uipc_usrreq.c revision 35823
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 1.33 1998/04/17 22:36:50 des Exp $ 35 */ 36 37#include <sys/param.h> 38#include <sys/systm.h> 39#include <sys/kernel.h> 40#include <sys/domain.h> 41#include <sys/fcntl.h> 42#include <sys/malloc.h> /* XXX must be before <sys/file.h> */ 43#include <sys/file.h> 44#include <sys/filedesc.h> 45#include <sys/mbuf.h> 46#include <sys/namei.h> 47#include <sys/proc.h> 48#include <sys/protosw.h> 49#include <sys/socket.h> 50#include <sys/socketvar.h> 51#include <sys/stat.h> 52#include <sys/sysctl.h> 53#include <sys/un.h> 54#include <sys/vnode.h> 55 56/* 57 * Unix communications domain. 58 * 59 * TODO: 60 * SEQPACKET, RDM 61 * rethink name space problems 62 * need a proper out-of-band 63 */ 64static struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL }; 65static ino_t unp_ino; /* prototype for fake inode numbers */ 66 67static int unp_attach __P((struct socket *)); 68static void unp_detach __P((struct unpcb *)); 69static int unp_bind __P((struct unpcb *,struct sockaddr *, struct proc *)); 70static int unp_connect __P((struct socket *,struct sockaddr *, 71 struct proc *)); 72static void unp_disconnect __P((struct unpcb *)); 73static void unp_shutdown __P((struct unpcb *)); 74static void unp_drop __P((struct unpcb *, int)); 75static void unp_gc __P((void)); 76static void unp_scan __P((struct mbuf *, void (*)(struct file *))); 77static void unp_mark __P((struct file *)); 78static void unp_discard __P((struct file *)); 79static int unp_internalize __P((struct mbuf *, struct proc *)); 80 81static int 82uipc_abort(struct socket *so) 83{ 84 struct unpcb *unp = sotounpcb(so); 85 86 if (unp == 0) 87 return EINVAL; 88 unp_drop(unp, ECONNABORTED); 89 return 0; 90} 91 92static int 93uipc_accept(struct socket *so, struct sockaddr **nam) 94{ 95 struct unpcb *unp = sotounpcb(so); 96 97 if (unp == 0) 98 return EINVAL; 99 100 /* 101 * Pass back name of connected socket, 102 * if it was bound and we are still connected 103 * (our peer may have closed already!). 104 */ 105 if (unp->unp_conn && unp->unp_conn->unp_addr) { 106 *nam = dup_sockaddr((struct sockaddr *)unp->unp_conn->unp_addr, 107 1); 108 } else { 109 *nam = dup_sockaddr((struct sockaddr *)&sun_noname, 1); 110 } 111 return 0; 112} 113 114static int 115uipc_attach(struct socket *so, int proto, struct proc *p) 116{ 117 struct unpcb *unp = sotounpcb(so); 118 119 if (unp != 0) 120 return EISCONN; 121 return unp_attach(so); 122} 123 124static int 125uipc_bind(struct socket *so, struct sockaddr *nam, struct proc *p) 126{ 127 struct unpcb *unp = sotounpcb(so); 128 129 if (unp == 0) 130 return EINVAL; 131 132 return unp_bind(unp, nam, p); 133} 134 135static int 136uipc_connect(struct socket *so, struct sockaddr *nam, struct proc *p) 137{ 138 struct unpcb *unp = sotounpcb(so); 139 140 if (unp == 0) 141 return EINVAL; 142 return unp_connect(so, nam, curproc); 143} 144 145static int 146uipc_connect2(struct socket *so1, struct socket *so2) 147{ 148 struct unpcb *unp = sotounpcb(so1); 149 150 if (unp == 0) 151 return EINVAL; 152 153 return unp_connect2(so1, so2); 154} 155 156/* control is EOPNOTSUPP */ 157 158static int 159uipc_detach(struct socket *so) 160{ 161 struct unpcb *unp = sotounpcb(so); 162 163 if (unp == 0) 164 