1 /* $NetBSD: uipc_usrreq.c,v 1.83 2005/06/16 14:36:42 yamt 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 1.83 2005/06/16 14:36:42 yamt 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, (const 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_uidinfo, 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 goto die; 309 } 310 switch (so->so_type) { 311 312 case SOCK_DGRAM: { 313 if (nam) { 314 if ((so->so_state & SS_ISCONNECTED) != 0) { 315 error = EISCONN; 316 goto die; 317 } 318 error = unp_connect(so, nam, p); 319 if (error) { 320 die: 321 m_freem(control); 322 m_freem(m); 323 break; 324 } 325 } else { 326 if ((so->so_state & SS_ISCONNECTED) == 0) { 327 error = ENOTCONN; 328 goto die; 329 } 330 } 331 error = unp_output(m, control, unp, p); 332 if (nam) 333 unp_disconnect(unp); 334 break; 335 } 336 337 case SOCK_STREAM: 338 #define rcv (&so2->so_rcv) 339 #define snd (&so->so_snd) 340 if (unp->unp_conn == 0) 341 panic("uipc 3"); 342 so2 = unp->unp_conn->unp_socket; 343 if (unp->unp_conn->unp_flags & UNP_WANTCRED) { 344 /* 345 * Credentials are passed only once on 346 * SOCK_STREAM. 347 */ 348 unp->unp_conn->unp_flags &= ~UNP_WANTCRED; 349 control = unp_addsockcred(p, control); 350 } 351 /* 352 * Send to paired receive port, and then reduce 353 * send buffer hiwater marks to maintain backpressure. 354 * Wake up readers. 355 */ 356 if (control) { 357 if (sbappendcontrol(rcv, m, control) == 0) 358 m_freem(control); 359 } else 360 sbappend(rcv, m); 361 snd->sb_mbmax -= 362 rcv->sb_mbcnt - unp->unp_conn->unp_mbcnt; 363 unp->unp_conn->unp_mbcnt = rcv->sb_mbcnt; 364 newhiwat = snd->sb_hiwat - 365 (rcv->sb_cc - unp->unp_conn->unp_cc); 366 (void)chgsbsize(so->so_uidinfo, 367 &snd->sb_hiwat, newhiwat, RLIM_INFINITY); 368 unp->unp_conn->unp_cc = rcv->sb_cc; 369 sorwakeup(so2); 370 #undef snd 371 #undef rcv 372 break; 373 374 default: 375 panic("uipc 4"); 376 } 377 break; 378 379 case PRU_ABORT: 380 unp_drop(unp, ECONNABORTED); 381 382 #ifdef DIAGNOSTIC 383 if (so->so_pcb == 0) 384 panic("uipc 5: drop killed pcb"); 385 #endif 386 unp_detach(unp); 387 break; 388 389 case PRU_SENSE: 390 ((struct stat *) m)->st_blksize = so->so_snd.sb_hiwat; 391 if (so->so_type == SOCK_STREAM && unp->unp_conn != 0) { 392 so2 = unp->unp_conn->unp_socket; 393 ((struct stat *) m)->st_blksize += so2->so_rcv.sb_cc; 394 } 395 ((struct stat *) m)->st_dev = NODEV; 396 if (unp->unp_ino == 0) 397 unp->unp_ino = unp_ino++; 398 ((struct stat *) m)->st_atimespec = 399 ((struct stat *) m)->st_mtimespec = 400 ((struct stat *) m)->st_ctimespec = unp->unp_ctime; 401 ((struct stat *) m)->st_ino = unp->unp_ino; 402 return (0); 403 404 case PRU_RCVOOB: 405 error = EOPNOTSUPP; 406 break; 407 408 case PRU_SENDOOB: 409 m_freem(control); 410 m_freem(m); 411 error = EOPNOTSUPP; 412 break; 413 414 case PRU_SOCKADDR: 415 unp_setsockaddr(unp, nam); 416 break; 417 418 case PRU_PEERADDR: 419 unp_setpeeraddr(unp, nam); 420 break; 421 422 default: 423 panic("piusrreq"); 424 } 425 426 release: 427 return (error); 428 } 429 430 /* 431 * Unix domain socket option processing. 