1 /* $NetBSD: uipc_usrreq.c,v 1.79 2004/09/03 18:14:09 darrenr Exp $ */ 2 3 /*- 4 * Copyright (c) 1998, 2000, 2004 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 1.79 2004/09/03 18:14:09 darrenr 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 const 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(struct proc *, struct mbuf *); 136 137 int 138 unp_output(struct mbuf *m, struct mbuf *control, struct unpcb *unp, 139 struct proc *p) 140 { 141 struct socket *so2; 142 const struct sockaddr_un *sun; 143 144 so2 = unp->unp_conn->unp_socket; 145 if (unp->unp_addr) 146 sun = unp->unp_addr; 147 else 148 sun = &sun_noname; 149 if (unp->unp_conn->unp_flags & UNP_WANTCRED) 150 control = unp_addsockcred(p, control); 151 if (sbappendaddr(&so2->so_rcv, (struct sockaddr *)sun, m, 152 control) == 0) { 153 m_freem(control); 154 m_freem(m); 155 so2->so_rcv.sb_overflowed++; 156 return (ENOBUFS); 157 } else { 158 sorwakeup(so2); 159 return (0); 160 } 161 } 162 163 void 164 unp_setsockaddr(struct unpcb *unp, struct mbuf *nam) 165 { 166 const struct sockaddr_un *sun; 167 168 if (unp->unp_addr) 169 sun = unp->unp_addr; 170 else 171 sun = &sun_noname; 172 nam->m_len = sun->sun_len; 173 if (nam->m_len > MLEN) 174 MEXTMALLOC(nam, nam->m_len, M_WAITOK); 175 memcpy(mtod(nam, caddr_t), sun, (size_t)nam->m_len); 176 } 177 178 void 179 unp_setpeeraddr(struct unpcb *unp, struct mbuf *nam) 180 { 181 const struct sockaddr_un *sun; 182 183 if (unp->unp_conn && unp->unp_conn->unp_addr) 184 sun = unp->unp_conn->unp_addr; 185 else 186 sun = &sun_noname; 187 nam->m_len = sun->sun_len; 188 if (nam->m_len > MLEN) 189 MEXTMALLOC(nam, nam->m_len, M_WAITOK); 190 memcpy(mtod(nam, caddr_t), sun, (size_t)nam->m_len); 191 } 192 193 /*ARGSUSED*/ 194 int 195 uipc_usrreq(struct socket *so, int req, struct mbuf *m, struct mbuf *nam, 196 struct mbuf *control, struct proc *p) 197 { 198 struct unpcb *unp = sotounpcb(so); 199 struct socket *so2; 200 u_int newhiwat; 201 int error = 0; 202 203 if (req == PRU_CONTROL) 204 return (EOPNOTSUPP); 205 206 #ifdef DIAGNOSTIC 207 if (req != PRU_SEND && req != PRU_SENDOOB && control) 208 panic("uipc_usrreq: unexpected control mbuf"); 209 #endif 210 if (unp == 0 && req != PRU_ATTACH) { 211 error = EINVAL; 212 goto release; 213 } 214 215 switch (req) { 216 217 case PRU_ATTACH: 218 if (unp != 0) { 219 error = EISCONN; 220 break; 221 } 222 error = unp_attach(so); 223 break; 224 225 case PRU_DETACH: 226 unp_detach(unp); 227 break; 228 229 case PRU_BIND: 230 error = unp_bind(unp, nam, p); 231 break; 232 233 case PRU_LISTEN: 234 if (unp->unp_vnode == 0) 235 error = EINVAL; 236 break; 237 238 case PRU_CONNECT: 239 error = unp_connect(so, nam, p); 240 break; 241 242 case PRU_CONNECT2: 243 error = unp_connect2(so, (struct socket *)nam, PRU_CONNECT2); 244 break; 245 246 case PRU_DISCONNECT: 247 unp_disconnect(unp); 248 break; 249 250 case PRU_ACCEPT: 251 unp_setpeeraddr(unp, nam); 252 /* 253 * Mark the initiating STREAM socket as connected *ONLY* 254 * after it's been accepted. This prevents a client from 255 * overrunning a server and receiving ECONNREFUSED. 256 */ 257 if (unp->unp_conn != NULL && 258 (unp->unp_conn->unp_socket->so_state & SS_ISCONNECTING)) 259 soisconnected(unp->unp_conn->unp_socket); 260 break; 261 262 case PRU_SHUTDOWN: 263 socantsendmore(so); 264 unp_shutdown(unp); 265 break; 266 267 case PRU_RCVD: 268 switch (so->so_type) { 269 270 case SOCK_DGRAM: 271 panic("uipc 1"); 272 /*NOTREACHED*/ 273 274 case SOCK_STREAM: 275 #define rcv (&so->so_rcv) 276 #define snd (&so2->so_snd) 277 if (unp->unp_conn == 0) 278 break; 279 so2 = unp->unp_conn->unp_socket; 280 /* 281 * Adjust backpressure on sender 282 * and wakeup any waiting to write. 