uipc_usrreq.c revision 51801
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 * $FreeBSD: head/sys/kern/uipc_usrreq.c 51801 1999-09-29 21:09:41Z guido $ 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 if (unp->unp_conn == 0) 230 break; 231 so2 = unp->unp_conn->unp_socket; 232 /* 233 * Adjust backpressure on sender 234 * and wakeup any waiting to write. 235 */ 236 so2->so_snd.sb_mbmax += unp->unp_mbcnt - so->so_rcv.sb_mbcnt; 237 unp->unp_mbcnt = so->so_rcv.sb_mbcnt; 238 so2->so_snd.sb_hiwat += unp->unp_cc - so->so_rcv.sb_cc; 239 unp->unp_cc = so->so_rcv.sb_cc; 240 sowwakeup(so2); 241 break; 242 243 default: 244 panic("uipc_rcvd unknown socktype"); 245 } 246 return 0; 247} 248 249/* pru_rcvoob is EOPNOTSUPP */ 250 251static int 252uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 253 struct mbuf *control, struct proc *p) 254{ 255 int error = 0; 256 struct unpcb *unp = sotounpcb(so); 257 struct socket *so2; 258 259 if (unp == 0) { 260 error = EINVAL; 261 goto release; 262 } 263 if (flags & PRUS_OOB) { 264 error = EOPNOTSUPP; 265 goto release; 266 } 267 268 if (control && (error = unp_internalize(control, p))) 269 goto release; 270 271 switch (so->so_type) { 272 case SOCK_DGRAM: 273 { 274 struct sockaddr *from; 275 276 if (nam) { 277 if (unp->unp_conn) { 278 error = EISCONN; 279 break; 280 } 281 error = unp_connect(so, nam, p); 282 if (error) 283 break; 284 } else { 285 if (unp->unp_conn == 0) { 286 error = ENOTCONN; 287 break; 288 } 289 } 290 so2 = unp->unp_conn->unp_socket; 291 if (unp->unp_addr) 292 from = (struct sockaddr *)unp->unp_addr; 293 else 294 from = &sun_noname; 295 if (sbappendaddr(&so2->so_rcv, from, m, control)) { 296 sorwakeup(so2); 297 m = 0; 298 control = 0; 299 } else 300 error = ENOBUFS; 301 if (nam) 302 unp_disconnect(unp); 303 break; 304 } 305 306 case SOCK_STREAM: 307 /* Connect if not connected yet. */ 308 /* 309 * Note: A better implementation would complain 310 * if not equal to the peer's address. 311 */ 312 if ((so->so_state & SS_ISCONNECTED) == 0) { 313 if (nam) { 314 error = unp_connect(so, nam, p); 315 if (error) 316 break; /* XXX */ 317 } else { 318 error = ENOTCONN; 319 break; 320 } 321 } 322 323 if (so->so_state & SS_CANTSENDMORE) { 324 error = EPIPE; 325 break; 326 } 327 if (unp->unp_conn == 0) 328 panic("uipc_send connected but no connection?"); 329 so2 = unp->unp_conn->unp_socket; 330 /* 331 * Send to paired receive port, and then reduce 332 * send buffer hiwater marks to maintain backpressure. 333 * Wake up readers. 334 */ 335 if (control) { 336 if (sbappendcontrol(&so2->so_rcv, m, control)) 337 control = 0; 338 } else 339 sbappend(&so2->so_rcv, m); 340 so->so_snd.sb_mbmax -= 341 so2->so_rcv.sb_mbcnt - unp->unp_conn->unp_mbcnt; 342 unp->unp_conn->unp_mbcnt = so2->so_rcv.sb_mbcnt; 343 so->so_snd.sb_hiwat -= 344 so2->so_rcv.sb_cc - unp->unp_conn->unp_cc; 345 unp->unp_conn->unp_cc = so2->so_rcv.sb_cc; 346 sorwakeup(so2); 347 m = 0; 348 break; 349 350 default: 351 panic("uipc_send unknown socktype"); 352 } 353 354 /* 355 * SEND_EOF is equivalent to a SEND followed by 356 * a SHUTDOWN. 