uipc_usrreq.c revision 8876
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 * 3. All advertising materials mentioning features or use of this software 14 * must display the following acknowledgement: 15 * This product includes software developed by the University of 16 * California, Berkeley and its contributors. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94 34 * $Id: uipc_usrreq.c,v 1.8 1995/05/11 00:13:06 wollman Exp $ 35 */ 36 37#include <sys/param.h> 38#include <sys/systm.h> 39#include <sys/proc.h> 40#include <sys/filedesc.h> 41#include <sys/domain.h> 42#include <sys/protosw.h> 43#include <sys/stat.h> 44#include <sys/socket.h> 45#include <sys/socketvar.h> 46#include <sys/unpcb.h> 47#include <sys/un.h> 48#include <sys/namei.h> 49#include <sys/vnode.h> 50#include <sys/file.h> 51#include <sys/stat.h> 52#include <sys/mbuf.h> 53 54/* 55 * Unix communications domain. 56 * 57 * TODO: 58 * SEQPACKET, RDM 59 * rethink name space problems 60 * need a proper out-of-band 61 */ 62struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL }; 63ino_t unp_ino; /* prototype for fake inode numbers */ 64 65/*ARGSUSED*/ 66int 67uipc_usrreq(so, req, m, nam, control) 68 struct socket *so; 69 int req; 70 struct mbuf *m, *nam, *control; 71{ 72 struct unpcb *unp = sotounpcb(so); 73 register struct socket *so2; 74 register int error = 0; 75 struct proc *p = curproc; /* XXX */ 76 77 if (req == PRU_CONTROL) 78 return (EOPNOTSUPP); 79 if (req != PRU_SEND && control && control->m_len) { 80 error = EOPNOTSUPP; 81 goto release; 82 } 83 if (unp == 0 && req != PRU_ATTACH) { 84 error = EINVAL; 85 goto release; 86 } 87 switch (req) { 88 89 case PRU_ATTACH: 90 if (unp) { 91 error = EISCONN; 92 break; 93 } 94 error = unp_attach(so); 95 break; 96 97 case PRU_DETACH: 98 unp_detach(unp); 99 break; 100 101 case PRU_BIND: 102 error = unp_bind(unp, nam, p); 103 break; 104 105 case PRU_LISTEN: 106 if (unp->unp_vnode == 0) 107 error = EINVAL; 108 break; 109 110 case PRU_CONNECT: 111 error = unp_connect(so, nam, p); 112 break; 113 114 case PRU_CONNECT2: 115 error = unp_connect2(so, (struct socket *)nam); 116 break; 117 118 case PRU_DISCONNECT: 119 unp_disconnect(unp); 120 break; 121 122 case PRU_ACCEPT: 123 /* 124 * Pass back name of connected socket, 125 * if it was bound and we are still connected 126 * (our peer may have closed already!). 127 */ 128 if (unp->unp_conn && unp->unp_conn->unp_addr) { 129 nam->m_len = unp->unp_conn->unp_addr->m_len; 130 bcopy(mtod(unp->unp_conn->unp_addr, caddr_t), 131 mtod(nam, caddr_t), (unsigned)nam->m_len); 132 } else { 133 nam->m_len = sizeof(sun_noname); 134 *(mtod(nam, struct sockaddr *)) = sun_noname; 135 } 136 break; 137 138 case PRU_SHUTDOWN: 139 socantsendmore(so); 140 unp_shutdown(unp); 141 break; 142 143 case PRU_RCVD: 144 switch (so->so_type) { 145 146 case SOCK_DGRAM: 147 panic("uipc 1"); 148 /*NOTREACHED*/ 149 150 case SOCK_STREAM: 151#define rcv (&so->so_rcv) 152#define snd (&so2->so_snd) 153 if (unp->unp_conn == 0) 154 break; 155 so2 = unp->unp_conn->unp_socket; 156 /* 157 * Adjust backpressure on sender 158 * and wakeup any waiting to write. 