uipc_usrreq.c revision 45620
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 1.41 1999/04/11 02:17:47 eivind 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/lock.h> 46#include <sys/mbuf.h> 47#include <sys/namei.h> 48#include <sys/proc.h> 49#include <sys/protosw.h> 50#include <sys/socket.h> 51#include <sys/socketvar.h> 52#include <sys/stat.h> 53#include <sys/sysctl.h> 54#include <sys/un.h> 55#include <sys/unpcb.h> 56#include <sys/vnode.h> 57 58#include <vm/vm_zone.h> 59 60static struct vm_zone *unp_zone; 61static unp_gen_t unp_gencnt; 62static u_int unp_count; 63 64static struct unp_head unp_shead, unp_dhead; 65 66/* 67 * Unix communications domain. 68 * 69 * TODO: 70 * SEQPACKET, RDM 71 * rethink name space problems 72 * need a proper out-of-band 73 * lock pushdown 74 */ 75static struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL }; 76static ino_t unp_ino; /* prototype for fake inode numbers */ 77 78static int unp_attach __P((struct socket *)); 79static void unp_detach __P((struct unpcb *)); 80static int unp_bind __P((struct unpcb *,struct sockaddr *, struct proc *)); 81static int unp_connect __P((struct socket *,struct sockaddr *, 82 struct proc *)); 83static void unp_disconnect __P((struct unpcb *)); 84static void unp_shutdown __P((struct unpcb *)); 85static void unp_drop __P((struct unpcb *, int)); 86static void unp_gc __P((void)); 87static void unp_scan __P((struct mbuf *, void (*)(struct file *))); 88static void unp_mark __P((struct file *)); 89static void unp_discard __P((struct file *)); 90static int unp_internalize __P((struct mbuf *, struct proc *)); 91 92static int 93uipc_abort(struct socket *so) 94{ 95 struct unpcb *unp = sotounpcb(so); 96 97 if (unp == 0) 98 return EINVAL; 99 unp_drop(unp, ECONNABORTED); 100 return 0; 101} 102 103static int 104uipc_accept(struct socket *so, struct sockaddr **nam) 105{ 106 struct unpcb *unp = sotounpcb(so); 107 108 if (unp == 0) 109 return EINVAL; 110 111 /* 112 * Pass back name of connected socket, 113 * if it was bound and we are still connected 114 * (our peer may have closed already!). 115 */ 116 if (unp->unp_conn && unp->unp_conn->unp_addr) { 117 *nam = dup_sockaddr((struct sockaddr *)unp->unp_conn->unp_addr, 118 1); 119 } else { 120 *nam = dup_sockaddr((struct sockaddr *)&sun_noname, 1); 121 } 122 return 0; 123} 124 125static int 126uipc_attach(struct socket *so, int proto, struct proc *p) 127{ 128 struct unpcb *unp = sotounpcb(so); 129 130 if (unp != 0) 131 return EISCONN; 132 return unp_attach(so); 133} 134 135static int 136uipc_bind(struct socket *so, struct sockaddr *nam, struct proc *p) 137{ 138 struct unpcb *unp = sotounpcb(so); 139 140 if (unp == 0) 141 return EINVAL; 142 143 return unp_bind(unp, nam, p); 144} 145 146static int 147uipc_connect(struct socket *so, struct sockaddr *nam, struct proc *p) 148{ 149 struct unpcb *unp = sotounpcb(so); 150 151 if (unp == 0) 152 return EINVAL; 153 return unp_connect(so, nam, curproc); 154} 155 156static int 157uipc_connect2(struct socket *so1, struct socket *so2) 158{ 159 struct unpcb *unp = sotounpcb(so1); 160 161 if (unp == 0) 162 return EINVAL; 163 164 return unp_connect2(so1, so2); 165} 166 167/* control is EOPNOTSUPP */ 168 169static int 170uipc_detach(struct socket *so) 171{ 172 struct unpcb *unp = sotounpcb(so); 173 174 if (unp == 0) 175 return EINVAL; 176 177 unp_detach(unp); 178 return 0; 179} 180 181static int 182uipc_disconnect(struct socket *so) 183{ 184 struct unpcb *unp = sotounpcb(so); 185 186 if (unp == 0) 187 return EINVAL; 188 unp_disconnect(unp); 189 return 0; 190} 191 192static int 193uipc_listen(struct socket *so, struct proc *p) 194{ 195 struct unpcb *unp = sotounpcb(so); 196 197 if (unp == 0 || unp->unp_vnode == 0) 198 return EINVAL; 199 return 0; 200} 201 202static int 203uipc_peeraddr(struct socket *so, struct sockaddr **nam) 204{ 205 struct unpcb *unp = sotounpcb(so); 206 207 if (unp == 0) 208 return EINVAL; 209 if (unp->unp_conn && unp->unp_conn->unp_addr) 210 *nam = dup_sockaddr((struct sockaddr *)unp->unp_conn->unp_addr, 211 1); 212 return 0; 213} 214 215static int 216uipc_rcvd(struct socket *so, int flags) 217{ 218 struct unpcb *unp = sotounpcb(so); 219 struct socket *so2; 220 221 if (unp == 0) 222 return EINVAL; 223 switch (so->so_type) { 224 case SOCK_DGRAM: 225 panic("uipc_rcvd DGRAM?"); 226 /*NOTREACHED*/ 227 228 case SOCK_STREAM: 229#define rcv (&so->so_rcv) 230#define snd (&so2->so_snd) 231 if (unp->unp_conn == 0) 232 break; 233 so2 = unp->unp_conn->unp_socket; 234 /* 235 * Adjust backpressure on sender 236 * and wakeup any waiting to write. 237 */ 238 snd->sb_mbmax += unp->unp_mbcnt - rcv->sb_mbcnt; 239 unp->unp_mbcnt = rcv->sb_mbcnt; 240 snd->sb_hiwat += unp->unp_cc - rcv->sb_cc; 241 unp->unp_cc = rcv->sb_cc; 242 sowwakeup(so2); 243#undef snd 244#undef rcv 245 break; 246 247 default: 248 panic("uipc_rcvd unknown socktype"); 249 } 250 return 0; 251} 252 253/* pru_rcvoob is EOPNOTSUPP */ 254 255static int 256uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 257 struct mbuf *control, struct proc *p) 258{ 259 int error = 0; 260 struct unpcb *unp = sotounpcb(so); 261 struct socket *so2; 262 263 if (unp == 0) { 264 error = EINVAL; 265 goto release; 266 } 267 if (flags & PRUS_OOB) { 268 error = EOPNOTSUPP; 269 goto release; 270 } 271 272 if (control && (error = unp_internalize(control, p))) 273 goto release; 274 275 switch (so->so_type) { 276 case SOCK_DGRAM: 277 { 278 struct sockaddr *from; 279 280 if (nam) { 281 if (unp->unp_conn) { 282 error = EISCONN; 283 break; 284 } 285 error = unp_connect(so, nam, p); 286 if (error) 287 break; 288 } else { 289 if (unp->unp_conn == 0) { 290 error = ENOTCONN; 291 break; 292 } 293 } 294 so2 = unp->unp_conn->unp_socket; 295 if (unp->unp_addr) 296 from = (struct sockaddr *)unp->unp_addr; 297 else 298 from = &sun_noname; 299 if (sbappendaddr(&so2->so_rcv, from, m, control)) { 300 sorwakeup(so2); 301 m = 0; 302 control = 0; 303 } else 304 error = ENOBUFS; 305 if (nam) 306 unp_disconnect(unp); 307 break; 308 } 309 310 case SOCK_STREAM: 311#define rcv (&so2->so_rcv) 312#define snd (&so->so_snd) 313 /* Connect if not connected yet. */ 314 /* 315 * Note: A better implementation would complain 316 * if not equal to the peer's address. 317 */ 318 if ((so->so_state & SS_ISCONNECTED) == 0) { 319 if (nam) { 320 error = unp_connect(so, nam, p); 321 if (error) 322 break; /* XXX */ 323 } else { 324 error = ENOTCONN; 325 break; 326 } 327 } 328 329 if (so->so_state & SS_CANTSENDMORE) { 330 error = EPIPE; 331 break; 332 } 333 if (unp->unp_conn == 0) 334 panic("uipc_send connected but no connection?"); 335 so2 = unp->unp_conn->unp_socket; 336 /* 337 * Send to paired receive port, and then reduce 338 * send buffer hiwater marks to maintain backpressure. 339 * Wake up readers. 340 */ 341 if (control) { 342 if (sbappendcontrol(rcv, m, control)) 343 control = 0; 344 } else 345 sbappend(rcv, m); 346 snd->sb_mbmax -= 347 rcv->sb_mbcnt - unp->unp_conn->unp_mbcnt; 348 unp->unp_conn->unp_mbcnt = rcv->sb_mbcnt; 349 snd->sb_hiwat -= rcv->sb_cc - unp->unp_conn->unp_cc; 350 unp->unp_conn->unp_cc = rcv->sb_cc; 351 sorwakeup(so2); 352 m = 0; 353#undef snd 354#undef rcv 355 break; 356 357 default: 358 panic("uipc_send unknown socktype"); 359 } 360 361 /* 362 * SEND_EOF is equivalent to a SEND followed by 363 * a SHUTDOWN. 