return EINVAL; 165 166 unp_detach(unp); 167 return 0; 168} 169 170static int 171uipc_disconnect(struct socket *so) 172{ 173 struct unpcb *unp = sotounpcb(so); 174 175 if (unp == 0) 176 return EINVAL; 177 unp_disconnect(unp); 178 return 0; 179} 180 181static int 182uipc_listen(struct socket *so, struct proc *p) 183{ 184 struct unpcb *unp = sotounpcb(so); 185 186 if (unp == 0 || unp->unp_vnode == 0) 187 return EINVAL; 188 return 0; 189} 190 191static int 192uipc_peeraddr(struct socket *so, struct sockaddr **nam) 193{ 194 struct unpcb *unp = sotounpcb(so); 195 196 if (unp == 0) 197 return EINVAL; 198 if (unp->unp_conn && unp->unp_conn->unp_addr) 199 *nam = dup_sockaddr((struct sockaddr *)unp->unp_conn->unp_addr, 200 1); 201 return 0; 202} 203 204static int 205uipc_rcvd(struct socket *so, int flags) 206{ 207 struct unpcb *unp = sotounpcb(so); 208 struct socket *so2; 209 210 if (unp == 0) 211 return EINVAL; 212 switch (so->so_type) { 213 case SOCK_DGRAM: 214 panic("uipc_rcvd DGRAM?"); 215 /*NOTREACHED*/ 216 217 case SOCK_STREAM: 218#define rcv (&so->so_rcv) 219#define snd (&so2->so_snd) 220 if (unp->unp_conn == 0) 221 break; 222 so2 = unp->unp_conn->unp_socket; 223 /* 224 * Adjust backpressure on sender 225 * and wakeup any waiting to write. 226 */ 227 snd->sb_mbmax += unp->unp_mbcnt - rcv->sb_mbcnt; 228 unp->unp_mbcnt = rcv->sb_mbcnt; 229 snd->sb_hiwat += unp->unp_cc - rcv->sb_cc; 230 unp->unp_cc = rcv->sb_cc; 231 sowwakeup(so2); 232#undef snd 233#undef rcv 234 break; 235 236 default: 237 panic("uipc_rcvd unknown socktype"); 238 } 239 return 0; 240} 241 242/* pru_rcvoob is EOPNOTSUPP */ 243 244static int 245uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 246 struct mbuf *control, struct proc *p) 247{ 248 int error = 0; 249 struct unpcb *unp = sotounpcb(so); 250 struct socket *so2; 251 252 if (unp == 0) { 253 error = EINVAL; 254 goto release; 255 } 256 if (flags & PRUS_OOB) { 257 error = EOPNOTSUPP; 258 goto release; 259 } 260 261 if (control && (error = unp_internalize(control, p))) 262 goto release; 263 264 switch (so->so_type) { 265 case SOCK_DGRAM: 266 { 267 struct sockaddr *from; 268 269 if (nam) { 270 if (unp->unp_conn) { 271 error = EISCONN; 272 break; 273 } 274 error = unp_connect(so, nam, p); 275 if (error) 276 break; 277 } else { 278 if (unp->unp_conn == 0) { 279 error = ENOTCONN; 280 break; 281 } 282 } 283 so2 = unp->unp_conn->unp_socket; 284 if (unp->unp_addr) 285 from = (struct sockaddr *)unp->unp_addr; 286 else 287 from = &sun_noname; 288 if (sbappendaddr(&so2->so_rcv, from, m, control)) { 289 sorwakeup(so2); 290 m = 0; 291 control = 0; 292 } else 293 error = ENOBUFS; 294 if (nam) 295 unp_disconnect(unp); 296 break; 297 } 298 299 case SOCK_STREAM: 300#define rcv (&so2->so_rcv) 301#define snd (&so->so_snd) 302 /* Connect if not connected yet. */ 303 /* 304 * Note: A better implementation would complain 305 * if not equal to the peer's address. 306 */ 307 if ((so->so_state & SS_ISCONNECTED) == 0) { 308 if (nam) { 309 error = unp_connect(so, nam, p); 310 if (error) 311 break; /* XXX */ 312 } else { 313 error = ENOTCONN; 314 break; 315 } 316 } 317 318 if (so->so_state & SS_CANTSENDMORE) { 319 error = EPIPE; 320 break; 321 } 322 if (unp->unp_conn == 0) 323 panic("uipc_send connected but no connection?"); 324 so2 = unp->unp_conn->unp_socket; 325 /* 326 * Send to paired receive port, and then reduce 327 * send buffer hiwater marks to maintain backpressure. 