1 /* $NetBSD: uipc_usrreq.c,v 1.73 2003/12/29 22:08:02 martin Exp $ */ 2 3 /*- 4 * Copyright (c) 1998, 2000 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility, 9 * NASA Ames Research Center. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 3. All advertising materials mentioning features or use of this software 20 * must display the following acknowledgement: 21 * This product includes software developed by the NetBSD 22 * Foundation, Inc. and its contributors. 23 * 4. Neither the name of The NetBSD Foundation nor the names of its 24 * contributors may be used to endorse or promote products derived 25 * from this software without specific prior written permission. 26 * 27 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 28 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 29 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 30 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 31 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 32 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 33 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 34 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 35 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 36 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 37 * POSSIBILITY OF SUCH DAMAGE. 38 */ 39 40 /* 41 * Copyright (c) 1982, 1986, 1989, 1991, 1993 42 * The Regents of the University of California. All rights reserved. 43 * 44 * Redistribution and use in source and binary forms, with or without 45 * modification, are permitted provided that the following conditions 46 * are met: 47 * 1. Redistributions of source code must retain the above copyright 48 * notice, this list of conditions and the following disclaimer. 49 * 2. Redistributions in binary form must reproduce the above copyright 50 * notice, this list of conditions and the following disclaimer in the 51 * documentation and/or other materials provided with the distribution. 52 * 3. Neither the name of the University nor the names of its contributors 53 * may be used to endorse or promote products derived from this software 54 * without specific prior written permission. 55 * 56 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 57 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 58 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 59 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 60 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 61 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 62 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 63 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 64 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 65 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 66 * SUCH DAMAGE. 67 * 68 * @(#)uipc_usrreq.c 8.9 (Berkeley) 5/14/95 69 */ 70 71 /* 72 * Copyright (c) 1997 Christopher G. Demetriou. All rights reserved. 73 * 74 * Redistribution and use in source and binary forms, with or without 75 * modification, are permitted provided that the following conditions 76 * are met: 77 * 1. Redistributions of source code must retain the above copyright 78 * notice, this list of conditions and the following disclaimer. 79 * 2. Redistributions in binary form must reproduce the above copyright 80 * notice, this list of conditions and the following disclaimer in the 81 * documentation and/or other materials provided with the distribution. 82 * 3. All advertising materials mentioning features or use of this software 83 * must display the following acknowledgement: 84 * This product includes software developed by the University of 85 * California, Berkeley and its contributors. 86 * 4. Neither the name of the University nor the names of its contributors 87 * may be used to endorse or promote products derived from this software 88 * without specific prior written permission. 89 * 90 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 91 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 92 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 93 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 94 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 95 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 96 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 97 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 98 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 99 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 100 * SUCH DAMAGE. 