364 */ 365 if (flags & PRUS_EOF) { 366 socantsendmore(so); 367 unp_shutdown(unp); 368 } 369 370release: 371 if (control) 372 m_freem(control); 373 if (m) 374 m_freem(m); 375 return error; 376} 377 378static int 379uipc_sense(struct socket *so, struct stat *sb) 380{ 381 struct unpcb *unp = sotounpcb(so); 382 struct socket *so2; 383 384 if (unp == 0) 385 return EINVAL; 386 sb->st_blksize = so->so_snd.sb_hiwat; 387 if (so->so_type == SOCK_STREAM && unp->unp_conn != 0) { 388 so2 = unp->unp_conn->unp_socket; 389 sb->st_blksize += so2->so_rcv.sb_cc; 390 } 391 sb->st_dev = NODEV; 392 if (unp->unp_ino == 0) 393 unp->unp_ino = unp_ino++; 394 sb->st_ino = unp->unp_ino; 395 return (0); 396} 397 398static int 399uipc_shutdown(struct socket *so) 400{ 401 struct unpcb *unp = sotounpcb(so); 402 403 if (unp == 0) 404 return EINVAL; 405 socantsendmore(so); 406 unp_shutdown(unp); 407 return 0; 408} 409 410static int 411uipc_sockaddr(struct socket *so, struct sockaddr **nam) 412{ 413 struct unpcb *unp = sotounpcb(so); 414 415 if (unp == 0) 416 return EINVAL; 417 if (unp->unp_addr) 418 *nam = dup_sockaddr((struct sockaddr *)unp->unp_addr, 1); 419 return 0; 420} 421 422struct pr_usrreqs uipc_usrreqs = { 423 uipc_abort, uipc_accept, uipc_attach, uipc_bind, uipc_connect, 424 uipc_connect2, pru_control_notsupp, uipc_detach, uipc_disconnect, 425 uipc_listen, uipc_peeraddr, uipc_rcvd, pru_rcvoob_notsupp, 426 uipc_send, uipc_sense, uipc_shutdown, uipc_sockaddr, 427 sosend, soreceive, sopoll 428}; 429 430/* 431 * Both send and receive buffers are allocated PIPSIZ bytes of buffering 432 * for stream sockets, although the total for sender and receiver is 433 * actually only PIPSIZ. 434 * Datagram sockets really use the sendspace as the maximum datagram size, 435 * and don't really want to reserve the sendspace. Their recvspace should 436 * be large enough for at least one max-size datagram plus address. 437 */ 438#ifndef PIPSIZ 439#define PIPSIZ 8192 440#endif 441static u_long unpst_sendspace = PIPSIZ; 442static u_long unpst_recvspace = PIPSIZ; 443static u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 444static u_long unpdg_recvspace = 4*1024; 445 446static int unp_rights; /* file descriptors in flight */ 447 448SYSCTL_DECL(_net_local_stream); 449SYSCTL_INT(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 450 &unpst_sendspace, 0, ""); 451SYSCTL_INT(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 452 &unpst_recvspace, 0, ""); 453SYSCTL_DECL(_net_local_dgram); 454SYSCTL_INT(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 455 &unpdg_sendspace, 0, ""); 456SYSCTL_INT(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 457 &unpdg_recvspace, 0, ""); 458SYSCTL_DECL(_net_local); 459SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, ""); 460 461static int 462unp_attach(so) 463 struct socket *so; 464{ 465 register struct unpcb *unp; 466 int error; 467 468 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 469 switch (so->so_type) { 470 471 case SOCK_STREAM: 472 error = soreserve(so, unpst_sendspace, unpst_recvspace); 473 break; 474 475 case SOCK_DGRAM: 476 error = soreserve(so, unpdg_sendspace, unpdg_recvspace); 477 break; 478 479 default: 480 panic("unp_attach"); 481 } 482 if (error) 483 return (error); 484 } 485 unp = zalloc(unp_zone); 486 if (unp == NULL) 487 return (ENOBUFS); 488 bzero(unp, sizeof *unp); 489 unp->unp_gencnt = ++unp_gencnt; 490 unp_count++; 491 LIST_INIT(&unp->unp_refs); 492 unp->unp_socket = so; 493 LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead 494 : &unp_shead, unp, unp_link); 495 so->so_pcb = (caddr_t)unp; 496 return (0); 497} 498 499static void 500unp_detach(unp) 501 register struct unpcb *unp; 502{ 503 LIST_REMOVE(unp, unp_link); 504 unp->unp_gencnt = ++unp_gencnt; 505 --unp_count; 