432 */ 433 int 434 uipc_ctloutput(int op, struct socket *so, int level, int optname, 435 struct mbuf **mp) 436 { 437 struct unpcb *unp = sotounpcb(so); 438 struct mbuf *m = *mp; 439 int optval = 0, error = 0; 440 441 if (level != 0) { 442 error = EINVAL; 443 if (op == PRCO_SETOPT && m) 444 (void) m_free(m); 445 } else switch (op) { 446 447 case PRCO_SETOPT: 448 switch (optname) { 449 case LOCAL_CREDS: 450 case LOCAL_CONNWAIT: 451 if (m == NULL || m->m_len != sizeof(int)) 452 error = EINVAL; 453 else { 454 optval = *mtod(m, int *); 455 switch (optname) { 456 #define OPTSET(bit) \ 457 if (optval) \ 458 unp->unp_flags |= (bit); \ 459 else \ 460 unp->unp_flags &= ~(bit); 461 462 case LOCAL_CREDS: 463 OPTSET(UNP_WANTCRED); 464 break; 465 case LOCAL_CONNWAIT: 466 OPTSET(UNP_CONNWAIT); 467 break; 468 } 469 } 470 break; 471 #undef OPTSET 472 473 default: 474 error = ENOPROTOOPT; 475 break; 476 } 477 if (m) 478 (void) m_free(m); 479 break; 480 481 case PRCO_GETOPT: 482 switch (optname) { 483 case LOCAL_CREDS: 484 *mp = m = m_get(M_WAIT, MT_SOOPTS); 485 m->m_len = sizeof(int); 486 switch (optname) { 487 488 #define OPTBIT(bit) (unp->unp_flags & (bit) ? 1 : 0) 489 490 case LOCAL_CREDS: 491 optval = OPTBIT(UNP_WANTCRED); 492 break; 493 } 494 *mtod(m, int *) = optval; 495 break; 496 #undef OPTBIT 497 498 default: 499 error = ENOPROTOOPT; 500 break; 501 } 502 break; 503 } 504 return (error); 505 } 506 507 /* 508 * Both send and receive buffers are allocated PIPSIZ bytes of buffering 509 * for stream sockets, although the total for sender and receiver is 510 * actually only PIPSIZ. 511 * Datagram sockets really use the sendspace as the maximum datagram size, 512 * and don't really want to reserve the sendspace. Their recvspace should 513 * be large enough for at least one max-size datagram plus address. 514 */ 515 #define PIPSIZ 4096 516 u_long unpst_sendspace = PIPSIZ; 517 u_long unpst_recvspace = PIPSIZ; 518 u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 519 u_long unpdg_recvspace = 4*1024; 520 521 int unp_rights; /* file descriptors in flight */ 522 523 int 524 unp_attach(struct socket *so) 525 { 526 struct unpcb *unp; 527 struct timeval tv; 528 int error; 529 530 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 531 switch (so->so_type) { 532 533 case SOCK_STREAM: 534 error = soreserve(so, unpst_sendspace, unpst_recvspace); 535 break; 536 537 case SOCK_DGRAM: 538 error = soreserve(so, unpdg_sendspace, unpdg_recvspace); 539 break; 540 541 default: 542 panic("unp_attach"); 543 } 544 if (error) 545 return (error); 546 } 547 unp = malloc(sizeof(*unp), M_PCB, M_NOWAIT); 548 if (unp == NULL) 549 return (ENOBUFS); 550 memset((caddr_t)unp, 0, sizeof(*unp)); 551 unp->unp_socket = so; 552 so->so_pcb = unp; 553 microtime(&tv); 554 TIMEVAL_TO_TIMESPEC(&tv, &unp->unp_ctime); 555 return (0); 556 } 557 558 void 559 unp_detach(struct unpcb *unp) 560 { 561 562 if (unp->unp_vnode) { 563 unp->unp_vnode->v_socket = 0; 564 vrele(unp->unp_vnode); 565 unp->unp_vnode = 0; 566 } 567 if (unp->unp_conn) 568 unp_disconnect(unp); 569 while (unp->unp_refs) 570 unp_drop(unp->unp_refs, ECONNRESET); 571 soisdisconnected(unp->unp_socket); 572 unp->unp_socket->so_pcb = 0; 573 if (unp->unp_addr) 574 free(unp->unp_addr, M_SONAME); 575 if (unp_rights) { 576 /* 577 * Normally the receive buffer is flushed later, 578 * in sofree, but if our receive buffer holds references 579 * to descriptors that are now garbage, we will dispose 580 * of those descriptor references after the garbage collector 581 * gets them (resulting in a "panic: closef: count < 0"). 