * already marked.. if so skip it 1540 */ 1541 if (fp->f_gcflag & FMARK) { 1542 FILE_UNLOCK(fp); 1543 continue; 1544 } 1545 /* 1546 * If all references are from messages 1547 * in transit, then skip it. it's not 1548 * externally accessible. 1549 */ 1550 if (fp->f_count == fp->f_msgcount) { 1551 FILE_UNLOCK(fp); 1552 continue; 1553 } 1554 /* 1555 * If it got this far then it must be 1556 * externally accessible. 1557 */ 1558 fp->f_gcflag |= FMARK; 1559 } 1560 /* 1561 * either it was defered, or it is externally 1562 * accessible and not already marked so. 1563 * Now check if it is possibly one of OUR sockets. 1564 */ 1565 if (fp->f_type != DTYPE_SOCKET || 1566 (so = fp->f_data) == NULL) { 1567 FILE_UNLOCK(fp); 1568 continue; 1569 } 1570 FILE_UNLOCK(fp); 1571 if (so->so_proto->pr_domain != &localdomain || 1572 (so->so_proto->pr_flags&PR_RIGHTS) == 0) 1573 continue; 1574#ifdef notdef 1575 if (so->so_rcv.sb_flags & SB_LOCK) { 1576 /* 1577 * This is problematical; it's not clear 1578 * we need to wait for the sockbuf to be 1579 * unlocked (on a uniprocessor, at least), 1580 * and it's also not clear what to do 1581 * if sbwait returns an error due to receipt 1582 * of a signal. If sbwait does return 1583 * an error, we'll go into an infinite 1584 * loop. Delete all of this for now. 1585 */ 1586 (void) sbwait(&so->so_rcv); 1587 goto restart; 1588 } 1589#endif 1590 /* 1591 * So, Ok, it's one of our sockets and it IS externally 1592 * accessible (or was defered). Now we look 1593 * to see if we hold any file descriptors in its 1594 * message buffers. Follow those links and mark them 1595 * as accessible too. 1596 */ 1597 SOCKBUF_LOCK(&so->so_rcv); 1598 unp_scan(so->so_rcv.sb_mb, unp_mark); 1599 SOCKBUF_UNLOCK(&so->so_rcv); 1600 } 1601 } while (unp_defer); 1602 sx_sunlock(&filelist_lock); 1603 /* 1604 * We grab an extra reference to each of the file table entries 1605 * that are not otherwise accessible and then free the rights 1606 * that are stored in messages on them. 1607 * 1608 * The bug in the orginal code is a little tricky, so I'll describe 1609 * what's wrong with it here. 1610 * 1611 * It is incorrect to simply unp_discard each entry for f_msgcount 1612 * times -- consider the case of sockets A and B that contain 1613 * references to each other. On a last close of some other socket, 1614 * we trigger a gc since the number of outstanding rights (unp_rights) 1615 * is non-zero. If during the sweep phase the gc code un_discards, 1616 * we end up doing a (full) closef on the descriptor. A closef on A 1617 * results in the following chain. Closef calls soo_close, which 1618 * calls soclose. Soclose calls first (through the switch 1619 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply 1620 * returns because the previous instance had set unp_gcing, and 1621 * we return all the way back to soclose, which marks the socket 1622 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush 1623 * to free up the rights that are queued in messages on the socket A, 1624 * i.e., the reference on B. The sorflush calls via the dom_dispose 1625 * switch unp_dispose, which unp_scans with unp_discard. This second 1626 * instance of unp_discard just calls closef on B. 1627 * 1628 * Well, a similar chain occurs on B, resulting in a sorflush on B, 1629 * which results in another closef on A. Unfortunately, A is already 1630 * being closed, and the descriptor has already been marked with 1631 * SS_NOFDREF, and soclose panics at this point. 