506 if (unp->unp_vnode) { 507 unp->unp_vnode->v_socket = 0; 508 vrele(unp->unp_vnode); 509 unp->unp_vnode = 0; 510 } 511 if (unp->unp_conn) 512 unp_disconnect(unp); 513 while (unp->unp_refs.lh_first) 514 unp_drop(unp->unp_refs.lh_first, ECONNRESET); 515 soisdisconnected(unp->unp_socket); 516 unp->unp_socket->so_pcb = 0; 517 if (unp_rights) { 518 /* 519 * Normally the receive buffer is flushed later, 520 * in sofree, but if our receive buffer holds references 521 * to descriptors that are now garbage, we will dispose 522 * of those descriptor references after the garbage collector 523 * gets them (resulting in a "panic: closef: count < 0"). 524 */ 525 sorflush(unp->unp_socket); 526 unp_gc(); 527 } 528 if (unp->unp_addr) 529 FREE(unp->unp_addr, M_SONAME); 530 zfree(unp_zone, unp); 531} 532 533static int 534unp_bind(unp, nam, p) 535 struct unpcb *unp; 536 struct sockaddr *nam; 537 struct proc *p; 538{ 539 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 540 register struct vnode *vp; 541 struct vattr vattr; 542 int error, namelen; 543 struct nameidata nd; 544 char buf[SOCK_MAXADDRLEN]; 545 546 if (unp->unp_vnode != NULL) 547 return (EINVAL); 548#define offsetof(s, e) ((char *)&((s *)0)->e - (char *)((s *)0)) 549 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 550 if (namelen <= 0) 551 return EINVAL; 552 strncpy(buf, soun->sun_path, namelen); 553 buf[namelen] = 0; /* null-terminate the string */ 554 NDINIT(&nd, CREATE, FOLLOW | LOCKPARENT, UIO_SYSSPACE, 555 buf, p); 556/* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 557 error = namei(&nd); 558 if (error) 559 return (error); 560 vp = nd.ni_vp; 561 if (vp != NULL) { 562 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 563 if (nd.ni_dvp == vp) 564 vrele(nd.ni_dvp); 565 else 566 vput(nd.ni_dvp); 567 vrele(vp); 568 return (EADDRINUSE); 569 } 570 VATTR_NULL(&vattr); 571 vattr.va_type = VSOCK; 572 vattr.va_mode = (ACCESSPERMS & ~p->p_fd->fd_cmask); 573 VOP_LEASE(nd.ni_dvp, p, p->p_ucred, LEASE_WRITE); 574 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 575 vput(nd.ni_dvp); 576 if (error) 577 return (error); 578 vp = nd.ni_vp; 579 vp->v_socket = unp->unp_socket; 580 unp->unp_vnode = vp; 581 unp->unp_addr = (struct sockaddr_un *)dup_sockaddr(nam, 1); 582 VOP_UNLOCK(vp, 0, p); 583 return (0); 584} 585 586static int 587unp_connect(so, nam, p) 588 struct socket *so; 589 struct sockaddr *nam; 590 struct proc *p; 591{ 592 register struct sockaddr_un *soun = (struct sockaddr_un *)nam; 593 register struct vnode *vp; 594 register struct socket *so2, *so3; 595 struct unpcb *unp2, *unp3; 596 int error, len; 597 struct nameidata nd; 598 char buf[SOCK_MAXADDRLEN]; 599 600 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 601 if (len <= 0) 602 return EINVAL; 603 strncpy(buf, soun->sun_path, len); 604 buf[len] = 0; 605 606 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, p); 607 error = namei(&nd); 608 if (error) 609 return (error); 610 vp = nd.ni_vp; 611 if (vp->v_type != VSOCK) { 612 error = ENOTSOCK; 613 goto bad; 614 } 615 error = VOP_ACCESS(vp, VWRITE, p->p_ucred, p); 616 if (error) 617 goto bad; 618 so2 = vp->v_socket; 619 if (so2 == 0) { 620 error = ECONNREFUSED; 621 goto bad; 622 } 623 if (so->so_type != so2->so_type) { 624 error = EPROTOTYPE; 625 goto bad; 626 } 627 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 628 if ((so2->so_options & SO_ACCEPTCONN) == 0 || 629 (so3 = sonewconn(so2, 0)) == 0) { 630 error = ECONNREFUSED; 631 goto bad; 632 } 633 unp2 = sotounpcb(so2); 634 unp3 = sotounpcb(so3); 635 if (unp2->unp_addr) 636 unp3->unp_addr = (struct sockaddr_un *) 637 dup_sockaddr((struct sockaddr *) 638 unp2->unp_addr, 1); 639 so2 = so3; 640 } 641 