283 */ 284 snd->sb_mbmax += unp->unp_mbcnt - rcv->sb_mbcnt; 285 unp->unp_mbcnt = rcv->sb_mbcnt; 286 newhiwat = snd->sb_hiwat + unp->unp_cc - rcv->sb_cc; 287 (void)chgsbsize(so2->so_uid, 288 &snd->sb_hiwat, newhiwat, RLIM_INFINITY); 289 unp->unp_cc = rcv->sb_cc; 290 sowwakeup(so2); 291 #undef snd 292 #undef rcv 293 break; 294 295 default: 296 panic("uipc 2"); 297 } 298 break; 299 300 case PRU_SEND: 301 /* 302 * Note: unp_internalize() rejects any control message 303 * other than SCM_RIGHTS, and only allows one. This 304 * has the side-effect of preventing a caller from 305 * forging SCM_CREDS. 306 */ 307 if (control && (error = unp_internalize(control, p))) 308 break; 309 switch (so->so_type) { 310 311 case SOCK_DGRAM: { 312 if (nam) { 313 if ((so->so_state & SS_ISCONNECTED) != 0) { 314 error = EISCONN; 315 goto die; 316 } 317 error = unp_connect(so, nam, p); 318 if (error) { 319 die: 320 m_freem(control); 321 m_freem(m); 322 break; 323 } 324 } else { 325 if ((so->so_state & SS_ISCONNECTED) == 0) { 326 error = ENOTCONN; 327 goto die; 328 } 329 } 330 error = unp_output(m, control, unp, p); 331 if (nam) 332 unp_disconnect(unp); 333 break; 334 } 335 336 case SOCK_STREAM: 337 #define rcv (&so2->so_rcv) 338 #define snd (&so->so_snd) 339 if (unp->unp_conn == 0) 340 panic("uipc 3"); 341 so2 = unp->unp_conn->unp_socket; 342 if (unp->unp_conn->unp_flags & UNP_WANTCRED) { 343 /* 344 * Credentials are passed only once on 345 * SOCK_STREAM. 346 */ 347 unp->unp_conn->unp_flags &= ~UNP_WANTCRED; 348 control = unp_addsockcred(p, control); 349 } 350 /* 351 * Send to paired receive port, and then reduce 352 * send buffer hiwater marks to maintain backpressure. 353 * Wake up readers. 354 */ 355 if (control) { 356 if (sbappendcontrol(rcv, m, control) == 0) 357 m_freem(control); 358 } else 359 sbappend(rcv, m); 360 snd->sb_mbmax -= 361 rcv->sb_mbcnt - unp->unp_conn->unp_mbcnt; 362 unp->unp_conn->unp_mbcnt = rcv->sb_mbcnt; 363 newhiwat = snd->sb_hiwat - 364 (rcv->sb_cc - unp->unp_conn->unp_cc); 365 (void)chgsbsize(so->so_uid, 366 &snd->sb_hiwat, newhiwat, RLIM_INFINITY); 367 unp->unp_conn->unp_cc = rcv->sb_cc; 368 sorwakeup(so2); 369 #undef snd 370 #undef rcv 371 break; 372 373 default: 374 panic("uipc 4"); 375 } 376 break; 377 378 case PRU_ABORT: 379 unp_drop(unp, ECONNABORTED); 380 381 #ifdef DIAGNOSTIC 382 if (so->so_pcb == 0) 383 panic("uipc 5: drop killed pcb"); 384 #endif 385 unp_detach(unp); 386 break; 387 388 case PRU_SENSE: 389 ((struct stat *) m)->st_blksize = so->so_snd.sb_hiwat; 390 if (so->so_type == SOCK_STREAM && unp->unp_conn != 0) { 391 so2 = unp->unp_conn->unp_socket; 392 ((struct stat *) m)->st_blksize += so2->so_rcv.sb_cc; 393 } 394 ((struct stat *) m)->st_dev = NODEV; 395 if (unp->unp_ino == 0) 396 unp->unp_ino = unp_ino++; 397 ((struct stat *) m)->st_atimespec = 398 ((struct stat *) m)->st_mtimespec = 399 ((struct stat *) m)->st_ctimespec = unp->unp_ctime; 400 ((struct stat *) m)->st_ino = unp->unp_ino; 401 return (0); 402 403 case PRU_RCVOOB: 404 error = EOPNOTSUPP; 405 break; 406 407 case PRU_SENDOOB: 408 m_freem(control); 409 m_freem(m); 410 error = EOPNOTSUPP; 411 break; 412 413 case PRU_SOCKADDR: 414 unp_setsockaddr(unp, nam); 415 break; 416 417 case PRU_PEERADDR: 418 unp_setpeeraddr(unp, nam); 419 break; 420 421 default: 422 panic("piusrreq"); 423 } 424 425 release: 426 return (error); 427 } 428 429 /* 430 * Unix domain socket option processing. 431 */ 432 int 433 uipc_ctloutput(int op, struct socket *so, int level, int optname, 434 struct mbuf **mp) 435 { 436 struct unpcb *unp = sotounpcb(so); 437 struct mbuf *m = *mp; 438 int optval = 0, error = 0; 439 440 if (level != 0) { 441 error = EINVAL; 442 if (op == PRCO_SETOPT && m) 443 (void) m_free(m); 444 } else switch (op) { 445 446 case PRCO_SETOPT: 447 switch (optname) { 448 case LOCAL_CREDS: 449 case LOCAL_CONNWAIT: 450 if (m == NULL || m->m_len != sizeof(int)) 451 error = EINVAL; 452 else { 453 optval = *mtod(m, int *); 454 switch (optname) { 455 #define OPTSET(bit) \ 456 if (optval) \ 457 unp->unp_flags |= (bit); \ 458 else \ 459 unp->unp_flags &= ~(bit); 460 461 case LOCAL_CREDS: 462 OPTSET(UNP_WANTCRED); 463 break; 464 case LOCAL_CONNWAIT: 465 OPTSET(UNP_CONNWAIT); 466 break; 467 } 468 } 469 break; 470 #undef OPTSET 471 472 default: 473 error = ENOPROTOOPT; 474 break; 475 } 476 if (m) 477 (void) m_free(m); 478 break; 479 480 case PRCO_GETOPT: 481 switch (optname) { 482 case LOCAL_CREDS: 483 *mp = m = m_get(M_WAIT, MT_SOOPTS); 484 m->m_len = sizeof(int); 485 switch (optname) { 486 487 #define OPTBIT(bit) (unp->unp_flags & (bit) ? 