uipc_usrreq.c revision 130387
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 130387 2004-06-12 20:47:32Z 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 /* 284 * Adjust backpressure on sender 285 * and wakeup any waiting to write. 286 */ 287 so2->so_snd.sb_mbmax += unp->unp_mbcnt - so->so_rcv.sb_mbcnt; 288 unp->unp_mbcnt = so->so_rcv.sb_mbcnt; 289 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - 290 so->so_rcv.sb_cc; 291 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat, 292 newhiwat, RLIM_INFINITY); 293 unp->unp_cc = so->so_rcv.sb_cc; 294 sowwakeup(so2); 295 break; 296 297 default: 298 panic("uipc_rcvd unknown socktype"); 299 } 300 UNP_UNLOCK(); 301 return (0); 302} 303 304/* pru_rcvoob is EOPNOTSUPP */ 305 306static int 307uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 308 struct mbuf *control, struct thread *td) 309{ 310 int error = 0; 311 struct unpcb *unp = sotounpcb(so); 312 struct socket *so2; 313 u_long newhiwat; 314 315 if (unp == NULL) { 316 error = EINVAL; 317 goto release; 318 } 319 if (flags & PRUS_OOB) { 320 error = EOPNOTSUPP; 321 goto release; 322 } 323 324 if (control != NULL && (error = unp_internalize(&control, td))) 325 goto release; 326 327 UNP_LOCK(); 328 switch (so->so_type) { 329 case SOCK_DGRAM: 330 { 331 const struct sockaddr *from; 332 333 if (nam != NULL) { 334 if (unp->unp_conn != NULL) { 335 error = EISCONN; 336 break; 337 } 338 error = unp_connect(so, nam, td); 339 if (error) 340 break; 341 } else { 342 if (unp->unp_conn == NULL) { 343 error = ENOTCONN; 344 break; 345 } 346 } 347 so2 = unp->unp_conn->unp_socket; 348 if (unp->unp_addr != NULL) 349 from = (struct sockaddr *)unp->unp_addr; 350 else 351 from = &sun_noname; 352 if (sbappendaddr(&so2->so_rcv, from, m, control)) { 353 sorwakeup(so2); 354 m = NULL; 355 control = NULL; 356 } else { 357 error = ENOBUFS; 358 } 359 if (nam != NULL) 360 unp_disconnect(unp); 361 break; 362 } 363 364 case SOCK_STREAM: 365 /* Connect if not connected yet. */ 366 /* 367 * Note: A better implementation would complain 368 * if not equal to the peer's address. 369 */ 370 if ((so->so_state & SS_ISCONNECTED) == 0) { 371 if (nam != NULL) { 372 error = unp_connect(so, nam, td); 373 if (error) 374 break; /* XXX */ 375 } else { 376 error = ENOTCONN; 377 break; 378 } 379 } 380 381 if (so->so_state & SS_CANTSENDMORE) { 382 error = EPIPE; 383 break; 384 } 385 if (unp->unp_conn == NULL) 386 panic("uipc_send connected but no connection?"); 387 so2 = unp->unp_conn->unp_socket; 388 /* 389 * Send to paired receive port, and then reduce 390 * send buffer hiwater marks to maintain backpressure. 391 * Wake up readers. 392 */ 393 if (control != NULL) { 394 if (sbappendcontrol(&so2->so_rcv, m, control)) 395 control = NULL; 396 } else { 397 sbappend(&so2->so_rcv, m); 398 } 399 so->so_snd.sb_mbmax -= 400 so2->so_rcv.sb_mbcnt - unp->unp_conn->unp_mbcnt; 401 unp->unp_conn->unp_mbcnt = so2->so_rcv.sb_mbcnt; 402 newhiwat = so->so_snd.sb_hiwat - 403 (so2->so_rcv.sb_cc - unp->unp_conn->unp_cc); 404 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat, 405 newhiwat, RLIM_INFINITY); 406 unp->unp_conn->unp_cc = so2->so_rcv.sb_cc; 407 sorwakeup(so2); 408 m = NULL; 409 break; 410 411 default: 412 panic("uipc_send unknown socktype"); 413 } 414 415 /* 416 * SEND_EOF is equivalent to a SEND followed by 417 * a SHUTDOWN. 418 */ 419 if (flags & PRUS_EOF) { 420 socantsendmore(so); 421 unp_shutdown(unp); 422 } 423 UNP_UNLOCK(); 424 425 if (control != NULL && error != 0) 426 unp_dispose(control); 427 428release: 429 if (control != NULL) 430 m_freem(control); 431 if (m != NULL) 432 m_freem(m); 433 return (error); 434} 435 436static int 437uipc_sense(struct socket *so, struct stat *sb) 438{ 439 struct unpcb *unp = sotounpcb(so); 440 struct socket *so2; 441 442 if (unp == NULL) 443 return (EINVAL); 444 UNP_LOCK(); 445 sb->st_blksize = so->so_snd.sb_hiwat; 446 if (so->so_type == SOCK_STREAM && unp->unp_conn != NULL) { 447 so2 = unp->unp_conn->unp_socket; 448 sb->st_blksize += so2->so_rcv.sb_cc; 449 } 450 sb->st_dev = NOUDEV; 451 if (unp->unp_ino == 0) 452 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino; 453 sb->st_ino = unp->unp_ino; 454 UNP_UNLOCK(); 455 return (0); 456} 457 458static int 459uipc_shutdown(struct socket *so) 460{ 461 struct unpcb *unp = sotounpcb(so); 462 463 if (unp == NULL) 464 return (EINVAL); 465 UNP_LOCK(); 466 socantsendmore(so); 467 unp_shutdown(unp); 468 UNP_UNLOCK(); 469 return (0); 470} 471 