159 */ 160 snd->sb_mbmax += unp->unp_mbcnt - rcv->sb_mbcnt; 161 unp->unp_mbcnt = rcv->sb_mbcnt; 162 snd->sb_hiwat += unp->unp_cc - rcv->sb_cc; 163 unp->unp_cc = rcv->sb_cc; 164 sowwakeup(so2); 165#undef snd 166#undef rcv 167 break; 168 169 default: 170 panic("uipc 2"); 171 } 172 break; 173 174 case PRU_SEND: 175 case PRU_SEND_EOF: 176 if (control && (error = unp_internalize(control, p))) 177 break; 178 switch (so->so_type) { 179 180 case SOCK_DGRAM: { 181 struct sockaddr *from; 182 183 if (nam) { 184 if (unp->unp_conn) { 185 error = EISCONN; 186 break; 187 } 188 error = unp_connect(so, nam, p); 189 if (error) 190 break; 191 } else { 192 if (unp->unp_conn == 0) { 193 error = ENOTCONN; 194 break; 195 } 196 } 197 so2 = unp->unp_conn->unp_socket; 198 if (unp->unp_addr) 199 from = mtod(unp->unp_addr, struct sockaddr *); 200 else 201 from = &sun_noname; 202 if (sbappendaddr(&so2->so_rcv, from, m, control)) { 203 sorwakeup(so2); 204 m = 0; 205 control = 0; 206 } else 207 error = ENOBUFS; 208 if (nam) 209 unp_disconnect(unp); 210 break; 211 } 212 213 case SOCK_STREAM: 214#define rcv (&so2->so_rcv) 215#define snd (&so->so_snd) 216 /* Connect if not connected yet. */ 217 /* 218 * Note: A better implementation would complain 219 * if not equal to the peer's address. 220 */ 221 if ((so->so_state & SS_ISCONNECTED) == 0) { 222 if (nam) { 223 error = unp_connect(so, nam, p); 224 if (error) 225 break; /* XXX */ 226 } else { 227 error = ENOTCONN; 228 break; 229 } 230 } 231 232 if (so->so_state & SS_CANTSENDMORE) { 233 error = EPIPE; 234 break; 235 } 236 if (unp->unp_conn == 0) 237 panic("uipc 3"); 238 so2 = unp->unp_conn->unp_socket; 239 /* 240 * Send to paired receive port, and then reduce 241 * send buffer hiwater marks to maintain backpressure. 242 * Wake up readers. 243 */ 244 if (control) { 245 if (sbappendcontrol(rcv, m, control)) 246 control = 0; 247 } else 248 sbappend(rcv, m); 249 snd->sb_mbmax -= 250 rcv->sb_mbcnt - unp->unp_conn->unp_mbcnt; 251 unp->unp_conn->unp_mbcnt = rcv->sb_mbcnt; 252 snd->sb_hiwat -= rcv->sb_cc - unp->unp_conn->unp_cc; 253 unp->unp_conn->unp_cc = rcv->sb_cc; 254 sorwakeup(so2); 255 m = 0; 256#undef snd 257#undef rcv 258 break; 259 260 default: 261 panic("uipc 4"); 262 } 263 /* 264 * SEND_EOF is equivalent to a SEND followed by 265 * a SHUTDOWN. 266 */ 267 if (req == PRU_SEND_EOF) { 268 socantsendmore(so); 269 unp_shutdown(unp); 270 } 271 break; 272 273 case PRU_ABORT: 274 unp_drop(unp, ECONNABORTED); 275 break; 276 277 case PRU_SENSE: 278 ((struct stat *) m)->st_blksize = so->so_snd.sb_hiwat; 279 if (so->so_type == SOCK_STREAM && unp->unp_conn != 0) { 280 so2 = unp->unp_conn->unp_socket; 281 ((struct stat *) m)->st_blksize += so2->so_rcv.sb_cc; 282 } 283 ((struct stat *) m)->st_dev = NODEV; 284 if (unp->unp_ino == 0) 285 unp->unp_ino = unp_ino++; 286 ((struct stat *) m)->st_ino = unp->unp_ino; 287 return (0); 288 289 case PRU_RCVOOB: 290 return (EOPNOTSUPP); 291 292 case PRU_SENDOOB: 293 error = EOPNOTSUPP; 294 break; 295 296 case PRU_SOCKADDR: 297 if (unp->unp_addr) { 298 nam->m_len = unp->unp_addr->m_len; 299 bcopy(mtod(unp->unp_addr, caddr_t), 300 mtod(nam, caddr_t), (unsigned)nam->m_len); 301 } else 302 nam->m_len = 0; 303 break; 304 305 case PRU_PEERADDR: 306 if (unp->unp_conn && unp->unp_conn->unp_addr) { 307 nam->m_len = unp->unp_conn->unp_addr->m_len; 308 bcopy(mtod(unp->unp_conn->unp_addr, caddr_t), 309 mtod(nam, caddr_t), (unsigned)nam->m_len); 310 } else 311 nam->m_len = 0; 312 break; 313 314 case PRU_SLOWTIMO: 315 break; 316 317 default: 318 panic("piusrreq"); 319 } 320release: 321 if (control) 322 m_freem(control); 323 if (m) 324 m_freem(m); 325 return (error); 326} 327 328/* 329 * Both send and receive buffers are allocated PIPSIZ bytes of buffering 330 * for stream sockets, although the total for sender and receiver is 331 * actually only PIPSIZ. 