1 /* $NetBSD: uipc_usrreq.c,v 1.86 2005/12/11 12:24:30 christos 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 1.86 2005/12/11 12:24:30 christos 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 lwp *l) 197 { 198 struct unpcb *unp = sotounpcb(so); 199 struct socket *so2; 200 struct proc *p; 201 u_int newhiwat; 202 int error = 0; 203 204 if (req == PRU_CONTROL) 205 return (EOPNOTSUPP); 206 207 #ifdef DIAGNOSTIC 208 if (req != PRU_SEND && req != PRU_SENDOOB && control) 209 panic("uipc_usrreq: unexpected control mbuf"); 210 #endif 211 p = l ? l->l_proc : NULL; 212 if (unp == 0 && req != PRU_ATTACH) { 213 error = EINVAL; 214 goto release; 215 } 216 217 switch (req) { 218 219 case PRU_ATTACH: 220 if (unp != 0) { 221 error = EISCONN; 222 break; 223 } 224 error = unp_attach(so); 225 break; 226 227 case PRU_DETACH: 228 unp_detach(unp); 229 break; 230 231 case PRU_BIND: 232 error = unp_bind(unp, nam, l); 233 break; 234 235 case PRU_LISTEN: 236 if (unp->unp_vnode == 0) 237 error = EINVAL; 238 break; 239 240 case PRU_CONNECT: 241 error = unp_connect(so, nam, l); 242 break; 243 244 case PRU_CONNECT2: 245 error = unp_connect2(so, (struct socket *)nam, PRU_CONNECT2); 246 break; 247 248 case PRU_DISCONNECT: 249 unp_disconnect(unp); 250 break; 251 252 case PRU_ACCEPT: 253 unp_setpeeraddr(unp, nam); 254 /* 255 * Mark the initiating STREAM socket as connected *ONLY* 256 * after it's been accepted. This prevents a client from 257 * overrunning a server and receiving ECONNREFUSED. 258 */ 259 if (unp->unp_conn != NULL && 260 (unp->unp_conn->unp_socket->so_state & SS_ISCONNECTING)) 261 soisconnected(unp->unp_conn->unp_socket); 262 break; 263 264 case PRU_SHUTDOWN: 265 socantsendmore(so); 266 unp_shutdown(unp); 267 break; 268 269 case PRU_RCVD: 270 switch (so->so_type) { 271 272 case SOCK_DGRAM: 273 panic("uipc 1"); 274 /*NOTREACHED*/ 275 276 case SOCK_STREAM: 277 #define rcv (&so->so_rcv) 278 #define snd (&so2->so_snd) 279 if (unp->unp_conn == 0) 280 break; 281 so2 = unp->unp_conn->unp_socket; 282 /* 283 * Adjust backpressure on sender 284 * and wakeup any waiting to write. 285 */ 286 snd->sb_mbmax += unp->unp_mbcnt - rcv->sb_mbcnt; 287 unp->unp_mbcnt = rcv->sb_mbcnt; 288 newhiwat = snd->sb_hiwat + unp->unp_cc - rcv->sb_cc; 289 (void)chgsbsize(so2->so_uidinfo, 290 &snd->sb_hiwat, newhiwat, RLIM_INFINITY); 291 unp->unp_cc = rcv->sb_cc; 292 sowwakeup(so2); 293 #undef snd 294 #undef rcv 295 break; 296 297 default: 298 panic("uipc 2"); 299 } 300 break; 301 302 case PRU_SEND: 303 /* 304 * Note: unp_internalize() rejects any control message 305 * other than SCM_RIGHTS, and only allows one. This 306 * has the side-effect of preventing a caller from 307 * forging SCM_CREDS. 