582 */ 583 sorflush(unp->unp_socket); 584 free(unp, M_PCB); 585 unp_gc(); 586 } else 587 free(unp, M_PCB); 588 } 589 590 int 591 unp_bind(struct unpcb *unp, struct mbuf *nam, struct proc *p) 592 { 593 struct sockaddr_un *sun; 594 struct vnode *vp; 595 struct mount *mp; 596 struct vattr vattr; 597 size_t addrlen; 598 int error; 599 struct nameidata nd; 600 601 if (unp->unp_vnode != 0) 602 return (EINVAL); 603 604 /* 605 * Allocate the new sockaddr. We have to allocate one 606 * extra byte so that we can ensure that the pathname 607 * is nul-terminated. 608 */ 609 addrlen = nam->m_len + 1; 610 sun = malloc(addrlen, M_SONAME, M_WAITOK); 611 m_copydata(nam, 0, nam->m_len, (caddr_t)sun); 612 *(((char *)sun) + nam->m_len) = '\0'; 613 614 restart: 615 NDINIT(&nd, CREATE, FOLLOW | LOCKPARENT, UIO_SYSSPACE, 616 sun->sun_path, p); 617 618 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 619 if ((error = namei(&nd)) != 0) 620 goto bad; 621 vp = nd.ni_vp; 622 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { 623 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 624 if (nd.ni_dvp == vp) 625 vrele(nd.ni_dvp); 626 else 627 vput(nd.ni_dvp); 628 vrele(vp); 629 if (vp != NULL) { 630 error = EADDRINUSE; 631 goto bad; 632 } 633 error = vn_start_write(NULL, &mp, 634 V_WAIT | V_SLEEPONLY | V_PCATCH); 635 if (error) 636 goto bad; 637 goto restart; 638 } 639 VATTR_NULL(&vattr); 640 vattr.va_type = VSOCK; 641 vattr.va_mode = ACCESSPERMS; 642 VOP_LEASE(nd.ni_dvp, p, p->p_ucred, LEASE_WRITE); 643 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 644 vn_finished_write(mp, 0); 645 if (error) 646 goto bad; 647 vp = nd.ni_vp; 648 vp->v_socket = unp->unp_socket; 649 unp->unp_vnode = vp; 650 unp->unp_addrlen = addrlen; 651 unp->unp_addr = sun; 652 VOP_UNLOCK(vp, 0); 653 return (0); 654 655 bad: 656 free(sun, M_SONAME); 657 return (error); 658 } 659 660 int 661 unp_connect(struct socket *so, struct mbuf *nam, struct proc *p) 662 { 663 struct sockaddr_un *sun; 664 struct vnode *vp; 665 struct socket *so2, *so3; 666 struct unpcb *unp2, *unp3; 667 size_t addrlen; 668 int error; 669 struct nameidata nd; 670 671 /* 672 * Allocate a temporary sockaddr. We have to allocate one extra 673 * byte so that we can ensure that the pathname is nul-terminated. 674 * When we establish the connection, we copy the other PCB's 675 * sockaddr to our own. 676 */ 677 addrlen = nam->m_len + 1; 678 sun = malloc(addrlen, M_SONAME, M_WAITOK); 679 m_copydata(nam, 0, nam->m_len, (caddr_t)sun); 680 *(((char *)sun) + nam->m_len) = '\0'; 681 682 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, sun->sun_path, p); 683 684 if ((error = namei(&nd)) != 0) 685 goto bad2; 686 vp = nd.ni_vp; 687 if (vp->v_type != VSOCK) { 688 error = ENOTSOCK; 689 goto bad; 690 } 691 if ((error = VOP_ACCESS(vp, VWRITE, p->p_ucred, p)) != 0) 692 goto bad; 693 so2 = vp->v_socket; 694 if (so2 == 0) { 695 error = ECONNREFUSED; 696 goto bad; 697 } 698 if (so->so_type != so2->so_type) { 699 error = EPROTOTYPE; 700 goto bad; 701 } 702 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 703 if ((so2->so_options & SO_ACCEPTCONN) == 0 || 704 (so3 = sonewconn(so2, 0)) == 0) { 705 error = ECONNREFUSED; 706 goto bad; 707 } 708 unp2 = sotounpcb(so2); 709 unp3 = sotounpcb(so3); 710 if (unp2->unp_addr) { 711 unp3->unp_addr = malloc(unp2->unp_addrlen, 712 M_SONAME, M_WAITOK); 713 memcpy(unp3->unp_addr, unp2->unp_addr, 714 unp2->unp_addrlen); 715 unp3->unp_addrlen = unp2->unp_addrlen; 716 } 717 unp3->unp_flags = unp2->unp_flags; 718 so2 = so3; 719 } 720 error = unp_connect2(so, so2, PRU_CONNECT); 721 bad: 722 vput(vp); 723 bad2: 724 free(sun, M_SONAME); 725 return (error); 726 } 727 728 int 729 unp_connect2(struct socket *so, struct socket *so2, int req) 730 { 731 struct unpcb *unp = sotounpcb(so); 732 struct unpcb *unp2; 733 734 if (so2->so_type != so->so_type) 735 return (EPROTOTYPE); 736 unp2 = sotounpcb(so2); 737 unp->unp_conn = unp2; 738 switch (so->so_type) { 739 740 case SOCK_DGRAM: 741 unp->unp_nextref = unp2->unp_refs; 742 unp2->unp_refs = unp; 743 soisconnected(so); 744 break; 745 746 case SOCK_STREAM: 747 unp2->unp_conn = unp; 748 if (req == PRU_CONNECT && 749 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 750 soisconnecting(so); 751 else 752 soisconnected(so); 753 soisconnected(so2); 754 break; 755 756 default: 757 panic("unp_connect2"); 758 } 759 return (0); 760 } 761 762 void 763 unp_disconnect(struct unpcb *unp) 764 { 765 struct unpcb *unp2 = unp->unp_conn; 766 767 if (unp2 == 0) 768 return; 769 unp->unp_conn = 0; 770 switch (unp->unp_socket->so_type) { 771 772 case SOCK_DGRAM: 773 if (unp2->unp_refs == unp) 774 unp2->unp_refs = unp->unp_nextref; 775 else { 776 unp2 = unp2->unp_refs; 777 for (;;) { 778 if (unp2 == 0) 779 panic("unp_disconnect"); 780 if (unp2->unp_nextref == unp) 781 break; 782 unp2 = unp2->unp_nextref; 783 } 784 unp2->unp_nextref = unp->unp_nextref; 785 } 786 unp->unp_nextref = 0; 787 unp->unp_socket->so_state &= ~SS_ISCONNECTED; 788 break; 789 790 case SOCK_STREAM: 791 soisdisconnected(unp->unp_socket); 792 unp2->unp_conn = 0; 793 soisdisconnected(unp2->unp_socket); 794 break; 795 } 796 } 797 798 #ifdef notdef 799 unp_abort(struct unpcb *unp) 800 { 801 unp_detach(unp); 802 } 803 #endif 804 805 void 806 unp_shutdown(struct unpcb *unp) 807 { 808 struct socket *so; 809 810 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn && 811 (so = unp->unp_conn->unp_socket)) 812 socantrcvmore(so); 813 } 814 815 void 816 unp_drop(struct unpcb *unp, int errno) 817 { 818 struct socket *so = unp->unp_socket; 819 820 so->so_error = errno; 821 unp_disconnect(unp); 822 if (so->so_head) { 823 so->so_pcb = 0; 824 sofree(so); 825 if (unp->unp_addr) 826 free(unp->unp_addr, M_SONAME); 827 free(unp, M_PCB); 828 } 829 } 830 831 #ifdef notdef 832 unp_drain(void) 833 { 834 835 } 836 #endif 837 838 int 839 unp_externalize(struct mbuf *rights, struct proc *p) 840 { 841 struct cmsghdr *cm = mtod(rights, struct cmsghdr *); 842 int i, *fdp; 843 struct file **rp; 844 struct file *fp; 845 int nfds, error = 0; 846 847 nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) / 848 sizeof(struct file *); 849 rp = (struct file **)CMSG_DATA(cm); 850 851 fdp = malloc(nfds * sizeof(int), M_TEMP, M_WAITOK); 852 853 /* Make sure the recipient should be able to see the descriptors.. */ 854 if (p->p_cwdi->cwdi_rdir != NULL) { 855 rp = (struct file **)CMSG_DATA(cm); 856 for (i = 0; i < nfds; i++) { 857 fp = *rp++; 858 /* 859 * If we are in a chroot'ed directory, and 860 * someone wants to pass us a directory, make 861 * sure it's inside the subtree we're allowed 862 * to access. 863 */ 864 if (fp->f_type == DTYPE_VNODE) { 865 struct vnode *vp = (struct vnode *)fp->f_data; 866 if ((vp->v_type == VDIR) && 867 !vn_isunder(vp, p->p_cwdi->cwdi_rdir, p)) { 868 error = EPERM; 869 break; 870 } 871 } 872 } 873 } 874 875 restart: 876 rp = (struct file **)CMSG_DATA(cm); 877 if (error != 0) { 878 for (i = 0; i < nfds; i++) { 879 fp = *rp; 880 /* 881 * zero the pointer before calling unp_discard, 882 * since it may end up in unp_gc().. 883 */ 884 *rp++ = 0; 885 unp_discard(fp); 886 } 887 goto out; 888 } 889 890 /* 891 * First loop -- allocate file descriptor table slots for the 892 * new descriptors. 