332 * Datagram sockets really use the sendspace as the maximum datagram size, 333 * and don't really want to reserve the sendspace. Their recvspace should 334 * be large enough for at least one max-size datagram plus address. 335 */ 336#define PIPSIZ 4096 337u_long unpst_sendspace = PIPSIZ; 338u_long unpst_recvspace = PIPSIZ; 339u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 340u_long unpdg_recvspace = 4*1024; 341 342int unp_rights; /* file descriptors in flight */ 343 344int 345unp_attach(so) 346 struct socket *so; 347{ 348 register struct mbuf *m; 349 register struct unpcb *unp; 350 int error; 351 352 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 353 switch (so->so_type) { 354 355 case SOCK_STREAM: 356 error = soreserve(so, unpst_sendspace, unpst_recvspace); 357 break; 358 359 case SOCK_DGRAM: 360 error = soreserve(so, unpdg_sendspace, unpdg_recvspace); 361 break; 362 363 default: 364 panic("unp_attach"); 365 } 366 if (error) 367 return (error); 368 } 369 m = m_getclr(M_DONTWAIT, MT_PCB); 370 if (m == NULL) 371 return (ENOBUFS); 372 unp = mtod(m, struct unpcb *); 373 so->so_pcb = (caddr_t)unp; 374 unp->unp_socket = so; 375 return (0); 376} 377 378void 379unp_detach(unp) 380 register struct unpcb *unp; 381{ 382 383 if (unp->unp_vnode) { 384 unp->unp_vnode->v_socket = 0; 385 vrele(unp->unp_vnode); 386 unp->unp_vnode = 0; 387 } 388 if (unp->unp_conn) 389 unp_disconnect(unp); 390 while (unp->unp_refs) 391 unp_drop(unp->unp_refs, ECONNRESET); 392 soisdisconnected(unp->unp_socket); 393 unp->unp_socket->so_pcb = 0; 394 m_freem(unp->unp_addr); 395 (void) m_free(dtom(unp)); 396 if (unp_rights) { 397 /* 398 * Normally the receive buffer is flushed later, 399 * in sofree, but if our receive buffer holds references 400 * to descriptors that are now garbage, we will dispose 401 * of those descriptor references after the garbage collector 402 * gets them (resulting in a "panic: closef: count < 0"). 403 */ 404 sorflush(unp->unp_socket); 405 unp_gc(); 406 } 407} 408 409int 410unp_bind(unp, nam, p) 411 struct unpcb *unp; 412 struct mbuf *nam; 413 struct proc *p; 414{ 415 struct sockaddr_un *soun = mtod(nam, struct sockaddr_un *); 416 register struct vnode *vp; 417 struct vattr vattr; 418 int error; 419 struct nameidata nd; 420 421 NDINIT(&nd, CREATE, FOLLOW | LOCKPARENT, UIO_SYSSPACE, 422 soun->sun_path, p); 423 if (unp->unp_vnode != NULL) 424 return (EINVAL); 425 if (nam->m_len == MLEN) { 426 if (*(mtod(nam, caddr_t) + nam->m_len - 1) != 0) 427 return (EINVAL); 428 } else 429 *(mtod(nam, caddr_t) + nam->m_len) = 0; 430/* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 431 error = namei(&nd); 432 if (error) 433 return (error); 434 vp = nd.ni_vp; 435 if (vp != NULL) { 436 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 437 if (nd.ni_dvp == vp) 438 vrele(nd.ni_dvp); 439 else 440 vput(nd.ni_dvp); 441 vrele(vp); 442 return (EADDRINUSE); 443 } 444 VATTR_NULL(&vattr); 445 vattr.va_type = VSOCK; 446 vattr.va_mode = ACCESSPERMS; 447 LEASE_CHECK(nd.ni_dvp, p, p->p_ucred, LEASE_WRITE); 448 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 449 if (error) 450 return (error); 451 vp = nd.ni_vp; 452 vp->v_socket = unp->unp_socket; 453 unp->unp_vnode = vp; 454 unp->unp_addr = m_copy(nam, 0, (int)M_COPYALL); 455 VOP_UNLOCK(vp); 456 return (0); 457} 458 459int 460unp_connect(so, nam, p) 461 struct socket *so; 462 struct