101 * 102 * @(#)uipc_usrreq.c 8.9 (Berkeley) 5/14/95 103 */ 104 105 #include <sys/cdefs.h> 106 __KERNEL_RCSID(0, "$NetBSD: uipc_usrreq.c,v 1.73 2003/12/29 22:08:02 martin Exp $"); 107 108 #include <sys/param.h> 109 #include <sys/systm.h> 110 #include <sys/proc.h> 111 #include <sys/filedesc.h> 112 #include <sys/domain.h> 113 #include <sys/protosw.h> 114 #include <sys/socket.h> 115 #include <sys/socketvar.h> 116 #include <sys/unpcb.h> 117 #include <sys/un.h> 118 #include <sys/namei.h> 119 #include <sys/vnode.h> 120 #include <sys/file.h> 121 #include <sys/stat.h> 122 #include <sys/mbuf.h> 123 124 /* 125 * Unix communications domain. 126 * 127 * TODO: 128 * SEQPACKET, RDM 129 * rethink name space problems 130 * need a proper out-of-band 131 */ 132 struct sockaddr_un sun_noname = { sizeof(sun_noname), AF_LOCAL }; 133 ino_t unp_ino; /* prototype for fake inode numbers */ 134 135 struct mbuf *unp_addsockcred __P((struct proc *, struct mbuf *)); 136 137 int 138 unp_output(m, control, unp, p) 139 struct mbuf *m, *control; 140 struct unpcb *unp; 141 struct proc *p; 142 { 143 struct socket *so2; 144 struct sockaddr_un *sun; 145 146 so2 = unp->unp_conn->unp_socket; 147 if (unp->unp_addr) 148 sun = unp->unp_addr; 149 else 150 sun = &sun_noname; 151 if (unp->unp_conn->unp_flags & UNP_WANTCRED) 152 control = unp_addsockcred(p, control); 153 if (sbappendaddr(&so2->so_rcv, (struct sockaddr *)sun, m, 154 control) == 0) { 155 m_freem(control); 156 m_freem(m); 157 return (ENOBUFS); 158 } else { 159 sorwakeup(so2); 160 return (0); 161 } 162 } 163 164 void 165 unp_setsockaddr(unp, nam) 166 struct unpcb *unp; 167 struct mbuf *nam; 168 { 169 struct sockaddr_un *sun; 170 171 if (unp->unp_addr) 172 sun = unp->unp_addr; 173 else 174 sun = &sun_noname; 175 nam->m_len = sun->sun_len; 176 if (nam->m_len > MLEN) 177 MEXTMALLOC(nam, nam->m_len, M_WAITOK); 178 memcpy(mtod(nam, caddr_t), sun, (size_t)nam->m_len); 179 } 180 181 void 182 unp_setpeeraddr(unp, nam) 183 struct unpcb *unp; 184 struct mbuf *nam; 185 { 186 struct sockaddr_un *sun; 187 188 if (unp->unp_conn && unp->unp_conn->unp_addr) 189 sun = unp->unp_conn->unp_addr; 190 else 191 sun = &sun_noname; 192 nam->m_len = sun->sun_len; 193 if (nam->m_len > MLEN) 194 MEXTMALLOC(nam, nam->m_len, M_WAITOK); 195 memcpy(mtod(nam, caddr_t), sun, (size_t)nam->m_len); 196 } 197 198 /*ARGSUSED*/ 199 int 200 uipc_usrreq(so, req, m, nam, control, p) 201 struct socket *so; 202 int req; 203 struct mbuf *m, *nam, *control; 204 struct proc *p; 205 { 206 struct unpcb *unp = sotounpcb(so); 207 struct socket *so2; 208 int error = 0; 209 210 if (req == PRU_CONTROL) 211 return (EOPNOTSUPP); 212 213 #ifdef DIAGNOSTIC 214 if (req != PRU_SEND && req != PRU_SENDOOB && control) 215 panic("uipc_usrreq: unexpected control mbuf"); 216 #endif 217 if (unp == 0 && req != PRU_ATTACH) { 218 error = EINVAL; 219 goto release; 220 } 221 222 switch (req) { 223 224 case PRU_ATTACH: 225 if (unp != 0) { 226 error = EISCONN; 227 break; 228 } 229 error = unp_attach(so); 230 break; 231 232 case PRU_DETACH: 233 unp_detach(unp); 234 break; 235 236 case PRU_BIND: 237 error = unp_bind(unp, nam, p); 238 break; 239 240 case PRU_LISTEN: 241 if (unp->unp_vnode == 0) 242 error = EINVAL; 243 break; 244 245 case PRU_CONNECT: 246 error = unp_connect(so, nam, p); 247 break; 248 249 case PRU_CONNECT2: 250 error = unp_connect2(so, (struct socket *)nam, PRU_CONNECT2); 251 break; 252 253 case PRU_DISCONNECT: 254 unp_disconnect(unp); 255 break; 256 257 case PRU_ACCEPT: 258 unp_setpeeraddr(unp, nam); 259 /* 260 * Mark the initiating STREAM socket as connected *ONLY* 261 * after it's been accepted. This prevents a client from 262 * overrunning a server and receiving ECONNREFUSED. 263 */ 264 if (unp->unp_conn != NULL && 265 (unp->unp_conn->unp_socket->so_state & SS_ISCONNECTING)) 266 soisconnected(unp->unp_conn->unp_socket); 267 break; 268 269 case PRU_SHUTDOWN: 270 socantsendmore(so); 271 unp_shutdown(unp); 272 break; 273 274 case PRU_RCVD: 275 switch (so->so_type) { 276 277 case SOCK_DGRAM: 278 panic("uipc 1"); 279 /*NOTREACHED*/ 280 281 case SOCK_STREAM: 282 #define rcv (&so->so_rcv) 283 #define snd (&so2->so_snd) 284 if (unp->unp_conn == 0) 285 break; 286 so2 = unp->unp_conn->unp_socket; 287 /* 288 * Adjust backpressure on sender 289 * and wakeup any waiting to write. 290 */ 291 snd->sb_mbmax += unp->unp_mbcnt - rcv->sb_mbcnt; 292 unp->unp_mbcnt = rcv->sb_mbcnt; 293 snd->sb_hiwat += unp->unp_cc - rcv->sb_cc; 294 unp->unp_cc = rcv->sb_cc; 295 sowwakeup(so2); 296 #undef snd 297 #undef rcv 298 break; 299 300 default: 301 panic("uipc 2"); 302 } 303 break; 304 305 case PRU_SEND: 306 /* 307 * Note: unp_internalize() rejects any control message 308 * other than SCM_RIGHTS, and only allows one. This 309 * has the side-effect of preventing a caller from 310 * forging SCM_CREDS. 