308 */ 309 if (control && (error = unp_internalize(control, l))) { 310 goto die; 311 } 312 switch (so->so_type) { 313 314 case SOCK_DGRAM: { 315 if (nam) { 316 if ((so->so_state & SS_ISCONNECTED) != 0) { 317 error = EISCONN; 318 goto die; 319 } 320 error = unp_connect(so, nam, l); 321 if (error) { 322 die: 323 m_freem(control); 324 m_freem(m); 325 break; 326 } 327 } else { 328 if ((so->so_state & SS_ISCONNECTED) == 0) { 329 error = ENOTCONN; 330 goto die; 331 } 332 } 333 error = unp_output(m, control, unp, p); 334 if (nam) 335 unp_disconnect(unp); 336 break; 337 } 338 339 case SOCK_STREAM: 340 #define rcv (&so2->so_rcv) 341 #define snd (&so->so_snd) 342 if (unp->unp_conn == 0) 343 panic("uipc 3"); 344 so2 = unp->unp_conn->unp_socket; 345 if (unp->unp_conn->unp_flags & UNP_WANTCRED) { 346 /* 347 * Credentials are passed only once on 348 * SOCK_STREAM. 349 */ 350 unp->unp_conn->unp_flags &= ~UNP_WANTCRED; 351 control = unp_addsockcred(p, control); 352 } 353 /* 354 * Send to paired receive port, and then reduce 355 * send buffer hiwater marks to maintain backpressure. 356 * Wake up readers. 357 */ 358 if (control) { 359 if (sbappendcontrol(rcv, m, control) == 0) 360 m_freem(control); 361 } else 362 sbappend(rcv, m); 363 snd->sb_mbmax -= 364 rcv->sb_mbcnt - unp->unp_conn->unp_mbcnt; 365 unp->unp_conn->unp_mbcnt = rcv->sb_mbcnt; 366 newhiwat = snd->sb_hiwat - 367 (rcv->sb_cc - unp->unp_conn->unp_cc); 368 (void)chgsbsize(so->so_uidinfo, 369 &snd->sb_hiwat, newhiwat, RLIM_INFINITY); 370 unp->unp_conn->unp_cc = rcv->sb_cc; 371 sorwakeup(so2); 372 #undef snd 373 #undef rcv 374 break; 375 376 default: 377 panic("uipc 4"); 378 } 379 break; 380 381 case PRU_ABORT: 382 unp_drop(unp, ECONNABORTED); 383 384 #ifdef DIAGNOSTIC 385 if (so->so_pcb == 0) 386 panic("uipc 5: drop killed pcb"); 387 #endif 388 unp_detach(unp); 389 break; 390 391 case PRU_SENSE: 392 ((struct stat *) m)->st_blksize = so->so_snd.sb_hiwat; 393 if (so->so_type == SOCK_STREAM && unp->unp_conn != 0) { 394 so2 = unp->unp_conn->unp_socket; 395 ((struct stat *) m)->st_blksize += so2->so_rcv.sb_cc; 396 } 397 ((struct stat *) m)->st_dev = NODEV; 398 if (unp->unp_ino == 0) 399 unp->unp_ino = unp_ino++; 400 ((struct stat *) m)->st_atimespec = 401 ((struct stat *) m)->st_mtimespec = 402 ((struct stat *) m)->st_ctimespec = unp->unp_ctime; 403 ((struct stat *) m)->st_ino = unp->unp_ino; 404 return (0); 405 406 case PRU_RCVOOB: 407 error = EOPNOTSUPP; 408 break; 409 410 case PRU_SENDOOB: 411 m_freem(control); 412 m_freem(m); 413 error = EOPNOTSUPP; 414 break; 415 416 case PRU_SOCKADDR: 417 unp_setsockaddr(unp, nam); 418 break; 419 420 case PRU_PEERADDR: 421 unp_setpeeraddr(unp, nam); 422 break; 423 424 default: 425 panic("piusrreq"); 426 } 427 428 release: 429 return (error); 430 } 431 432 /* 433 * Unix domain socket option processing. 