472static int 473uipc_sockaddr(struct socket *so, struct sockaddr **nam) 474{ 475 struct unpcb *unp = sotounpcb(so); 476 const struct sockaddr *sa; 477 478 if (unp == NULL) 479 return (EINVAL); 480 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 481 UNP_LOCK(); 482 if (unp->unp_addr != NULL) 483 sa = (struct sockaddr *) unp->unp_addr; 484 else 485 sa = &sun_noname; 486 bcopy(sa, *nam, sa->sa_len); 487 UNP_UNLOCK(); 488 return (0); 489} 490 491struct pr_usrreqs uipc_usrreqs = { 492 uipc_abort, uipc_accept, uipc_attach, uipc_bind, uipc_connect, 493 uipc_connect2, pru_control_notsupp, uipc_detach, uipc_disconnect, 494 uipc_listen, uipc_peeraddr, uipc_rcvd, pru_rcvoob_notsupp, 495 uipc_send, uipc_sense, uipc_shutdown, uipc_sockaddr, 496 sosend, soreceive, sopoll, pru_sosetlabel_null 497}; 498 499int 500uipc_ctloutput(so, sopt) 501 struct socket *so; 502 struct sockopt *sopt; 503{ 504 struct unpcb *unp = sotounpcb(so); 505 struct xucred xu; 506 int error; 507 508 switch (sopt->sopt_dir) { 509 case SOPT_GET: 510 switch (sopt->sopt_name) { 511 case LOCAL_PEERCRED: 512 error = 0; 513 UNP_LOCK(); 514 if (unp->unp_flags & UNP_HAVEPC) 515 xu = unp->unp_peercred; 516 else { 517 if (so->so_type == SOCK_STREAM) 518 error = ENOTCONN; 519 else 520 error = EINVAL; 521 } 522 UNP_UNLOCK(); 523 if (error == 0) 524 error = sooptcopyout(sopt, &xu, sizeof(xu)); 525 break; 526 default: 527 error = EOPNOTSUPP; 528 break; 529 } 530 break; 531 case SOPT_SET: 532 default: 533 error = EOPNOTSUPP; 534 break; 535 } 536 return (error); 537} 538 539/* 540 * Both send and receive buffers are allocated PIPSIZ bytes of buffering 541 * for stream sockets, although the total for sender and receiver is 542 * actually only PIPSIZ. 543 * Datagram sockets really use the sendspace as the maximum datagram size, 544 * and don't really want to reserve the sendspace. Their recvspace should 545 * be large enough for at least one max-size datagram plus address. 546 */ 547#ifndef PIPSIZ 548#define PIPSIZ 8192 549#endif 550static u_long unpst_sendspace = PIPSIZ; 551static u_long unpst_recvspace = PIPSIZ; 552static u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 553static u_long unpdg_recvspace = 4*1024; 554 555static int unp_rights; /* file descriptors in flight */ 556 557SYSCTL_DECL(_net_local_stream); 558SYSCTL_INT(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 559 &unpst_sendspace, 0, ""); 560SYSCTL_INT(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 561 &unpst_recvspace, 0, ""); 562SYSCTL_DECL(_net_local_dgram); 563SYSCTL_INT(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 564 &unpdg_sendspace, 0, ""); 565SYSCTL_INT(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 566 &unpdg_recvspace, 0, ""); 567SYSCTL_DECL(_net_local); 568SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, ""); 569 570static int 571unp_attach(so) 572 struct socket *so; 573{ 574 register struct unpcb *unp; 575 int error; 576 577 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 578 switch (so->so_type) { 579 580 case SOCK_STREAM: 581 error = soreserve(so, unpst_sendspace, unpst_recvspace); 582 break; 583 584 case SOCK_DGRAM: 585 error = soreserve(so, unpdg_sendspace, unpdg_recvspace); 586 break; 587 588 default: 589 panic("unp_attach"); 590 } 591 if (error) 592 return (error); 593 } 594 unp = uma_zalloc(unp_zone, M_WAITOK); 595 if (unp == NULL) 596 return (ENOBUFS); 597 bzero(unp, sizeof *unp); 598 LIST_INIT(&unp->unp_refs); 599 unp->unp_socket = so; 600 601 UNP_LOCK(); 602 unp->unp_gencnt = ++unp_gencnt; 603 unp_count++; 604 LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead 605 : &unp_shead, unp, unp_link); 606 UNP_UNLOCK(); 607 608 so->so_pcb = unp; 609 return (0); 610} 611 612static void 613unp_detach(unp) 614 register struct unpcb *unp; 615{ 616 struct vnode *vp; 617 618 UNP_LOCK_ASSERT(); 619 620 LIST_REMOVE(unp, unp_link); 621 unp->unp_gencnt = ++unp_gencnt; 622 --unp_count; 623 if ((vp = unp->unp_vnode) != NULL) { 624 /* 625 * XXXRW: should v_socket be frobbed only while holding 626 * Giant? 627 */ 628 unp->unp_vnode->v_socket = NULL; 629 unp->unp_vnode = NULL; 630 } 631 if (unp->unp_conn != NULL) 632 unp_disconnect(unp); 633 while (!LIST_EMPTY(&unp->unp_refs)) { 634 struct unpcb *ref = LIST_FIRST(&unp->unp_refs); 635 unp_drop(ref, ECONNRESET); 636 } 637 soisdisconnected(unp->unp_socket); 638 unp->unp_socket->so_pcb = NULL; 639 if (unp_rights) { 640 /* 641 * Normally the receive buffer is flushed later, 642 * in sofree, but if our receive buffer holds references 643 * to descriptors that are now garbage, we will dispose 644 * of those descriptor references after the garbage collector 645 * gets them (resulting in a "panic: closef: count < 0"). 