328 * Wake up readers. 329 */ 330 if (control) { 331 if (sbappendcontrol(rcv, m, control)) 332 control = 0; 333 } else 334 sbappend(rcv, m); 335 snd->sb_mbmax -= 336 rcv->sb_mbcnt - unp->unp_conn->unp_mbcnt; 337 unp->unp_conn->unp_mbcnt = rcv->sb_mbcnt; 338 snd->sb_hiwat -= rcv->sb_cc - unp->unp_conn->unp_cc; 339 unp->unp_conn->unp_cc = rcv->sb_cc; 340 sorwakeup(so2); 341 m = 0; 342#undef snd 343#undef rcv 344 break; 345 346 default: 347 panic("uipc_send unknown socktype"); 348 } 349 350 /* 351 * SEND_EOF is equivalent to a SEND followed by 352 * a SHUTDOWN. 353 */ 354 if (flags & PRUS_EOF) { 355 socantsendmore(so); 356 unp_shutdown(unp); 357 } 358 359release: 360 if (control) 361 m_freem(control); 362 if (m) 363 m_freem(m); 364 return error; 365} 366 367static int 368uipc_sense(struct socket *so, struct stat *sb) 369{ 370 struct unpcb *unp = sotounpcb(so); 371 struct socket *so2; 372 373 if (unp == 0) 374 return EINVAL; 375 sb->st_blksize = so->so_snd.sb_hiwat; 376 if (so->so_type == SOCK_STREAM && unp->unp_conn != 0) { 377 so2 = unp->unp_conn->unp_socket; 378 sb->st_blksize += so2->so_rcv.sb_cc; 379 } 380 sb->st_dev = NODEV; 381 if (unp->unp_ino == 0) 382 unp->unp_ino = unp_ino++; 383 sb->st_ino = unp->unp_ino; 384 return (0); 385} 386 387static int 388uipc_shutdown(struct socket *so) 389{ 390 struct unpcb *unp = sotounpcb(so); 391 392 if (unp == 0) 393 return EINVAL; 394 socantsendmore(so); 395 unp_shutdown(unp); 396 return 0; 397} 398 399static int 400uipc_sockaddr(struct socket *so, struct sockaddr **nam) 401{ 402 struct unpcb *unp = sotounpcb(so); 403 404 if (unp == 0) 405 return EINVAL; 406 if (unp->unp_addr) 407 *nam = dup_sockaddr((struct sockaddr *)unp->unp_addr, 1); 408 return 0; 409} 410 411struct pr_usrreqs uipc_usrreqs = { 412 uipc_abort, uipc_accept, uipc_attach, uipc_bind, uipc_connect, 413 uipc_connect2, pru_control_notsupp, uipc_detach, uipc_disconnect, 414 uipc_listen, uipc_peeraddr, uipc_rcvd, pru_rcvoob_notsupp, 415 uipc_send, uipc_sense, uipc_shutdown, uipc_sockaddr, 416 sosend, soreceive, sopoll 417}; 418 419/* 420 * Both send and receive buffers are allocated PIPSIZ bytes of buffering 421 * for stream sockets, although the total for sender and receiver is 422 * actually only PIPSIZ. 423 * Datagram sockets really use the sendspace as the maximum datagram size, 424 * and don't really want to reserve the sendspace. Their recvspace should 425 * be large enough for at least one max-size datagram plus address. 426 */ 427#ifndef PIPSIZ 428#define PIPSIZ 8192 429#endif 430static u_long unpst_sendspace = PIPSIZ; 431static u_long unpst_recvspace = PIPSIZ; 432static u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 433static u_long unpdg_recvspace = 4*1024; 434 435static int unp_rights; /* file descriptors in flight */ 436 437SYSCTL_INT(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 438 &unpst_sendspace, 0, ""); 439SYSCTL_INT(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 440 &unpst_recvspace, 0, ""); 441SYSCTL_INT(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 442 &unpdg_sendspace, 0, ""); 443SYSCTL_INT(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 