434 */ 435 int 436 uipc_ctloutput(int op, struct socket *so, int level, int optname, 437 struct mbuf **mp) 438 { 439 struct unpcb *unp = sotounpcb(so); 440 struct mbuf *m = *mp; 441 int optval = 0, error = 0; 442 443 if (level != 0) { 444 error = EINVAL; 445 if (op == PRCO_SETOPT && m) 446 (void) m_free(m); 447 } else switch (op) { 448 449 case PRCO_SETOPT: 450 switch (optname) { 451 case LOCAL_CREDS: 452 case LOCAL_CONNWAIT: 453 if (m == NULL || m->m_len != sizeof(int)) 454 error = EINVAL; 455 else { 456 optval = *mtod(m, int *); 457 switch (optname) { 458 #define OPTSET(bit) \ 459 if (optval) \ 460 unp->unp_flags |= (bit); \ 461 else \ 462 unp->unp_flags &= ~(bit); 463 464 case LOCAL_CREDS: 465 OPTSET(UNP_WANTCRED); 466 break; 467 case LOCAL_CONNWAIT: 468 OPTSET(UNP_CONNWAIT); 469 break; 470 } 471 } 472 break; 473 #undef OPTSET 474 475 default: 476 error = ENOPROTOOPT; 477 break; 478 } 479 if (m) 480 (void) m_free(m); 481 break; 482 483 case PRCO_GETOPT: 484 switch (optname) { 485 case LOCAL_CREDS: 486 *mp = m = m_get(M_WAIT, MT_SOOPTS); 487 m->m_len = sizeof(int); 488 switch (optname) { 489 490 #define OPTBIT(bit) (unp->unp_flags & (bit) ? 1 : 0) 491 492 case LOCAL_CREDS: 493 optval = OPTBIT(UNP_WANTCRED); 494 break; 495 } 496 *mtod(m, int *) = optval; 497 break; 498 #undef OPTBIT 499 500 default: 501 error = ENOPROTOOPT; 502 break; 503 } 504 break; 505 } 506 return (error); 507 } 508 509 /* 510 * Both send and receive buffers are allocated PIPSIZ bytes of buffering 511 * for stream sockets, although the total for sender and receiver is 512 * actually only PIPSIZ. 513 * Datagram sockets really use the sendspace as the maximum datagram size, 514 * and don't really want to reserve the sendspace. Their recvspace should 515 * be large enough for at least one max-size datagram plus address. 516 */ 517 #define PIPSIZ 4096 518 u_long unpst_sendspace = PIPSIZ; 519 u_long unpst_recvspace = PIPSIZ; 520 u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 521 u_long unpdg_recvspace = 4*1024; 522 523 int unp_rights; /* file descriptors in flight */ 524 525 int 526 unp_attach(struct socket *so) 527 { 528 struct unpcb *unp; 529 int error; 530 531 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 532 switch (so->so_type) { 533 534 case SOCK_STREAM: 535 error = soreserve(so, unpst_sendspace, unpst_recvspace); 536 break; 537 538 case SOCK_DGRAM: 539 error = soreserve(so, unpdg_sendspace, unpdg_recvspace); 540 break; 541 542 default: 543 panic("unp_attach"); 544 } 545 if (error) 546 return (error); 547 } 548 unp = malloc(sizeof(*unp), M_PCB, M_NOWAIT); 549 if (unp == NULL) 550 return (ENOBUFS); 551 memset((caddr_t)unp, 0, sizeof(*unp)); 552 unp->unp_socket = so; 553 so->so_pcb = unp; 554 nanotime(&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 lwp *l) 592 { 593 struct sockaddr_un *sun; 594 struct vnode *vp; 595 struct mount *mp; 596 struct vattr vattr; 597 size_t addrlen; 598 struct proc *p; 599 int error; 600 struct nameidata nd; 601 602 if (unp->unp_vnode != 0) 603 return (EINVAL); 604 605 p = l->l_proc; 606 /* 607 * Allocate the new sockaddr. We have to allocate one 608 * extra byte so that we can ensure that the pathname 609 * is nul-terminated. 