error = unp_connect2(so, so2); 642bad: 643 vput(vp); 644 return (error); 645} 646 647int 648unp_connect2(so, so2) 649 register struct socket *so; 650 register struct socket *so2; 651{ 652 register struct unpcb *unp = sotounpcb(so); 653 register struct unpcb *unp2; 654 655 if (so2->so_type != so->so_type) 656 return (EPROTOTYPE); 657 unp2 = sotounpcb(so2); 658 unp->unp_conn = unp2; 659 switch (so->so_type) { 660 661 case SOCK_DGRAM: 662 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 663 soisconnected(so); 664 break; 665 666 case SOCK_STREAM: 667 unp2->unp_conn = unp; 668 soisconnected(so); 669 soisconnected(so2); 670 break; 671 672 default: 673 panic("unp_connect2"); 674 } 675 return (0); 676} 677 678static void 679unp_disconnect(unp) 680 struct unpcb *unp; 681{ 682 register struct unpcb *unp2 = unp->unp_conn; 683 684 if (unp2 == 0) 685 return; 686 unp->unp_conn = 0; 687 switch (unp->unp_socket->so_type) { 688 689 case SOCK_DGRAM: 690 LIST_REMOVE(unp, unp_reflink); 691 unp->unp_socket->so_state &= ~SS_ISCONNECTED; 692 break; 693 694 case SOCK_STREAM: 695 soisdisconnected(unp->unp_socket); 696 unp2->unp_conn = 0; 697 soisdisconnected(unp2->unp_socket); 698 break; 699 } 700} 701 702#ifdef notdef 703void 704unp_abort(unp) 705 struct unpcb *unp; 706{ 707 708 unp_detach(unp); 709} 710#endif 711 712static int 713unp_pcblist SYSCTL_HANDLER_ARGS 714{ 715 int error, i, n; 716 struct unpcb *unp, **unp_list; 717 unp_gen_t gencnt; 718 struct xunpgen xug; 719 struct unp_head *head; 720 721 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead); 722 723 /* 724 * The process of preparing the PCB list is too time-consuming and 725 * resource-intensive to repeat twice on every request. 726 */ 727 if (req->oldptr == 0) { 728 n = unp_count; 729 req->oldidx = 2 * (sizeof xug) 730 + (n + n/8) * sizeof(struct xunpcb); 731 return 0; 732 } 733 734 if (req->newptr != 0) 735 return EPERM; 736 737 /* 738 * OK, now we're committed to doing something. 739 */ 740 gencnt = unp_gencnt; 741 n = unp_count; 742 743 xug.xug_len = sizeof xug; 744 xug.xug_count = n; 745 xug.xug_gen = gencnt; 746 xug.xug_sogen = so_gencnt; 747 error = SYSCTL_OUT(req, &xug, sizeof xug); 748 if (error) 749 return error; 750 751 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 752 if (unp_list == 0) 753 return ENOMEM; 754 755 for (unp = head->lh_first, i = 0; unp && i < n; 756 unp = unp->unp_link.le_next) { 757 if (unp->unp_gencnt <= gencnt) 758 unp_list[i++] = unp; 759 } 760 n = i; /* in case we lost some during malloc */ 761 762 error = 0; 763 for (i = 0; i < n; i++) { 764 unp = unp_list[i]; 765 if (unp->unp_gencnt <= gencnt) { 766 struct xunpcb xu; 767 xu.xu_len = sizeof xu; 768 xu.xu_unpp = unp; 769 /* 770 * XXX - need more locking here to protect against 771 * connect/disconnect races for SMP. 772 */ 773 if (unp->unp_addr) 774 bcopy(unp->unp_addr, &xu.xu_addr, 775 unp->unp_addr->sun_len); 776 if (unp->unp_conn && unp->unp_conn->unp_addr) 777 bcopy(unp->unp_conn->unp_addr, 778 &xu.xu_caddr, 779 unp->unp_conn->unp_addr->sun_len); 780 bcopy(unp, &xu.xu_unp, sizeof *unp); 781 sotoxsocket(unp->unp_socket, &xu.xu_socket); 782 error = SYSCTL_OUT(req, &xu, sizeof xu); 783 } 784 } 785 if (!error) { 786 /* 787 * Give the user an updated idea of our state. 788 * If the generation differs from what we told 789 * her before, she knows that something happened 790 * while we were processing this request, and it 791 * might be necessary to retry. 