311 */ 312 if (control && (error = unp_internalize(control, p))) 313 break; 314 switch (so->so_type) { 315 316 case SOCK_DGRAM: { 317 if (nam) { 318 if ((so->so_state & SS_ISCONNECTED) != 0) { 319 error = EISCONN; 320 goto die; 321 } 322 error = unp_connect(so, nam, p); 323 if (error) { 324 die: 325 m_freem(control); 326 m_freem(m); 327 break; 328 } 329 } else { 330 if ((so->so_state & SS_ISCONNECTED) == 0) { 331 error = ENOTCONN; 332 goto die; 333 } 334 } 335 error = unp_output(m, control, unp, p); 336 if (nam) 337 unp_disconnect(unp); 338 break; 339 } 340 341 case SOCK_STREAM: 342 #define rcv (&so2->so_rcv) 343 #define snd (&so->so_snd) 344 if (unp->unp_conn == 0) 345 panic("uipc 3"); 346 so2 = unp->unp_conn->unp_socket; 347 if (unp->unp_conn->unp_flags & UNP_WANTCRED) { 348 /* 349 * Credentials are passed only once on 350 * SOCK_STREAM. 351 */ 352 unp->unp_conn->unp_flags &= ~UNP_WANTCRED; 353 control = unp_addsockcred(p, control); 354 } 355 /* 356 * Send to paired receive port, and then reduce 357 * send buffer hiwater marks to maintain backpressure. 358 * Wake up readers. 359 */ 360 if (control) { 361 if (sbappendcontrol(rcv, m, control) == 0) 362 m_freem(control); 363 } else 364 sbappend(rcv, m); 365 snd->sb_mbmax -= 366 rcv->sb_mbcnt - unp->unp_conn->unp_mbcnt; 367 unp->unp_conn->unp_mbcnt = rcv->sb_mbcnt; 368 snd->sb_hiwat -= rcv->sb_cc - unp->unp_conn->unp_cc; 369 unp->unp_conn->unp_cc = rcv->sb_cc; 370 sorwakeup(so2); 371 #undef snd 372 #undef rcv 373 break; 374 375 default: 376 panic("uipc 4"); 377 } 378 break; 379 380 case PRU_ABORT: 381 unp_drop(unp, ECONNABORTED); 382 383 #ifdef DIAGNOSTIC 384 if (so->so_pcb == 0) 385 panic("uipc 5: drop killed pcb"); 386 #endif 387 unp_detach(unp); 388 break; 389 390 case PRU_SENSE: 391 ((struct stat *) m)->st_blksize = so->so_snd.sb_hiwat; 392 if (so->so_type == SOCK_STREAM && unp->unp_conn != 0) { 393 so2 = unp->unp_conn->unp_socket; 394 ((struct stat *) m)->st_blksize += so2->so_rcv.sb_cc; 395 } 396 ((struct stat *) m)->st_dev = NODEV; 397 if (unp->unp_ino == 0) 398 unp->unp_ino = unp_ino++; 399 ((struct stat *) m)->st_atimespec = 400 ((struct stat *) m)->st_mtimespec = 401 ((struct stat *) m)->st_ctimespec = unp->unp_ctime; 402 ((struct stat *) m)->st_ino = unp->unp_ino; 403 return (0); 404 405 case PRU_RCVOOB: 406 error = EOPNOTSUPP; 407 break; 408 409 case PRU_SENDOOB: 410 m_freem(control); 411 m_freem(m); 412 error = EOPNOTSUPP; 413 break; 414 415 case PRU_SOCKADDR: 416 unp_setsockaddr(unp, nam); 417 break; 418 419 case PRU_PEERADDR: 420 unp_setpeeraddr(unp, nam); 421 break; 422 423 default: 424 panic("piusrreq"); 425 } 426 427 release: 428 return (error); 429 } 430 431 /* 432 * Unix domain socket option processing. 433 */ 434 int 435 uipc_ctloutput(op, so, level, optname, mp) 436 int op; 437 struct socket *so; 438 int level, optname; 439 struct mbuf **mp; 440 { 441 struct unpcb *unp = sotounpcb(so); 442 struct mbuf *m = *mp; 443 int optval = 0, error = 0; 444 445 if (level != 0) { 446 error = EINVAL; 447 if (op == PRCO_SETOPT && m) 448 (void) m_free(m); 449 } else switch (op) { 450 451 case PRCO_SETOPT: 452 switch (optname) { 453 case LOCAL_CREDS: 454 case LOCAL_CONNWAIT: 455 if (m == NULL || m->m_len != sizeof(int)) 456 error = EINVAL; 457 else { 458 optval = *mtod(m, int *); 459 switch (optname) { 460 #define OPTSET(bit) \ 461 if (optval) \ 462 unp->unp_flags |= (bit); \ 463 else \ 464 unp->unp_flags &= ~(bit); 465 466 case LOCAL_CREDS: 467 OPTSET(UNP_WANTCRED); 468 break; 469 case LOCAL_CONNWAIT: 470 OPTSET(UNP_CONNWAIT); 471 break; 472 } 473 } 474 break; 475 #undef OPTSET 476 477 default: 478 error = ENOPROTOOPT; 479 break; 480 } 481 if (m) 482 (void) m_free(m); 483 break; 484 485 case PRCO_GETOPT: 486 switch (optname) { 487 case LOCAL_CREDS: 488 *mp = m = m_get(M_WAIT, MT_SOOPTS); 489 m->m_len = sizeof(int); 490 switch (optname) { 491 492 #define OPTBIT(bit) (unp->unp_flags & (bit) ? 1 : 0) 493 494 case LOCAL_CREDS: 495 optval = OPTBIT(UNP_WANTCRED); 496 break; 497 } 498 *mtod(m, int *) = optval; 499 break; 500 #undef OPTBIT 501 502 default: 503 error = ENOPROTOOPT; 504 break; 505 } 506 break; 507 } 508 return (error); 509 } 510 511 /* 512 * Both send and receive buffers are allocated PIPSIZ bytes of buffering 513 * for stream sockets, although the total for sender and receiver is 514 * actually only PIPSIZ. 515 * Datagram sockets really use the sendspace as the maximum datagram size, 516 * and don't really want to reserve the sendspace. Their recvspace should 517 * be large enough for at least one max-size datagram plus address. 