357 */ 358 if (flags & PRUS_EOF) { 359 socantsendmore(so); 360 unp_shutdown(unp); 361 } 362 363 if (control && error != 0) 364 unp_dispose(control); 365 366release: 367 if (control) 368 m_freem(control); 369 if (m) 370 m_freem(m); 371 return error; 372} 373 374static int 375uipc_sense(struct socket *so, struct stat *sb) 376{ 377 struct unpcb *unp = sotounpcb(so); 378 struct socket *so2; 379 380 if (unp == 0) 381 return EINVAL; 382 sb->st_blksize = so->so_snd.sb_hiwat; 383 if (so->so_type == SOCK_STREAM && unp->unp_conn != 0) { 384 so2 = unp->unp_conn->unp_socket; 385 sb->st_blksize += so2->so_rcv.sb_cc; 386 } 387 sb->st_dev = NOUDEV; 388 if (unp->unp_ino == 0) 389 unp->unp_ino = unp_ino++; 390 sb->st_ino = unp->unp_ino; 391 return (0); 392} 393 394static int 395uipc_shutdown(struct socket *so) 396{ 397 struct unpcb *unp = sotounpcb(so); 398 399 if (unp == 0) 400 return EINVAL; 401 socantsendmore(so); 402 unp_shutdown(unp); 403 return 0; 404} 405 406static int 407uipc_sockaddr(struct socket *so, struct sockaddr **nam) 408{ 409 struct unpcb *unp = sotounpcb(so); 410 411 if (unp == 0) 412 return EINVAL; 413 if (unp->unp_addr) 414 *nam = dup_sockaddr((struct sockaddr *)unp->unp_addr, 1); 415 return 0; 416} 417 418struct pr_usrreqs uipc_usrreqs = { 419 uipc_abort, uipc_accept, uipc_attach, uipc_bind, uipc_connect, 420 uipc_connect2, pru_control_notsupp, uipc_detach, uipc_disconnect, 421 uipc_listen, uipc_peeraddr, uipc_rcvd, pru_rcvoob_notsupp, 422 uipc_send, uipc_sense, uipc_shutdown, uipc_sockaddr, 423 sosend, soreceive, sopoll 424}; 425 426/* 427 * Both send and receive buffers are allocated PIPSIZ bytes of buffering 428 * for stream sockets, although the total for sender and receiver is 429 * actually only PIPSIZ. 430 * Datagram sockets really use the sendspace as the maximum datagram size, 431 * and don't really want to reserve the sendspace. Their recvspace should 432 * be large enough for at least one max-size datagram plus address. 433 */ 434#ifndef PIPSIZ 435#define PIPSIZ 8192 436#endif 437static u_long unpst_sendspace = PIPSIZ; 438static u_long unpst_recvspace = PIPSIZ; 439static u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 440static u_long unpdg_recvspace = 4*1024; 441 442static int unp_rights; /* file descriptors in flight */ 443 444SYSCTL_DECL(_net_local_stream); 445SYSCTL_INT(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 446 &unpst_sendspace, 0, ""); 447SYSCTL_INT(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 448 &unpst_recvspace, 0, ""); 449SYSCTL_DECL(_net_local_dgram); 450SYSCTL_INT(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 451 &unpdg_sendspace, 0, ""); 452SYSCTL_INT(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 453 &unpdg_recvspace, 0, ""); 454SYSCTL_DECL(_net_local); 455SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, ""); 456 457static int 458unp_attach(so) 459 struct socket *so; 460{ 461 register struct unpcb *unp; 462 int error; 463 464 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 465 switch (so->so_type) { 466 467 case SOCK_STREAM: 468 error = soreserve(so, unpst_sendspace, unpst_recvspace); 469 break; 470 471 case SOCK_DGRAM: 472 error = soreserve(so, unpdg_sendspace, unpdg_recvspace); 473 break; 474 475 default: 476 panic("unp_attach"); 477 } 478 if (error) 479 return (error); 480 } 481 unp = zalloc(unp_zone); 482 if (unp == NULL) 483 return (ENOBUFS); 484 bzero(unp, sizeof *unp); 485 unp->unp_gencnt = ++unp_gencnt; 486 unp_count++; 487 LIST_INIT(&unp->unp_refs); 488 unp->unp_socket = so; 489 unp->unp_rvnode = curproc->p_fd->fd_rdir; 490 LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead 491 : &unp_shead, unp, unp_link); 492 so->so_pcb = (caddr_t)unp; 493 return (0); 494} 495 496static void 497unp_detach(unp) 498 register struct unpcb *unp; 499{ 500 LIST_REMOVE(unp, unp_link); 501 unp->unp_gencnt = ++unp_gencnt; 502 --unp_count; 503 if (unp->unp_vnode) { 504 unp->unp_vnode->v_socket = 0; 505 vrele(unp->unp_vnode); 506 unp->unp_vnode = 0; 507 } 508 if (unp->unp_conn) 509 unp_disconnect(unp); 510 while (unp->unp_refs.lh_first) 511 unp_drop(unp->unp_refs.lh_first, ECONNRESET); 512 soisdisconnected(unp->unp_socket); 513 unp->unp_socket->so_pcb = 0; 514 if (unp_rights) { 515 /* 516 * Normally the receive buffer is flushed later, 517 * in sofree, but if our receive buffer holds references 518 * to descriptors that are now garbage, we will dispose 519 * of those descriptor references after the garbage collector 520 * gets them (resulting in a "panic: closef: count < 0"). 521 */ 522 sorflush(unp->unp_socket); 523 unp_gc(); 524 } 525 if (unp->unp_addr) 526 FREE(unp->unp_addr, M_SONAME); 527 zfree(unp_zone, unp); 528} 529 530static int 531unp_bind(unp, nam, p) 532 struct unpcb *unp; 533 struct sockaddr *nam; 534 struct proc *p; 535{ 536 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 537 register struct vnode *vp; 538 struct vattr vattr; 539 int error, namelen; 540 struct nameidata nd; 541 char buf[SOCK_MAXADDRLEN]; 542 543 if (unp->unp_vnode != NULL) 544 return (EINVAL); 545#define offsetof(s, e) ((char *)&((s *)0)->e - (char *)((s *)0)) 546 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 547 if (namelen <= 0) 548 return EINVAL; 549 strncpy(buf, soun->sun_path, namelen); 550 buf[namelen] = 0; /* null-terminate the string */ 551 NDINIT(&nd, CREATE, NOFOLLOW | LOCKPARENT, UIO_SYSSPACE, 552 buf, p); 553/* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 554 error = namei(&nd); 555 if (error) 556 return (error); 557 vp = nd.ni_vp; 558 if (vp != NULL) { 559 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 560 if (nd.ni_dvp == vp) 561 vrele(nd.ni_dvp); 562 else 563 vput(nd.ni_dvp); 564 vrele(vp); 565 return (EADDRINUSE); 566 } 567 VATTR_NULL(&vattr); 568 vattr.va_type = VSOCK; 569 vattr.va_mode = (ACCESSPERMS & ~p->p_fd->fd_cmask); 570 VOP_LEASE(nd.ni_dvp, p, p->p_ucred, LEASE_WRITE); 571 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 572 vput(nd.ni_dvp); 573 if (error) 574 return (error); 575 vp = nd.ni_vp; 576 vp->v_socket = unp->unp_socket; 577 unp->unp_vnode = vp; 578 unp->unp_addr = (struct sockaddr_un *)dup_sockaddr(nam, 1); 579 VOP_UNLOCK(vp, 0, p); 580 return (0); 581} 582 583static int 584unp_connect(so, nam, p) 585 struct socket *so; 586 struct sockaddr *nam; 587 struct proc *p; 588{ 589 register struct sockaddr_un *soun = (struct sockaddr_un *)nam; 590 register struct vnode *vp; 591 register struct socket *so2, *so3; 592 struct unpcb *unp2, *unp3; 593 int error, len; 594 struct nameidata nd; 595 char buf[SOCK_MAXADDRLEN]; 