uipc_usrreq.c revision 130820
1/* 2 * Copyright (c) 1982, 1986, 1989, 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94 30 */ 31 32#include <sys/cdefs.h> 33__FBSDID("$FreeBSD: head/sys/kern/uipc_usrreq.c 130820 2004-06-20 21:43:13Z rwatson $"); 34 35#include "opt_mac.h" 36 37#include <sys/param.h> 38#include <sys/domain.h> 39#include <sys/fcntl.h> 40#include <sys/malloc.h> /* XXX must be before <sys/file.h> */ 41#include <sys/file.h> 42#include <sys/filedesc.h> 43#include <sys/jail.h> 44#include <sys/kernel.h> 45#include <sys/lock.h> 46#include <sys/mac.h> 47#include <sys/mbuf.h> 48#include <sys/mutex.h> 49#include <sys/namei.h> 50#include <sys/proc.h> 51#include <sys/protosw.h> 52#include <sys/resourcevar.h> 53#include <sys/socket.h> 54#include <sys/socketvar.h> 55#include <sys/signalvar.h> 56#include <sys/stat.h> 57#include <sys/sx.h> 58#include <sys/sysctl.h> 59#include <sys/systm.h> 60#include <sys/un.h> 61#include <sys/unpcb.h> 62#include <sys/vnode.h> 63 64#include <vm/uma.h> 65 66static uma_zone_t unp_zone; 67static unp_gen_t unp_gencnt; 68static u_int unp_count; 69 70static struct unp_head unp_shead, unp_dhead; 71 72/* 73 * Unix communications domain. 74 * 75 * TODO: 76 * SEQPACKET, RDM 77 * rethink name space problems 78 * need a proper out-of-band 79 * lock pushdown 80 */ 81static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL }; 82static ino_t unp_ino; /* prototype for fake inode numbers */ 83 84static struct mtx unp_mtx; 85#define UNP_LOCK_INIT() \ 86 mtx_init(&unp_mtx, "unp", NULL, MTX_DEF) 87#define UNP_LOCK() mtx_lock(&unp_mtx) 88#define UNP_UNLOCK() mtx_unlock(&unp_mtx) 89#define UNP_LOCK_ASSERT() mtx_assert(&unp_mtx, MA_OWNED) 90 91static int unp_attach(struct socket *); 92static void unp_detach(struct unpcb *); 93static int unp_bind(struct unpcb *,struct sockaddr *, struct thread *); 94static int unp_connect(struct socket *,struct sockaddr *, struct thread *); 95static int unp_connect2(struct socket *so, struct socket *so2); 96static void unp_disconnect(struct unpcb *); 97static void unp_shutdown(struct unpcb *); 98static void unp_drop(struct unpcb *, int); 99static void unp_gc(void); 100static void unp_scan(struct mbuf *, void (*)(struct file *)); 101static void unp_mark(struct file *); 102static void unp_discard(struct file *); 103static void unp_freerights(struct file **, int); 104static int unp_internalize(struct mbuf **, struct thread *); 105static int unp_listen(struct unpcb *, struct thread *); 106 107static int 108uipc_abort(struct socket *so) 109{ 110 struct unpcb *unp = sotounpcb(so); 111 112 if (unp == NULL) 113 return (EINVAL); 114 UNP_LOCK(); 115 unp_drop(unp, ECONNABORTED); 116 unp_detach(unp); /* NB: unlocks */ 117 SOCK_LOCK(so); 118 sotryfree(so); 119 return (0); 120} 121 122static int 123uipc_accept(struct socket *so, struct sockaddr **nam) 124{ 125 struct unpcb *unp = sotounpcb(so); 126 const struct sockaddr *sa; 127 128 if (unp == NULL) 129 return (EINVAL); 130 131 /* 132 * Pass back name of connected socket, 133 * if it was bound and we are still connected 134 * (our peer may have closed already!). 135 */ 136 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 137 UNP_LOCK(); 138 if (unp->unp_conn != NULL && unp->unp_conn->unp_addr != NULL) 139 sa = (struct sockaddr *) unp->unp_conn->unp_addr; 140 else 141 sa = &sun_noname; 142 bcopy(sa, *nam, sa->sa_len); 143 UNP_UNLOCK(); 144 return (0); 145} 146 147static int 148uipc_attach(struct socket *so, int proto, struct thread *td) 149{ 150 struct unpcb *unp = sotounpcb(so); 151 152 if (unp != NULL) 153 return (EISCONN); 154 return (unp_attach(so)); 155} 156 157static int 158uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 159{ 160 struct unpcb *unp = sotounpcb(so); 161 162 if (unp == NULL) 163 return (EINVAL); 164 165 return (unp_bind(unp, nam, td)); 166} 167 168static int 169uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 170{ 171 struct unpcb *unp = sotounpcb(so); 172 int error; 173 174 if (unp == NULL) 175 return (EINVAL); 176 UNP_LOCK(); 177 error = unp_connect(so, nam, curthread); 178 UNP_UNLOCK(); 179 return (error); 180} 181 182int 183uipc_connect2(struct socket *so1, struct socket *so2) 184{ 185 struct unpcb *unp = sotounpcb(so1); 186 int error; 187 188 if (unp == NULL) 189 return (EINVAL); 190 191 UNP_LOCK(); 192 error = unp_connect2(so1, so2); 193 UNP_UNLOCK(); 194 return (error); 195} 196 197/* control is EOPNOTSUPP */ 198 199static int 200uipc_detach(struct socket *so) 201{ 202 struct unpcb *unp = sotounpcb(so); 203 204 if (unp == NULL) 205 return (EINVAL); 206 207 UNP_LOCK(); 208 unp_detach(unp); /* NB: unlocks unp */ 209 return (0); 210} 211 212static int 213uipc_disconnect(struct socket *so) 214{ 215 struct unpcb *unp = sotounpcb(so); 216 217 if (unp == NULL) 218 return (EINVAL); 219 UNP_LOCK(); 220 unp_disconnect(unp); 221 UNP_UNLOCK(); 222 return (0); 223} 224 225static int 226uipc_listen(struct socket *so, struct thread *td) 227{ 228 struct unpcb *unp = sotounpcb(so); 229 int error; 230 231 if (unp == NULL || unp->unp_vnode == NULL) 232 return (EINVAL); 233 UNP_LOCK(); 234 error = unp_listen(unp, td); 235 UNP_UNLOCK(); 236 return (error); 237} 238 239static int 240uipc_peeraddr(struct socket *so, struct sockaddr **nam) 241{ 242 struct unpcb *unp = sotounpcb(so); 243 const struct sockaddr *sa; 244 245 if (unp == NULL) 246 return (EINVAL); 247 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 248 UNP_LOCK(); 249 if (unp->unp_conn != NULL && unp->unp_conn->unp_addr!