518 */ 519 #define PIPSIZ 4096 520 u_long unpst_sendspace = PIPSIZ; 521 u_long unpst_recvspace = PIPSIZ; 522 u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 523 u_long unpdg_recvspace = 4*1024; 524 525 int unp_rights; /* file descriptors in flight */ 526 527 int 528 unp_attach(so) 529 struct socket *so; 530 { 531 struct unpcb *unp; 532 struct timeval tv; 533 int error; 534 535 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 536 switch (so->so_type) { 537 538 case SOCK_STREAM: 539 error = soreserve(so, unpst_sendspace, unpst_recvspace); 540 break; 541 542 case SOCK_DGRAM: 543 error = soreserve(so, unpdg_sendspace, unpdg_recvspace); 544 break; 545 546 default: 547 panic("unp_attach"); 548 } 549 if (error) 550 return (error); 551 } 552 unp = malloc(sizeof(*unp), M_PCB, M_NOWAIT); 553 if (unp == NULL) 554 return (ENOBUFS); 555 memset((caddr_t)unp, 0, sizeof(*unp)); 556 unp->unp_socket = so; 557 so->so_pcb = unp; 558 microtime(&tv); 559 TIMEVAL_TO_TIMESPEC(&tv, &unp->unp_ctime); 560 return (0); 561 } 562 563 void 564 unp_detach(unp) 565 struct unpcb *unp; 566 { 567 568 if (unp->unp_vnode) { 569 unp->unp_vnode->v_socket = 0; 570 vrele(unp->unp_vnode); 571 unp->unp_vnode = 0; 572 } 573 if (unp->unp_conn) 574 unp_disconnect(unp); 575 while (unp->unp_refs) 576 unp_drop(unp->unp_refs, ECONNRESET); 577 soisdisconnected(unp->unp_socket); 578 unp->unp_socket->so_pcb = 0; 579 if (unp->unp_addr) 580 free(unp->unp_addr, M_SONAME); 581 if (unp_rights) { 582 /* 583 * Normally the receive buffer is flushed later, 584 * in sofree, but if our receive buffer holds references 585 * to descriptors that are now garbage, we will dispose 586 * of those descriptor references after the garbage collector 587 * gets them (resulting in a "panic: closef: count < 0"). 588 */ 589 sorflush(unp->unp_socket); 590 free(unp, M_PCB); 591 unp_gc(); 592 } else 593 free(unp, M_PCB); 594 } 595 596 int 597 unp_bind(unp, nam, p) 598 struct unpcb *unp; 599 struct mbuf *nam; 600 struct proc *p; 601 { 602 struct sockaddr_un *sun; 603 struct vnode *vp; 604 struct mount *mp; 605 struct vattr vattr; 606 size_t addrlen; 607 int error; 608 struct nameidata nd; 609 610 if (unp->unp_vnode != 0) 611 return (EINVAL); 612 613 /* 614 * Allocate the new sockaddr. We have to allocate one 615 * extra byte so that we can ensure that the pathname 616 * is nul-terminated. 617 */ 618 addrlen = nam->m_len + 1; 619 sun = malloc(addrlen, M_SONAME, M_WAITOK); 620 m_copydata(nam, 0, nam->m_len, (caddr_t)sun); 621 *(((char *)sun) + nam->m_len) = '\0'; 622 623 restart: 624 NDINIT(&nd, CREATE, FOLLOW | LOCKPARENT, UIO_SYSSPACE, 625 sun->sun_path, p); 626 627 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 628 if ((error = namei(&nd)) != 0) 629 goto bad; 630 vp = nd.ni_vp; 631 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { 632 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 633 if (nd.ni_dvp == vp) 634 vrele(nd.ni_dvp); 635 else 636 vput(nd.ni_dvp); 637 vrele(vp); 638 if (vp != NULL) { 639 error = EADDRINUSE; 640 goto bad; 641 } 642 error = vn_start_write(NULL, &mp, 643 V_WAIT | V_SLEEPONLY | V_PCATCH); 644 if (error) 645 goto bad; 646 goto restart; 647 } 648 VATTR_NULL(&vattr); 649 vattr.va_type = VSOCK; 650 vattr.va_mode = ACCESSPERMS; 651 VOP_LEASE(nd.ni_dvp, p, p->p_ucred, LEASE_WRITE); 652 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 653 vn_finished_write(mp, 0); 654 if (error) 655 goto bad; 656 vp = nd.ni_vp; 657 vp->v_socket = unp->unp_socket; 658 unp->unp_vnode = vp; 659 unp->unp_addrlen = addrlen; 660 unp->unp_addr = sun; 661 VOP_UNLOCK(vp, 0); 662 return (0); 663 664 bad: 665 free(sun, M_SONAME); 666 return (error); 667 } 668 669 int 670 unp_connect(so, nam, p) 671 struct socket *so; 672 struct mbuf *nam; 673 struct proc *p; 674 { 675 struct sockaddr_un *sun; 676 struct vnode *vp; 677 struct socket *so2, *so3; 678 struct unpcb *unp2, *unp3; 679 size_t addrlen; 680 int error; 681 struct nameidata nd; 682 683 /* 684 * Allocate a temporary sockaddr. We have to allocate one extra 685 * byte so that we can ensure that the pathname is nul-terminated. 