646 */ 647 sorflush(unp->unp_socket); 648 unp_gc(); 649 } 650 if (unp->unp_addr != NULL) 651 FREE(unp->unp_addr, M_SONAME); 652 UNP_UNLOCK(); 653 uma_zfree(unp_zone, unp); 654 if (vp) { 655 mtx_lock(&Giant); 656 vrele(vp); 657 mtx_unlock(&Giant); 658 } 659} 660 661static int 662unp_bind(unp, nam, td) 663 struct unpcb *unp; 664 struct sockaddr *nam; 665 struct thread *td; 666{ 667 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 668 struct vnode *vp; 669 struct mount *mp; 670 struct vattr vattr; 671 int error, namelen; 672 struct nameidata nd; 673 char *buf; 674 675 /* 676 * XXXRW: This test-and-set of unp_vnode is non-atomic; the 677 * unlocked read here is fine, but the value of unp_vnode needs 678 * to be tested again after we do all the lookups to see if the 679 * pcb is still unbound? 680 */ 681 if (unp->unp_vnode != NULL) 682 return (EINVAL); 683 684 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 685 if (namelen <= 0) 686 return (EINVAL); 687 688 buf = malloc(namelen + 1, M_TEMP, M_WAITOK); 689 strlcpy(buf, soun->sun_path, namelen + 1); 690 691 mtx_lock(&Giant); 692restart: 693 mtx_assert(&Giant, MA_OWNED); 694 NDINIT(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME, UIO_SYSSPACE, 695 buf, td); 696/* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 697 error = namei(&nd); 698 if (error) 699 goto done; 700 vp = nd.ni_vp; 701 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { 702 NDFREE(&nd, NDF_ONLY_PNBUF); 703 if (nd.ni_dvp == vp) 704 vrele(nd.ni_dvp); 705 else 706 vput(nd.ni_dvp); 707 if (vp != NULL) { 708 vrele(vp); 709 error = EADDRINUSE; 710 goto done; 711 } 712 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH); 713 if (error) 714 goto done; 715 goto restart; 716 } 717 VATTR_NULL(&vattr); 718 vattr.va_type = VSOCK; 719 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask); 720#ifdef MAC 721 error = mac_check_vnode_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, 722 &vattr); 723#endif 724 if (error == 0) { 725 VOP_LEASE(nd.ni_dvp, td, td->td_ucred, LEASE_WRITE); 726 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 727 } 728 NDFREE(&nd, NDF_ONLY_PNBUF); 729 vput(nd.ni_dvp); 730 if (error) 731 goto done; 732 vp = nd.ni_vp; 733 ASSERT_VOP_LOCKED(vp, "unp_bind"); 734 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK); 735 UNP_LOCK(); 736 vp->v_socket = unp->unp_socket; 737 unp->unp_vnode = vp; 738 unp->unp_addr = soun; 739 UNP_UNLOCK(); 740 VOP_UNLOCK(vp, 0, td); 741 vn_finished_write(mp); 742done: 743 mtx_unlock(&Giant); 744 free(buf, M_TEMP); 745 return (error); 746} 747 748static int 749unp_connect(so, nam, td) 750 struct socket *so; 751 struct sockaddr *nam; 752 struct thread *td; 753{ 754 register struct sockaddr_un *soun = (struct sockaddr_un *)nam; 755 register struct vnode *vp; 756 register struct socket *so2, *so3; 757 struct unpcb *unp = sotounpcb(so); 758 struct unpcb *unp2, *unp3; 759 int error, len; 760 struct nameidata nd; 761 char buf[SOCK_MAXADDRLEN]; 762 struct sockaddr *sa; 763 764 UNP_LOCK_ASSERT(); 765 766 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 767 if (len <= 0) 768 return (EINVAL); 769 strlcpy(buf, soun->sun_path, len + 1); 770 UNP_UNLOCK(); 771 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 772 mtx_lock(&Giant); 773 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, td); 774 error = namei(&nd); 775 if (error) 776 vp = NULL; 777 else 778 vp = nd.ni_vp; 779 ASSERT_VOP_LOCKED(vp, "unp_connect"); 780 NDFREE(&nd, NDF_ONLY_PNBUF); 781 if (error) 782 goto bad; 783 784 if (vp->v_type != VSOCK) { 785 error = ENOTSOCK; 786 goto bad; 787 } 788 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); 789 if (error) 790 goto bad; 791 so2 = vp->v_socket; 792 if (so2 == NULL) { 793 error = ECONNREFUSED; 794 goto bad; 795 } 796 if (so->so_type != so2->so_type) { 797 error = EPROTOTYPE; 798 goto bad; 799 } 800 mtx_unlock(&Giant); 801 UNP_LOCK(); 802 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 803 if (so2->so_options & SO_ACCEPTCONN) { 804 /* 805 * NB: drop locks here so unp_attach is entered 806 * w/o locks; this avoids a recursive lock 807 * of the head and holding sleep locks across 808 * a (potentially) blocking malloc. 