596 597 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 598 if (len <= 0) 599 return EINVAL; 600 strncpy(buf, soun->sun_path, len); 601 buf[len] = 0; 602 603 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, p); 604 error = namei(&nd); 605 if (error) 606 return (error); 607 vp = nd.ni_vp; 608 if (vp->v_type != VSOCK) { 609 error = ENOTSOCK; 610 goto bad; 611 } 612 error = VOP_ACCESS(vp, VWRITE, p->p_ucred, p); 613 if (error) 614 goto bad; 615 so2 = vp->v_socket; 616 if (so2 == 0) { 617 error = ECONNREFUSED; 618 goto bad; 619 } 620 if (so->so_type != so2->so_type) { 621 error = EPROTOTYPE; 622 goto bad; 623 } 624 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 625 if ((so2->so_options & SO_ACCEPTCONN) == 0 || 626 (so3 = sonewconn3(so2, 0, p)) == 0) { 627 error = ECONNREFUSED; 628 goto bad; 629 } 630 unp2 = sotounpcb(so2); 631 unp3 = sotounpcb(so3); 632 if (unp2->unp_addr) 633 unp3->unp_addr = (struct sockaddr_un *) 634 dup_sockaddr((struct sockaddr *) 635 unp2->unp_addr, 1); 636 so2 = so3; 637 } 638 error = unp_connect2(so, so2); 639bad: 640 vput(vp); 641 return (error); 642} 643 644int 645unp_connect2(so, so2) 646 register struct socket *so; 647 register struct socket *so2; 648{ 649 register struct unpcb *unp = sotounpcb(so); 650 register struct unpcb *unp2; 651 652 if (so2->so_type != so->so_type) 653 return (EPROTOTYPE); 654 unp2 = sotounpcb(so2); 655 unp->unp_conn = unp2; 656 switch (so->so_type) { 657 658 case SOCK_DGRAM: 659 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 660 soisconnected(so); 661 break; 662 663 case SOCK_STREAM: 664 unp2->unp_conn = unp; 665 soisconnected(so); 666 soisconnected(so2); 667 break; 668 669 default: 670 panic("unp_connect2"); 671 } 672 return (0); 673} 674 675static void 676unp_disconnect(unp) 677 struct unpcb *unp; 678{ 679 register struct unpcb *unp2 = unp->unp_conn; 680 681 if (unp2 == 0) 682 return; 683 unp->unp_conn = 0; 684 switch (unp->unp_socket->so_type) { 685 686 case SOCK_DGRAM: 687 LIST_REMOVE(unp, unp_reflink); 688 unp->unp_socket->so_state &= ~SS_ISCONNECTED; 689 break; 690 691 case SOCK_STREAM: 692 soisdisconnected(unp->unp_socket); 693 unp2->unp_conn = 0; 694 soisdisconnected(unp2->unp_socket); 695 break; 696 } 697} 698 699#ifdef notdef 700void 701unp_abort(unp) 702 struct unpcb *unp; 703{ 704 705 unp_detach(unp); 706} 707#endif 708 709static int 710prison_unpcb(struct proc *p, struct unpcb *unp) 711{ 712 if (!p->p_prison) 713 return (0); 714 if (p->p_fd->fd_rdir == unp->unp_rvnode) 715 return (0); 716 return (1); 717} 718 719static int 720unp_pcblist SYSCTL_HANDLER_ARGS 721{ 722 int error, i, n; 723 struct unpcb *unp, **unp_list; 724 unp_gen_t gencnt; 725 struct xunpgen xug; 726 struct unp_head *head; 727 728 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead); 729 730 /* 731 * The process of preparing the PCB list is too time-consuming and 732 * resource-intensive to repeat twice on every request. 733 */ 734 if (req->oldptr == 0) { 735 n = unp_count; 736 req->oldidx = 2 * (sizeof xug) 737 + (n + n/8) * sizeof(struct xunpcb); 738 return 0; 739 } 740 741 if (req->newptr != 0) 742 return EPERM; 743 744 /* 745 * OK, now we're committed to doing something. 