792 */ 793 xug.xug_gen = unp_gencnt; 794 xug.xug_sogen = so_gencnt; 795 xug.xug_count = unp_count; 796 error = SYSCTL_OUT(req, &xug, sizeof xug); 797 } 798 free(unp_list, M_TEMP); 799 return error; 800} 801 802SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD, 803 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 804 "List of active local datagram sockets"); 805SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD, 806 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 807 "List of active local stream sockets"); 808 809static void 810unp_shutdown(unp) 811 struct unpcb *unp; 812{ 813 struct socket *so; 814 815 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn && 816 (so = unp->unp_conn->unp_socket)) 817 socantrcvmore(so); 818} 819 820static void 821unp_drop(unp, errno) 822 struct unpcb *unp; 823 int errno; 824{ 825 struct socket *so = unp->unp_socket; 826 827 so->so_error = errno; 828 unp_disconnect(unp); 829 if (so->so_head) { 830 LIST_REMOVE(unp, unp_link); 831 unp->unp_gencnt = ++unp_gencnt; 832 unp_count--; 833 so->so_pcb = (caddr_t) 0; 834 if (unp->unp_addr) 835 FREE(unp->unp_addr, M_SONAME); 836 zfree(unp_zone, unp); 837 sofree(so); 838 } 839} 840 841#ifdef notdef 842void 843unp_drain() 844{ 845 846} 847#endif 848 849int 850unp_externalize(rights) 851 struct mbuf *rights; 852{ 853 struct proc *p = curproc; /* XXX */ 854 register int i; 855 register struct cmsghdr *cm = mtod(rights, struct cmsghdr *); 856 register struct file **rp = (struct file **)(cm + 1); 857 register struct file *fp; 858 int newfds = (cm->cmsg_len - sizeof(*cm)) / sizeof (int); 859 int f; 860 861 /* 862 * if the new FD's will not fit, then we free them all 863 */ 864 if (!fdavail(p, newfds)) { 865 for (i = 0; i < newfds; i++) { 866 fp = *rp; 867 unp_discard(fp); 868 *rp++ = 0; 869 } 870 return (EMSGSIZE); 871 } 872 /* 873 * now change each pointer to an fd in the global table to 874 * an integer that is the index to the local fd table entry 875 * that we set up to point to the global one we are transferring. 876 * XXX this assumes a pointer and int are the same size...! 877 */ 878 for (i = 0; i < newfds; i++) { 879 if (fdalloc(p, 0, &f)) 880 panic("unp_externalize"); 881 fp = *rp; 882 p->p_fd->fd_ofiles[f] = fp; 883 fp->f_msgcount--; 884 unp_rights--; 885 *(int *)rp++ = f; 886 } 887 return (0); 888} 889 890void 891unp_init(void) 892{ 893 unp_zone = zinit("unpcb", sizeof(struct unpcb), nmbclusters, 0, 0); 894 if (unp_zone == 0) 895 panic("unp_init"); 896 LIST_INIT(&unp_dhead); 897 LIST_INIT(&unp_shead); 898} 899 900#ifndef MIN 901#define MIN(a,b) (((a)<(b))?(a):(b)) 902#endif 903 904static int 905unp_internalize(control, p) 906 struct mbuf *control; 907 struct proc *p; 908{ 909 struct filedesc *fdp = p->p_fd; 910 register struct cmsghdr *cm = mtod(control, struct cmsghdr *); 911 register struct file **rp; 912 register struct file *fp; 913 register int i, fd; 914 register struct cmsgcred *cmcred; 915 int oldfds; 916 917 if ((cm->cmsg_type != SCM_RIGHTS && cm->cmsg_type != SCM_CREDS) || 918 cm->cmsg_level != SOL_SOCKET || cm->cmsg_len != control->m_len) 919 return (EINVAL); 920 921 /* 922 * Fill in credential information. 923 */ 924 if (cm->cmsg_type == SCM_CREDS) { 925 cmcred = (struct cmsgcred *)(cm + 1); 926 cmcred->cmcred_pid = p->p_pid; 927 cmcred->cmcred_uid = p->p_cred->p_ruid; 928 cmcred->cmcred_gid = p->p_cred->p_rgid; 929 cmcred->cmcred_euid = p->p_ucred->cr_uid; 930 cmcred->cmcred_ngroups = MIN(p->p_ucred->cr_ngroups, 931 CMGROUP_MAX); 932 for (i = 0; i < cmcred->cmcred_ngroups; i++) 933 cmcred->cmcred_groups[i] = p->p_ucred->cr_groups[i]; 934 return(0); 935 } 936 937 oldfds = (cm->cmsg_len - sizeof (*cm)) / sizeof (int); 938 /* 939 * check that all the FDs passed in refer to legal OPEN files 940 * If not, reject the entire operation. 