893 */ 894 for (i = 0; i < nfds; i++) { 895 fp = *rp++; 896 if ((error = fdalloc(p, 0, &fdp[i])) != 0) { 897 /* 898 * Back out what we've done so far. 899 */ 900 for (--i; i >= 0; i--) 901 fdremove(p->p_fd, fdp[i]); 902 903 if (error == ENOSPC) { 904 fdexpand(p); 905 error = 0; 906 } else { 907 /* 908 * This is the error that has historically 909 * been returned, and some callers may 910 * expect it. 911 */ 912 error = EMSGSIZE; 913 } 914 goto restart; 915 } 916 917 /* 918 * Make the slot reference the descriptor so that 919 * fdalloc() works properly.. We finalize it all 920 * in the loop below. 921 */ 922 p->p_fd->fd_ofiles[fdp[i]] = fp; 923 } 924 925 /* 926 * Now that adding them has succeeded, update all of the 927 * descriptor passing state. 928 */ 929 rp = (struct file **)CMSG_DATA(cm); 930 for (i = 0; i < nfds; i++) { 931 fp = *rp++; 932 fp->f_msgcount--; 933 unp_rights--; 934 } 935 936 /* 937 * Copy temporary array to message and adjust length, in case of 938 * transition from large struct file pointers to ints. 939 */ 940 memcpy(CMSG_DATA(cm), fdp, nfds * sizeof(int)); 941 cm->cmsg_len = CMSG_LEN(nfds * sizeof(int)); 942 rights->m_len = CMSG_SPACE(nfds * sizeof(int)); 943 out: 944 free(fdp, M_TEMP); 945 return (error); 946 } 947 948 int 949 unp_internalize(struct mbuf *control, struct proc *p) 950 { 951 struct filedesc *fdescp = p->p_fd; 952 struct cmsghdr *newcm, *cm = mtod(control, struct cmsghdr *); 953 struct file **rp, **files; 954 struct file *fp; 955 int i, fd, *fdp; 956 int nfds; 957 u_int neededspace; 958 959 /* Sanity check the control message header */ 960 if (cm->cmsg_type != SCM_RIGHTS || cm->cmsg_level != SOL_SOCKET || 961 cm->cmsg_len != control->m_len) 962 return (EINVAL); 963 964 /* Verify that the file descriptors are valid */ 965 nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) / sizeof(int); 966 fdp = (int *)CMSG_DATA(cm); 967 for (i = 0; i < nfds; i++) { 968 fd = *fdp++; 969 if ((fp = fd_getfile(fdescp, fd)) == NULL) 970 return (EBADF); 971 simple_unlock(&fp->f_slock); 972 } 973 974 /* Make sure we have room for the struct file pointers */ 975 neededspace = CMSG_SPACE(nfds * sizeof(struct file *)) - 976 control->m_len; 977 if (neededspace > M_TRAILINGSPACE(control)) { 978 979 /* allocate new space and copy header into it */ 980 newcm = malloc( 981 CMSG_SPACE(nfds * sizeof(struct file *)), 982 M_MBUF, M_WAITOK); 983 if (newcm == NULL) 984 return (E2BIG); 985 memcpy(newcm, cm, sizeof(struct cmsghdr)); 986 files = (struct file **)CMSG_DATA(newcm); 987 } else { 988 /* we can convert in-place */ 989 newcm = NULL; 990 files = (struct file **)CMSG_DATA(cm); 991 } 992 993 /* 994 * Transform the file descriptors into struct file pointers, in 995 * reverse order so that if pointers are bigger than ints, the 996 * int won't get until we're done. 997 */ 998 fdp = (int *)CMSG_DATA(cm) + nfds - 1; 999 rp = files + nfds - 1; 1000 for (i = 0; i < nfds; i++) { 1001 fp = fdescp->fd_ofiles[*fdp--]; 1002 simple_lock(&fp->f_slock); 1003 #ifdef DIAGNOSTIC 1004 if (fp->f_iflags & FIF_WANTCLOSE) 1005 panic("unp_internalize: file already closed"); 1006 #endif 1007 *rp-- = fp; 1008 fp->f_count++; 1009 fp->f_msgcount++; 1010 simple_unlock(&fp->f_slock); 1011 unp_rights++; 1012 } 1013 1014 if (newcm) { 1015 if (control->m_flags & M_EXT) 1016 MEXTREMOVE(control); 1017 MEXTADD(control, newcm, 1018 CMSG_SPACE(nfds * sizeof(struct