= NULL) 250 sa = (struct sockaddr *) unp->unp_conn->unp_addr; 251 else { 252 /* 253 * XXX: It seems that this test always fails even when 254 * connection is established. So, this else clause is 255 * added as workaround to return PF_LOCAL sockaddr. 256 */ 257 sa = &sun_noname; 258 } 259 bcopy(sa, *nam, sa->sa_len); 260 UNP_UNLOCK(); 261 return (0); 262} 263 264static int 265uipc_rcvd(struct socket *so, int flags) 266{ 267 struct unpcb *unp = sotounpcb(so); 268 struct socket *so2; 269 u_long newhiwat; 270 271 if (unp == NULL) 272 return (EINVAL); 273 UNP_LOCK(); 274 switch (so->so_type) { 275 case SOCK_DGRAM: 276 panic("uipc_rcvd DGRAM?"); 277 /*NOTREACHED*/ 278 279 case SOCK_STREAM: 280 if (unp->unp_conn == NULL) 281 break; 282 so2 = unp->unp_conn->unp_socket; 283 SOCKBUF_LOCK(&so2->so_snd); 284 SOCKBUF_LOCK(&so->so_rcv); 285 /* 286 * Adjust backpressure on sender 287 * and wakeup any waiting to write. 288 */ 289 so2->so_snd.sb_mbmax += unp->unp_mbcnt - so->so_rcv.sb_mbcnt; 290 unp->unp_mbcnt = so->so_rcv.sb_mbcnt; 291 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - 292 so->so_rcv.sb_cc; 293 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat, 294 newhiwat, RLIM_INFINITY); 295 unp->unp_cc = so->so_rcv.sb_cc; 296 SOCKBUF_UNLOCK(&so->so_rcv); 297 SOCKBUF_UNLOCK(&so2->so_snd); 298 sowwakeup(so2); 299 break; 300 301 default: 302 panic("uipc_rcvd unknown socktype"); 303 } 304 UNP_UNLOCK(); 305 return (0); 306} 307 308/* pru_rcvoob is EOPNOTSUPP */ 309 310static int 311uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 312 struct mbuf *control, struct thread *td) 313{ 314 int error = 0; 315 struct unpcb *unp = sotounpcb(so); 316 struct socket *so2; 317 u_long newhiwat; 318 319 if (unp == NULL) { 320 error = EINVAL; 321 goto release; 322 } 323 if (flags & PRUS_OOB) { 324 error = EOPNOTSUPP; 325 goto release; 326 } 327 328 if (control != NULL && (error = unp_internalize(&control, td))) 329 goto release; 330 331 UNP_LOCK(); 332 switch (so->so_type) { 333 case SOCK_DGRAM: 334 { 335 const struct sockaddr *from; 336 337 if (nam != NULL) { 338 if (unp->unp_conn != NULL) { 339 error = EISCONN; 340 break; 341 } 342 error = unp_connect(so, nam, td); 343 if (error) 344 break; 345 } else { 346 if (unp->unp_conn == NULL) { 347 error = ENOTCONN; 348 break; 349 } 350 } 351 so2 = unp->unp_conn->unp_socket; 352 if (unp->unp_addr != NULL) 353 from = (struct sockaddr *)unp->unp_addr; 354 else 355 from = &sun_noname; 356 if (sbappendaddr(&so2->so_rcv, from, m, control)) { 357 sorwakeup(so2); 358 m = NULL; 359 control = NULL; 360 } else { 361 error = ENOBUFS; 362 } 363 if (nam != NULL) 364 unp_disconnect(unp); 365 break; 366 } 367 368 case SOCK_STREAM: 369 /* Connect if not connected yet. */ 370 /* 371 * Note: A better implementation would complain 372 * if not equal to the peer's address. 373 */ 374 if ((so->so_state & SS_ISCONNECTED) == 0) { 375 if (nam != NULL) { 376 error = unp_connect(so, nam, td); 377 if (error) 378 break; /* XXX */ 379 } else { 380 error = ENOTCONN; 381 break; 382 } 383 } 384 385 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 386 error = EPIPE; 387 break; 388 } 389 if (unp->unp_conn == NULL) 390 panic("uipc_send connected but no connection?"); 391 so2 = unp->unp_conn->unp_socket; 392 /* 393 * Send to paired receive port, and then reduce 394 * send buffer hiwater marks to maintain backpressure. 395 * Wake up readers. 396 */ 397 if (control != NULL) { 398 if (sbappendcontrol(&so2->so_rcv, m, control)) 399 control = NULL; 400 } else { 401 sbappend(&so2->so_rcv, m); 402 } 403 so->so_snd.sb_mbmax -= 404 so2->so_rcv.sb_mbcnt - unp->unp_conn->unp_mbcnt; 405 unp->unp_conn->unp_mbcnt = so2->so_rcv.sb_mbcnt; 406 newhiwat = so->so_snd.sb_hiwat - 407 (so2->so_rcv.sb_cc - unp->unp_conn->unp_cc); 408 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat, 409 newhiwat, RLIM_INFINITY); 410 unp->unp_conn->unp_cc = so2->so_rcv.sb_cc; 411 sorwakeup(so2); 412 m = NULL; 413 break; 414 415 default: 416 panic("uipc_send unknown socktype"); 417 } 418 419 /* 420 * SEND_EOF is equivalent to a SEND followed by 421 * a SHUTDOWN. 422 */ 423 if (flags & PRUS_EOF) { 424 socantsendmore(so); 425 unp_shutdown(unp); 426 } 427 UNP_UNLOCK(); 428 429 if (control != NULL && error != 0) 430 unp_dispose(control); 431 432release: 433 if (control != NULL) 434 m_freem(control); 435 if (m != NULL) 436 m_freem(m); 437 return (error); 438} 439 440static int 441uipc_sense(struct socket *so, struct stat *sb) 442{ 443 struct unpcb *unp = sotounpcb(so); 444 struct socket *so2; 445 446 if (unp == NULL) 447 return (EINVAL); 448 UNP_LOCK(); 449 sb->st_blksize = so->so_snd.sb_hiwat; 450 if (so->so_type == SOCK_STREAM && unp->unp_conn != NULL) { 451 so2 = unp->unp_conn->unp_socket; 452 sb->st_blksize += so2->so_rcv.sb_cc; 453 } 454 sb->st_dev = NODEV; 455 if (unp->unp_ino == 0) 456 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino; 457 sb->st_ino = unp->unp_ino; 458 UNP_UNLOCK(); 459 return (0); 460} 461 462static int 463uipc_shutdown(struct socket *so) 464{ 