941 */ 942 rp = (struct file **)(cm + 1); 943 for (i = 0; i < oldfds; i++) { 944 fd = *(int *)rp++; 945 if ((unsigned)fd >= fdp->fd_nfiles || 946 fdp->fd_ofiles[fd] == NULL) 947 return (EBADF); 948 } 949 /* 950 * Now replace the integer FDs with pointers to 951 * the associated global file table entry.. 952 * XXX this assumes a pointer and an int are the same size! 953 */ 954 rp = (struct file **)(cm + 1); 955 for (i = 0; i < oldfds; i++) { 956 fp = fdp->fd_ofiles[*(int *)rp]; 957 *rp++ = fp; 958 fp->f_count++; 959 fp->f_msgcount++; 960 unp_rights++; 961 } 962 return (0); 963} 964 965static int unp_defer, unp_gcing; 966 967static void 968unp_gc() 969{ 970 register struct file *fp, *nextfp; 971 register struct socket *so; 972 struct file **extra_ref, **fpp; 973 int nunref, i; 974 975 if (unp_gcing) 976 return; 977 unp_gcing = 1; 978 unp_defer = 0; 979 /* 980 * before going through all this, set all FDs to 981 * be NOT defered and NOT externally accessible 982 */ 983 for (fp = filehead.lh_first; fp != 0; fp = fp->f_list.le_next) 984 fp->f_flag &= ~(FMARK|FDEFER); 985 do { 986 for (fp = filehead.lh_first; fp != 0; fp = fp->f_list.le_next) { 987 /* 988 * If the file is not open, skip it 989 */ 990 if (fp->f_count == 0) 991 continue; 992 /* 993 * If we already marked it as 'defer' in a 994 * previous pass, then try process it this time 995 * and un-mark it 996 */ 997 if (fp->f_flag & FDEFER) { 998 fp->f_flag &= ~FDEFER; 999 unp_defer--; 1000 } else { 1001 /* 1002 * if it's not defered, then check if it's 1003 * already marked.. if so skip it 1004 */ 1005 if (fp->f_flag & FMARK) 1006 continue; 1007 /* 1008 * If all references are from messages 1009 * in transit, then skip it. it's not 1010 * externally accessible. 1011 */ 1012 if (fp->f_count == fp->f_msgcount) 1013 continue; 1014 /* 1015 * If it got this far then it must be 1016 * externally accessible. 1017 */ 1018 fp->f_flag |= FMARK; 1019 } 1020 /* 1021 * either it was defered, or it is externally 1022 * accessible and not already marked so. 1023 * Now check if it is possibly one of OUR sockets. 1024 */ 1025 if (fp->f_type != DTYPE_SOCKET || 1026 (so = (struct socket *)fp->f_data) == 0) 1027 continue; 1028 if (so->so_proto->pr_domain != &localdomain || 1029 (so->so_proto->pr_flags&PR_RIGHTS) == 0) 1030 continue; 1031#ifdef notdef 1032 if (so->so_rcv.sb_flags & SB_LOCK) { 1033 /* 1034 * This is problematical; it's not clear 1035 * we need to wait for the sockbuf to be 1036 * unlocked (on a uniprocessor, at least), 1037 * and it's also not clear what to do 1038 * if sbwait returns an error due to receipt 1039 * of a signal. If sbwait does return 1040 * an error, we'll go into an infinite 1041 * loop. Delete all of this for now. 1042 */ 1043 (void) sbwait(&so->so_rcv); 1044 goto restart; 1045 } 1046#endif 1047 /* 1048 * So, Ok, it's one of our sockets and it IS externally 1049 * accessible (or was defered). Now we look 1050 * to see if we hold any file descriptors in its 1051 * message buffers. Follow those links and mark them 1052 * as accessible too. 1053 */ 1054 unp_scan(so->so_rcv.sb_mb, unp_mark); 1055 } 1056 } while (unp_defer); 1057 /* 1058 * We grab an extra reference to each of the file table entries 1059 * that are not otherwise accessible and then free the rights 1060 * that are stored in messages on them. 1061 * 1062 * The bug in the orginal code is a little tricky, so I'll describe 1063 * what's wrong with it here. 1064 * 1065 * It is incorrect to simply unp_discard each entry for f_msgcount 1066 * times -- consider the case of sockets A and B that contain 1067 * references to each other. On a last close of some other socket, 1068 * we trigger a gc since the number of outstanding rights (unp_rights) 1069 * is non-zero. If during the sweep phase the gc code un_discards, 1070 * we end up doing a (full) closef