809 */ 810 UNP_UNLOCK(); 811 so3 = sonewconn(so2, 0); 812 UNP_LOCK(); 813 } else 814 so3 = NULL; 815 if (so3 == NULL) { 816 error = ECONNREFUSED; 817 goto bad2; 818 } 819 unp = sotounpcb(so); 820 unp2 = sotounpcb(so2); 821 unp3 = sotounpcb(so3); 822 if (unp2->unp_addr != NULL) { 823 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); 824 unp3->unp_addr = (struct sockaddr_un *) sa; 825 sa = NULL; 826 } 827 /* 828 * unp_peercred management: 829 * 830 * The connecter's (client's) credentials are copied 831 * from its process structure at the time of connect() 832 * (which is now). 833 */ 834 cru2x(td->td_ucred, &unp3->unp_peercred); 835 unp3->unp_flags |= UNP_HAVEPC; 836 /* 837 * The receiver's (server's) credentials are copied 838 * from the unp_peercred member of socket on which the 839 * former called listen(); unp_listen() cached that 840 * process's credentials at that time so we can use 841 * them now. 842 */ 843 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED, 844 ("unp_connect: listener without cached peercred")); 845 memcpy(&unp->unp_peercred, &unp2->unp_peercred, 846 sizeof(unp->unp_peercred)); 847 unp->unp_flags |= UNP_HAVEPC; 848#ifdef MAC 849 mac_set_socket_peer_from_socket(so, so3); 850 mac_set_socket_peer_from_socket(so3, so); 851#endif 852 853 so2 = so3; 854 } 855 error = unp_connect2(so, so2); 856bad2: 857 UNP_UNLOCK(); 858 mtx_lock(&Giant); 859bad: 860 mtx_assert(&Giant, MA_OWNED); 861 if (vp != NULL) 862 vput(vp); 863 mtx_unlock(&Giant); 864 free(sa, M_SONAME); 865 UNP_LOCK(); 866 return (error); 867} 868 869static int 870unp_connect2(so, so2) 871 register struct socket *so; 872 register struct socket *so2; 873{ 874 register struct unpcb *unp = sotounpcb(so); 875 register struct unpcb *unp2; 876 877 UNP_LOCK_ASSERT(); 878 879 if (so2->so_type != so->so_type) 880 return (EPROTOTYPE); 881 unp2 = sotounpcb(so2); 882 unp->unp_conn = unp2; 883 switch (so->so_type) { 884 885 case SOCK_DGRAM: 886 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 887 soisconnected(so); 888 break; 889 890 case SOCK_STREAM: 891 unp2->unp_conn = unp; 892 soisconnected(so); 893 soisconnected(so2); 894 break; 895 896 default: 897 panic("unp_connect2"); 898 } 899 return (0); 900} 901 902static void 903unp_disconnect(unp) 904 struct unpcb *unp; 905{ 906 register struct unpcb *unp2 = unp->unp_conn; 907 908 UNP_LOCK_ASSERT(); 909 910 if (unp2 == NULL) 911 return; 912 unp->unp_conn = NULL; 913 switch (unp->unp_socket->so_type) { 914 915 case SOCK_DGRAM: 916 LIST_REMOVE(unp, unp_reflink); 917 unp->unp_socket->so_state &= ~SS_ISCONNECTED; 918 break; 919 920 case SOCK_STREAM: 921 soisdisconnected(unp->unp_socket); 922 unp2->unp_conn = NULL; 923 soisdisconnected(unp2->unp_socket); 924 break; 925 } 926} 927 928#ifdef notdef 929void 930unp_abort(unp) 931 struct unpcb *unp; 932{ 933 934 unp_detach(unp); 935} 936#endif 937 938/* 939 * unp_pcblist() assumes that UNIX domain socket memory is never reclaimed 940 * by the zone (UMA_ZONE_NOFREE), and as such potentially stale pointers 941 * are safe to reference. It first scans the list of struct unpcb's to 942 * generate a pointer list, then it rescans its list one entry at a time to 943 * externalize and copyout. It checks the generation number to see if a 944 * struct unpcb has been reused, and will skip it if so. 945 */ 946static int 947unp_pcblist(SYSCTL_HANDLER_ARGS) 948{ 949 int error, i, n; 950 struct unpcb *unp, **unp_list; 951 unp_gen_t gencnt; 952 struct xunpgen *xug; 953 struct unp_head *head; 954 struct xunpcb *xu; 955 956 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead); 957 958 /* 959 * The process of preparing the PCB list is too time-consuming and 960 * resource-intensive to repeat twice on every request. 961 */ 962 if (req->oldptr == NULL) { 963 n = unp_count; 964 req->oldidx = 2 * (sizeof *xug) 965 + (n + n/8) * sizeof(struct xunpcb); 966 return (0); 967 } 968 969 if (req->newptr != NULL) 970 return (EPERM); 971 972 /* 973 * OK, now we're committed to doing something. 