file *)), 1019 M_MBUF, NULL, NULL); 1020 cm = newcm; 1021 } 1022 1023 /* adjust message & mbuf to note amount of space actually used. */ 1024 cm->cmsg_len = CMSG_LEN(nfds * sizeof(struct file *)); 1025 control->m_len = CMSG_SPACE(nfds * sizeof(struct file *)); 1026 1027 return (0); 1028 } 1029 1030 struct mbuf * 1031 unp_addsockcred(struct proc *p, struct mbuf *control) 1032 { 1033 struct cmsghdr *cmp; 1034 struct sockcred *sc; 1035 struct mbuf *m, *n; 1036 int len, space, i; 1037 1038 len = CMSG_LEN(SOCKCREDSIZE(p->p_ucred->cr_ngroups)); 1039 space = CMSG_SPACE(SOCKCREDSIZE(p->p_ucred->cr_ngroups)); 1040 1041 m = m_get(M_WAIT, MT_CONTROL); 1042 if (space > MLEN) { 1043 if (space > MCLBYTES) 1044 MEXTMALLOC(m, space, M_WAITOK); 1045 else 1046 m_clget(m, M_WAIT); 1047 if ((m->m_flags & M_EXT) == 0) { 1048 m_free(m); 1049 return (control); 1050 } 1051 } 1052 1053 m->m_len = space; 1054 m->m_next = NULL; 1055 cmp = mtod(m, struct cmsghdr *); 1056 sc = (struct sockcred *)CMSG_DATA(cmp); 1057 cmp->cmsg_len = len; 1058 cmp->cmsg_level = SOL_SOCKET; 1059 cmp->cmsg_type = SCM_CREDS; 1060 sc->sc_uid = p->p_cred->p_ruid; 1061 sc->sc_euid = p->p_ucred->cr_uid; 1062 sc->sc_gid = p->p_cred->p_rgid; 1063 sc->sc_egid = p->p_ucred->cr_gid; 1064 sc->sc_ngroups = p->p_ucred->cr_ngroups; 1065 for (i = 0; i < sc->sc_ngroups; i++) 1066 sc->sc_groups[i] = p->p_ucred->cr_groups[i]; 1067 1068 /* 1069 * If a control message already exists, append us to the end. 1070 */ 1071 if (control != NULL) { 1072 for (n = control; n->m_next != NULL; n = n->m_next) 1073 ; 1074 n->m_next = m; 1075 } else 1076 control = m; 1077 1078 return (control); 1079 } 1080 1081 int unp_defer, unp_gcing; 1082 extern struct domain unixdomain; 1083 1084 /* 1085 * Comment added long after the fact explaining what's going on here. 1086 * Do a mark-sweep GC of file descriptors on the system, to free up 1087 * any which are caught in flight to an about-to-be-closed socket. 1088 * 1089 * Traditional mark-sweep gc's start at the "root", and mark 1090 * everything reachable from the root (which, in our case would be the 1091 * process table). The mark bits are cleared during the sweep. 1092 * 1093 * XXX For some inexplicable reason (perhaps because the file 1094 * descriptor tables used to live in the u area which could be swapped 1095 * out and thus hard to reach), we do multiple scans over the set of 1096 * descriptors, using use *two* mark bits per object (DEFER and MARK). 1097 * Whenever we find a descriptor which references other descriptors, 1098 * the ones it references are marked with both bits, and we iterate 1099 * over the whole file table until there are no more DEFER bits set. 1100 * We also make an extra pass *before* the GC to clear the mark bits, 1101 * which could have been cleared at almost no cost during the previous 1102 * sweep. 1103 * 1104 * XXX MP: this needs to run with locks such that no other thread of 1105 * control can create or destroy references to file descriptors. it 1106 * may be necessary to defer the GC until later (when the locking 1107 * situation is more hospitable); it may be necessary to push this 1108 * into a separate thread. 1109 */ 1110 void 1111 unp_gc(void) 1112 { 1113 struct file *fp, *nextfp; 1114 struct socket *so, *so1; 1115 struct file **extra_ref, **fpp; 1116 int nunref, i; 1117 1118 if (unp_gcing) 1119 return; 1120 unp_gcing = 1; 1121 unp_defer = 0; 1122 1123 /* Clear mark bits */ 1124 LIST_FOREACH(fp, &filehead, f_list) 1125 fp->f_flag &= ~(FMARK|FDEFER); 1126 1127 /* 1128 * Iterate over the set of descriptors, marking ones believed 1129 * (based on refcount) to be referenced from a process, and 1130 * marking for rescan descriptors which are queued on a socket. 