mbuf *nam; 463 struct proc *p; 464{ 465 register struct sockaddr_un *soun = mtod(nam, struct sockaddr_un *); 466 register struct vnode *vp; 467 register struct socket *so2, *so3; 468 struct unpcb *unp2, *unp3; 469 int error; 470 struct nameidata nd; 471 472 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, soun->sun_path, p); 473 if (nam->m_data + nam->m_len == &nam->m_dat[MLEN]) { /* XXX */ 474 if (*(mtod(nam, caddr_t) + nam->m_len - 1) != 0) 475 return (EMSGSIZE); 476 } else 477 *(mtod(nam, caddr_t) + nam->m_len) = 0; 478 error = namei(&nd); 479 if (error) 480 return (error); 481 vp = nd.ni_vp; 482 if (vp->v_type != VSOCK) { 483 error = ENOTSOCK; 484 goto bad; 485 } 486 error = VOP_ACCESS(vp, VWRITE, p->p_ucred, p); 487 if (error) 488 goto bad; 489 so2 = vp->v_socket; 490 if (so2 == 0) { 491 error = ECONNREFUSED; 492 goto bad; 493 } 494 if (so->so_type != so2->so_type) { 495 error = EPROTOTYPE; 496 goto bad; 497 } 498 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 499 if ((so2->so_options & SO_ACCEPTCONN) == 0 || 500 (so3 = sonewconn(so2, 0)) == 0) { 501 error = ECONNREFUSED; 502 goto bad; 503 } 504 unp2 = sotounpcb(so2); 505 unp3 = sotounpcb(so3); 506 if (unp2->unp_addr) 507 unp3->unp_addr = 508 m_copy(unp2->unp_addr, 0, (int)M_COPYALL); 509 so2 = so3; 510 } 511 error = unp_connect2(so, so2); 512bad: 513 vput(vp); 514 return (error); 515} 516 517int 518unp_connect2(so, so2) 519 register struct socket *so; 520 register struct socket *so2; 521{ 522 register struct unpcb *unp = sotounpcb(so); 523 register struct unpcb *unp2; 524 525 if (so2->so_type != so->so_type) 526 return (EPROTOTYPE); 527 unp2 = sotounpcb(so2); 528 unp->unp_conn = unp2; 529 switch (so->so_type) { 530 531 case SOCK_DGRAM: 532 unp->unp_nextref = unp2->unp_refs; 533 unp2->unp_refs = unp; 534 soisconnected(so); 535 break; 536 537 case SOCK_STREAM: 538 unp2->unp_conn = unp; 539 soisconnected(so); 540 soisconnected(so2); 541 break; 542 543 default: 544 panic("unp_connect2"); 545 } 546 return (0); 547} 548 549void 550unp_disconnect(unp) 551 struct unpcb *unp; 552{ 553 register struct unpcb *unp2 = unp->unp_conn; 554 555 if (unp2 == 0) 556 return; 557 unp->unp_conn = 0; 558 switch (unp->unp_socket->so_type) { 559 560 case SOCK_DGRAM: 561 if (unp2->unp_refs == unp) 562 unp2->unp_refs = unp->unp_nextref; 563 else { 564 unp2 = unp2->unp_refs; 565 for (;;) { 566 if (unp2 == 0) 567 panic("unp_disconnect"); 568 if (unp2->unp_nextref == unp) 569 break; 570 unp2 = unp2->unp_nextref; 571 } 572 unp2->unp_nextref = unp->unp_nextref; 573 } 574 unp->unp_nextref = 0; 575 unp->unp_socket->so_state &= ~SS_ISCONNECTED; 576 break; 577 578 case SOCK_STREAM: 579 soisdisconnected(unp->unp_socket); 580 unp2->unp_conn = 0; 581 soisdisconnected(unp2->unp_socket); 582 break; 583 } 584} 585 586#ifdef notdef 587void 588unp_abort(unp) 589 struct unpcb *unp; 590{ 591 592 unp_detach(unp); 593} 594#endif 595 596void 597unp_shutdown(unp) 598 struct unpcb *unp; 599{ 600 struct socket *so; 601 602 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn && 603 (so = unp->unp_conn->unp_socket)) 604 socantrcvmore(so); 605} 606 607void 608unp_drop(unp, errno) 609 struct unpcb *unp; 610 int errno; 611{ 612 struct socket *so = unp->unp_socket; 613 614 so->so_error = errno; 615 unp_disconnect(unp); 616 if (so->so_head) { 617 so->so_pcb = (caddr_t) 0; 618 m_freem(unp->unp_addr); 619 (void) m_free(dtom(unp)); 620 sofree(so); 621 } 622} 623 624#ifdef notdef 