610 */ 611 addrlen = nam->m_len + 1; 612 sun = malloc(addrlen, M_SONAME, M_WAITOK); 613 m_copydata(nam, 0, nam->m_len, (caddr_t)sun); 614 *(((char *)sun) + nam->m_len) = '\0'; 615 616 restart: 617 NDINIT(&nd, CREATE, FOLLOW | LOCKPARENT, UIO_SYSSPACE, 618 sun->sun_path, l); 619 620 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 621 if ((error = namei(&nd)) != 0) 622 goto bad; 623 vp = nd.ni_vp; 624 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { 625 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 626 if (nd.ni_dvp == vp) 627 vrele(nd.ni_dvp); 628 else 629 vput(nd.ni_dvp); 630 vrele(vp); 631 if (vp != NULL) { 632 error = EADDRINUSE; 633 goto bad; 634 } 635 error = vn_start_write(NULL, &mp, 636 V_WAIT | V_SLEEPONLY | V_PCATCH); 637 if (error) 638 goto bad; 639 goto restart; 640 } 641 VATTR_NULL(&vattr); 642 vattr.va_type = VSOCK; 643 vattr.va_mode = ACCESSPERMS & ~(p->p_cwdi->cwdi_cmask); 644 VOP_LEASE(nd.ni_dvp, l, p->p_ucred, LEASE_WRITE); 645 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 646 vn_finished_write(mp, 0); 647 if (error) 648 goto bad; 649 vp = nd.ni_vp; 650 vp->v_socket = unp->unp_socket; 651 unp->unp_vnode = vp; 652 unp->unp_addrlen = addrlen; 653 unp->unp_addr = sun; 654 VOP_UNLOCK(vp, 0); 655 return (0); 656 657 bad: 658 free(sun, M_SONAME); 659 return (error); 660 } 661 662 int 663 unp_connect(struct socket *so, struct mbuf *nam, struct lwp *l) 664 { 665 struct sockaddr_un *sun; 666 struct vnode *vp; 667 struct socket *so2, *so3; 668 struct unpcb *unp2, *unp3; 669 size_t addrlen; 670 int error; 671 struct nameidata nd; 672 673 /* 674 * Allocate a temporary sockaddr. We have to allocate one extra 675 * byte so that we can ensure that the pathname is nul-terminated. 676 * When we establish the connection, we copy the other PCB's 677 * sockaddr to our own. 678 */ 679 addrlen = nam->m_len + 1; 680 sun = malloc(addrlen, M_SONAME, M_WAITOK); 681 m_copydata(nam, 0, nam->m_len, (caddr_t)sun); 682 *(((char *)sun) + nam->m_len) = '\0'; 683 684 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, sun->sun_path, l); 685 686 if ((error = namei(&nd)) != 0) 687 goto bad2; 688 vp = nd.ni_vp; 689 if (vp->v_type != VSOCK) { 690 error = ENOTSOCK; 691 goto bad; 692 } 693 if ((error = VOP_ACCESS(vp, VWRITE, l->l_proc->p_ucred, l)) != 0) 694 goto bad; 695 so2 = vp->v_socket; 696 if (so2 == 0) { 697 error = ECONNREFUSED; 698 goto bad; 699 } 700 if (so->so_type != so2->so_type) { 701 error = EPROTOTYPE; 702 goto bad; 703 } 704 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 705 if ((so2->so_options & SO_ACCEPTCONN) == 0 || 706 (so3 = sonewconn(so2, 0)) == 0) { 707 error = ECONNREFUSED; 708 goto bad; 709 } 710 unp2 = sotounpcb(so2); 711 unp3 = sotounpcb(so3); 712 if (unp2->unp_addr) { 713 unp3->unp_addr = malloc(unp2->unp_addrlen, 714 M_SONAME, M_WAITOK); 715 memcpy(unp3->unp_addr, unp2->unp_addr, 716 unp2->unp_addrlen); 717 unp3->unp_addrlen = unp2->unp_addrlen; 718 } 719 unp3->unp_flags = unp2->unp_flags; 720 so2 = so3; 721 } 722 error = unp_connect2(so, so2, PRU_CONNECT); 723 bad: 724 vput(vp); 725 bad2: 726 free(sun, M_SONAME); 727 return (error); 728 } 729 730 