1131 */ 1132 do { 1133 LIST_FOREACH(fp, &filehead, f_list) { 1134 if (fp->f_flag & FDEFER) { 1135 fp->f_flag &= ~FDEFER; 1136 unp_defer--; 1137 #ifdef DIAGNOSTIC 1138 if (fp->f_count == 0) 1139 panic("unp_gc: deferred unreferenced socket"); 1140 #endif 1141 } else { 1142 if (fp->f_count == 0) 1143 continue; 1144 if (fp->f_flag & FMARK) 1145 continue; 1146 if (fp->f_count == fp->f_msgcount) 1147 continue; 1148 } 1149 fp->f_flag |= FMARK; 1150 1151 if (fp->f_type != DTYPE_SOCKET || 1152 (so = (struct socket *)fp->f_data) == 0) 1153 continue; 1154 if (so->so_proto->pr_domain != &unixdomain || 1155 (so->so_proto->pr_flags&PR_RIGHTS) == 0) 1156 continue; 1157 #ifdef notdef 1158 if (so->so_rcv.sb_flags & SB_LOCK) { 1159 /* 1160 * This is problematical; it's not clear 1161 * we need to wait for the sockbuf to be 1162 * unlocked (on a uniprocessor, at least), 1163 * and it's also not clear what to do 1164 * if sbwait returns an error due to receipt 1165 * of a signal. If sbwait does return 1166 * an error, we'll go into an infinite 1167 * loop. Delete all of this for now. 1168 */ 1169 (void) sbwait(&so->so_rcv); 1170 goto restart; 1171 } 1172 #endif 1173 unp_scan(so->so_rcv.sb_mb, unp_mark, 0); 1174 /* 1175 * mark descriptors referenced from sockets queued on the accept queue as well. 1176 */ 1177 if (so->so_options & SO_ACCEPTCONN) { 1178 TAILQ_FOREACH(so1, &so->so_q0, so_qe) { 1179 unp_scan(so1->so_rcv.sb_mb, unp_mark, 0); 1180 } 1181 TAILQ_FOREACH(so1, &so->so_q, so_qe) { 1182 unp_scan(so1->so_rcv.sb_mb, unp_mark, 0); 1183 } 1184 } 1185 1186 } 1187 } while (unp_defer); 1188 /* 1189 * Sweep pass. Find unmarked descriptors, and free them. 1190 * 1191 * We grab an extra reference to each of the file table entries 1192 * that are not otherwise accessible and then free the rights 1193 * that are stored in messages on them. 1194 * 1195 * The bug in the original code is a little tricky, so I'll describe 1196 * what's wrong with it here. 1197 * 1198 * It is incorrect to simply unp_discard each entry for f_msgcount 1199 * times -- consider the case of sockets A and B that contain 1200 * references to each other. On a last close of some other socket, 1201 * we trigger a gc since the number of outstanding rights (unp_rights) 1202 * is non-zero. If during the sweep phase the gc code un_discards, 1203 * we end up doing a (full) closef on the descriptor. A closef on A 1204 * results in the following chain. Closef calls soo_close, which 1205 * calls soclose. Soclose calls first (through the switch 1206 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply 1207 * returns because the previous instance had set unp_gcing, and 1208 * we return all the way back to soclose, which marks the socket 1209 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush 1210 * to free up the rights that are queued in messages on the socket A, 1211 * i.e., the reference on B. The sorflush calls via the dom_dispose 1212 * switch unp_dispose, which unp_scans with unp_discard. This second 1213 * instance of unp_discard just calls closef on B. 1214 * 1215 * Well, a similar chain occurs on B, resulting in a sorflush on B, 1216 * which results in another closef on A. Unfortunately, A is already 1217 * being closed, and the descriptor has already been marked with 1218 * SS_NOFDREF, and soclose panics at this point. 