444 &unpdg_recvspace, 0, ""); 445SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, ""); 446 447static int 448unp_attach(so) 449 struct socket *so; 450{ 451 register struct unpcb *unp; 452 int error; 453 454 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 455 switch (so->so_type) { 456 457 case SOCK_STREAM: 458 error = soreserve(so, unpst_sendspace, unpst_recvspace); 459 break; 460 461 case SOCK_DGRAM: 462 error = soreserve(so, unpdg_sendspace, unpdg_recvspace); 463 break; 464 465 default: 466 panic("unp_attach"); 467 } 468 if (error) 469 return (error); 470 } 471 MALLOC(unp, struct unpcb *, sizeof *unp, M_PCB, M_NOWAIT); 472 if (unp == NULL) 473 return (ENOBUFS); 474 bzero(unp, sizeof *unp); 475 so->so_pcb = (caddr_t)unp; 476 unp->unp_socket = so; 477 return (0); 478} 479 480static void 481unp_detach(unp) 482 register struct unpcb *unp; 483{ 484 if (unp->unp_vnode) { 485 unp->unp_vnode->v_socket = 0; 486 vrele(unp->unp_vnode); 487 unp->unp_vnode = 0; 488 } 489 if (unp->unp_conn) 490 unp_disconnect(unp); 491 while (unp->unp_refs) 492 unp_drop(unp->unp_refs, ECONNRESET); 493 soisdisconnected(unp->unp_socket); 494 unp->unp_socket->so_pcb = 0; 495 if (unp_rights) { 496 /* 497 * Normally the receive buffer is flushed later, 498 * in sofree, but if our receive buffer holds references 499 * to descriptors that are now garbage, we will dispose 500 * of those descriptor references after the garbage collector 501 * gets them (resulting in a "panic: closef: count < 0"). 502 */ 503 sorflush(unp->unp_socket); 504 unp_gc(); 505 } 506 if (unp->unp_addr) 507 FREE(unp->unp_addr, M_SONAME); 508 FREE(unp, M_PCB); 509} 510 511static int 512unp_bind(unp, nam, p) 513 struct unpcb *unp; 514 struct sockaddr *nam; 515 struct proc *p; 516{ 517 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 518 register struct vnode *vp; 519 struct vattr vattr; 520 int error, namelen; 521 struct nameidata nd; 522 char buf[SOCK_MAXADDRLEN]; 523 524 if (unp->unp_vnode != NULL) 525 return (EINVAL); 526#define offsetof(s, e) ((char *)&((s *)0)->e - (char *)((s *)0)) 527 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 528 if (namelen <= 0) 529 return EINVAL; 530 strncpy(buf, soun->sun_path, namelen); 531 buf[namelen] = 0; /* null-terminate the string */ 532 NDINIT(&nd, CREATE, FOLLOW | LOCKPARENT, UIO_SYSSPACE, 533 buf, p); 534/* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 535 error = namei(&nd); 536 if (error) 537 return (error); 538 vp = nd.ni_vp; 539 if (vp != NULL) { 540 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 541 if (nd.ni_dvp == vp) 542 vrele(nd.ni_dvp); 543 else 544 vput(nd.ni_dvp); 545 vrele(vp); 546 return (EADDRINUSE); 547 } 548 VATTR_NULL(&vattr); 549 vattr.va_type = VSOCK; 550 vattr.va_mode = (ACCESSPERMS & ~p->p_fd->fd_cmask); 551 VOP_LEASE(nd.ni_dvp, p, p->p_ucred, LEASE_WRITE); 552 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 553 vput(nd.ni_dvp); 554 if (error) 555 return (error); 556 vp = nd.ni_vp; 557 vp->v_socket = unp->unp_socket; 558 unp->unp_vnode = vp; 559 unp->unp_addr = (struct sockaddr_un *)dup_sockaddr(nam, 1); 560 VOP_UNLOCK(vp, 0, p); 561 return (0); 562} 563 564static int 565unp_connect(so, nam, p) 566 struct socket *so; 567 struct sockaddr *nam; 568 struct proc *p; 569{ 570 register struct