465 struct unpcb *unp = sotounpcb(so); 466 467 if (unp == NULL) 468 return (EINVAL); 469 UNP_LOCK(); 470 socantsendmore(so); 471 unp_shutdown(unp); 472 UNP_UNLOCK(); 473 return (0); 474} 475 476static int 477uipc_sockaddr(struct socket *so, struct sockaddr **nam) 478{ 479 struct unpcb *unp = sotounpcb(so); 480 const struct sockaddr *sa; 481 482 if (unp == NULL) 483 return (EINVAL); 484 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 485 UNP_LOCK(); 486 if (unp->unp_addr != NULL) 487 sa = (struct sockaddr *) unp->unp_addr; 488 else 489 sa = &sun_noname; 490 bcopy(sa, *nam, sa->sa_len); 491 UNP_UNLOCK(); 492 return (0); 493} 494 495struct pr_usrreqs uipc_usrreqs = { 496 uipc_abort, uipc_accept, uipc_attach, uipc_bind, uipc_connect, 497 uipc_connect2, pru_control_notsupp, uipc_detach, uipc_disconnect, 498 uipc_listen, uipc_peeraddr, uipc_rcvd, pru_rcvoob_notsupp, 499 uipc_send, uipc_sense, uipc_shutdown, uipc_sockaddr, 500 sosend, soreceive, sopoll, pru_sosetlabel_null 501}; 502 503int 504uipc_ctloutput(so, sopt) 505 struct socket *so; 506 struct sockopt *sopt; 507{ 508 struct unpcb *unp = sotounpcb(so); 509 struct xucred xu; 510 int error; 511 512 switch (sopt->sopt_dir) { 513 case SOPT_GET: 514 switch (sopt->sopt_name) { 515 case LOCAL_PEERCRED: 516 error = 0; 517 UNP_LOCK(); 518 if (unp->unp_flags & UNP_HAVEPC) 519 xu = unp->unp_peercred; 520 else { 521 if (so->so_type == SOCK_STREAM) 522 error = ENOTCONN; 523 else 524 error = EINVAL; 525 } 526 UNP_UNLOCK(); 527 if (error == 0) 528 error = sooptcopyout(sopt, &xu, sizeof(xu)); 529 break; 530 default: 531 error = EOPNOTSUPP; 532 break; 533 } 534 break; 535 case SOPT_SET: 536 default: 537 error = EOPNOTSUPP; 538 break; 539 } 540 return (error); 541} 542 543/* 544 * Both send and receive buffers are allocated PIPSIZ bytes of buffering 545 * for stream sockets, although the total for sender and receiver is 546 * actually only PIPSIZ. 547 * Datagram sockets really use the sendspace as the maximum datagram size, 548 * and don't really want to reserve the sendspace. Their recvspace should 549 * be large enough for at least one max-size datagram plus address. 550 */ 551#ifndef PIPSIZ 552#define PIPSIZ 8192 553#endif 554static u_long unpst_sendspace = PIPSIZ; 555static u_long unpst_recvspace = PIPSIZ; 556static u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 557static u_long unpdg_recvspace = 4*1024; 558 559static int unp_rights; /* file descriptors in flight */ 560 561SYSCTL_DECL(_net_local_stream); 562SYSCTL_INT(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 563 &unpst_sendspace, 0, ""); 564SYSCTL_INT(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 565 &unpst_recvspace, 0, ""); 566SYSCTL_DECL(_net_local_dgram); 567SYSCTL_INT(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 568 &unpdg_sendspace, 0, ""); 569SYSCTL_INT(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 570 &unpdg_recvspace, 0, ""); 571SYSCTL_DECL(_net_local); 572SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, ""); 573 574static int 575unp_attach(so) 576 struct socket *so; 577{ 578 register struct unpcb *unp; 579 int error; 580 581 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 582 switch (so->so_type) { 583 584 case SOCK_STREAM: 585 error = soreserve(so, unpst_sendspace, unpst_recvspace); 586 break; 587 588 case SOCK_DGRAM: 589 error = soreserve(so, unpdg_sendspace, unpdg_recvspace); 590 break; 591 592 default: 593 panic("unp_attach"); 594 } 595 if (error) 596 return (error); 597 } 598 unp = uma_zalloc(unp_zone, M_WAITOK); 599 if (unp == NULL) 600 return (ENOBUFS); 601 bzero(unp, sizeof *unp); 602 LIST_INIT(&unp->unp_refs); 603 unp->unp_socket = so; 604 605 UNP_LOCK(); 606 unp->unp_gencnt = ++unp_gencnt; 607 unp_count++; 608 LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead 609 : &unp_shead, unp, unp_link); 610 UNP_UNLOCK(); 611 612 so->so_pcb = unp; 613 return (0); 614} 615 616static void 617unp_detach(unp) 618 register struct unpcb *unp; 619{ 620 struct vnode *vp; 621 622 UNP_LOCK_ASSERT(); 623 624 LIST_REMOVE(unp, unp_link); 625 unp->unp_gencnt = ++unp_gencnt; 626 --unp_count; 627 if ((vp = unp->unp_vnode) != NULL) { 628 /* 629 * XXXRW: should v_socket be frobbed only while holding 630 * Giant? 631 */ 632 unp->unp_vnode->v_socket = NULL; 633 unp->unp_vnode = NULL; 634 } 635 if (unp->unp_conn != NULL) 636 unp_disconnect(unp); 637 while (!LIST_EMPTY(&unp->unp_refs)) { 638 struct unpcb *ref = LIST_FIRST(&unp->unp_refs); 639 unp_drop(ref, ECONNRESET); 640 } 641 soisdisconnected(unp->unp_socket); 642 unp->unp_socket->so_pcb = NULL; 643 if (unp_rights) { 644 /* 645 * Normally the receive buffer is flushed later, 646 * in sofree, but if our receive buffer holds references 647 * to descriptors that are now garbage, we will dispose 648 * of those descriptor references after the garbage collector 649 * gets them (resulting in a "panic: closef: count < 0"). 