686 * When we establish the connection, we copy the other PCB's 687 * sockaddr to our own. 688 */ 689 addrlen = nam->m_len + 1; 690 sun = malloc(addrlen, M_SONAME, M_WAITOK); 691 m_copydata(nam, 0, nam->m_len, (caddr_t)sun); 692 *(((char *)sun) + nam->m_len) = '\0'; 693 694 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, sun->sun_path, p); 695 696 if ((error = namei(&nd)) != 0) 697 goto bad2; 698 vp = nd.ni_vp; 699 if (vp->v_type != VSOCK) { 700 error = ENOTSOCK; 701 goto bad; 702 } 703 if ((error = VOP_ACCESS(vp, VWRITE, p->p_ucred, p)) != 0) 704 goto bad; 705 so2 = vp->v_socket; 706 if (so2 == 0) { 707 error = ECONNREFUSED; 708 goto bad; 709 } 710 if (so->so_type != so2->so_type) { 711 error = EPROTOTYPE; 712 goto bad; 713 } 714 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 715 if ((so2->so_options & SO_ACCEPTCONN) == 0 || 716 (so3 = sonewconn(so2, 0)) == 0) { 717 error = ECONNREFUSED; 718 goto bad; 719 } 720 unp2 = sotounpcb(so2); 721 unp3 = sotounpcb(so3); 722 if (unp2->unp_addr) { 723 unp3->unp_addr = malloc(unp2->unp_addrlen, 724 M_SONAME, M_WAITOK); 725 memcpy(unp3->unp_addr, unp2->unp_addr, 726 unp2->unp_addrlen); 727 unp3->unp_addrlen = unp2->unp_addrlen; 728 } 729 unp3->unp_flags = unp2->unp_flags; 730 so2 = so3; 731 } 732 error = unp_connect2(so, so2, PRU_CONNECT); 733 bad: 734 vput(vp); 735 bad2: 736 free(sun, M_SONAME); 737 return (error); 738 } 739 740 int 741 unp_connect2(so, so2, req) 742 struct socket *so; 743 struct socket *so2; 744 int req; 745 { 746 struct unpcb *unp = sotounpcb(so); 747 struct unpcb *unp2; 748 749 if (so2->so_type != so->so_type) 750 return (EPROTOTYPE); 751 unp2 = sotounpcb(so2); 752 unp->unp_conn = unp2; 753 switch (so->so_type) { 754 755 case SOCK_DGRAM: 756 unp->unp_nextref = unp2->unp_refs; 757 unp2->unp_refs = unp; 758 soisconnected(so); 759 break; 760 761 case SOCK_STREAM: 762 unp2->unp_conn = unp; 763 if (req == PRU_CONNECT && 764 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 765 soisconnecting(so); 766 else 767 soisconnected(so); 768 soisconnected(so2); 769 break; 770 771 default: 772 panic("unp_connect2"); 773 } 774 return (0); 775 } 776 777 void 778 unp_disconnect(unp) 779 struct unpcb *unp; 780 { 781 struct unpcb *unp2 = unp->unp_conn; 782 783 if (unp2 == 0) 784 return; 785 unp->unp_conn = 0; 786 switch (unp->unp_socket->so_type) { 787 788 case SOCK_DGRAM: 789 if (unp2->unp_refs == unp) 790 unp2->unp_refs = unp->unp_nextref; 791 else { 792 unp2 = unp2->unp_refs; 793 for (;;) { 794 if (unp2 == 0) 795 panic("unp_disconnect"); 796 if (unp2->unp_nextref == unp) 797 break; 798 unp2 = unp2->unp_nextref; 799 } 800 unp2->unp_nextref = unp->unp_nextref; 801 } 802 unp->unp_nextref = 0; 803 unp->unp_socket->so_state &= ~SS_ISCONNECTED; 804 break; 805 806 case SOCK_STREAM: 807 soisdisconnected(unp->unp_socket); 808 unp2->unp_conn = 0; 809 soisdisconnected(unp2->unp_socket); 810 break; 811 } 812 } 813 814 #ifdef notdef 815 unp_abort(unp) 816 struct unpcb *unp; 817 { 818 819 unp_detach(unp); 820 } 821 #endif 822 823 void 824 unp_shutdown(unp) 825 struct unpcb *unp; 826 { 827 struct socket *so; 828 829 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn && 830 (so = unp->unp_conn->unp_socket)) 831 socantrcvmore(so); 832 } 833 834 void 835 unp_drop(unp, errno) 836 struct unpcb *unp; 837 int errno; 838 { 839 struct socket *so = unp->unp_socket; 840 841 so->so_error = errno; 842 unp_disconnect(unp); 843 if (so->so_head) { 844 so->so_pcb = 0; 845 sofree(so); 846 if (unp->unp_addr) 847 free(unp->unp_addr, M_SONAME); 848 free(unp, M_PCB); 849 } 850 } 851 852 #ifdef notdef 853 unp_drain() 854 { 855 856 } 857 #endif 858 859 int 860 unp_externalize(rights) 861 struct mbuf *rights; 862 { 863 struct proc *p = curproc; /* XXX */ 864 struct cmsghdr *cm = mtod(rights, struct cmsghdr *); 865 int i, *fdp; 866 struct file **rp; 867 struct file *fp; 868 int nfds, error = 0; 869 870 nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) / 871 sizeof(struct file *); 872 rp = (struct file **)CMSG_DATA(cm); 873 874 fdp = malloc(nfds * sizeof(int), M_TEMP, M_WAITOK); 875 876 /* Make sure the recipient should be able to see the descriptors.. */ 877 if (p->p_cwdi->cwdi_rdir != NULL) { 878 rp = (struct file **)CMSG_DATA(cm); 879 for (i = 0; i < nfds; i++) { 880 fp = *rp++; 881 /* 882 * If we are in a chroot'ed directory, and 883 * someone wants to pass us a directory, make 884 * sure it's inside the subtree we're allowed 885 * to access. 886 */ 887 if (fp->f_type == DTYPE_VNODE) { 888 struct vnode *vp = (struct vnode *)fp->f_data; 889 if ((vp->v_type == VDIR) && 890 !vn_isunder(vp, p->p_cwdi->cwdi_rdir, p)) { 891 error = EPERM; 892 break; 893 } 894 } 895 } 896 } 897 898 restart: 899 rp = (struct file **)CMSG_DATA(cm); 900 if (error != 0) { 901 for (i = 0; i < nfds; i++) { 902 fp = *rp; 903 /* 904 * zero the pointer before calling unp_discard, 905 * since it may end up in unp_gc().. 906 */ 907 *rp++ = 0; 908 unp_discard(fp); 909 } 910 goto out; 911 } 912 913 /* 914 * First loop -- allocate file descriptor table slots for the 915 * new descriptors. 