1 : 0) 488 489 case LOCAL_CREDS: 490 optval = OPTBIT(UNP_WANTCRED); 491 break; 492 } 493 *mtod(m, int *) = optval; 494 break; 495 #undef OPTBIT 496 497 default: 498 error = ENOPROTOOPT; 499 break; 500 } 501 break; 502 } 503 return (error); 504 } 505 506 /* 507 * Both send and receive buffers are allocated PIPSIZ bytes of buffering 508 * for stream sockets, although the total for sender and receiver is 509 * actually only PIPSIZ. 510 * Datagram sockets really use the sendspace as the maximum datagram size, 511 * and don't really want to reserve the sendspace. Their recvspace should 512 * be large enough for at least one max-size datagram plus address. 513 */ 514 #define PIPSIZ 4096 515 u_long unpst_sendspace = PIPSIZ; 516 u_long unpst_recvspace = PIPSIZ; 517 u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 518 u_long unpdg_recvspace = 4*1024; 519 520 int unp_rights; /* file descriptors in flight */ 521 522 int 523 unp_attach(struct socket *so) 524 { 525 struct unpcb *unp; 526 struct timeval tv; 527 int error; 528 529 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 530 switch (so->so_type) { 531 532 case SOCK_STREAM: 533 error = soreserve(so, unpst_sendspace, unpst_recvspace); 534 break; 535 536 case SOCK_DGRAM: 537 error = soreserve(so, unpdg_sendspace, unpdg_recvspace); 538 break; 539 540 default: 541 panic("unp_attach"); 542 } 543 if (error) 544 return (error); 545 } 546 unp = malloc(sizeof(*unp), M_PCB, M_NOWAIT); 547 if (unp == NULL) 548 return (ENOBUFS); 549 memset((caddr_t)unp, 0, sizeof(*unp)); 550 unp->unp_socket = so; 551 so->so_pcb = unp; 552 microtime(&tv); 553 TIMEVAL_TO_TIMESPEC(&tv, &unp->unp_ctime); 554 return (0); 555 } 556 557 void 558 unp_detach(struct unpcb *unp) 559 { 560 561 if (unp->unp_vnode) { 562 unp->unp_vnode->v_socket = 0; 563 vrele(unp->unp_vnode); 564 unp->unp_vnode = 0; 565 } 566 if (unp->unp_conn) 567 unp_disconnect(unp); 568 while (unp->unp_refs) 569 unp_drop(unp->unp_refs, ECONNRESET); 570 soisdisconnected(unp->unp_socket); 571 unp->unp_socket->so_pcb = 0; 572 if (unp->unp_addr) 573 free(unp->unp_addr, M_SONAME); 574 if (unp_rights) { 575 /* 576 * Normally the receive buffer is flushed later, 577 * in sofree, but if our receive buffer holds references 578 * to descriptors that are now garbage, we will dispose 579 * of those descriptor references after the garbage collector 580 * gets them (resulting in a "panic: closef: count < 0"). 581 */ 582 sorflush(unp->unp_socket); 583 free(unp, M_PCB); 584 unp_gc(); 585 } else 586 free(unp, M_PCB); 587 } 588 589 int 590 unp_bind(struct unpcb *unp, struct mbuf *nam, struct proc *p) 591 { 592 struct sockaddr_un *sun; 593 struct vnode *vp; 594 struct mount *mp; 595 struct vattr vattr; 596 size_t addrlen; 597 int error; 598 struct nameidata nd; 599 600 if (unp->unp_vnode != 0) 601 return (EINVAL); 602 603 /* 604 * Allocate the new sockaddr. We have to allocate one 605 * extra byte so that we can ensure that the pathname 606 * is nul-terminated. 607 */ 608 addrlen = nam->m_len + 1; 609 sun = malloc(addrlen, M_SONAME, M_WAITOK); 610 m_copydata(nam, 0, nam->m_len, (caddr_t)sun); 611 *(((char *)sun) + nam->m_len) = '\0'; 612 613 restart: 614 NDINIT(&nd, CREATE, FOLLOW | LOCKPARENT, UIO_SYSSPACE, 615 sun->sun_path, p); 616 617 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 618 if ((error = namei(&nd)) != 0) 619 goto bad; 620 vp = nd.ni_vp; 621 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { 622 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 623 if (nd.ni_dvp == vp) 624 vrele(nd.ni_dvp); 625 else 626 vput(nd.ni_dvp); 627 vrele(vp); 628 if (vp != NULL) { 629 error = EADDRINUSE; 630 goto bad; 631 } 632 error = vn_start_write(NULL, &mp, 633 V_WAIT | V_SLEEPONLY | V_PCATCH); 634 if (error) 635 goto bad; 636 goto restart; 637 } 638 VATTR_NULL(&vattr); 639 vattr.va_type = VSOCK; 640 vattr.va_mode = ACCESSPERMS; 641 VOP_LEASE(nd.ni_dvp, p, p->p_ucred, LEASE_WRITE); 642 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 643 vn_finished_write(mp, 0); 644 if (error) 645 goto bad; 646 vp = nd.ni_vp; 647 vp->v_socket = unp->unp_socket; 648 unp->unp_vnode = vp; 649 unp->unp_addrlen = addrlen; 650 unp->unp_addr = sun; 651 VOP_UNLOCK(vp, 0); 652 return (0); 653 654 bad: 655 free(sun, M_SONAME); 656 return (error); 657 } 658 659 int 660 unp_connect(struct socket *so, struct mbuf *nam, struct proc *p) 661 { 662 struct sockaddr_un *sun; 663 struct vnode *vp; 664 struct socket *so2, *so3; 665 struct unpcb *unp2, *unp3; 666 size_t addrlen; 667 int error; 668 struct nameidata nd; 669 670 /* 671 * Allocate a temporary sockaddr. We have to allocate one extra 672 * byte so that we can ensure that the pathname is nul-terminated. 673 * When we establish the connection, we copy the other PCB's 674 * sockaddr to our own. 675 */ 676 addrlen = nam->m_len + 1; 677 sun = malloc(addrlen, M_SONAME, M_WAITOK); 678 m_copydata(nam, 0, nam->m_len, (caddr_t)sun); 679 *(((char *)sun) + nam->m_len) = '\0'; 680 681 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, sun->sun_path, p); 682 683 if ((error = namei(&nd)) != 0) 684 goto bad2; 685 vp = nd.ni_vp; 686 if (vp->v_type != VSOCK) { 687 error = ENOTSOCK; 688 goto bad; 689 } 690 if ((error = VOP_ACCESS(vp, VWRITE, p->p_ucred, p)) != 0) 691 goto bad; 692 so2 = vp->v_socket; 693 if (so2 == 0) { 694 error = ECONNREFUSED; 695 goto bad; 696 } 697 if (so->so_type != so2->so_type) { 698 error = EPROTOTYPE; 699 goto bad; 700 } 701 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 702 if ((so2->so_options & SO_ACCEPTCONN) == 0 || 703 (so3 = sonewconn(so2, 0)) == 0) { 704 error = ECONNREFUSED; 705 goto bad; 706 } 707 unp2 = sotounpcb(so2); 708 unp3 = sotounpcb(so3); 709 if (unp2->unp_addr) { 710 unp3->unp_addr = malloc(unp2->unp_addrlen, 711 M_SONAME, M_WAITOK); 712 memcpy(unp3->unp_addr, unp2->unp_addr, 713 unp2->unp_addrlen); 714 unp3->unp_addrlen = unp2->unp_addrlen; 715 } 716 unp3->unp_flags = unp2->unp_flags; 717 so2 = so3; 718 } 719 error = unp_connect2(so, so2, PRU_CONNECT); 720 bad: 721 vput(vp); 722 bad2: 723 free(sun, M_SONAME); 724 return (error); 725 } 726 727 int 728 unp_connect2(struct socket *so, struct socket *so2, int req) 729 { 730 struct unpcb *unp = sotounpcb(so); 731 struct unpcb *unp2; 732 733 if (so2->so_type != so->so_type) 734 return (EPROTOTYPE); 735 unp2 = sotounpcb(so2); 736 unp->unp_conn = unp2; 737 switch (so->so_type) { 738 739 case SOCK_DGRAM: 740 unp->unp_nextref = unp2->unp_refs; 741 unp2->unp_refs = unp; 742 soisconnected(so); 743 break; 744 745 case SOCK_STREAM: 746 unp2->unp_conn = unp; 747 if (req == PRU_CONNECT && 748 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 749 soisconnecting(so); 750 else 751 soisconnected(so); 752 soisconnected(so2); 753 break; 754 755 default: 756 panic("unp_connect2"); 757 } 758 return (0); 759 } 760 761 void 762 unp_disconnect(struct unpcb *unp) 763 { 764 struct unpcb *unp2 = unp->unp_conn; 765 766 if (unp2 == 0) 767 return; 768 unp->unp_conn = 0; 769 switch (unp->unp_socket->so_type) { 770 771 case SOCK_DGRAM: 772 if (unp2->unp_refs == unp) 773 unp2->unp_refs = unp->unp_nextref; 774 else { 775 unp2 = unp2->unp_refs; 776 for (;;) { 777 if (unp2 == 0) 778 panic("unp_disconnect"); 779 if (unp2->unp_nextref == unp) 780 break; 781 unp2 = unp2->unp_nextref; 782 } 783 unp2->unp_nextref = unp->unp_nextref; 784 } 785 unp->unp_nextref = 0; 786 unp->unp_socket->so_state &= ~SS_ISCONNECTED; 787 break; 788 789 case SOCK_STREAM: 790 soisdisconnected(unp->unp_socket); 791 unp2->unp_conn = 0; 792 soisdisconnected(unp2->unp_socket); 793 break; 794 } 795 } 796 797 #ifdef notdef 798 unp_abort(struct unpcb *unp) 799 { 800 unp_detach(unp); 801 } 802 #endif 803 804 void 805 unp_shutdown(struct unpcb *unp) 806 { 807 struct socket *so; 808 809 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn && 810 (so = unp->unp_conn->unp_socket)) 811 socantrcvmore(so); 812 } 813 814 void 815 unp_drop(struct unpcb *unp, int errno) 816 { 817 struct socket *so = unp->unp_socket; 818 819 so->so_error = errno; 820 unp_disconnect(unp); 821 if (so->so_head) { 822 so->so_pcb = 0; 823 sofree(so); 824 if (unp->unp_addr) 825 free(unp->unp_addr, M_SONAME); 826 free(unp, M_PCB); 827 } 828 } 829 830 #ifdef notdef 831 unp_drain(void) 832 { 833 834 } 835 #endif 836 837 int 838 unp_externalize(struct mbuf *rights, struct proc *p) 839 { 840 struct cmsghdr *cm = mtod(rights, struct cmsghdr *); 841 int i, *fdp; 842 struct file **rp; 843 struct file *fp; 844 int nfds, error = 0; 845 846 nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) / 847 sizeof(struct file *); 848 rp = (struct file **)CMSG_DATA(cm); 849 850 fdp = malloc(nfds * sizeof(int), M_TEMP, M_WAITOK); 851 852 /* Make sure the recipient should be able to see the descriptors.. */ 853 if (p->p_cwdi->cwdi_rdir != NULL) { 854 rp = (struct file **)CMSG_DATA(cm); 855 for (i = 0; i < nfds; i++) { 856 fp = *rp++; 857 /* 858 * If we are in a chroot'ed directory, and 859 * someone wants to pass us a directory, make 860 * sure it's inside the subtree we're allowed 861 * to access. 862 */ 863 if (fp->f_type == DTYPE_VNODE) { 864 struct vnode *vp = (struct vnode *)fp->f_data; 865 if ((vp->v_type == VDIR) && 866 !vn_isunder(vp, p->p_cwdi->cwdi_rdir, p)) { 867 error = EPERM; 868 break; 869 } 870 } 871 } 872 } 873 874 restart: 875 rp = (struct file **)CMSG_DATA(cm); 876 if (error != 0) { 877 for (i = 0; i < nfds; i++) { 878 fp = *rp; 879 /* 880 * zero the pointer before calling unp_discard, 881 * since it may end up in unp_gc().. 882 */ 883 *rp++ = 0; 884 unp_discard(fp); 885 } 886 goto out; 887 } 888 889 /* 890 * First loop -- allocate file descriptor table slots for the 891 * new descriptors. 892 */ 893 for (i = 0; i < nfds; i++) { 894 fp = *rp++; 895 if ((error = fdalloc(p, 0, &fdp[i])) != 0) { 896 /* 897 * Back out what we've done so far. 898 */ 899 for (--i; i >= 0; i--) 900 fdremove(p->p_fd, fdp[i]); 901 902 if (error == ENOSPC) { 903 fdexpand(p); 904 error = 0; 905 } else { 906 /* 907 * This is the error that has historically 908 * been returned, and some callers may 909 * expect it. 910 */ 911 error = EMSGSIZE; 912 } 913 goto restart; 914 } 915 916 /* 917 * Make the slot reference the descriptor so that 918 * fdalloc() works properly.. We finalize it all 919 * in the loop below. 920 */ 921 p->p_fd->fd_ofiles[fdp[i]] = fp; 922 } 923 924 /* 925 * Now that adding them has succeeded, update all of the 926 * descriptor passing state. 927 */ 928 rp = (struct file **)CMSG_DATA(cm); 929 for (i = 0; i < nfds; i++) { 930 fp = *rp++; 931 fp->f_msgcount--; 932 unp_rights--; 933 } 934 935 /* 936 * Copy temporary array to message and adjust length, in case of 937 * transition from large struct file pointers to ints. 938 */ 939 memcpy(CMSG_DATA(cm), fdp, nfds * sizeof(int)); 940 cm->cmsg_len = CMSG_LEN(nfds * sizeof(int)); 941 rights->m_len = CMSG_SPACE(nfds * sizeof(int)); 942 out: 943 free(fdp, M_TEMP); 944 return (error); 945 } 946 947 int 948 unp_internalize(struct mbuf *control, struct proc *p) 949 { 950 struct filedesc *fdescp = p->p_fd; 951 struct cmsghdr *newcm, *cm = mtod(control, struct cmsghdr *); 952 struct file **rp, **files; 953 struct file *fp; 954 int i, fd, *fdp; 955 int nfds; 956 u_int neededspace; 957 958 /* Sanity check the control message header */ 959 if (cm->cmsg_type != SCM_RIGHTS || cm->cmsg_level != SOL_SOCKET || 960 cm->cmsg_len != control->m_len) 961 return (EINVAL); 962 963 /* Verify that the file descriptors are valid */ 964 nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) / sizeof(int); 965 fdp = (int *)CMSG_DATA(cm); 966 for (i = 0; i < nfds; i++) { 967 fd = *fdp++; 968 if ((fp = fd_getfile(fdescp, fd)) == NULL) 969 return (EBADF); 970 simple_unlock(&fp->f_slock); 971 } 972 973 /* Make sure we have room for the struct file pointers */ 974 neededspace = CMSG_SPACE(nfds * sizeof(struct file *)) - 975 control->m_len; 976 if (neededspace > M_TRAILINGSPACE(control)) { 977 978 /* allocate new space and copy header into it */ 979 newcm = malloc( 980 CMSG_SPACE(nfds * sizeof(struct file *)), 981 M_MBUF, M_WAITOK); 982 if (newcm == NULL) 983 return (E2BIG); 984 memcpy(newcm, cm, sizeof(struct cmsghdr)); 985 files = (struct file **)CMSG_DATA(newcm); 986 } else { 987 /* we can convert in-place */ 988 newcm = NULL; 989 files = (struct file **)CMSG_DATA(cm); 990 } 991 992 /* 993 * Transform the file descriptors into struct file pointers, in 994 * reverse order so that if pointers are bigger than ints, the 995 * int won't get until we're done. 996 */ 997 fdp = (int *)CMSG_DATA(cm) + nfds - 1; 998 rp = files + nfds - 1; 999 for (i = 0; i < nfds; i++) { 1000 fp = fdescp->fd_ofiles[*fdp--]; 1001 simple_lock(&fp->f_slock); 1002 #ifdef DIAGNOSTIC 1003 if (fp->f_iflags & FIF_WANTCLOSE) 1004 panic("unp_internalize: file already closed"); 1005 #endif 1006 *rp-- = fp; 1007 fp->f_count++; 1008 fp->f_msgcount++; 1009 simple_unlock(&fp->f_slock); 1010 unp_rights++; 1011 } 1012 1013 if (newcm) { 1014 if (control->m_flags & M_EXT) 1015 MEXTREMOVE(control); 1016 MEXTADD(control, newcm, 1017 CMSG_SPACE(nfds * sizeof(struct file *)), 1018 M_MBUF, NULL, NULL); 1019 cm = newcm; 1020 } 1021 1022 /* adjust message & mbuf to note amount of space actually used. */ 1023 cm->cmsg_len = CMSG_LEN(nfds * sizeof(struct file *)); 1024 control->m_len = CMSG_SPACE(nfds * sizeof(struct file *)); 1025 1026 return (0); 1027 } 1028 1029 struct mbuf * 1030 unp_addsockcred(struct proc *p, struct mbuf *control) 1031 { 1032 struct cmsghdr *cmp; 1033 struct sockcred *sc; 1034 struct mbuf *m, *n; 1035 int len, space, i; 1036 1037 len = CMSG_LEN(SOCKCREDSIZE(p->p_ucred->cr_ngroups)); 1038 space = CMSG_SPACE(SOCKCREDSIZE(p->p_ucred->cr_ngroups)); 1039 1040 m = m_get(M_WAIT, MT_CONTROL); 1041 if (space > MLEN) { 1042 if (space > MCLBYTES) 1043 MEXTMALLOC(m, space, M_WAITOK); 1044 else 1045 m_clget(m, M_WAIT); 1046 if ((m->m_flags & M_EXT) == 0) { 1047 m_free(m); 1048 return (control); 1049 } 1050 } 1051 1052 m->m_len = space; 1053 m->m_next = NULL; 1054 cmp = mtod(m, struct cmsghdr *); 1055 sc = (struct sockcred *)CMSG_DATA(cmp); 1056 cmp->cmsg_len = len; 1057 cmp->cmsg_level = SOL_SOCKET; 1058 cmp->cmsg_type = SCM_CREDS; 1059 sc->sc_uid = p->p_cred->p_ruid; 1060 sc->sc_euid = p->p_ucred->cr_uid; 1061 sc->sc_gid = p->p_cred->p_rgid; 1062 sc->sc_egid = p->p_ucred->cr_gid; 1063 sc->sc_ngroups = p->p_ucred->cr_ngroups; 1064 for (i = 0; i < sc->sc_ngroups; i++) 1065 sc->sc_groups[i] = p->p_ucred->cr_groups[i]; 1066 1067 /* 1068 * If a control message already exists, append us to the end. 1069 */ 1070 if (control != NULL) { 1071 for (n = control; n->m_next != NULL; n = n->m_next) 1072 ; 1073 n->m_next = m; 1074 } else 1075 control = m; 1076 1077 return (control); 1078 } 1079 1080 int unp_defer, unp_gcing; 1081 extern struct domain unixdomain; 1082 1083 /* 1084 * Comment added long after the fact explaining what's going on here. 1085 * Do a mark-sweep GC of file descriptors on the system, to free up 1086 * any which are caught in flight to an about-to-be-closed socket. 1087 * 1088 * Traditional mark-sweep gc's start at the "root", and mark 1089 * everything reachable from the root (which, in our case would be the 1090 * process table). The mark bits are cleared during the sweep. 1091 * 1092 * XXX For some inexplicable reason (perhaps because the file 1093 * descriptor tables used to live in the u area which could be swapped 1094 * out and thus hard to reach), we do multiple scans over the set of 1095 * descriptors, using use *two* mark bits per object (DEFER and MARK). 