650 */ 651 sorflush(unp->unp_socket); 652 unp_gc(); 653 } 654 if (unp->unp_addr != NULL) 655 FREE(unp->unp_addr, M_SONAME); 656 UNP_UNLOCK(); 657 uma_zfree(unp_zone, unp); 658 if (vp) { 659 mtx_lock(&Giant); 660 vrele(vp); 661 mtx_unlock(&Giant); 662 } 663} 664 665static int 666unp_bind(unp, nam, td) 667 struct unpcb *unp; 668 struct sockaddr *nam; 669 struct thread *td; 670{ 671 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 672 struct vnode *vp; 673 struct mount *mp; 674 struct vattr vattr; 675 int error, namelen; 676 struct nameidata nd; 677 char *buf; 678 679 /* 680 * XXXRW: This test-and-set of unp_vnode is non-atomic; the 681 * unlocked read here is fine, but the value of unp_vnode needs 682 * to be tested again after we do all the lookups to see if the 683 * pcb is still unbound? 684 */ 685 if (unp->unp_vnode != NULL) 686 return (EINVAL); 687 688 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 689 if (namelen <= 0) 690 return (EINVAL); 691 692 buf = malloc(namelen + 1, M_TEMP, M_WAITOK); 693 strlcpy(buf, soun->sun_path, namelen + 1); 694 695 mtx_lock(&Giant); 696restart: 697 mtx_assert(&Giant, MA_OWNED); 698 NDINIT(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME, UIO_SYSSPACE, 699 buf, td); 700/* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 701 error = namei(&nd); 702 if (error) 703 goto done; 704 vp = nd.ni_vp; 705 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { 706 NDFREE(&nd, NDF_ONLY_PNBUF); 707 if (nd.ni_dvp == vp) 708 vrele(nd.ni_dvp); 709 else 710 vput(nd.ni_dvp); 711 if (vp != NULL) { 712 vrele(vp); 713 error = EADDRINUSE; 714 goto done; 715 } 716 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH); 717 if (error) 718 goto done; 719 goto restart; 720 } 721 VATTR_NULL(&vattr); 722 vattr.va_type = VSOCK; 723 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask); 724#ifdef MAC 725 error = mac_check_vnode_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, 726 &vattr); 727#endif 728 if (error == 0) { 729 VOP_LEASE(nd.ni_dvp, td, td->td_ucred, LEASE_WRITE); 730 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 731 } 732 NDFREE(&nd, NDF_ONLY_PNBUF); 733 vput(nd.ni_dvp); 734 if (error) 735 goto done; 736 vp = nd.ni_vp; 737 ASSERT_VOP_LOCKED(vp, "unp_bind"); 738 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK); 739 UNP_LOCK(); 740 vp->v_socket = unp->unp_socket; 741 unp->unp_vnode = vp; 742 unp->unp_addr = soun; 743 UNP_UNLOCK(); 744 VOP_UNLOCK(vp, 0, td); 745 vn_finished_write(mp); 746done: 747 mtx_unlock(&Giant); 748 free(buf, M_TEMP); 749 return (error); 750} 751 752static int 753unp_connect(so, nam, td) 754 struct socket *so; 755 struct sockaddr *nam; 756 struct thread *td; 757{ 758 register struct sockaddr_un *soun = (struct sockaddr_un *)nam; 759 register struct vnode *vp; 760 register struct socket *so2, *so3; 761 struct unpcb *unp = sotounpcb(so); 762 struct unpcb *unp2, *unp3; 763 int error, len; 764 struct nameidata nd; 765 char buf[SOCK_MAXADDRLEN]; 766 struct sockaddr *sa; 767 768 UNP_LOCK_ASSERT(); 769 770 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 771 if (len <= 0) 772 return (EINVAL); 773 strlcpy(buf, soun->sun_path, len + 1); 774 UNP_UNLOCK(); 775 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 776 mtx_lock(&Giant); 777 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, td); 778 error = namei(&nd); 779 if (error) 780 vp = NULL; 781 else 782 vp = nd.ni_vp; 783 ASSERT_VOP_LOCKED(vp, "unp_connect"); 784 NDFREE(&nd, NDF_ONLY_PNBUF); 785 if (error) 786 goto bad; 787 788 if (vp->v_type != VSOCK) { 789 error = ENOTSOCK; 790 goto bad; 791 } 792 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); 793 if (error) 794 goto bad; 795 so2 = vp->v_socket; 796 if (so2 == NULL) { 797 error = ECONNREFUSED; 798 goto bad; 799 } 800 if (so->so_type != so2->so_type) { 801 error = EPROTOTYPE; 802 goto bad; 803 } 804 mtx_unlock(&Giant); 805 UNP_LOCK(); 806 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 807 if (so2->so_options & SO_ACCEPTCONN) { 808 /* 809 * NB: drop locks here so unp_attach is entered 810 * w/o locks; this avoids a recursive lock 811 * of the head and holding sleep locks across 812 * a (potentially) blocking malloc. 813 */ 814 UNP_UNLOCK(); 815 so3 = sonewconn(so2, 0); 816 UNP_LOCK(); 817 } else 818 so3 = NULL; 819 if (so3 == NULL) { 820 error = ECONNREFUSED; 821 goto bad2; 822 } 823 unp = sotounpcb(so); 824 unp2 = sotounpcb(so2); 825 unp3 = sotounpcb(so3); 826 if (unp2->unp_addr != NULL) { 827 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); 828 unp3->unp_addr = (struct sockaddr_un *) sa; 829 sa = NULL; 830 } 831 /* 832 * unp_peercred management: 833 * 834 * The connecter's (client's) credentials are copied 835 * from its process structure at the time of connect() 836 * (which is now). 837 */ 838 cru2x(td->td_ucred, &unp3->unp_peercred); 839 unp3->unp_flags |= UNP_HAVEPC; 840 /* 841 * The receiver's (server's) credentials are copied 842 * from the unp_peercred member of socket on which the 843 * former called listen(); unp_listen() cached that 844 * process's credentials at that time so we can use 845 * them now. 846 */ 847 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED, 848 ("unp_connect: listener without cached peercred")); 849 memcpy(&unp->unp_peercred, &unp2->unp_peercred, 850 sizeof(unp->unp_peercred)); 851 unp->unp_flags |= UNP_HAVEPC; 852#ifdef MAC 853 SOCK_LOCK(so); 854 mac_set_socket_peer_from_socket(so, so3); 855 mac_set_socket_peer_from_socket(so3, so); 856 SOCK_UNLOCK(so); 857#endif 858 859 so2 = so3; 860 } 861 error = unp_connect2(so, so2); 862bad2: 863 UNP_UNLOCK(); 864 