916 */ 917 for (i = 0; i < nfds; i++) { 918 fp = *rp++; 919 if ((error = fdalloc(p, 0, &fdp[i])) != 0) { 920 /* 921 * Back out what we've done so far. 922 */ 923 for (--i; i >= 0; i--) 924 fdremove(p->p_fd, fdp[i]); 925 926 if (error == ENOSPC) { 927 fdexpand(p); 928 error = 0; 929 } else { 930 /* 931 * This is the error that has historically 932 * been returned, and some callers may 933 * expect it. 934 */ 935 error = EMSGSIZE; 936 } 937 goto restart; 938 } 939 940 /* 941 * Make the slot reference the descriptor so that 942 * fdalloc() works properly.. We finalize it all 943 * in the loop below. 944 */ 945 p->p_fd->fd_ofiles[fdp[i]] = fp; 946 } 947 948 /* 949 * Now that adding them has succeeded, update all of the 950 * descriptor passing state. 951 */ 952 rp = (struct file **)CMSG_DATA(cm); 953 for (i = 0; i < nfds; i++) { 954 fp = *rp++; 955 fp->f_msgcount--; 956 unp_rights--; 957 } 958 959 /* 960 * Copy temporary array to message and adjust length, in case of 961 * transition from large struct file pointers to ints. 962 */ 963 memcpy(CMSG_DATA(cm), fdp, nfds * sizeof(int)); 964 cm->cmsg_len = CMSG_LEN(nfds * sizeof(int)); 965 rights->m_len = CMSG_SPACE(nfds * sizeof(int)); 966 out: 967 free(fdp, M_TEMP); 968 return (error); 969 } 970 971 int 972 unp_internalize(control, p) 973 struct mbuf *control; 974 struct proc *p; 975 { 976 struct filedesc *fdescp = p->p_fd; 977 struct cmsghdr *newcm, *cm = mtod(control, struct cmsghdr *); 978 struct file **rp, **files; 979 struct file *fp; 980 int i, fd, *fdp; 981 int nfds; 982 u_int neededspace; 983 984 /* Sanity check the control message header */ 985 if (cm->cmsg_type != SCM_RIGHTS || cm->cmsg_level != SOL_SOCKET || 986 cm->cmsg_len != control->m_len) 987 return (EINVAL); 988 989 /* Verify that the file descriptors are valid */ 990 nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) / sizeof(int); 991 fdp = (int *)CMSG_DATA(cm); 992 for (i = 0; i < nfds; i++) { 993 fd = *fdp++; 994 if ((fp = fd_getfile(fdescp, fd)) == NULL) 995 return (EBADF); 996 simple_unlock(&fp->f_slock); 997 } 998 999 /* Make sure we have room for the struct file pointers */ 1000 neededspace = CMSG_SPACE(nfds * sizeof(struct file *)) - 1001 control->m_len; 1002 if (neededspace > M_TRAILINGSPACE(control)) { 1003 1004 /* allocate new space and copy header into it */ 1005 newcm = malloc( 1006 CMSG_SPACE(nfds * sizeof(struct file *)), 1007 M_MBUF, M_WAITOK); 1008 if (newcm == NULL) 1009 return (E2BIG); 1010 memcpy(newcm, cm, sizeof(struct cmsghdr)); 1011 files = (struct file **)CMSG_DATA(newcm); 1012 } else { 1013 /* we can convert in-place */ 1014 newcm = NULL; 1015 files = (struct file **)CMSG_DATA(cm); 1016 } 1017 1018 /* 1019 * Transform the file descriptors into struct file pointers, in 1020 * reverse order so that if pointers are bigger than ints, the 1021 * int won't get until we're done. 1022 */ 1023 fdp = (int *)CMSG_DATA(cm) + nfds - 1; 1024 rp = files + nfds - 1; 1025 for (i = 0; i < nfds; i++) { 1026 fp = fdescp->fd_ofiles[*fdp--]; 1027 simple_lock(&fp->f_slock); 1028 #ifdef DIAGNOSTIC 1029 if (fp->f_iflags & FIF_WANTCLOSE) 1030 panic("unp_internalize: file already closed"); 1031 #endif 1032 *rp-- = fp; 1033 fp->f_count++; 1034 fp->f_msgcount++; 1035 simple_unlock(&fp->f_slock); 1036 unp_rights++; 1037 } 1038 1039 if (newcm) { 1040 if (control->m_flags & M_EXT) 1041 MEXTREMOVE(control); 1042 MEXTADD(control, newcm, 1043 CMSG_SPACE(nfds * sizeof(struct file *)), 1044 M_MBUF, NULL, NULL); 1045 cm = newcm; 1046 } 1047 1048 /* adjust message & mbuf to note amount of space actually used. */ 1049 cm->cmsg_len = CMSG_LEN(nfds * sizeof(struct file *)); 1050 control->m_len = CMSG_SPACE(nfds * sizeof(struct file *)); 1051 1052 return (0); 1053 } 1054 1055 struct mbuf * 1056 unp_addsockcred(p, control) 1057 struct proc *p; 1058 struct mbuf *control; 1059 { 1060 struct cmsghdr *cmp; 1061 struct sockcred *sc; 1062 struct mbuf *m, *n; 1063 int len, space, i; 1064 1065 len = CMSG_LEN(SOCKCREDSIZE(p->p_ucred->cr_ngroups)); 1066 space = CMSG_SPACE(SOCKCREDSIZE(p->p_ucred->cr_ngroups)); 1067 1068 m = m_get(M_WAIT, MT_CONTROL); 1069 if (space > MLEN) { 1070 if (space > MCLBYTES) 1071 MEXTMALLOC(m, space, M_WAITOK); 1072 else 1073 m_clget(m, M_WAIT); 1074 if ((m->m_flags & M_EXT) == 0) { 1075 m_free(m); 1076 return (control); 1077 } 1078 } 1079 1080 m->m_len = space; 1081 m->m_next = NULL; 1082 cmp = mtod(m, struct cmsghdr *); 1083 sc = (struct sockcred *)CMSG_DATA(cmp); 1084 cmp->cmsg_len = len; 1085 cmp->cmsg_level = SOL_SOCKET; 1086 cmp->cmsg_type = SCM_CREDS; 1087 sc->sc_uid = p->p_cred->p_ruid; 1088 sc->sc_euid = p->p_ucred->cr_uid; 1089 sc->sc_gid = p->p_cred->p_rgid; 