974 */ 975 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK); 976 UNP_LOCK(); 977 gencnt = unp_gencnt; 978 n = unp_count; 979 UNP_UNLOCK(); 980 981 xug->xug_len = sizeof *xug; 982 xug->xug_count = n; 983 xug->xug_gen = gencnt; 984 xug->xug_sogen = so_gencnt; 985 error = SYSCTL_OUT(req, xug, sizeof *xug); 986 if (error) { 987 free(xug, M_TEMP); 988 return (error); 989 } 990 991 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 992 993 UNP_LOCK(); 994 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 995 unp = LIST_NEXT(unp, unp_link)) { 996 if (unp->unp_gencnt <= gencnt) { 997 if (cr_cansee(req->td->td_ucred, 998 unp->unp_socket->so_cred)) 999 continue; 1000 unp_list[i++] = unp; 1001 } 1002 } 1003 UNP_UNLOCK(); 1004 n = i; /* in case we lost some during malloc */ 1005 1006 error = 0; 1007 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK); 1008 for (i = 0; i < n; i++) { 1009 unp = unp_list[i]; 1010 if (unp->unp_gencnt <= gencnt) { 1011 xu->xu_len = sizeof *xu; 1012 xu->xu_unpp = unp; 1013 /* 1014 * XXX - need more locking here to protect against 1015 * connect/disconnect races for SMP. 1016 */ 1017 if (unp->unp_addr != NULL) 1018 bcopy(unp->unp_addr, &xu->xu_addr, 1019 unp->unp_addr->sun_len); 1020 if (unp->unp_conn != NULL && 1021 unp->unp_conn->unp_addr != NULL) 1022 bcopy(unp->unp_conn->unp_addr, 1023 &xu->xu_caddr, 1024 unp->unp_conn->unp_addr->sun_len); 1025 bcopy(unp, &xu->xu_unp, sizeof *unp); 1026 sotoxsocket(unp->unp_socket, &xu->xu_socket); 1027 error = SYSCTL_OUT(req, xu, sizeof *xu); 1028 } 1029 } 1030 free(xu, M_TEMP); 1031 if (!error) { 1032 /* 1033 * Give the user an updated idea of our state. 1034 * If the generation differs from what we told 1035 * her before, she knows that something happened 1036 * while we were processing this request, and it 1037 * might be necessary to retry. 1038 */ 1039 xug->xug_gen = unp_gencnt; 1040 xug->xug_sogen = so_gencnt; 1041 xug->xug_count = unp_count; 1042 error = SYSCTL_OUT(req, xug, sizeof *xug); 1043 } 1044 free(unp_list, M_TEMP); 1045 free(xug, M_TEMP); 1046 return (error); 1047} 1048 1049SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD, 1050 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 1051 "List of active local datagram sockets"); 1052SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD, 1053 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 1054 "List of active local stream sockets"); 1055 1056static void 1057unp_shutdown(unp) 1058 struct unpcb *unp; 1059{ 1060 struct socket *so; 1061 1062 UNP_LOCK_ASSERT(); 1063 1064 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn && 1065 (so = unp->unp_conn->unp_socket)) 1066 socantrcvmore(so); 1067} 1068 1069static void 1070unp_drop(unp, errno) 1071 struct unpcb *unp; 1072 int errno; 1073{ 1074 struct socket *so = unp->unp_socket; 1075 1076 UNP_LOCK_ASSERT(); 1077 1078 so->so_error = errno; 1079 unp_disconnect(unp); 1080} 1081 1082#ifdef notdef 1083void 1084unp_drain() 1085{ 1086 1087} 1088#endif 1089 1090static void 1091unp_freerights(rp, fdcount) 1092 struct file **rp; 1093 int fdcount; 1094{ 1095 int i; 1096 struct file *fp; 1097 1098 for (i = 0; i < fdcount; i++) { 1099 fp = *rp; 1100 /* 1101 * zero the pointer before calling 1102 * unp_discard since it may end up 1103 * in unp_gc().. 1104 */ 1105 *rp++ = 0; 1106 unp_discard(fp); 1107 } 1108} 1109 1110int 1111unp_externalize(control, controlp) 1112 struct mbuf *control, **controlp; 1113{ 1114 struct thread *td = curthread; /* XXX */ 1115 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1116 int i; 1117 int *fdp; 1118 struct file **rp; 1119 struct file *fp; 1120 void *data; 1121 socklen_t clen = control->m_len, datalen; 1122 int error, newfds; 1123 int f; 1124 u_int newlen; 1125 1126 error = 0; 1127 if (controlp != NULL) /* controlp == NULL => free control messages */ 1128 *controlp = NULL; 1129 1130 while (cm != NULL) { 1131 if (sizeof(*cm) > clen || cm->cmsg_len > clen) { 1132 error = EINVAL; 1133 break; 1134 } 1135 1136 data = CMSG_DATA(cm); 1137 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1138 1139 if (cm->cmsg_level == SOL_SOCKET 1140 && cm->cmsg_type == SCM_RIGHTS) { 1141 newfds = datalen / sizeof(struct file *); 1142 rp = data; 1143 1144 /* If we're not outputting the descriptors free them. */ 1145 if (error || controlp == NULL) { 1146 unp_freerights(rp, newfds); 1147 goto next; 1148 } 1149 FILEDESC_LOCK(td->td_proc->p_fd); 1150 /* if the new FD's will not fit free them. */ 1151 if (!fdavail(td, newfds)) { 1152 FILEDESC_UNLOCK(td->td_proc->p_fd); 1153 error = EMSGSIZE; 1154 unp_freerights(rp, newfds); 1155 goto next; 1156 } 1157 /* 1158 * now change each pointer to an fd in the global 1159 * table to an integer that is the index to the 1160 * local fd table entry that we set up to point 1161 * to the global one we are transferring. 