on the descriptor. A closef on A 1071 * results in the following chain. Closef calls soo_close, which 1072 * calls soclose. Soclose calls first (through the switch 1073 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply 1074 * returns because the previous instance had set unp_gcing, and 1075 * we return all the way back to soclose, which marks the socket 1076 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush 1077 * to free up the rights that are queued in messages on the socket A, 1078 * i.e., the reference on B. The sorflush calls via the dom_dispose 1079 * switch unp_dispose, which unp_scans with unp_discard. This second 1080 * instance of unp_discard just calls closef on B. 1081 * 1082 * Well, a similar chain occurs on B, resulting in a sorflush on B, 1083 * which results in another closef on A. Unfortunately, A is already 1084 * being closed, and the descriptor has already been marked with 1085 * SS_NOFDREF, and soclose panics at this point. 1086 * 1087 * Here, we first take an extra reference to each inaccessible 1088 * descriptor. Then, we call sorflush ourself, since we know 1089 * it is a Unix domain socket anyhow. After we destroy all the 1090 * rights carried in messages, we do a last closef to get rid 1091 * of our extra reference. This is the last close, and the 1092 * unp_detach etc will shut down the socket. 1093 * 1094 * 91/09/19, bsy@cs.cmu.edu 1095 */ 1096 extra_ref = malloc(nfiles * sizeof(struct file *), M_FILE, M_WAITOK); 1097 for (nunref = 0, fp = filehead.lh_first, fpp = extra_ref; fp != 0; 1098 fp = nextfp) { 1099 nextfp = fp->f_list.le_next; 1100 /* 1101 * If it's not open, skip it 1102 */ 1103 if (fp->f_count == 0) 1104 continue; 1105 /* 1106 * If all refs are from msgs, and it's not marked accessible 1107 * then it must be referenced from some unreachable cycle 1108 * of (shut-down) FDs, so include it in our 1109 * list of FDs to remove 1110 */ 1111 if (fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) { 1112 *fpp++ = fp; 1113 nunref++; 1114 fp->f_count++; 1115 } 1116 } 1117 /* 1118 * for each FD on our hit list, do the following two things 1119 */ 1120 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1121 struct file *tfp = *fpp; 1122 if (tfp->f_type == DTYPE_SOCKET && tfp->f_data != NULL) 1123 sorflush((struct socket *)(tfp->f_data)); 1124 } 1125 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) 1126 closef(*fpp, (struct proc *) NULL); 1127 free((caddr_t)extra_ref, M_FILE); 1128 unp_gcing = 0; 1129} 1130 1131void 1132unp_dispose(m) 1133 struct mbuf *m; 1134{ 1135 1136 if (m) 1137 unp_scan(m, unp_discard); 1138} 1139 1140static void 1141unp_scan(m0, op) 1142 register struct mbuf *m0; 1143 void (*op) __P((struct file *)); 1144{ 1145 register struct mbuf *m; 1146 register struct file **rp; 1147 register struct cmsghdr *cm; 1148 register int i; 1149 int qfds; 1150 1151 while (m0) { 1152 for (m = m0; m; m = m->m_next) 1153 if (m->m_type == MT_CONTROL && 1154 m->m_len >= sizeof(*cm)) { 1155 cm = mtod(m, struct cmsghdr *); 1156 if (cm->cmsg_level != SOL_SOCKET || 1157 cm->cmsg_type != SCM_RIGHTS) 1158 continue; 1159 qfds = (cm->cmsg_len - sizeof *cm) 1160 / sizeof (struct file *); 1161 rp = (struct file **)(cm + 1); 1162 for (i = 0; i < qfds; i++) 1163 (*op)(*rp++); 1164 break; /* XXX, but saves time */ 1165 } 1166 m0 = m0->m_act; 1167 } 1168} 1169 1170static void 1171unp_mark(fp) 1172 struct file *fp; 1173{ 1174 1175 if (fp->f_flag & FMARK) 1176 return; 1177 unp_defer++; 1178 fp->f_flag |= (FMARK|FDEFER); 1179} 1180 1181static void 1182unp_discard(fp) 1183 struct file *fp; 1184{ 1185 1186 fp->f_msgcount--; 1187 unp_rights--; 1188 (void) closef(fp, (struct proc *)NULL); 1189} 1190