mtx_lock(&Giant); 865bad: 866 mtx_assert(&Giant, MA_OWNED); 867 if (vp != NULL) 868 vput(vp); 869 mtx_unlock(&Giant); 870 free(sa, M_SONAME); 871 UNP_LOCK(); 872 return (error); 873} 874 875static int 876unp_connect2(so, so2) 877 register struct socket *so; 878 register struct socket *so2; 879{ 880 register struct unpcb *unp = sotounpcb(so); 881 register struct unpcb *unp2; 882 883 UNP_LOCK_ASSERT(); 884 885 if (so2->so_type != so->so_type) 886 return (EPROTOTYPE); 887 unp2 = sotounpcb(so2); 888 unp->unp_conn = unp2; 889 switch (so->so_type) { 890 891 case SOCK_DGRAM: 892 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 893 soisconnected(so); 894 break; 895 896 case SOCK_STREAM: 897 unp2->unp_conn = unp; 898 soisconnected(so); 899 soisconnected(so2); 900 break; 901 902 default: 903 panic("unp_connect2"); 904 } 905 return (0); 906} 907 908static void 909unp_disconnect(unp) 910 struct unpcb *unp; 911{ 912 register struct unpcb *unp2 = unp->unp_conn; 913 struct socket *so; 914 915 UNP_LOCK_ASSERT(); 916 917 if (unp2 == NULL) 918 return; 919 unp->unp_conn = NULL; 920 switch (unp->unp_socket->so_type) { 921 922 case SOCK_DGRAM: 923 LIST_REMOVE(unp, unp_reflink); 924 so = unp->unp_socket; 925 SOCK_LOCK(so); 926 so->so_state &= ~SS_ISCONNECTED; 927 SOCK_UNLOCK(so); 928 break; 929 930 case SOCK_STREAM: 931 soisdisconnected(unp->unp_socket); 932 unp2->unp_conn = NULL; 933 soisdisconnected(unp2->unp_socket); 934 break; 935 } 936} 937 938#ifdef notdef 939void 940unp_abort(unp) 941 struct unpcb *unp; 942{ 943 944 unp_detach(unp); 945} 946#endif 947 948/* 949 * unp_pcblist() assumes that UNIX domain socket memory is never reclaimed 950 * by the zone (UMA_ZONE_NOFREE), and as such potentially stale pointers 951 * are safe to reference. It first scans the list of struct unpcb's to 952 * generate a pointer list, then it rescans its list one entry at a time to 953 * externalize and copyout. It checks the generation number to see if a 954 * struct unpcb has been reused, and will skip it if so. 955 */ 956static int 957unp_pcblist(SYSCTL_HANDLER_ARGS) 958{ 959 int error, i, n; 960 struct unpcb *unp, **unp_list; 961 unp_gen_t gencnt; 962 struct xunpgen *xug; 963 struct unp_head *head; 964 struct xunpcb *xu; 965 966 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead); 967 968 /* 969 * The process of preparing the PCB list is too time-consuming and 970 * resource-intensive to repeat twice on every request. 971 */ 972 if (req->oldptr == NULL) { 973 n = unp_count; 974 req->oldidx = 2 * (sizeof *xug) 975 + (n + n/8) * sizeof(struct xunpcb); 976 return (0); 977 } 978 979 if (req->newptr != NULL) 980 return (EPERM); 981 982 /* 983 * OK, now we're committed to doing something. 984 */ 985 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK); 986 UNP_LOCK(); 987 gencnt = unp_gencnt; 988 n = unp_count; 989 UNP_UNLOCK(); 990 991 xug->xug_len = sizeof *xug; 992 xug->xug_count = n; 993 xug->xug_gen = gencnt; 994 xug->xug_sogen = so_gencnt; 995 error = SYSCTL_OUT(req, xug, sizeof *xug); 996 if (error) { 997 free(xug, M_TEMP); 998 return (error); 999 } 1000 1001 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 1002 1003 UNP_LOCK(); 1004 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 1005 unp = LIST_NEXT(unp, unp_link)) { 1006 if (unp->unp_gencnt <= gencnt) { 1007 if (cr_cansee(req->td->td_ucred, 1008 unp->unp_socket->so_cred)) 1009 continue; 1010 unp_list[i++] = unp; 1011 } 1012 } 1013 UNP_UNLOCK(); 1014 n = i; /* in case we lost some during malloc */ 1015 1016 error = 0; 1017 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK); 1018 for (i = 0; i < n; i++) { 1019 unp = unp_list[i]; 1020 if (unp->unp_gencnt <= gencnt) { 1021 xu->xu_len = sizeof *xu; 1022 xu->xu_unpp = unp; 1023 /* 1024 * XXX - need more locking here to protect against 1025 * connect/disconnect races for SMP. 1026 */ 1027 if (unp->unp_addr != NULL) 1028 bcopy(unp->unp_addr, &xu->xu_addr, 1029 unp->unp_addr->sun_len); 1030 if (unp->unp_conn != NULL && 1031 unp->unp_conn->unp_addr != NULL) 1032 bcopy(unp->unp_conn->unp_addr, 1033 &xu->xu_caddr, 1034 unp->unp_conn->unp_addr->sun_len); 1035 bcopy(unp, &xu->xu_unp, sizeof *unp); 1036 sotoxsocket(unp->unp_socket, &xu->xu_socket); 1037 error = SYSCTL_OUT(req, xu, sizeof *xu); 1038 } 1039 } 1040 free(xu, M_TEMP); 1041 if (!error) { 1042 /* 1043 * Give the user an updated idea of our state. 1044 * If the generation differs from what we told 1045 * her before, she knows that something happened 1046 * while we were processing this request, and it 1047 * might be necessary to retry. 