1162 */ 1163 newlen = newfds * sizeof(int); 1164 *controlp = sbcreatecontrol(NULL, newlen, 1165 SCM_RIGHTS, SOL_SOCKET); 1166 if (*controlp == NULL) { 1167 FILEDESC_UNLOCK(td->td_proc->p_fd); 1168 error = E2BIG; 1169 unp_freerights(rp, newfds); 1170 goto next; 1171 } 1172 1173 fdp = (int *) 1174 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1175 for (i = 0; i < newfds; i++) { 1176 if (fdalloc(td, 0, &f)) 1177 panic("unp_externalize fdalloc failed"); 1178 fp = *rp++; 1179 td->td_proc->p_fd->fd_ofiles[f] = fp; 1180 FILE_LOCK(fp); 1181 fp->f_msgcount--; 1182 FILE_UNLOCK(fp); 1183 unp_rights--; 1184 *fdp++ = f; 1185 } 1186 FILEDESC_UNLOCK(td->td_proc->p_fd); 1187 } else { /* We can just copy anything else across */ 1188 if (error || controlp == NULL) 1189 goto next; 1190 *controlp = sbcreatecontrol(NULL, datalen, 1191 cm->cmsg_type, cm->cmsg_level); 1192 if (*controlp == NULL) { 1193 error = ENOBUFS; 1194 goto next; 1195 } 1196 bcopy(data, 1197 CMSG_DATA(mtod(*controlp, struct cmsghdr *)), 1198 datalen); 1199 } 1200 1201 controlp = &(*controlp)->m_next; 1202 1203next: 1204 if (CMSG_SPACE(datalen) < clen) { 1205 clen -= CMSG_SPACE(datalen); 1206 cm = (struct cmsghdr *) 1207 ((caddr_t)cm + CMSG_SPACE(datalen)); 1208 } else { 1209 clen = 0; 1210 cm = NULL; 1211 } 1212 } 1213 1214 m_freem(control); 1215 1216 return (error); 1217} 1218 1219void 1220unp_init(void) 1221{ 1222 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL, 1223 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 1224 if (unp_zone == NULL) 1225 panic("unp_init"); 1226 uma_zone_set_max(unp_zone, nmbclusters); 1227 LIST_INIT(&unp_dhead); 1228 LIST_INIT(&unp_shead); 1229 1230 UNP_LOCK_INIT(); 1231} 1232 1233static int 1234unp_internalize(controlp, td) 1235 struct mbuf **controlp; 1236 struct thread *td; 1237{ 1238 struct mbuf *control = *controlp; 1239 struct proc *p = td->td_proc; 1240 struct filedesc *fdescp = p->p_fd; 1241 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1242 struct cmsgcred *cmcred; 1243 struct file **rp; 1244 struct file *fp; 1245 struct timeval *tv; 1246 int i, fd, *fdp; 1247 void *data; 1248 socklen_t clen = control->m_len, datalen; 1249 int error, oldfds; 1250 u_int newlen; 1251 1252 error = 0; 1253 *controlp = NULL; 1254 1255 while (cm != NULL) { 1256 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET 1257 || cm->cmsg_len > clen) { 1258 error = EINVAL; 1259 goto out; 1260 } 1261 1262 data = CMSG_DATA(cm); 1263 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1264 1265 switch (cm->cmsg_type) { 1266 /* 1267 * Fill in credential information. 1268 */ 1269 case SCM_CREDS: 1270 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred), 1271 SCM_CREDS, SOL_SOCKET); 1272 if (*controlp == NULL) { 1273 error = ENOBUFS; 1274 goto out; 1275 } 1276 1277 cmcred = (struct cmsgcred *) 1278 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1279 cmcred->cmcred_pid = p->p_pid; 1280 cmcred->cmcred_uid = td->td_ucred->cr_ruid; 1281 cmcred->cmcred_gid = td->td_ucred->cr_rgid; 1282 cmcred->cmcred_euid = td->td_ucred->cr_uid; 1283 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups, 1284 CMGROUP_MAX); 1285 for (i = 0; i < cmcred->cmcred_ngroups; i++) 1286 cmcred->cmcred_groups[i] = 1287 td->td_ucred->cr_groups[i]; 1288 break; 1289 1290 case SCM_RIGHTS: 1291 oldfds = datalen / sizeof (int); 1292 /* 1293 * check that all the FDs passed in refer to legal files 1294 * If not, reject the entire operation. 1295 */ 1296 fdp = data; 1297 FILEDESC_LOCK(fdescp); 1298 for (i = 0; i < oldfds; i++) { 1299 fd = *fdp++; 1300 if ((unsigned)fd >= fdescp->fd_nfiles || 1301 fdescp->fd_ofiles[fd] == NULL) { 1302 FILEDESC_UNLOCK(fdescp); 1303 error = EBADF; 1304 goto out; 1305 } 1306 fp = fdescp->fd_ofiles[fd]; 1307 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { 1308 FILEDESC_UNLOCK(fdescp); 1309 error = EOPNOTSUPP; 1310 goto out; 1311 } 1312 1313 } 1314 /* 1315 * Now replace the integer FDs with pointers to 1316 * the associated global file table entry.. 