sockaddr_un *soun = (struct sockaddr_un *)nam; 571 register struct vnode *vp; 572 register struct socket *so2, *so3; 573 struct unpcb *unp2, *unp3; 574 int error, len; 575 struct nameidata nd; 576 char buf[SOCK_MAXADDRLEN]; 577 578 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 579 if (len <= 0) 580 return EINVAL; 581 strncpy(buf, soun->sun_path, len); 582 buf[len] = 0; 583 584 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, p); 585 error = namei(&nd); 586 if (error) 587 return (error); 588 vp = nd.ni_vp; 589 if (vp->v_type != VSOCK) { 590 error = ENOTSOCK; 591 goto bad; 592 } 593 error = VOP_ACCESS(vp, VWRITE, p->p_ucred, p); 594 if (error) 595 goto bad; 596 so2 = vp->v_socket; 597 if (so2 == 0) { 598 error = ECONNREFUSED; 599 goto bad; 600 } 601 if (so->so_type != so2->so_type) { 602 error = EPROTOTYPE; 603 goto bad; 604 } 605 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 606 if ((so2->so_options & SO_ACCEPTCONN) == 0 || 607 (so3 = sonewconn(so2, 0)) == 0) { 608 error = ECONNREFUSED; 609 goto bad; 610 } 611 unp2 = sotounpcb(so2); 612 unp3 = sotounpcb(so3); 613 if (unp2->unp_addr) 614 unp3->unp_addr = (struct sockaddr_un *) 615 dup_sockaddr((struct sockaddr *) 616 unp2->unp_addr, 1); 617 so2 = so3; 618 } 619 error = unp_connect2(so, so2); 620bad: 621 vput(vp); 622 return (error); 623} 624 625int 626unp_connect2(so, so2) 627 register struct socket *so; 628 register struct socket *so2; 629{ 630 register struct unpcb *unp = sotounpcb(so); 631 register struct unpcb *unp2; 632 633 if (so2->so_type != so->so_type) 634 return (EPROTOTYPE); 635 unp2 = sotounpcb(so2); 636 unp->unp_conn = unp2; 637 switch (so->so_type) { 638 639 case SOCK_DGRAM: 640 unp->unp_nextref = unp2->unp_refs; 641 unp2->unp_refs = unp; 642 soisconnected(so); 643 break; 644 645 case SOCK_STREAM: 646 unp2->unp_conn = unp; 647 soisconnected(so); 648 soisconnected(so2); 649 break; 650 651 default: 652 panic("unp_connect2"); 653 } 654 return (0); 655} 656 657static void 658unp_disconnect(unp) 659 struct unpcb *unp; 660{ 661 register struct unpcb *unp2 = unp->unp_conn; 662 663 if (unp2 == 0) 664 return; 665 unp->unp_conn = 0; 666 switch (unp->unp_socket->so_type) { 667 668 case SOCK_DGRAM: 669 if (unp2->unp_refs == unp) 670 unp2->unp_refs = unp->unp_nextref; 671 else { 672 unp2 = unp2->unp_refs; 673 for (;;) { 674 if (unp2 == 0) 675 panic("unp_disconnect"); 676 if (unp2->unp_nextref == unp) 677 break; 678 unp2 = unp2->unp_nextref; 679 } 680 unp2->unp_nextref = unp->unp_nextref; 681 } 682 unp->unp_nextref = 0; 683 unp->unp_socket->so_state &= ~SS_ISCONNECTED; 684 break; 685 686 case SOCK_STREAM: 687 soisdisconnected(unp->unp_socket); 688 unp2->unp_conn = 0; 689 soisdisconnected(unp2->unp_socket); 690 break; 691 } 692} 693 694#ifdef notdef 695void 696unp_abort(unp) 697 struct unpcb *unp; 698{ 699 700 unp_detach(unp); 701} 702#endif 703 704static void 705unp_shutdown(unp) 706 struct unpcb *unp; 707{ 708 struct socket *so; 709 710 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn && 711 (so = unp->unp_conn->unp_socket)) 712 socantrcvmore(so); 713} 714 715static void 716unp_drop(unp, errno) 717 struct unpcb *unp; 718 int errno; 719{ 720 struct socket *so = unp->unp_socket; 721 722 so->so_error = errno; 723 unp_disconnect(unp); 724 if (so->so_head) { 