746 */ 747 gencnt = unp_gencnt; 748 n = unp_count; 749 750 xug.xug_len = sizeof xug; 751 xug.xug_count = n; 752 xug.xug_gen = gencnt; 753 xug.xug_sogen = so_gencnt; 754 error = SYSCTL_OUT(req, &xug, sizeof xug); 755 if (error) 756 return error; 757 758 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 759 if (unp_list == 0) 760 return ENOMEM; 761 762 for (unp = head->lh_first, i = 0; unp && i < n; 763 unp = unp->unp_link.le_next) { 764 if (unp->unp_gencnt <= gencnt && !prison_unpcb(req->p, unp)) 765 unp_list[i++] = unp; 766 } 767 n = i; /* in case we lost some during malloc */ 768 769 error = 0; 770 for (i = 0; i < n; i++) { 771 unp = unp_list[i]; 772 if (unp->unp_gencnt <= gencnt) { 773 struct xunpcb xu; 774 xu.xu_len = sizeof xu; 775 xu.xu_unpp = unp; 776 /* 777 * XXX - need more locking here to protect against 778 * connect/disconnect races for SMP. 779 */ 780 if (unp->unp_addr) 781 bcopy(unp->unp_addr, &xu.xu_addr, 782 unp->unp_addr->sun_len); 783 if (unp->unp_conn && unp->unp_conn->unp_addr) 784 bcopy(unp->unp_conn->unp_addr, 785 &xu.xu_caddr, 786 unp->unp_conn->unp_addr->sun_len); 787 bcopy(unp, &xu.xu_unp, sizeof *unp); 788 sotoxsocket(unp->unp_socket, &xu.xu_socket); 789 error = SYSCTL_OUT(req, &xu, sizeof xu); 790 } 791 } 792 if (!error) { 793 /* 794 * Give the user an updated idea of our state. 795 * If the generation differs from what we told 796 * her before, she knows that something happened 797 * while we were processing this request, and it 798 * might be necessary to retry. 799 */ 800 xug.xug_gen = unp_gencnt; 801 xug.xug_sogen = so_gencnt; 802 xug.xug_count = unp_count; 803 error = SYSCTL_OUT(req, &xug, sizeof xug); 804 } 805 free(unp_list, M_TEMP); 806 return error; 807} 808 809SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD, 810 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 811 "List of active local datagram sockets"); 812SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD, 813 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 814 "List of active local stream sockets"); 815 816static void 817unp_shutdown(unp) 818 struct unpcb *unp; 819{ 820 struct socket *so; 821 822 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn && 823 (so = unp->unp_conn->unp_socket)) 824 socantrcvmore(so); 825} 826 827static void 828unp_drop(unp, errno) 829 struct unpcb *unp; 830 int errno; 831{ 832 struct socket *so = unp->unp_socket; 833 834 so->so_error = errno; 835 unp_disconnect(unp); 836 if (so->so_head) { 837 LIST_REMOVE(unp, unp_link); 838 unp->unp_gencnt = ++unp_gencnt; 839 unp_count--; 840 so->so_pcb = (caddr_t) 0; 841 if (unp->unp_addr) 842 FREE(unp->unp_addr, M_SONAME); 843 zfree(unp_zone, unp); 844 sofree(so); 845 } 846} 847 848#ifdef notdef 849void 850unp_drain() 851{ 852 853} 854#endif 855 856int 857unp_externalize(rights) 858 struct mbuf *rights; 859{ 860 struct proc *p = curproc; /* XXX */ 861 register int i; 862 register struct cmsghdr *cm = mtod(rights, struct cmsghdr *); 863 register struct file **rp = (struct file **)(cm + 1); 864 register struct file *fp; 865 int newfds = (cm->cmsg_len - sizeof(*cm)) / sizeof (int); 866 int f; 867 868 /* 869 * if the new FD's will not fit, then we free them all 870 */ 871 if (!fdavail(p, newfds)) { 872 for (i = 0; i < newfds; i++) { 873 fp = *rp; 874 unp_discard(fp); 875 *rp++ = 0; 876 } 877 return (EMSGSIZE); 878 } 879 /* 880 * now change each pointer to an fd in the global table to 881 * an integer that is the index to the local fd table entry 882 * that we set up to point to the global one we are transferring. 