1632 * 1633 * Here, we first take an extra reference to each inaccessible 1634 * descriptor. Then, we call sorflush ourself, since we know 1635 * it is a Unix domain socket anyhow. After we destroy all the 1636 * rights carried in messages, we do a last closef to get rid 1637 * of our extra reference. This is the last close, and the 1638 * unp_detach etc will shut down the socket. 1639 * 1640 * 91/09/19, bsy@cs.cmu.edu 1641 */ 1642again: 1643 nfiles_snap = openfiles + nfiles_slack; /* some slack */ 1644 extra_ref = malloc(nfiles_snap * sizeof(struct file *), M_TEMP, 1645 M_WAITOK); 1646 sx_slock(&filelist_lock); 1647 if (nfiles_snap < openfiles) { 1648 sx_sunlock(&filelist_lock); 1649 free(extra_ref, M_TEMP); 1650 nfiles_slack += 20; 1651 goto again; 1652 } 1653 for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref; 1654 fp != NULL; fp = nextfp) { 1655 nextfp = LIST_NEXT(fp, f_list); 1656 FILE_LOCK(fp); 1657 /* 1658 * If it's not open, skip it 1659 */ 1660 if (fp->f_count == 0) { 1661 FILE_UNLOCK(fp); 1662 continue; 1663 } 1664 /* 1665 * If all refs are from msgs, and it's not marked accessible 1666 * then it must be referenced from some unreachable cycle 1667 * of (shut-down) FDs, so include it in our 1668 * list of FDs to remove 1669 */ 1670 if (fp->f_count == fp->f_msgcount && !(fp->f_gcflag & FMARK)) { 1671 *fpp++ = fp; 1672 nunref++; 1673 fp->f_count++; 1674 } 1675 FILE_UNLOCK(fp); 1676 } 1677 sx_sunlock(&filelist_lock); 1678 /* 1679 * for each FD on our hit list, do the following two things 1680 */ 1681 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1682 struct file *tfp = *fpp; 1683 FILE_LOCK(tfp); 1684 if (tfp->f_type == DTYPE_SOCKET && 1685 tfp->f_data != NULL) { 1686 FILE_UNLOCK(tfp); 1687 sorflush(tfp->f_data); 1688 } else { 1689 FILE_UNLOCK(tfp); 1690 } 1691 } 1692 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) 1693 closef(*fpp, (struct thread *) NULL); 1694 free(extra_ref, M_TEMP); 1695 unp_gcing = 0; 1696 1697 UNP_UNLOCK_ASSERT(); 1698} 1699 1700void 1701unp_dispose(struct mbuf *m) 1702{ 1703 1704 if (m) 1705 unp_scan(m, unp_discard); 1706} 1707 1708static int 1709unp_listen(struct unpcb *unp, struct thread *td) 1710{ 1711 UNP_LOCK_ASSERT(); 1712 1713 /* 1714 * XXXRW: Why populate the local peer cred with our own credential? 1715 */ 1716 cru2x(td->td_ucred, &unp->unp_peercred); 1717 unp->unp_flags |= UNP_HAVEPCCACHED; 1718 return (0); 1719} 1720 1721static void 1722unp_scan(struct mbuf *m0, void (*op)(struct file *)) 1723{ 1724 struct mbuf *m; 1725 struct file **rp; 1726 struct cmsghdr *cm; 1727 void *data; 1728 int i; 1729 socklen_t clen, datalen; 1730 int qfds; 1731 1732 while (m0 != NULL) { 1733 for (m = m0; m; m = m->m_next) { 1734 if (m->m_type != MT_CONTROL) 1735 continue; 1736 1737 cm = mtod(m, struct cmsghdr *); 1738 clen = m->m_len; 1739 1740 while (cm != NULL) { 1741 if (sizeof(*cm) > clen || cm->cmsg_len > clen) 1742 break; 1743 1744 data = CMSG_DATA(cm); 1745 datalen = (caddr_t)cm + cm->cmsg_len 1746 - (caddr_t)data; 1747 1748 if (cm->cmsg_level == SOL_SOCKET && 1749 cm->cmsg_type == SCM_RIGHTS) { 1750 qfds = datalen / sizeof (struct file *); 1751 rp = data; 1752 for (i = 0; i < qfds; i++) 1753 (*op)(*rp++); 1754 } 1755 1756 if (CMSG_SPACE(datalen) < clen) { 1757 clen -= CMSG_SPACE(datalen); 1758 cm = (struct cmsghdr *) 1759 ((caddr_t)cm + CMSG_SPACE(datalen)); 1760 } else { 1761 clen = 0; 1762 cm = NULL; 1763 } 1764 } 1765 } 1766 m0 = m0->m_act; 1767 } 1768} 1769 1770static void 1771unp_mark(struct file *fp) 1772{ 1773 if (fp->f_gcflag & FMARK) 1774 return; 1775 unp_defer++; 1776 fp->f_gcflag |= (FMARK|FDEFER); 1777} 1778 1779static void 1780unp_discard(struct file *fp) 1781{ 1782 FILE_LOCK(fp); 1783 fp->f_msgcount--; 1784 unp_rights--; 1785 FILE_UNLOCK(fp); 1786 (void) closef(fp, (struct thread *)NULL); 1787} 1788