725 so->so_pcb = (caddr_t) 0; 726 if (unp->unp_addr) 727 FREE(unp->unp_addr, M_SONAME); 728 FREE(unp, M_PCB); 729 sofree(so); 730 } 731} 732 733#ifdef notdef 734void 735unp_drain() 736{ 737 738} 739#endif 740 741int 742unp_externalize(rights) 743 struct mbuf *rights; 744{ 745 struct proc *p = curproc; /* XXX */ 746 register int i; 747 register struct cmsghdr *cm = mtod(rights, struct cmsghdr *); 748 register struct file **rp = (struct file **)(cm + 1); 749 register struct file *fp; 750 int newfds = (cm->cmsg_len - sizeof(*cm)) / sizeof (int); 751 int f; 752 753 /* 754 * if the new FD's will not fit, then we free them all 755 */ 756 if (!fdavail(p, newfds)) { 757 for (i = 0; i < newfds; i++) { 758 fp = *rp; 759 unp_discard(fp); 760 *rp++ = 0; 761 } 762 return (EMSGSIZE); 763 } 764 /* 765 * now change each pointer to an fd in the global table to 766 * an integer that is the index to the local fd table entry 767 * that we set up to point to the global one we are transferring. 768 * XXX this assumes a pointer and int are the same size...! 769 */ 770 for (i = 0; i < newfds; i++) { 771 if (fdalloc(p, 0, &f)) 772 panic("unp_externalize"); 773 fp = *rp; 774 p->p_fd->fd_ofiles[f] = fp; 775 fp->f_msgcount--; 776 unp_rights--; 777 *(int *)rp++ = f; 778 } 779 return (0); 780} 781 782#ifndef MIN 783#define MIN(a,b) (((a)<(b))?(a):(b)) 784#endif 785 786static int 787unp_internalize(control, p) 788 struct mbuf *control; 789 struct proc *p; 790{ 791 struct filedesc *fdp = p->p_fd; 792 register struct cmsghdr *cm = mtod(control, struct cmsghdr *); 793 register struct file **rp; 794 register struct file *fp; 795 register int i, fd; 796 register struct cmsgcred *cmcred; 797 int oldfds; 798 799 if ((cm->cmsg_type != SCM_RIGHTS && cm->cmsg_type != SCM_CREDS) || 800 cm->cmsg_level != SOL_SOCKET || cm->cmsg_len != control->m_len) 801 return (EINVAL); 802 803 /* 804 * Fill in credential information. 805 */ 806 if (cm->cmsg_type == SCM_CREDS) { 807 cmcred = (struct cmsgcred *)(cm + 1); 808 cmcred->cmcred_pid = p->p_pid; 809 cmcred->cmcred_uid = p->p_cred->p_ruid; 810 cmcred->cmcred_gid = p->p_cred->p_rgid; 811 cmcred->cmcred_euid = p->p_ucred->cr_uid; 812 cmcred->cmcred_ngroups = MIN(p->p_ucred->cr_ngroups, 813 CMGROUP_MAX); 814 for (i = 0; i < cmcred->cmcred_ngroups; i++) 815 cmcred->cmcred_groups[i] = p->p_ucred->cr_groups[i]; 816 return(0); 817 } 818 819 oldfds = (cm->cmsg_len - sizeof (*cm)) / sizeof (int); 820 /* 821 * check that all the FDs passed in refer to legal OPEN files 822 * If not, reject the entire operation. 823 */ 824 rp = (struct file **)(cm + 1); 825 for (i = 0; i < oldfds; i++) { 826 fd = *(int *)rp++; 827 if ((unsigned)fd >= fdp->fd_nfiles || 828 fdp->fd_ofiles[fd] == NULL) 829 return (EBADF); 830 } 831 /* 832 * Now replace the integer FDs with pointers to 833 * the associated global file table entry.. 