883 * XXX this assumes a pointer and int are the same size...! 884 */ 885 for (i = 0; i < newfds; i++) { 886 if (fdalloc(p, 0, &f)) 887 panic("unp_externalize"); 888 fp = *rp; 889 p->p_fd->fd_ofiles[f] = fp; 890 fp->f_msgcount--; 891 unp_rights--; 892 *(int *)rp++ = f; 893 } 894 return (0); 895} 896 897void 898unp_init(void) 899{ 900 unp_zone = zinit("unpcb", sizeof(struct unpcb), nmbclusters, 0, 0); 901 if (unp_zone == 0) 902 panic("unp_init"); 903 LIST_INIT(&unp_dhead); 904 LIST_INIT(&unp_shead); 905} 906 907#ifndef MIN 908#define MIN(a,b) (((a)<(b))?(a):(b)) 909#endif 910 911static int 912unp_internalize(control, p) 913 struct mbuf *control; 914 struct proc *p; 915{ 916 struct filedesc *fdp = p->p_fd; 917 register struct cmsghdr *cm = mtod(control, struct cmsghdr *); 918 register struct file **rp; 919 register struct file *fp; 920 register int i, fd; 921 register struct cmsgcred *cmcred; 922 int oldfds; 923 924 if ((cm->cmsg_type != SCM_RIGHTS && cm->cmsg_type != SCM_CREDS) || 925 cm->cmsg_level != SOL_SOCKET || cm->cmsg_len != control->m_len) 926 return (EINVAL); 927 928 /* 929 * Fill in credential information. 930 */ 931 if (cm->cmsg_type == SCM_CREDS) { 932 cmcred = (struct cmsgcred *)(cm + 1); 933 cmcred->cmcred_pid = p->p_pid; 934 cmcred->cmcred_uid = p->p_cred->p_ruid; 935 cmcred->cmcred_gid = p->p_cred->p_rgid; 936 cmcred->cmcred_euid = p->p_ucred->cr_uid; 937 cmcred->cmcred_ngroups = MIN(p->p_ucred->cr_ngroups, 938 CMGROUP_MAX); 939 for (i = 0; i < cmcred->cmcred_ngroups; i++) 940 cmcred->cmcred_groups[i] = p->p_ucred->cr_groups[i]; 941 return(0); 942 } 943 944 oldfds = (cm->cmsg_len - sizeof (*cm)) / sizeof (int); 945 /* 946 * check that all the FDs passed in refer to legal OPEN files 947 * If not, reject the entire operation. 948 */ 949 rp = (struct file **)(cm + 1); 950 for (i = 0; i < oldfds; i++) { 951 fd = *(int *)rp++; 952 if ((unsigned)fd >= fdp->fd_nfiles || 953 fdp->fd_ofiles[fd] == NULL) 954 return (EBADF); 955 } 956 /* 957 * Now replace the integer FDs with pointers to 958 * the associated global file table entry.. 959 * XXX this assumes a pointer and an int are the same size! 960 */ 961 rp = (struct file **)(cm + 1); 962 for (i = 0; i < oldfds; i++) { 963 fp = fdp->fd_ofiles[*(int *)rp]; 964 *rp++ = fp; 965 fp->f_count++; 966 fp->f_msgcount++; 967 unp_rights++; 968 } 969 return (0); 970} 971 972static int unp_defer, unp_gcing; 973 974static void 975unp_gc() 976{ 977 register struct file *fp, *nextfp; 978 register struct socket *so; 979 struct file **extra_ref, **fpp; 980 int nunref, i; 981 982 if (unp_gcing) 983 return; 984 unp_gcing = 1; 985 unp_defer = 0; 986 /* 987 * before going through all this, set all FDs to 988 * be NOT defered and NOT externally accessible 989 */ 990 for (fp = filehead.lh_first; fp != 0; fp = fp->f_list.le_next) 991 fp->f_flag &= ~(FMARK|FDEFER); 992 do { 993 for (fp = filehead.lh_first; fp != 0; fp = fp->f_list.le_next) { 994 /* 995 * If the file is not open, skip it 996 */ 997 if (fp->f_count == 0) 998 continue; 999 /* 1000 * If we already marked it as 'defer' in a 1001 * previous pass, then try process it this time 1002 * and un-mark it 1003 */ 1004 if (fp->f_flag & FDEFER) { 1005 fp->f_flag &= ~FDEFER; 1006 unp_defer--; 1007 } else { 1008 /* 1009 * if it's not defered, then check if it's 1010 * already marked.. if so skip it 1011 */ 1012 if (fp->f_flag & FMARK) 1013 continue; 1014 /* 1015 * If all references are from messages 1016 * in transit, then skip it. it's not 1017 * externally accessible. 1018 */ 1019 if (fp->f_count == fp->f_msgcount) 1020 continue; 1021 /* 1022 * If it got this far then it must be 1023 * externally accessible. 1024 */ 1025 fp->f_flag |= FMARK; 1026 } 1027 /* 1028 * either it was defered, or it is externally 1029 * accessible and not already marked so. 1030 * Now check if it is possibly one of OUR sockets. 1031 */ 1032 if (fp->f_type != DTYPE_SOCKET || 1033 (so = (struct socket *)fp->f_data) == 0) 1034 continue; 1035 if (so->so_proto->pr_domain != &localdomain || 1036 (so->so_proto->pr_flags&PR_RIGHTS) == 0) 1037 continue; 1038#ifdef notdef 1039 if (so->so_rcv.sb_flags & SB_LOCK) { 1040 /* 1041 * This is problematical; it's not clear 1042 * we need to wait for the sockbuf to be 1043 * unlocked (on a uniprocessor, at least), 1044 * and it's also not clear what to do 1045 * if sbwait returns an error due to receipt 1046 * of a signal. If sbwait does return 1047 * an error, we'll go into an infinite 1048 * loop. Delete all of this for now. 1049 */ 1050 (void) sbwait(&so->so_rcv); 1051 goto restart; 1052 } 1053#endif 1054 /* 1055 * So, Ok, it's one of our sockets and it IS externally 1056 * accessible (or was defered). Now we look 1057 * to see if we hold any file descriptors in its 1058 * message buffers. Follow those links and mark them 1059 * as accessible too. 1060 */ 1061 unp_scan(so->so_rcv.sb_mb, unp_mark); 1062 } 1063 } while (unp_defer); 1064 /* 1065 * We grab an extra reference to each of the file table entries 1066 * that are not otherwise accessible and then free the rights 1067 * that are stored in messages on them. 1068 * 1069 * The bug in the orginal code is a little tricky, so I'll describe 1070 * what's wrong with it here. 1071 * 1072 * It is incorrect to simply unp_discard each entry for f_msgcount 1073 * times -- consider the case of sockets A and B that contain 1074 * references to each other. On a last close of some other socket, 1075 * we trigger a gc since the number of outstanding rights (unp_rights) 1076 * is non-zero. If during the sweep phase the gc code un_discards, 1077 * we end up doing a (full) closef on the descriptor. A closef on A 1078 * results in the following chain. Closef calls soo_close, which 1079 * calls soclose. Soclose calls first (through the switch 1080 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply 1081 * returns because the previous instance had set unp_gcing, and 1082 * we return all the way back to soclose, which marks the socket 1083 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush 1084 * to free up the rights that are queued in messages