1090 sc->sc_egid = p->p_ucred->cr_gid; 1091 sc->sc_ngroups = p->p_ucred->cr_ngroups; 1092 for (i = 0; i < sc->sc_ngroups; i++) 1093 sc->sc_groups[i] = p->p_ucred->cr_groups[i]; 1094 1095 /* 1096 * If a control message already exists, append us to the end. 1097 */ 1098 if (control != NULL) { 1099 for (n = control; n->m_next != NULL; n = n->m_next) 1100 ; 1101 n->m_next = m; 1102 } else 1103 control = m; 1104 1105 return (control); 1106 } 1107 1108 int unp_defer, unp_gcing; 1109 extern struct domain unixdomain; 1110 1111 /* 1112 * Comment added long after the fact explaining what's going on here. 1113 * Do a mark-sweep GC of file descriptors on the system, to free up 1114 * any which are caught in flight to an about-to-be-closed socket. 1115 * 1116 * Traditional mark-sweep gc's start at the "root", and mark 1117 * everything reachable from the root (which, in our case would be the 1118 * process table). The mark bits are cleared during the sweep. 1119 * 1120 * XXX For some inexplicable reason (perhaps because the file 1121 * descriptor tables used to live in the u area which could be swapped 1122 * out and thus hard to reach), we do multiple scans over the set of 1123 * descriptors, using use *two* mark bits per object (DEFER and MARK). 1124 * Whenever we find a descriptor which references other descriptors, 1125 * the ones it references are marked with both bits, and we iterate 1126 * over the whole file table until there are no more DEFER bits set. 1127 * We also make an extra pass *before* the GC to clear the mark bits, 1128 * which could have been cleared at almost no cost during the previous 1129 * sweep. 1130 * 1131 * XXX MP: this needs to run with locks such that no other thread of 1132 * control can create or destroy references to file descriptors. it 1133 * may be necessary to defer the GC until later (when the locking 1134 * situation is more hospitable); it may be necessary to push this 1135 * into a separate thread. 1136 */ 1137 void 1138 unp_gc() 1139 { 1140 struct file *fp, *nextfp; 1141 struct socket *so, *so1; 1142 struct file **extra_ref, **fpp; 1143 int nunref, i; 1144 1145 if (unp_gcing) 1146 return; 1147 unp_gcing = 1; 1148 unp_defer = 0; 1149 1150 /* Clear mark bits */ 1151 LIST_FOREACH(fp, &filehead, f_list) 1152 fp->f_flag &= ~(FMARK|FDEFER); 1153 1154 /* 1155 * Iterate over the set of descriptors, marking ones believed 1156 * (based on refcount) to be referenced from a process, and 1157 * marking for rescan descriptors which are queued on a socket. 1158 */ 1159 do { 1160 LIST_FOREACH(fp, &filehead, f_list) { 1161 if (fp->f_flag & FDEFER) { 1162 fp->f_flag &= ~FDEFER; 1163 unp_defer--; 1164 #ifdef DIAGNOSTIC 1165 if (fp->f_count == 0) 1166 panic("unp_gc: deferred unreferenced socket"); 1167 #endif 1168 } else { 1169 if (fp->f_count == 0) 1170 continue; 1171 if (fp->f_flag & FMARK) 1172 continue; 1173 if (fp->f_count == fp->f_msgcount) 1174 continue; 1175 } 1176 fp->f_flag |= FMARK; 1177 1178 if (fp->f_type != DTYPE_SOCKET || 1179 (so = (struct socket *)fp->f_data) == 0) 1180 continue; 1181 if (so->so_proto->pr_domain != &unixdomain || 1182 (so->so_proto->pr_flags&PR_RIGHTS) == 0) 1183 continue; 1184 #ifdef notdef 1185 if (so->so_rcv.sb_flags & SB_LOCK) { 1186 /* 1187 * This is problematical; it's not clear 1188 * we need to wait for the sockbuf to be 1189 * unlocked (on a uniprocessor, at least), 1190 * and it's also not clear what to do 1191 * if sbwait returns an error due to receipt 1192 * of a signal. If sbwait does return 1193 * an error, we'll go into an infinite 1194 * loop. Delete all of this for now. 1195 */ 1196 (void) sbwait(&so->so_rcv); 1197 goto restart; 1198 } 1199 #endif 1200 unp_scan(so->so_rcv.sb_mb, unp_mark, 0); 1201 /* 1202 * mark descriptors referenced from sockets queued on the accept queue as well. 1203 */ 1204 if (so->so_options & SO_ACCEPTCONN) { 1205 TAILQ_FOREACH(so1, &so->so_q0, so_qe) { 1206 unp_scan(so1->so_rcv.sb_mb, unp_mark, 0); 1207 } 1208 TAILQ_FOREACH(so1, &so->so_q, so_qe) { 1209 unp_scan(so1->so_rcv.sb_mb, unp_mark, 0); 1210 } 1211 } 1212 1213 } 1214 } while (unp_defer); 1215 /* 1216 * Sweep pass. Find unmarked descriptors, and free them. 1217 * 1218 * We grab an extra reference to each of the file table entries 1219 * that are not otherwise accessible and then free the rights 1220 * that are stored in messages on them. 1221 * 1222 * The bug in the original code is a little tricky, so I'll describe 1223 * what's wrong with it here. 1224 * 1225 * It is incorrect to simply unp_discard each entry for f_msgcount 1226 * times -- consider the case of sockets A and B that