834 * XXX this assumes a pointer and an int are the same size! 835 */ 836 rp = (struct file **)(cm + 1); 837 for (i = 0; i < oldfds; i++) { 838 fp = fdp->fd_ofiles[*(int *)rp]; 839 *rp++ = fp; 840 fp->f_count++; 841 fp->f_msgcount++; 842 unp_rights++; 843 } 844 return (0); 845} 846 847static int unp_defer, unp_gcing; 848 849static void 850unp_gc() 851{ 852 register struct file *fp, *nextfp; 853 register struct socket *so; 854 struct file **extra_ref, **fpp; 855 int nunref, i; 856 857 if (unp_gcing) 858 return; 859 unp_gcing = 1; 860 unp_defer = 0; 861 /* 862 * before going through all this, set all FDs to 863 * be NOT defered and NOT externally accessible 864 */ 865 for (fp = filehead.lh_first; fp != 0; fp = fp->f_list.le_next) 866 fp->f_flag &= ~(FMARK|FDEFER); 867 do { 868 for (fp = filehead.lh_first; fp != 0; fp = fp->f_list.le_next) { 869 /* 870 * If the file is not open, skip it 871 */ 872 if (fp->f_count == 0) 873 continue; 874 /* 875 * If we already marked it as 'defer' in a 876 * previous pass, then try process it this time 877 * and un-mark it 878 */ 879 if (fp->f_flag & FDEFER) { 880 fp->f_flag &= ~FDEFER; 881 unp_defer--; 882 } else { 883 /* 884 * if it's not defered, then check if it's 885 * already marked.. if so skip it 886 */ 887 if (fp->f_flag & FMARK) 888 continue; 889 /* 890 * If all references are from messages 891 * in transit, then skip it. it's not 892 * externally accessible. 893 */ 894 if (fp->f_count == fp->f_msgcount) 895 continue; 896 /* 897 * If it got this far then it must be 898 * externally accessible. 899 */ 900 fp->f_flag |= FMARK; 901 } 902 /* 903 * either it was defered, or it is externally 904 * accessible and not already marked so. 905 * Now check if it is possibly one of OUR sockets. 906 */ 907 if (fp->f_type != DTYPE_SOCKET || 908 (so = (struct socket *)fp->f_data) == 0) 909 continue; 910 if (so->so_proto->pr_domain != &localdomain || 911 (so->so_proto->pr_flags&PR_RIGHTS) == 0) 912 continue; 913#ifdef notdef 914 if (so->so_rcv.sb_flags & SB_LOCK) { 915 /* 916 * This is problematical; it's not clear 917 * we need to wait for the sockbuf to be 918 * unlocked (on a uniprocessor, at least), 919 * and it's also not clear what to do 920 * if sbwait returns an error due to receipt 921 * of a signal. If sbwait does return 922 * an error, we'll go into an infinite 923 * loop. Delete all of this for now. 924 */ 925 (void) sbwait(&so->so_rcv); 926 goto restart; 927 } 928#endif 929 /* 930 * So, Ok, it's one of our sockets and it IS externally 931 * accessible (or was defered). Now we look 932 * to see if we hold any file descriptors in its 933 * message buffers. Follow those links and mark them 934 * as accessible too. 935 */ 936 unp_scan(so->so_rcv.sb_mb, unp_mark); 937 } 938 } while (unp_defer); 939 /* 940 * We grab an extra reference to each of the file table entries 941 * that are not otherwise accessible and then free the rights 942 * that are stored in messages on them. 943 * 944 * The bug in the orginal code is a little tricky, so I'll describe 945 * what's wrong with it here. 946 * 947 * It is incorrect to simply unp_discard each entry for f_msgcount 948 * times -- consider the case of sockets A and B that contain 949 * references to each other. On a last close of some other socket, 950 * we trigger a gc since the number of outstanding rights (unp_rights) 951 * is non-zero. If during the sweep phase the gc code un_discards, 952 * we end up doing a (full) closef on the descriptor. A closef on A 953 * results in the following chain. Closef calls soo_close, which 954 * calls soclose. Soclose calls first (through the switch 955 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply 956 * returns because the previous instance