on the socket A, 1085 * i.e., the reference on B. The sorflush calls via the dom_dispose 1086 * switch unp_dispose, which unp_scans with unp_discard. This second 1087 * instance of unp_discard just calls closef on B. 1088 * 1089 * Well, a similar chain occurs on B, resulting in a sorflush on B, 1090 * which results in another closef on A. Unfortunately, A is already 1091 * being closed, and the descriptor has already been marked with 1092 * SS_NOFDREF, and soclose panics at this point. 1093 * 1094 * Here, we first take an extra reference to each inaccessible 1095 * descriptor. Then, we call sorflush ourself, since we know 1096 * it is a Unix domain socket anyhow. After we destroy all the 1097 * rights carried in messages, we do a last closef to get rid 1098 * of our extra reference. This is the last close, and the 1099 * unp_detach etc will shut down the socket. 1100 * 1101 * 91/09/19, bsy@cs.cmu.edu 1102 */ 1103 extra_ref = malloc(nfiles * sizeof(struct file *), M_FILE, M_WAITOK); 1104 for (nunref = 0, fp = filehead.lh_first, fpp = extra_ref; fp != 0; 1105 fp = nextfp) { 1106 nextfp = fp->f_list.le_next; 1107 /* 1108 * If it's not open, skip it 1109 */ 1110 if (fp->f_count == 0) 1111 continue; 1112 /* 1113 * If all refs are from msgs, and it's not marked accessible 1114 * then it must be referenced from some unreachable cycle 1115 * of (shut-down) FDs, so include it in our 1116 * list of FDs to remove 1117 */ 1118 if (fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) { 1119 *fpp++ = fp; 1120 nunref++; 1121 fp->f_count++; 1122 } 1123 } 1124 /* 1125 * for each FD on our hit list, do the following two things 1126 */ 1127 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1128 struct file *tfp = *fpp; 1129 if (tfp->f_type == DTYPE_SOCKET && tfp->f_data != NULL) 1130 sorflush((struct socket *)(tfp->f_data)); 1131 } 1132 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) 1133 closef(*fpp, (struct proc *) NULL); 1134 free((caddr_t)extra_ref, M_FILE); 1135 unp_gcing = 0; 1136} 1137 1138void 1139unp_dispose(m) 1140 struct mbuf *m; 1141{ 1142 1143 if (m) 1144 unp_scan(m, unp_discard); 1145} 1146 1147static void 1148unp_scan(m0, op) 1149 register struct mbuf *m0; 1150 void (*op) __P((struct file *)); 1151{ 1152 register struct mbuf *m; 1153 register struct file **rp; 1154 register struct cmsghdr *cm; 1155 register int i; 1156 int qfds; 1157 1158 while (m0) { 1159 for (m = m0; m; m = m->m_next) 1160 if (m->m_type == MT_CONTROL && 1161 m->m_len >= sizeof(*cm)) { 1162 cm = mtod(m, struct cmsghdr *); 1163 if (cm->cmsg_level != SOL_SOCKET || 1164 cm->cmsg_type != SCM_RIGHTS) 1165 continue; 1166 qfds = (cm->cmsg_len - sizeof *cm) 1167 / sizeof (struct file *); 1168 rp = (struct file **)(cm + 1); 1169 for (i = 0; i < qfds; i++) 1170 (*op)(*rp++); 1171 break; /* XXX, but saves time */ 1172 } 1173 m0 = m0->m_act; 1174 } 1175} 1176 1177static void 1178unp_mark(fp) 1179 struct file *fp; 1180{ 1181 1182 if (fp->f_flag & FMARK) 1183 return; 1184 unp_defer++; 1185 fp->f_flag |= (FMARK|FDEFER); 1186} 1187 1188static void 1189unp_discard(fp) 1190 struct file *fp; 1191{ 1192 1193 fp->f_msgcount--; 1194 unp_rights--; 1195 (void) closef(fp, (struct proc *)NULL); 1196} 1197