1048 */ 1049 xug->xug_gen = unp_gencnt; 1050 xug->xug_sogen = so_gencnt; 1051 xug->xug_count = unp_count; 1052 error = SYSCTL_OUT(req, xug, sizeof *xug); 1053 } 1054 free(unp_list, M_TEMP); 1055 free(xug, M_TEMP); 1056 return (error); 1057} 1058 1059SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD, 1060 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 1061 "List of active local datagram sockets"); 1062SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD, 1063 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 1064 "List of active local stream sockets"); 1065 1066static void 1067unp_shutdown(unp) 1068 struct unpcb *unp; 1069{ 1070 struct socket *so; 1071 1072 UNP_LOCK_ASSERT(); 1073 1074 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn && 1075 (so = unp->unp_conn->unp_socket)) 1076 socantrcvmore(so); 1077} 1078 1079static void 1080unp_drop(unp, errno) 1081 struct unpcb *unp; 1082 int errno; 1083{ 1084 struct socket *so = unp->unp_socket; 1085 1086 UNP_LOCK_ASSERT(); 1087 1088 so->so_error = errno; 1089 unp_disconnect(unp); 1090} 1091 1092#ifdef notdef 1093void 1094unp_drain() 1095{ 1096 1097} 1098#endif 1099 1100static void 1101unp_freerights(rp, fdcount) 1102 struct file **rp; 1103 int fdcount; 1104{ 1105 int i; 1106 struct file *fp; 1107 1108 for (i = 0; i < fdcount; i++) { 1109 fp = *rp; 1110 /* 1111 * zero the pointer before calling 1112 * unp_discard since it may end up 1113 * in unp_gc().. 1114 */ 1115 *rp++ = 0; 1116 unp_discard(fp); 1117 } 1118} 1119 1120int 1121unp_externalize(control, controlp) 1122 struct mbuf *control, **controlp; 1123{ 1124 struct thread *td = curthread; /* XXX */ 1125 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1126 int i; 1127 int *fdp; 1128 struct file **rp; 1129 struct file *fp; 1130 void *data; 1131 socklen_t clen = control->m_len, datalen; 1132 int error, newfds; 1133 int f; 1134 u_int newlen; 1135 1136 error = 0; 1137 if (controlp != NULL) /* controlp == NULL => free control messages */ 1138 *controlp = NULL; 1139 1140 while (cm != NULL) { 1141 if (sizeof(*cm) > clen || cm->cmsg_len > clen) { 1142 error = EINVAL; 1143 break; 1144 } 1145 1146 data = CMSG_DATA(cm); 1147 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1148 1149 if (cm->cmsg_level == SOL_SOCKET 1150 && cm->cmsg_type == SCM_RIGHTS) { 1151 newfds = datalen / sizeof(struct file *); 1152 rp = data; 1153 1154 /* If we're not outputting the descriptors free them. */ 1155 if (error || controlp == NULL) { 1156 unp_freerights(rp, newfds); 1157 goto next; 1158 } 1159 FILEDESC_LOCK(td->td_proc->p_fd); 1160 /* if the new FD's will not fit free them. */ 1161 if (!fdavail(td, newfds)) { 1162 FILEDESC_UNLOCK(td->td_proc->p_fd); 1163 error = EMSGSIZE; 1164 unp_freerights(rp, newfds); 1165 goto next; 1166 } 1167 /* 1168 * now change each pointer to an fd in the global 1169 * table to an integer that is the index to the 1170 * local fd table entry that we set up to point 1171 * to the global one we are transferring. 1172 */ 1173 newlen = newfds * sizeof(int); 1174 *controlp = sbcreatecontrol(NULL, newlen, 1175 SCM_RIGHTS, SOL_SOCKET); 1176 if (*controlp == NULL) { 1177 FILEDESC_UNLOCK(td->td_proc->p_fd); 1178 error = E2BIG; 1179 unp_freerights(rp, newfds); 1180 goto next; 1181 } 1182 1183 fdp = (int *) 1184 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1185 for (i = 0; i < newfds; i++) { 1186 if (fdalloc(td, 0, &f)) 1187 panic("unp_externalize fdalloc failed"); 1188 fp = *rp++; 1189 td->td_proc->p_fd->fd_ofiles[f] = fp; 1190 FILE_LOCK(fp); 1191 fp->f_msgcount--; 1192 FILE_UNLOCK(fp); 1193 unp_rights--; 1194 *fdp++ = f; 1195 } 1196 FILEDESC_UNLOCK(td->td_proc->p_fd); 1197 } else { /* We can just copy anything else across */ 1198 if (error || controlp == NULL) 1199 goto next; 1200 *controlp = sbcreatecontrol(NULL, datalen, 1201 cm->cmsg_type, cm->cmsg_level); 1202 if (*controlp == NULL) { 1203 error = ENOBUFS; 1204 goto next; 1205 } 1206 bcopy(data, 1207 CMSG_DATA(mtod(*controlp, struct cmsghdr *)), 1208 datalen); 1209 } 1210 1211 controlp = &(*controlp)->m_next; 1212 1213next: 1214 if (CMSG_SPACE(datalen) < clen) { 1215 clen -= CMSG_SPACE(datalen); 1216 cm = (struct cmsghdr *) 1217 ((caddr_t)cm + CMSG_SPACE(datalen)); 1218 } else { 1219 clen = 0; 1220 cm = NULL; 1221 } 1222 } 1223 1224 m_freem(control); 1225 1226 return (error); 1227} 1228 1229void 1230unp_init(void) 1231{ 1232 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL, 1233 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 1234 if (unp_zone == NULL) 1235 panic("unp_init"); 1236 uma_zone_set_max(unp_zone, nmbclusters); 1237 LIST_INIT(&unp_dhead); 1238 LIST_INIT(&unp_shead); 1239 1240 UNP_LOCK_INIT(); 1241} 1242 1243static int 1244unp_internalize(controlp, td) 1245 struct mbuf **controlp; 1246 struct thread *td; 1247{ 1248 struct mbuf *control = *controlp; 1249 struct proc *p = td->td_proc; 1250 struct filedesc *fdescp = p->p_fd; 1251 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1252 struct cmsgcred *cmcred; 1253 struct file **rp; 1254 struct file *fp; 1255 struct timeval *tv; 1256 int i, fd, *fdp; 1257 void *data; 1258 socklen_t clen = control->m_len, datalen; 1259 int error, oldfds; 1260 u_int newlen; 1261 1262 error = 0; 1263 *controlp = NULL; 1264 1265 while (cm != NULL) { 1266 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET 1267 || cm->cmsg_len > clen) { 1268 error = EINVAL; 1269 goto out; 1270 } 1271 1272 data = CMSG_DATA(cm); 1273 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1274 1275 switch (cm->cmsg_type) { 1276 /* 1277 * Fill in credential information. 