had set unp_gcing, and 957 * we return all the way back to soclose, which marks the socket 958 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush 959 * to free up the rights that are queued in messages on the socket A, 960 * i.e., the reference on B. The sorflush calls via the dom_dispose 961 * switch unp_dispose, which unp_scans with unp_discard. This second 962 * instance of unp_discard just calls closef on B. 963 * 964 * Well, a similar chain occurs on B, resulting in a sorflush on B, 965 * which results in another closef on A. Unfortunately, A is already 966 * being closed, and the descriptor has already been marked with 967 * SS_NOFDREF, and soclose panics at this point. 968 * 969 * Here, we first take an extra reference to each inaccessible 970 * descriptor. Then, we call sorflush ourself, since we know 971 * it is a Unix domain socket anyhow. After we destroy all the 972 * rights carried in messages, we do a last closef to get rid 973 * of our extra reference. This is the last close, and the 974 * unp_detach etc will shut down the socket. 975 * 976 * 91/09/19, bsy@cs.cmu.edu 977 */ 978 extra_ref = malloc(nfiles * sizeof(struct file *), M_FILE, M_WAITOK); 979 for (nunref = 0, fp = filehead.lh_first, fpp = extra_ref; fp != 0; 980 fp = nextfp) { 981 nextfp = fp->f_list.le_next; 982 /* 983 * If it's not open, skip it 984 */ 985 if (fp->f_count == 0) 986 continue; 987 /* 988 * If all refs are from msgs, and it's not marked accessible 989 * then it must be referenced from some unreachable cycle 990 * of (shut-down) FDs, so include it in our 991 * list of FDs to remove 992 */ 993 if (fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) { 994 *fpp++ = fp; 995 nunref++; 996 fp->f_count++; 997 } 998 } 999 /* 1000 * for each FD on our hit list, do the following two things 1001 */ 1002 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) 1003 sorflush((struct socket *)(*fpp)->f_data); 1004 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) 1005 closef(*fpp, (struct proc *) NULL); 1006 free((caddr_t)extra_ref, M_FILE); 1007 unp_gcing = 0; 1008} 1009 1010void 1011unp_dispose(m) 1012 struct mbuf *m; 1013{ 1014 1015 if (m) 1016 unp_scan(m, unp_discard); 1017} 1018 1019static void 1020unp_scan(m0, op) 1021 register struct mbuf *m0; 1022 void (*op) __P((struct file *)); 1023{ 1024 register struct mbuf *m; 1025 register struct file **rp; 1026 register struct cmsghdr *cm; 1027 register int i; 1028 int qfds; 1029 1030 while (m0) { 1031 for (m = m0; m; m = m->m_next) 1032 if (m->m_type == MT_CONTROL && 1033 m->m_len >= sizeof(*cm)) { 1034 cm = mtod(m, struct cmsghdr *); 1035 if (cm->cmsg_level != SOL_SOCKET || 1036 cm->cmsg_type != SCM_RIGHTS) 1037 continue; 1038 qfds = (cm->cmsg_len - sizeof *cm) 1039 / sizeof (struct file *); 1040 rp = (struct file **)(cm + 1); 1041 for (i = 0; i < qfds; i++) 1042 (*op)(*rp++); 1043 break; /* XXX, but saves time */ 1044 } 1045 m0 = m0->m_act; 1046 } 1047} 1048 1049static void 1050unp_mark(fp) 1051 struct file *fp; 1052{ 1053 1054 if (fp->f_flag & FMARK) 1055 return; 1056 unp_defer++; 1057 fp->f_flag |= (FMARK|FDEFER); 1058} 1059 1060static void 1061unp_discard(fp) 1062 struct file *fp; 1063{ 1064 1065 fp->f_msgcount--; 1066 unp_rights--; 1067 (void) closef(fp, (struct proc *)NULL); 1068} 1069