int 731 unp_connect2(struct socket *so, struct socket *so2, int req) 732 { 733 struct unpcb *unp = sotounpcb(so); 734 struct unpcb *unp2; 735 736 if (so2->so_type != so->so_type) 737 return (EPROTOTYPE); 738 unp2 = sotounpcb(so2); 739 unp->unp_conn = unp2; 740 switch (so->so_type) { 741 742 case SOCK_DGRAM: 743 unp->unp_nextref = unp2->unp_refs; 744 unp2->unp_refs = unp; 745 soisconnected(so); 746 break; 747 748 case SOCK_STREAM: 749 unp2->unp_conn = unp; 750 if (req == PRU_CONNECT && 751 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 752 soisconnecting(so); 753 else 754 soisconnected(so); 755 soisconnected(so2); 756 break; 757 758 default: 759 panic("unp_connect2"); 760 } 761 return (0); 762 } 763 764 void 765 unp_disconnect(struct unpcb *unp) 766 { 767 struct unpcb *unp2 = unp->unp_conn; 768 769 if (unp2 == 0) 770 return; 771 unp->unp_conn = 0; 772 switch (unp->unp_socket->so_type) { 773 774 case SOCK_DGRAM: 775 if (unp2->unp_refs == unp) 776 unp2->unp_refs = unp->unp_nextref; 777 else { 778 unp2 = unp2->unp_refs; 779 for (;;) { 780 if (unp2 == 0) 781 panic("unp_disconnect"); 782 if (unp2->unp_nextref == unp) 783 break; 784 unp2 = unp2->unp_nextref; 785 } 786 unp2->unp_nextref = unp->unp_nextref; 787 } 788 unp->unp_nextref = 0; 789 unp->unp_socket->so_state &= ~SS_ISCONNECTED; 790 break; 791 792 case SOCK_STREAM: 793 soisdisconnected(unp->unp_socket); 794 unp2->unp_conn = 0; 795 soisdisconnected(unp2->unp_socket); 796 break; 797 } 798 } 799 800 #ifdef notdef 801 unp_abort(struct unpcb *unp) 802 { 803 unp_detach(unp); 804 } 805 #endif 806 807 void 808 unp_shutdown(struct unpcb *unp) 809 { 810 struct socket *so; 811 812 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn && 813 (so = unp->unp_conn->unp_socket)) 814 socantrcvmore(so); 815 } 816 817 void 818 unp_drop(struct unpcb *unp, int errno) 819 { 820 struct socket *so = unp->unp_socket; 821 822 so->so_error = errno; 823 unp_disconnect(unp); 824 if (so->so_head) { 825 so->so_pcb = 0; 826 sofree(so); 827 if (unp->unp_addr) 828 free(unp->unp_addr, M_SONAME); 829 free(unp, M_PCB); 830 } 831 } 832 833 #ifdef notdef 834 unp_drain(void) 835 { 836 837 } 838 #endif 839 840 int 841 unp_externalize(struct mbuf *rights, struct lwp *l) 842 { 843 struct cmsghdr *cm = mtod(rights, struct cmsghdr *); 844 struct proc *p = l->l_proc; 845 int i, *fdp; 846 struct file **rp; 847 struct file *fp; 848 int nfds, error = 0; 849 850 nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) / 851 sizeof(struct file *); 852 rp = (struct file **)CMSG_DATA(cm); 853 854 fdp = malloc(nfds * sizeof(int), M_TEMP, M_WAITOK); 855 856 /* Make sure the recipient should be able to see the descriptors.. */ 857 if (p->p_cwdi->cwdi_rdir != NULL) { 858 rp = (struct file **)CMSG_DATA(cm); 859 for (i = 0; i < nfds; i++) { 860 fp = *rp++; 861 /* 862 * If we are in a chroot'ed directory, and 863 * someone wants to pass us a directory, make 864 * sure it's inside the subtree we're allowed 865 * to access. 