1278 */ 1279 case SCM_CREDS: 1280 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred), 1281 SCM_CREDS, SOL_SOCKET); 1282 if (*controlp == NULL) { 1283 error = ENOBUFS; 1284 goto out; 1285 } 1286 1287 cmcred = (struct cmsgcred *) 1288 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1289 cmcred->cmcred_pid = p->p_pid; 1290 cmcred->cmcred_uid = td->td_ucred->cr_ruid; 1291 cmcred->cmcred_gid = td->td_ucred->cr_rgid; 1292 cmcred->cmcred_euid = td->td_ucred->cr_uid; 1293 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups, 1294 CMGROUP_MAX); 1295 for (i = 0; i < cmcred->cmcred_ngroups; i++) 1296 cmcred->cmcred_groups[i] = 1297 td->td_ucred->cr_groups[i]; 1298 break; 1299 1300 case SCM_RIGHTS: 1301 oldfds = datalen / sizeof (int); 1302 /* 1303 * check that all the FDs passed in refer to legal files 1304 * If not, reject the entire operation. 1305 */ 1306 fdp = data; 1307 FILEDESC_LOCK(fdescp); 1308 for (i = 0; i < oldfds; i++) { 1309 fd = *fdp++; 1310 if ((unsigned)fd >= fdescp->fd_nfiles || 1311 fdescp->fd_ofiles[fd] == NULL) { 1312 FILEDESC_UNLOCK(fdescp); 1313 error = EBADF; 1314 goto out; 1315 } 1316 fp = fdescp->fd_ofiles[fd]; 1317 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { 1318 FILEDESC_UNLOCK(fdescp); 1319 error = EOPNOTSUPP; 1320 goto out; 1321 } 1322 1323 } 1324 /* 1325 * Now replace the integer FDs with pointers to 1326 * the associated global file table entry.. 1327 */ 1328 newlen = oldfds * sizeof(struct file *); 1329 *controlp = sbcreatecontrol(NULL, newlen, 1330 SCM_RIGHTS, SOL_SOCKET); 1331 if (*controlp == NULL) { 1332 FILEDESC_UNLOCK(fdescp); 1333 error = E2BIG; 1334 goto out; 1335 } 1336 1337 fdp = data; 1338 rp = (struct file **) 1339 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1340 for (i = 0; i < oldfds; i++) { 1341 fp = fdescp->fd_ofiles[*fdp++]; 1342 *rp++ = fp; 1343 FILE_LOCK(fp); 1344 fp->f_count++; 1345 fp->f_msgcount++; 1346 FILE_UNLOCK(fp); 1347 unp_rights++; 1348 } 1349 FILEDESC_UNLOCK(fdescp); 1350 break; 1351 1352 case SCM_TIMESTAMP: 1353 *controlp = sbcreatecontrol(NULL, sizeof(*tv), 1354 SCM_TIMESTAMP, SOL_SOCKET); 1355 if (*controlp == NULL) { 1356 error = ENOBUFS; 1357 goto out; 1358 } 1359 tv = (struct timeval *) 1360 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1361 microtime(tv); 1362 break; 1363 1364 default: 1365 error = EINVAL; 1366 goto out; 1367 } 1368 1369 controlp = &(*controlp)->m_next; 1370 1371 if (CMSG_SPACE(datalen) < clen) { 1372 clen -= CMSG_SPACE(datalen); 1373 cm = (struct cmsghdr *) 1374 ((caddr_t)cm + CMSG_SPACE(datalen)); 1375 } else { 1376 clen = 0; 1377 cm = NULL; 1378 } 1379 } 1380 1381out: 1382 m_freem(control); 1383 1384 return (error); 1385} 1386 1387static int unp_defer, unp_gcing; 1388 1389static void 1390unp_gc() 1391{ 1392 register struct file *fp, *nextfp; 1393 register struct socket *so; 1394 struct file **extra_ref, **fpp; 1395 int nunref, i; 1396 1397 UNP_LOCK_ASSERT(); 1398 1399 if (unp_gcing) 1400 return; 1401 unp_gcing = 1; 1402 unp_defer = 0; 1403 /* 1404 * before going through all this, set all FDs to 1405 * be NOT defered and NOT externally accessible 1406 */ 1407 /* 1408 * XXXRW: Acquiring a sleep lock while holding UNP 1409 * mutex cannot be a good thing. 1410 */ 1411 sx_slock(&filelist_lock); 1412 LIST_FOREACH(fp, &filehead, f_list) 1413 fp->f_gcflag &= ~(FMARK|FDEFER); 1414 do { 1415 LIST_FOREACH(fp, &filehead, f_list) { 1416 FILE_LOCK(fp); 1417 /* 1418 * If the file is not open, skip it 1419 */ 1420 if (fp->f_count == 0) { 1421 FILE_UNLOCK(fp); 1422 continue; 1423 } 1424 /* 1425 * If we already marked it as 'defer' in a 1426 * previous pass, then try process it this time 1427 * and un-mark it 1428 */ 1429 if (fp->f_gcflag & FDEFER) { 1430 fp->f_gcflag &= ~FDEFER; 1431 unp_defer--; 1432 } else { 1433 /* 1434 * if it's not defered, then check if it's 1435 * already marked.. if so skip it 1436 */ 1437 if (fp->f_gcflag & FMARK) { 1438 FILE_UNLOCK(fp); 1439 continue; 1440 } 1441 /* 1442 * If all references are from messages 1443 * in transit, then skip it. it's not 1444 * externally accessible. 1445 */ 1446 if (fp->f_count == fp->f_msgcount) { 1447 FILE_UNLOCK(fp); 1448 continue; 1449 } 1450 /* 1451 * If it got this far then it must be 1452 * externally accessible. 1453 */ 1454 fp->f_gcflag |= FMARK; 1455 } 1456 /* 1457 * either it was defered, or it is externally 1458 * accessible and not already marked so. 1459 * Now check if it is possibly one of OUR sockets. 1460 */ 1461 if (fp->f_type != DTYPE_SOCKET || 1462 (so = fp->f_data) == NULL) { 1463 FILE_UNLOCK(fp); 1464 continue; 1465 } 1466 FILE_UNLOCK(fp); 1467 if (so->so_proto->pr_domain != &localdomain || 1468 (so->so_proto->pr_flags&PR_RIGHTS) == 0) 1469 continue; 1470#ifdef notdef 1471 if (so->so_rcv.sb_flags & SB_LOCK) { 1472 /* 1473 * This is problematical; it's not clear 1474 * we need to wait for the sockbuf to be 1475 * unlocked (on a uniprocessor, at least), 1476 * and it's also not clear what to do 1477 * if sbwait returns an error due to receipt 1478 * of a signal. If sbwait does return 1479 * an error, we'll go into an infinite 1480 * loop. Delete all of this for now. 1481 */ 1482 (void) sbwait(&so->so_rcv); 1483 goto restart; 1484 } 1485#endif 1486 /* 1487 * So, Ok, it's one of our sockets and it IS externally 1488 * accessible (or was defered). Now we look 1489 * to see if we hold any file descriptors in its 1490 * message buffers. Follow those links and mark them 1491 * as accessible too. 1492 */ 1493 unp_scan(so->so_rcv.sb_mb, unp_mark); 1494 } 1495 } while (unp_defer); 1496 sx_sunlock(&filelist_lock); 1497 /* 1498 * We grab an extra reference to each of