1317 */ 1318 newlen = oldfds * sizeof(struct file *); 1319 *controlp = sbcreatecontrol(NULL, newlen, 1320 SCM_RIGHTS, SOL_SOCKET); 1321 if (*controlp == NULL) { 1322 FILEDESC_UNLOCK(fdescp); 1323 error = E2BIG; 1324 goto out; 1325 } 1326 1327 fdp = data; 1328 rp = (struct file **) 1329 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1330 for (i = 0; i < oldfds; i++) { 1331 fp = fdescp->fd_ofiles[*fdp++]; 1332 *rp++ = fp; 1333 FILE_LOCK(fp); 1334 fp->f_count++; 1335 fp->f_msgcount++; 1336 FILE_UNLOCK(fp); 1337 unp_rights++; 1338 } 1339 FILEDESC_UNLOCK(fdescp); 1340 break; 1341 1342 case SCM_TIMESTAMP: 1343 *controlp = sbcreatecontrol(NULL, sizeof(*tv), 1344 SCM_TIMESTAMP, SOL_SOCKET); 1345 if (*controlp == NULL) { 1346 error = ENOBUFS; 1347 goto out; 1348 } 1349 tv = (struct timeval *) 1350 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1351 microtime(tv); 1352 break; 1353 1354 default: 1355 error = EINVAL; 1356 goto out; 1357 } 1358 1359 controlp = &(*controlp)->m_next; 1360 1361 if (CMSG_SPACE(datalen) < clen) { 1362 clen -= CMSG_SPACE(datalen); 1363 cm = (struct cmsghdr *) 1364 ((caddr_t)cm + CMSG_SPACE(datalen)); 1365 } else { 1366 clen = 0; 1367 cm = NULL; 1368 } 1369 } 1370 1371out: 1372 m_freem(control); 1373 1374 return (error); 1375} 1376 1377static int unp_defer, unp_gcing; 1378 1379static void 1380unp_gc() 1381{ 1382 register struct file *fp, *nextfp; 1383 register struct socket *so; 1384 struct file **extra_ref, **fpp; 1385 int nunref, i; 1386 1387 UNP_LOCK_ASSERT(); 1388 1389 if (unp_gcing) 1390 return; 1391 unp_gcing = 1; 1392 unp_defer = 0; 1393 /* 1394 * before going through all this, set all FDs to 1395 * be NOT defered and NOT externally accessible 1396 */ 1397 /* 1398 * XXXRW: Acquiring a sleep lock while holding UNP 1399 * mutex cannot be a good thing. 1400 */ 1401 sx_slock(&filelist_lock); 1402 LIST_FOREACH(fp, &filehead, f_list) 1403 fp->f_gcflag &= ~(FMARK|FDEFER); 1404 do { 1405 LIST_FOREACH(fp, &filehead, f_list) { 1406 FILE_LOCK(fp); 1407 /* 1408 * If the file is not open, skip it 1409 */ 1410 if (fp->f_count == 0) { 1411 FILE_UNLOCK(fp); 1412 continue; 1413 } 1414 /* 1415 * If we already marked it as 'defer' in a 1416 * previous pass, then try process it this time 1417 * and un-mark it 1418 */ 1419 if (fp->f_gcflag & FDEFER) { 1420 fp->f_gcflag &= ~FDEFER; 1421 unp_defer--; 1422 } else { 1423 /* 1424 * if it's not defered, then check if it's 1425 * already marked.. if so skip it 1426 */ 1427 if (fp->f_gcflag & FMARK) { 1428 FILE_UNLOCK(fp); 1429 continue; 1430 } 1431 /* 1432 * If all references are from messages 1433 * in transit, then skip it. it's not 1434 * externally accessible. 1435 */ 1436 if (fp->f_count == fp->f_msgcount) { 1437 FILE_UNLOCK(fp); 1438 continue; 1439 } 1440 /* 1441 * If it got this far then it must be 1442 * externally accessible. 1443 */ 1444 fp->f_gcflag |= FMARK; 1445 } 1446 /* 1447 * either it was defered, or it is externally 1448 * accessible and not already marked so. 1449 * Now check if it is possibly one of OUR sockets. 1450 */ 1451 if (fp->f_type != DTYPE_SOCKET || 1452 (so = fp->f_data) == NULL) { 1453 FILE_UNLOCK(fp); 1454 continue; 1455 } 1456 FILE_UNLOCK(fp); 1457 if (so->so_proto->pr_domain != &localdomain || 1458 (so->so_proto->pr_flags&PR_RIGHTS) == 0) 1459 continue; 1460#ifdef notdef 1461 if (so->so_rcv.sb_flags & SB_LOCK) { 1462 /* 1463 * This is problematical; it's not clear 1464 * we need to wait for the sockbuf to be 1465 * unlocked (on a uniprocessor, at least), 1466 * and it's also not clear what to do 1467 * if sbwait returns an error due to receipt 1468 * of a signal. If sbwait does return 1469 * an error, we'll go into an infinite 1470 * loop. Delete all of this for now. 1471 */ 1472 (void) sbwait(&so->so_rcv); 1473 goto restart; 1474 } 1475#endif 1476 /* 1477 * So, Ok, it's one of our sockets and it IS externally 1478 * accessible (or was defered). Now we look 1479 * to see if we hold any file descriptors in its 1480 * message buffers. Follow those links and mark them 1481 * as accessible too. 1482 */ 1483 unp_scan(so->so_rcv.sb_mb, unp_mark); 1484 } 1485 } while (unp_defer); 1486 sx_sunlock(&filelist_lock); 1487 /* 1488 * We grab an extra reference to each of the file table entries 1489 * that are not otherwise accessible and then free the rights 1490 * that are stored in messages on them. 1491 * 1492 * The bug in the orginal code is a little tricky, so I'll describe 1493 * what's wrong with it here. 1494 * 1495 * It is incorrect to simply unp_discard each entry for f_msgcount 1496 * times -- consider the case of sockets A and B that contain 1497 * references to each other. On a last close of some other socket, 1498 * we trigger a gc since the number of outstanding rights (unp_rights) 1499 * is non-zero. If during the sweep phase the gc code un_discards, 1500 * we end up doing a (full) closef on the descriptor. A closef on A 1501 * results in the following chain. Closef calls soo_close, which 1502 * calls soclose. Soclose calls first (through