1096 * Whenever we find a descriptor which references other descriptors, 1097 * the ones it references are marked with both bits, and we iterate 1098 * over the whole file table until there are no more DEFER bits set. 1099 * We also make an extra pass *before* the GC to clear the mark bits, 1100 * which could have been cleared at almost no cost during the previous 1101 * sweep. 1102 * 1103 * XXX MP: this needs to run with locks such that no other thread of 1104 * control can create or destroy references to file descriptors. it 1105 * may be necessary to defer the GC until later (when the locking 1106 * situation is more hospitable); it may be necessary to push this 1107 * into a separate thread. 1108 */ 1109 void 1110 unp_gc(void) 1111 { 1112 struct file *fp, *nextfp; 1113 struct socket *so, *so1; 1114 struct file **extra_ref, **fpp; 1115 int nunref, i; 1116 1117 if (unp_gcing) 1118 return; 1119 unp_gcing = 1; 1120 unp_defer = 0; 1121 1122 /* Clear mark bits */ 1123 LIST_FOREACH(fp, &filehead, f_list) 1124 fp->f_flag &= ~(FMARK|FDEFER); 1125 1126 /* 1127 * Iterate over the set of descriptors, marking ones believed 1128 * (based on refcount) to be referenced from a process, and 1129 * marking for rescan descriptors which are queued on a socket. 1130 */ 1131 do { 1132 LIST_FOREACH(fp, &filehead, f_list) { 1133 if (fp->f_flag & FDEFER) { 1134 fp->f_flag &= ~FDEFER; 1135 unp_defer--; 1136 #ifdef DIAGNOSTIC 1137 if (fp->f_count == 0) 1138 panic("unp_gc: deferred unreferenced socket"); 1139 #endif 1140 } else { 1141 if (fp->f_count == 0) 1142 continue; 1143 if (fp->f_flag & FMARK) 1144 continue; 1145 if (fp->f_count == fp->f_msgcount) 1146 continue; 1147 } 1148 fp->f_flag |= FMARK; 1149 1150 if (fp->f_type != DTYPE_SOCKET || 1151 (so = (struct socket *)fp->f_data) == 0) 1152 continue; 1153 if (so->so_proto->pr_domain != &unixdomain || 1154 (so->so_proto->pr_flags&PR_RIGHTS) == 0) 1155 continue; 1156 #ifdef notdef 1157 if (so->so_rcv.sb_flags & SB_LOCK) { 1158 /* 1159 * This is problematical; it's not clear 1160 * we need to wait for the sockbuf to be 1161 * unlocked (on a uniprocessor, at least), 1162 * and it's also not clear what to do 1163 * if sbwait returns an error due to receipt 1164 * of a signal. If sbwait does return 1165 * an error, we'll go into an infinite 1166 * loop. Delete all of this for now. 1167 */ 1168 (void) sbwait(&so->so_rcv); 1169 goto restart; 1170 } 1171 #endif 1172 unp_scan(so->so_rcv.sb_mb, unp_mark, 0); 1173 /* 1174 * mark descriptors referenced from sockets queued on the accept queue as well. 1175 */ 1176 if (so->so_options & SO_ACCEPTCONN) { 1177 TAILQ_FOREACH(so1, &so->so_q0, so_qe) { 1178 unp_scan(so1->so_rcv.sb_mb, unp_mark, 0); 1179 } 1180 TAILQ_FOREACH(so1, &so->so_q, so_qe) { 1181 unp_scan(so1->so_rcv.sb_mb, unp_mark, 0); 1182 } 1183 } 1184 1185 } 1186 } while (unp_defer); 1187 /* 1188 * Sweep pass. Find unmarked descriptors, and free them. 1189 * 1190 * We grab an extra reference to each of the file table entries 1191 * that are not otherwise accessible and then free the rights 1192 * that are stored in messages on them. 1193 * 1194 * The bug in the original code is a little tricky, so I'll describe 1195 * what's wrong with it here. 1196 * 1197 * It is incorrect to simply unp_discard each entry for f_msgcount 1198 * times -- consider the case of sockets A and B that contain 1199 * references to each other. On a last close of some other socket, 1200 * we trigger a gc since the number of outstanding rights (unp_rights) 1201 * is non-zero. If during the sweep phase the gc code un_discards, 1202 * we end up doing a (full) closef on the descriptor. A closef on A 1203 * results in the following chain. Closef calls soo_close, which 1204 * calls soclose. Soclose calls first (through the switch 1205 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply 1206 * returns because the previous instance had set unp_gcing, and 1207 * we return all the way