the file table entries 1499 * that are not otherwise accessible and then free the rights 1500 * that are stored in messages on them. 1501 * 1502 * The bug in the orginal code is a little tricky, so I'll describe 1503 * what's wrong with it here. 1504 * 1505 * It is incorrect to simply unp_discard each entry for f_msgcount 1506 * times -- consider the case of sockets A and B that contain 1507 * references to each other. On a last close of some other socket, 1508 * we trigger a gc since the number of outstanding rights (unp_rights) 1509 * is non-zero. If during the sweep phase the gc code un_discards, 1510 * we end up doing a (full) closef on the descriptor. A closef on A 1511 * results in the following chain. Closef calls soo_close, which 1512 * calls soclose. Soclose calls first (through the switch 1513 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply 1514 * returns because the previous instance had set unp_gcing, and 1515 * we return all the way back to soclose, which marks the socket 1516 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush 1517 * to free up the rights that are queued in messages on the socket A, 1518 * i.e., the reference on B. The sorflush calls via the dom_dispose 1519 * switch unp_dispose, which unp_scans with unp_discard. This second 1520 * instance of unp_discard just calls closef on B. 1521 * 1522 * Well, a similar chain occurs on B, resulting in a sorflush on B, 1523 * which results in another closef on A. Unfortunately, A is already 1524 * being closed, and the descriptor has already been marked with 1525 * SS_NOFDREF, and soclose panics at this point. 1526 * 1527 * Here, we first take an extra reference to each inaccessible 1528 * descriptor. Then, we call sorflush ourself, since we know 1529 * it is a Unix domain socket anyhow. After we destroy all the 1530 * rights carried in messages, we do a last closef to get rid 1531 * of our extra reference. This is the last close, and the 1532 * unp_detach etc will shut down the socket. 1533 * 1534 * 91/09/19, bsy@cs.cmu.edu 1535 */ 1536 extra_ref = malloc(nfiles * sizeof(struct file *), M_TEMP, M_WAITOK); 1537 sx_slock(&filelist_lock); 1538 for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref; 1539 fp != NULL; fp = nextfp) { 1540 nextfp = LIST_NEXT(fp, f_list); 1541 FILE_LOCK(fp); 1542 /* 1543 * If it's not open, skip it 1544 */ 1545 if (fp->f_count == 0) { 1546 FILE_UNLOCK(fp); 1547 continue; 1548 } 1549 /* 1550 * If all refs are from msgs, and it's not marked accessible 1551 * then it must be referenced from some unreachable cycle 1552 * of (shut-down) FDs, so include it in our 1553 * list of FDs to remove 1554 */ 1555 if (fp->f_count == fp->f_msgcount && !(fp->f_gcflag & FMARK)) { 1556 *fpp++ = fp; 1557 nunref++; 1558 fp->f_count++; 1559 } 1560 FILE_UNLOCK(fp); 1561 } 1562 sx_sunlock(&filelist_lock); 1563 /* 1564 * for each FD on our hit list, do the following two things 1565 */ 1566 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1567 struct file *tfp = *fpp; 1568 FILE_LOCK(tfp); 1569 if (tfp->f_type == DTYPE_SOCKET && 1570 tfp->f_data != NULL) { 1571 FILE_UNLOCK(tfp); 1572 sorflush(tfp->f_data); 1573 } else { 1574 FILE_UNLOCK(tfp); 1575 } 1576 } 1577 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) 1578 closef(*fpp, (struct thread *) NULL); 1579 free(extra_ref, M_TEMP); 1580 unp_gcing = 0; 1581} 1582 1583void 1584unp_dispose(m) 1585 struct mbuf *m; 1586{ 1587 1588 if (m) 1589 unp_scan(m, unp_discard); 1590} 1591 1592static int 1593unp_listen(unp, td) 1594 struct unpcb *unp; 1595 struct thread *td; 1596{ 1597 UNP_LOCK_ASSERT(); 1598 1599 /* 1600 * XXXRW: Why populate the local peer cred with our own credential? 1601 */ 1602 cru2x(td->td_ucred, &unp->unp_peercred); 1603 unp->unp_flags |= UNP_HAVEPCCACHED; 1604 return (0); 1605} 1606 1607static void 1608unp_scan(m0, op) 1609 register struct mbuf *m0; 1610 void (*op)(struct file *); 1611{ 1612 struct mbuf *m; 1613 struct file **rp; 1614 struct cmsghdr *cm; 1615 void *data; 1616 int i; 1617 socklen_t clen, datalen; 1618 int qfds; 1619 1620 while (m0 != NULL) { 1621 for (m = m0; m; m = m->m_next) { 1622 if (m->m_type != MT_CONTROL) 1623 continue; 1624 1625 cm = mtod(m, struct cmsghdr *); 1626 clen = m->m_len; 1627 1628 while (cm != NULL) { 1629 if (sizeof(*cm) > clen || cm->cmsg_len > clen) 1630 break; 1631 1632 data = CMSG_DATA(cm); 1633 datalen = (caddr_t)cm + cm->cmsg_len 1634 - (caddr_t)data; 1635 1636 if (cm->cmsg_level == SOL_SOCKET && 1637 cm->cmsg_type == SCM_RIGHTS) { 1638 qfds = datalen / sizeof (struct file *); 1639 rp = data; 1640 for (i = 0; i < qfds; i++) 1641 (*op)(*rp++); 1642 } 1643 1644 if (CMSG_SPACE(datalen) < clen) { 1645 clen -= CMSG_SPACE(datalen); 1646 cm = (struct cmsghdr *) 1647 ((caddr_t)cm + CMSG_SPACE(datalen)); 1648 } else { 1649 clen = 0; 1650 cm = NULL; 1651 } 1652 } 1653 } 1654 m0 = m0->m_act; 1655 } 1656} 1657 1658static void 1659unp_mark(fp) 1660 struct file *fp; 1661{ 1662 if (fp->f_gcflag & FMARK) 1663 return; 1664 unp_defer++; 1665 fp->f_gcflag |= (FMARK|FDEFER); 1666} 1667 1668static void 1669unp_discard(fp) 1670 struct file *fp; 1671{ 1672 FILE_LOCK(fp); 1673 fp->f_msgcount--; 1674 unp_rights--; 1675 FILE_UNLOCK(fp); 1676 (void) closef(fp, (struct thread *)NULL); 1677} 1678