contain 1227 * references to each other. On a last close of some other socket, 1228 * we trigger a gc since the number of outstanding rights (unp_rights) 1229 * is non-zero. If during the sweep phase the gc code un_discards, 1230 * we end up doing a (full) closef on the descriptor. A closef on A 1231 * results in the following chain. Closef calls soo_close, which 1232 * calls soclose. Soclose calls first (through the switch 1233 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply 1234 * returns because the previous instance had set unp_gcing, and 1235 * we return all the way back to soclose, which marks the socket 1236 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush 1237 * to free up the rights that are queued in messages on the socket A, 1238 * i.e., the reference on B. The sorflush calls via the dom_dispose 1239 * switch unp_dispose, which unp_scans with unp_discard. This second 1240 * instance of unp_discard just calls closef on B. 1241 * 1242 * Well, a similar chain occurs on B, resulting in a sorflush on B, 1243 * which results in another closef on A. Unfortunately, A is already 1244 * being closed, and the descriptor has already been marked with 1245 * SS_NOFDREF, and soclose panics at this point. 1246 * 1247 * Here, we first take an extra reference to each inaccessible 1248 * descriptor. Then, if the inaccessible descriptor is a 1249 * socket, we call sorflush in case it is a Unix domain 1250 * socket. After we destroy all the rights carried in 1251 * messages, we do a last closef to get rid of our extra 1252 * reference. This is the last close, and the unp_detach etc 1253 * will shut down the socket. 1254 * 1255 * 91/09/19, bsy@cs.cmu.edu 1256 */ 1257 extra_ref = malloc(nfiles * sizeof(struct file *), M_FILE, M_WAITOK); 1258 for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref; fp != 0; 1259 fp = nextfp) { 1260 nextfp = LIST_NEXT(fp, f_list); 1261 simple_lock(&fp->f_slock); 1262 if (fp->f_count != 0 && 1263 fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) { 1264 *fpp++ = fp; 1265 nunref++; 1266 fp->f_count++; 1267 } 1268 simple_unlock(&fp->f_slock); 1269 } 1270 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1271 fp = *fpp; 1272 simple_lock(&fp->f_slock); 1273 FILE_USE(fp); 1274 if (fp->f_type == DTYPE_SOCKET) 1275 sorflush((struct socket *)fp->f_data); 1276 FILE_UNUSE(fp, NULL); 1277 } 1278 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1279 fp = *fpp; 1280 simple_lock(&fp->f_slock); 1281 FILE_USE(fp); 1282 (void) closef(fp, (struct proc *)0); 1283 } 1284 free((caddr_t)extra_ref, M_FILE); 1285 unp_gcing = 0; 1286 } 1287 1288 void 1289 unp_dispose(m) 1290 struct mbuf *m; 1291 { 1292 1293 if (m) 1294 unp_scan(m, unp_discard, 1); 1295 } 1296 1297 void 1298 unp_scan(m0, op, discard) 1299 struct mbuf *m0; 1300 void (*op) __P((struct file *)); 1301 int discard; 1302 { 1303 struct mbuf *m; 1304 struct file **rp; 1305 struct cmsghdr *cm; 1306 int i; 1307 int qfds; 1308 1309 while (m0) { 1310 for (m = m0; m; m = m->m_next) { 1311 if (m->m_type == MT_CONTROL && 1312 m->m_len >= sizeof(*cm)) { 1313 cm = mtod(m, struct cmsghdr *); 1314 if (cm->cmsg_level != SOL_SOCKET || 1315 cm->cmsg_type != SCM_RIGHTS) 1316 continue; 1317 qfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) 1318 / sizeof(struct file *); 1319 rp = (struct file **)CMSG_DATA(cm); 1320 for (i = 0; i < qfds; i++) { 1321 struct file *fp = *rp; 1322 if (discard) 1323 *rp = 0; 1324 (*op)(fp); 1325 rp++; 1326 } 1327 break; /* XXX, but saves time */ 1328 } 1329 } 1330 m0 = m0->m_nextpkt; 1331 } 1332 } 1333 1334 void 1335 unp_mark(fp) 1336 struct file *fp; 1337 { 1338 if (fp == NULL) 1339 return; 1340 1341 if (fp->f_flag & FMARK) 1342 return; 1343 1344 /* If we're already deferred, don't screw up the defer count */ 1345 if (fp->f_flag & FDEFER) 1346 return; 1347 1348 /* 1349 * Minimize the number of deferrals... Sockets are the only 1350 * type of descriptor which can hold references to another 1351 * descriptor, so just mark other descriptors, and defer 1352 * unmarked sockets for the next pass. 1353 */ 1354 if (fp->f_type == DTYPE_SOCKET) { 1355 unp_defer++; 1356 if (fp->f_count == 0) 1357 panic("unp_mark: queued unref"); 1358 fp->f_flag |= FDEFER; 1359 } else { 1360 fp->f_flag |= FMARK; 1361 } 1362 return; 1363 } 1364 1365 void 1366 unp_discard(fp) 1367 struct file *fp; 1368 { 1369 if (fp == NULL) 1370 return; 1371 simple_lock(&fp->f_slock); 1372 fp->f_usecount++; /* i.e. FILE_USE(fp) sans locking */ 1373 fp->f_msgcount--; 1374 simple_unlock(&fp->f_slock); 1375 unp_rights--; 1376 (void) closef(fp, (struct proc *)0); 1377 } 1378