1219 * 1220 * Here, we first take an extra reference to each inaccessible 1221 * descriptor. Then, if the inaccessible descriptor is a 1222 * socket, we call sorflush in case it is a Unix domain 1223 * socket. After we destroy all the rights carried in 1224 * messages, we do a last closef to get rid of our extra 1225 * reference. This is the last close, and the unp_detach etc 1226 * will shut down the socket. 1227 * 1228 * 91/09/19, bsy@cs.cmu.edu 1229 */ 1230 extra_ref = malloc(nfiles * sizeof(struct file *), M_FILE, M_WAITOK); 1231 for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref; fp != 0; 1232 fp = nextfp) { 1233 nextfp = LIST_NEXT(fp, f_list); 1234 simple_lock(&fp->f_slock); 1235 if (fp->f_count != 0 && 1236 fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) { 1237 *fpp++ = fp; 1238 nunref++; 1239 fp->f_count++; 1240 } 1241 simple_unlock(&fp->f_slock); 1242 } 1243 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1244 fp = *fpp; 1245 simple_lock(&fp->f_slock); 1246 FILE_USE(fp); 1247 if (fp->f_type == DTYPE_SOCKET) 1248 sorflush((struct socket *)fp->f_data); 1249 FILE_UNUSE(fp, NULL); 1250 } 1251 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1252 fp = *fpp; 1253 simple_lock(&fp->f_slock); 1254 FILE_USE(fp); 1255 (void) closef(fp, (struct proc *)0); 1256 } 1257 free((caddr_t)extra_ref, M_FILE); 1258 unp_gcing = 0; 1259 } 1260 1261 void 1262 unp_dispose(struct mbuf *m) 1263 { 1264 1265 if (m) 1266 unp_scan(m, unp_discard, 1); 1267 } 1268 1269 void 1270 unp_scan(struct mbuf *m0, void (*op)(struct file *), int discard) 1271 { 1272 struct mbuf *m; 1273 struct file **rp; 1274 struct cmsghdr *cm; 1275 int i; 1276 int qfds; 1277 1278 while (m0) { 1279 for (m = m0; m; m = m->m_next) { 1280 if (m->m_type == MT_CONTROL && 1281 m->m_len >= sizeof(*cm)) { 1282 cm = mtod(m, struct cmsghdr *); 1283 if (cm->cmsg_level != SOL_SOCKET || 1284 cm->cmsg_type != SCM_RIGHTS) 1285 continue; 1286 qfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) 1287 / sizeof(struct file *); 1288 rp = (struct file **)CMSG_DATA(cm); 1289 for (i = 0; i < qfds; i++) { 1290 struct file *fp = *rp; 1291 if (discard) 1292 *rp = 0; 1293 (*op)(fp); 1294 rp++; 1295 } 1296 break; /* XXX, but saves time */ 1297 } 1298 } 1299 m0 = m0->m_nextpkt; 1300 } 1301 } 1302 1303 void 1304 unp_mark(struct file *fp) 1305 { 1306 if (fp == NULL) 1307 return; 1308 1309 if (fp->f_flag & FMARK) 1310 return; 1311 1312 /* If we're already deferred, don't screw up the defer count */ 1313 if (fp->f_flag & FDEFER) 1314 return; 1315 1316 /* 1317 * Minimize the number of deferrals... Sockets are the only 1318 * type of descriptor which can hold references to another 1319 * descriptor, so just mark other descriptors, and defer 1320 * unmarked sockets for the next pass. 1321 */ 1322 if (fp->f_type == DTYPE_SOCKET) { 1323 unp_defer++; 1324 if (fp->f_count == 0) 1325 panic("unp_mark: queued unref"); 1326 fp->f_flag |= FDEFER; 1327 } else { 1328 fp->f_flag |= FMARK; 1329 } 1330 return; 1331 } 1332 1333 void 1334 unp_discard(struct file *fp) 1335 { 1336 if (fp == NULL) 1337 return; 1338 simple_lock(&fp->f_slock); 1339 fp->f_usecount++; /* i.e. FILE_USE(fp) sans locking */ 1340 fp->f_msgcount--; 1341 simple_unlock(&fp->f_slock); 1342 unp_rights--; 1343 (void) closef(fp, (struct proc *)0); 1344 } 1345