625void 626unp_drain() 627{ 628 629} 630#endif 631 632int 633unp_externalize(rights) 634 struct mbuf *rights; 635{ 636 struct proc *p = curproc; /* XXX */ 637 register int i; 638 register struct cmsghdr *cm = mtod(rights, struct cmsghdr *); 639 register struct file **rp = (struct file **)(cm + 1); 640 register struct file *fp; 641 int newfds = (cm->cmsg_len - sizeof(*cm)) / sizeof (int); 642 int f; 643 644 if (!fdavail(p, newfds)) { 645 for (i = 0; i < newfds; i++) { 646 fp = *rp; 647 unp_discard(fp); 648 *rp++ = 0; 649 } 650 return (EMSGSIZE); 651 } 652 for (i = 0; i < newfds; i++) { 653 if (fdalloc(p, 0, &f)) 654 panic("unp_externalize"); 655 fp = *rp; 656 p->p_fd->fd_ofiles[f] = fp; 657 fp->f_msgcount--; 658 unp_rights--; 659 *(int *)rp++ = f; 660 } 661 return (0); 662} 663 664int 665unp_internalize(control, p) 666 struct mbuf *control; 667 struct proc *p; 668{ 669 struct filedesc *fdp = p->p_fd; 670 register struct cmsghdr *cm = mtod(control, struct cmsghdr *); 671 register struct file **rp; 672 register struct file *fp; 673 register int i, fd; 674 int oldfds; 675 676 if (cm->cmsg_type != SCM_RIGHTS || cm->cmsg_level != SOL_SOCKET || 677 cm->cmsg_len != control->m_len) 678 return (EINVAL); 679 oldfds = (cm->cmsg_len - sizeof (*cm)) / sizeof (int); 680 rp = (struct file **)(cm + 1); 681 for (i = 0; i < oldfds; i++) { 682 fd = *(int *)rp++; 683 if ((unsigned)fd >= fdp->fd_nfiles || 684 fdp->fd_ofiles[fd] == NULL) 685 return (EBADF); 686 } 687 rp = (struct file **)(cm + 1); 688 for (i = 0; i < oldfds; i++) { 689 fp = fdp->fd_ofiles[*(int *)rp]; 690 *rp++ = fp; 691 fp->f_count++; 692 fp->f_msgcount++; 693 unp_rights++; 694 } 695 return (0); 696} 697 698int unp_defer, unp_gcing; 699extern struct domain localdomain; 700 701void 702unp_gc() 703{ 704 register struct file *fp, *nextfp; 705 register struct socket *so; 706 struct file **extra_ref, **fpp; 707 int nunref, i; 708 709 if (unp_gcing) 710 return; 711 unp_gcing = 1; 712 unp_defer = 0; 713 for (fp = filehead; fp; fp = fp->f_filef) 714 fp->f_flag &= ~(FMARK|FDEFER); 715 do { 716 for (fp = filehead; fp; fp = fp->f_filef) { 717 if (fp->f_count == 0) 718 continue; 719 if (fp->f_flag & FDEFER) { 720 fp->f_flag &= ~FDEFER; 721 unp_defer--; 722 } else { 723 if (fp->f_flag & FMARK) 724 continue; 725 if (fp->f_count == fp->f_msgcount) 726 continue; 727 fp->f_flag |= FMARK; 728 } 729 if (fp->f_type != DTYPE_SOCKET || 730 (so = (struct socket *)fp->f_data) == 0) 731 continue; 732 if (so->so_proto->pr_domain != &localdomain || 733 (so->so_proto->pr_flags&PR_RIGHTS) == 0) 734 continue; 735#ifdef notdef 736 if (so->so_rcv.sb_flags & SB_LOCK) { 737 /* 738 * This is problematical; it's not clear 739 * we need to wait for the sockbuf to be 740 * unlocked (on a uniprocessor, at least), 741 * and it's also not clear what to do 742 * if sbwait returns an error due to receipt 743 * of a signal. If sbwait does return 744 * an error, we'll go into an infinite 745 * loop. Delete all of this for now. 746 */ 747 (void) sbwait(&so->so_rcv); 748 goto restart; 749 } 750#endif 751 unp_scan(so->so_rcv.sb_mb, unp_mark); 752 } 753 } while (unp_defer); 754 /* 755 * We grab an extra reference to each of the file table entries 756 * that are not otherwise accessible and then free the rights 757 * that are stored in messages on them. 758 * 759 * The bug in the orginal code is a little tricky, so I'll describe 760 * what's wrong with it here. 761 * 762 * It is incorrect to