back to soclose, which marks the socket 1208 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush 1209 * to free up the rights that are queued in messages on the socket A, 1210 * i.e., the reference on B. The sorflush calls via the dom_dispose 1211 * switch unp_dispose, which unp_scans with unp_discard. This second 1212 * instance of unp_discard just calls closef on B. 1213 * 1214 * Well, a similar chain occurs on B, resulting in a sorflush on B, 1215 * which results in another closef on A. Unfortunately, A is already 1216 * being closed, and the descriptor has already been marked with 1217 * SS_NOFDREF, and soclose panics at this point. 1218 * 1219 * Here, we first take an extra reference to each inaccessible 1220 * descriptor. Then, if the inaccessible descriptor is a 1221 * socket, we call sorflush in case it is a Unix domain 1222 * socket. After we destroy all the rights carried in 1223 * messages, we do a last closef to get rid of our extra 1224 * reference. This is the last close, and the unp_detach etc 1225 * will shut down the socket. 1226 * 1227 * 91/09/19, bsy@cs.cmu.edu 1228 */ 1229 extra_ref = malloc(nfiles * sizeof(struct file *), M_FILE, M_WAITOK); 1230 for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref; fp != 0; 1231 fp = nextfp) { 1232 nextfp = LIST_NEXT(fp, f_list); 1233 simple_lock(&fp->f_slock); 1234 if (fp->f_count != 0 && 1235 fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) { 1236 *fpp++ = fp; 1237 nunref++; 1238 fp->f_count++; 1239 } 1240 simple_unlock(&fp->f_slock); 1241 } 1242 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1243 fp = *fpp; 1244 simple_lock(&fp->f_slock); 1245 FILE_USE(fp); 1246 if (fp->f_type == DTYPE_SOCKET) 1247 sorflush((struct socket *)fp->f_data); 1248 FILE_UNUSE(fp, NULL); 1249 } 1250 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1251 fp = *fpp; 1252 simple_lock(&fp->f_slock); 1253 FILE_USE(fp); 1254 (void) closef(fp, (struct proc *)0); 1255 } 1256 free((caddr_t)extra_ref, M_FILE); 1257 unp_gcing = 0; 1258 } 1259 1260 void 1261 unp_dispose(struct mbuf *m) 1262 { 1263 1264 if (m) 1265 unp_scan(m, unp_discard, 1); 1266 } 1267 1268 void 1269 unp_scan(struct mbuf *m0, void (*op)(struct file *), int discard) 1270 { 1271 struct mbuf *m; 1272 struct file **rp; 1273 struct cmsghdr *cm; 1274 int i; 1275 int qfds; 1276 1277 while (m0) { 1278 for (m = m0; m; m = m->m_next) { 1279 if (m->m_type == MT_CONTROL && 1280 m->m_len >= sizeof(*cm)) { 1281 cm = mtod(m, struct cmsghdr *); 1282 if (cm->cmsg_level != SOL_SOCKET || 1283 cm->cmsg_type != SCM_RIGHTS) 1284 continue; 1285 qfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) 1286 / sizeof(struct file *); 1287 rp = (struct file **)CMSG_DATA(cm); 1288 for (i = 0; i < qfds; i++) { 1289 struct file *fp = *rp; 1290 if (discard) 1291 *rp = 0; 1292 (*op)(fp); 1293 rp++; 1294 } 1295 break; /* XXX, but saves time */ 1296 } 1297 } 1298 m0 = m0->m_nextpkt; 1299 } 1300 } 1301 1302 void 1303 unp_mark(struct file *fp) 1304 { 1305 if (fp == NULL) 1306 return; 1307 1308 if (fp->f_flag & FMARK) 1309 return; 1310 1311 /* If we're already deferred, don't screw up the defer count */ 1312 if (fp->f_flag & FDEFER) 1313 return; 1314 1315 /* 1316 * Minimize the number of deferrals... Sockets are the only 1317 * type of descriptor which can hold references to another 1318 * descriptor, so just mark other descriptors, and defer 1319 * unmarked sockets for the next pass. 1320 */ 1321 if (fp->f_type == DTYPE_SOCKET) { 1322 unp_defer++; 1323 if (fp->f_count == 0) 1324 panic("unp_mark: queued unref"); 1325 fp->f_flag |= FDEFER; 1326 } else { 1327 fp->f_flag |= FMARK; 1328 } 1329 return; 1330 } 1331 1332 void 1333 unp_discard(struct file *fp) 1334 { 1335 if (fp == NULL) 1336 return; 1337 simple_lock(&fp->f_slock); 1338 fp->f_usecount++; /* i.e. FILE_USE(fp) sans locking */ 1339 fp->f_msgcount--; 1340 simple_unlock(&fp->f_slock); 1341 unp_rights--; 1342 (void) closef(fp, (struct proc *)0); 1343 } 1344