the switch 1503 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply 1504 * returns because the previous instance had set unp_gcing, and 1505 * we return all the way back to soclose, which marks the socket 1506 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush 1507 * to free up the rights that are queued in messages on the socket A, 1508 * i.e., the reference on B. The sorflush calls via the dom_dispose 1509 * switch unp_dispose, which unp_scans with unp_discard. This second 1510 * instance of unp_discard just calls closef on B. 1511 * 1512 * Well, a similar chain occurs on B, resulting in a sorflush on B, 1513 * which results in another closef on A. Unfortunately, A is already 1514 * being closed, and the descriptor has already been marked with 1515 * SS_NOFDREF, and soclose panics at this point. 1516 * 1517 * Here, we first take an extra reference to each inaccessible 1518 * descriptor. Then, we call sorflush ourself, since we know 1519 * it is a Unix domain socket anyhow. After we destroy all the 1520 * rights carried in messages, we do a last closef to get rid 1521 * of our extra reference. This is the last close, and the 1522 * unp_detach etc will shut down the socket. 1523 * 1524 * 91/09/19, bsy@cs.cmu.edu 1525 */ 1526 extra_ref = malloc(nfiles * sizeof(struct file *), M_TEMP, M_WAITOK); 1527 sx_slock(&filelist_lock); 1528 for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref; 1529 fp != NULL; fp = nextfp) { 1530 nextfp = LIST_NEXT(fp, f_list); 1531 FILE_LOCK(fp); 1532 /* 1533 * If it's not open, skip it 1534 */ 1535 if (fp->f_count == 0) { 1536 FILE_UNLOCK(fp); 1537 continue; 1538 } 1539 /* 1540 * If all refs are from msgs, and it's not marked accessible 1541 * then it must be referenced from some unreachable cycle 1542 * of (shut-down) FDs, so include it in our 1543 * list of FDs to remove 1544 */ 1545 if (fp->f_count == fp->f_msgcount && !(fp->f_gcflag & FMARK)) { 1546 *fpp++ = fp; 1547 nunref++; 1548 fp->f_count++; 1549 } 1550 FILE_UNLOCK(fp); 1551 } 1552 sx_sunlock(&filelist_lock); 1553 /* 1554 * for each FD on our hit list, do the following two things 1555 */ 1556 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1557 struct file *tfp = *fpp; 1558 FILE_LOCK(tfp); 1559 if (tfp->f_type == DTYPE_SOCKET && 1560 tfp->f_data != NULL) { 1561 FILE_UNLOCK(tfp); 1562 sorflush(tfp->f_data); 1563 } else { 1564 FILE_UNLOCK(tfp); 1565 } 1566 } 1567 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) 1568 closef(*fpp, (struct thread *) NULL); 1569 free(extra_ref, M_TEMP); 1570 unp_gcing = 0; 1571} 1572 1573void 1574unp_dispose(m) 1575 struct mbuf *m; 1576{ 1577 1578 if (m) 1579 unp_scan(m, unp_discard); 1580} 1581 1582static int 1583unp_listen(unp, td) 1584 struct unpcb *unp; 1585 struct thread *td; 1586{ 1587 UNP_LOCK_ASSERT(); 1588 1589 /* 1590 * XXXRW: Why populate the local peer cred with our own credential? 1591 */ 1592 cru2x(td->td_ucred, &unp->unp_peercred); 1593 unp->unp_flags |= UNP_HAVEPCCACHED; 1594 return (0); 1595} 1596 1597static void 1598unp_scan(m0, op) 1599 register struct mbuf *m0; 1600 void (*op)(struct file *); 1601{ 1602 struct mbuf *m; 1603 struct file **rp; 1604 struct cmsghdr *cm; 1605 void *data; 1606 int i; 1607 socklen_t clen, datalen; 1608 int qfds; 1609 1610 while (m0 != NULL) { 1611 for (m = m0; m; m = m->m_next) { 1612 if (m->m_type != MT_CONTROL) 1613 continue; 1614 1615 cm = mtod(m, struct cmsghdr *); 1616 clen = m->m_len; 1617 1618 while (cm != NULL) { 1619 if (sizeof(*cm) > clen || cm->cmsg_len > clen) 1620 break; 1621 1622 data = CMSG_DATA(cm); 1623 datalen = (caddr_t)cm + cm->cmsg_len 1624 - (caddr_t)data; 1625 1626 if (cm->cmsg_level == SOL_SOCKET && 1627 cm->cmsg_type == SCM_RIGHTS) { 1628 qfds = datalen / sizeof (struct file *); 1629 rp = data; 1630 for (i = 0; i < qfds; i++) 1631 (*op)(*rp++); 1632 } 1633 1634 if (CMSG_SPACE(datalen) < clen) { 1635 clen -= CMSG_SPACE(datalen); 1636 cm = (struct cmsghdr *) 1637 ((caddr_t)cm + CMSG_SPACE(datalen)); 1638 } else { 1639 clen = 0; 1640 cm = NULL; 1641 } 1642 } 1643 } 1644 m0 = m0->m_act; 1645 } 1646} 1647 1648static void 1649unp_mark(fp) 1650 struct file *fp; 1651{ 1652 if (fp->f_gcflag & FMARK) 1653 return; 1654 unp_defer++; 1655 fp->f_gcflag |= (FMARK|FDEFER); 1656} 1657 1658static void 1659unp_discard(fp) 1660 struct file *fp; 1661{ 1662 FILE_LOCK(fp); 1663 fp->f_msgcount--; 1664 unp_rights--; 1665 FILE_UNLOCK(fp); 1666 (void) closef(fp, (struct thread *)NULL); 1667} 1668