simply unp_discard each entry for f_msgcount 763 * times -- consider the case of sockets A and B that contain 764 * references to each other. On a last close of some other socket, 765 * we trigger a gc since the number of outstanding rights (unp_rights) 766 * is non-zero. If during the sweep phase the gc code un_discards, 767 * we end up doing a (full) closef on the descriptor. A closef on A 768 * results in the following chain. Closef calls soo_close, which 769 * calls soclose. Soclose calls first (through the switch 770 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply 771 * returns because the previous instance had set unp_gcing, and 772 * we return all the way back to soclose, which marks the socket 773 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush 774 * to free up the rights that are queued in messages on the socket A, 775 * i.e., the reference on B. The sorflush calls via the dom_dispose 776 * switch unp_dispose, which unp_scans with unp_discard. This second 777 * instance of unp_discard just calls closef on B. 778 * 779 * Well, a similar chain occurs on B, resulting in a sorflush on B, 780 * which results in another closef on A. Unfortunately, A is already 781 * being closed, and the descriptor has already been marked with 782 * SS_NOFDREF, and soclose panics at this point. 783 * 784 * Here, we first take an extra reference to each inaccessible 785 * descriptor. Then, we call sorflush ourself, since we know 786 * it is a Unix domain socket anyhow. After we destroy all the 787 * rights carried in messages, we do a last closef to get rid 788 * of our extra reference. This is the last close, and the 789 * unp_detach etc will shut down the socket. 790 * 791 * 91/09/19, bsy@cs.cmu.edu 792 */ 793 extra_ref = malloc(nfiles * sizeof(struct file *), M_FILE, M_WAITOK); 794 for (nunref = 0, fp = filehead, fpp = extra_ref; fp; fp = nextfp) { 795 nextfp = fp->f_filef; 796 if (fp->f_count == 0) 797 continue; 798 if (fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) { 799 *fpp++ = fp; 800 nunref++; 801 fp->f_count++; 802 } 803 } 804 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) 805 sorflush((struct socket *)(*fpp)->f_data); 806 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) 807 closef(*fpp,(struct proc*) NULL); 808 free((caddr_t)extra_ref, M_FILE); 809 unp_gcing = 0; 810} 811 812void 813unp_dispose(m) 814 struct mbuf *m; 815{ 816 if (m) 817 unp_scan(m, unp_discard); 818} 819 820void 821unp_scan(m0, op) 822 register struct mbuf *m0; 823 void (*op)(struct file *); 824{ 825 register struct mbuf *m; 826 register struct file **rp; 827 register struct cmsghdr *cm; 828 register int i; 829 int qfds; 830 831 while (m0) { 832 for (m = m0; m; m = m->m_next) 833 if (m->m_type == MT_CONTROL && 834 m->m_len >= sizeof(*cm)) { 835 cm = mtod(m, struct cmsghdr *); 836 if (cm->cmsg_level != SOL_SOCKET || 837 cm->cmsg_type != SCM_RIGHTS) 838 continue; 839 qfds = (cm->cmsg_len - sizeof *cm) 840 / sizeof (struct file *); 841 rp = (struct file **)(cm + 1); 842 for (i = 0; i < qfds; i++) 843 (*op)(*rp++); 844 break; /* XXX, but saves time */ 845 } 846 m0 = m0->m_act; 847 } 848} 849 850void 851unp_mark(fp) 852 struct file *fp; 853{ 854 855 if (fp->f_flag & FMARK) 856 return; 857 unp_defer++; 858 fp->f_flag |= (FMARK|FDEFER); 859} 860 861void 862unp_discard(fp) 863 struct file *fp; 864{ 865 866 fp->f_msgcount--; 867 unp_rights--; 868 (void) closef(fp, (struct proc *)NULL); 869} 870