866 */ 867 if (fp->f_type == DTYPE_VNODE) { 868 struct vnode *vp = (struct vnode *)fp->f_data; 869 if ((vp->v_type == VDIR) && 870 !vn_isunder(vp, p->p_cwdi->cwdi_rdir, l)) { 871 error = EPERM; 872 break; 873 } 874 } 875 } 876 } 877 878 restart: 879 rp = (struct file **)CMSG_DATA(cm); 880 if (error != 0) { 881 for (i = 0; i < nfds; i++) { 882 fp = *rp; 883 /* 884 * zero the pointer before calling unp_discard, 885 * since it may end up in unp_gc().. 886 */ 887 *rp++ = 0; 888 unp_discard(fp); 889 } 890 goto out; 891 } 892 893 /* 894 * First loop -- allocate file descriptor table slots for the 895 * new descriptors. 896 */ 897 for (i = 0; i < nfds; i++) { 898 fp = *rp++; 899 if ((error = fdalloc(p, 0, &fdp[i])) != 0) { 900 /* 901 * Back out what we've done so far. 902 */ 903 for (--i; i >= 0; i--) 904 fdremove(p->p_fd, fdp[i]); 905 906 if (error == ENOSPC) { 907 fdexpand(p); 908 error = 0; 909 } else { 910 /* 911 * This is the error that has historically 912 * been returned, and some callers may 913 * expect it. 914 */ 915 error = EMSGSIZE; 916 } 917 goto restart; 918 } 919 920 /* 921 * Make the slot reference the descriptor so that 922 * fdalloc() works properly.. We finalize it all 923 * in the loop below. 924 */ 925 p->p_fd->fd_ofiles[fdp[i]] = fp; 926 } 927 928 /* 929 * Now that adding them has succeeded, update all of the 930 * descriptor passing state. 931 */ 932 rp = (struct file **)CMSG_DATA(cm); 933 for (i = 0; i < nfds; i++) { 934 fp = *rp++; 935 fp->f_msgcount--; 936 unp_rights--; 937 } 938 939 /* 940 * Copy temporary array to message and adjust length, in case of 941 * transition from large struct file pointers to ints. 942 */ 943 memcpy(CMSG_DATA(cm), fdp, nfds * sizeof(int)); 944 cm->cmsg_len = CMSG_LEN(nfds * sizeof(int)); 945 rights->m_len = CMSG_SPACE(nfds * sizeof(int)); 946 out: 947 free(fdp, M_TEMP); 948 return (error); 949 } 950 951 int 952 unp_internalize(struct mbuf *control, struct lwp *l) 953 { 954 struct proc *p = l->l_proc; 955 struct filedesc *fdescp = p->p_fd; 956 struct cmsghdr *newcm, *cm = mtod(control, struct cmsghdr *); 957 struct file **rp, **files; 958 struct file *fp; 959 int i, fd, *fdp; 960 int nfds; 961 u_int neededspace; 962 963 /* Sanity check the control message header */ 964 if (cm->cmsg_type != SCM_RIGHTS || cm->cmsg_level != SOL_SOCKET || 965 cm->cmsg_len != control->m_len) 966 return (EINVAL); 967 968 /* Verify that the file descriptors are valid */ 969 nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) / sizeof(int); 970 fdp = (int *)CMSG_DATA(cm); 971 for (i = 0; i < nfds; i++) { 972 fd = *fdp++; 973 if ((fp = fd_getfile(fdescp, fd)) == NULL) 974 return (EBADF); 975 simple_unlock(&fp->f_slock); 976 } 977 978 /* Make sure we have room for the struct file pointers */ 979 neededspace = CMSG_SPACE(nfds * sizeof(struct file *)) - 980 control->m_len; 981 if (neededspace > M_TRAILINGSPACE(control)) { 982 983 /* allocate new space and copy header into it */ 984 newcm = malloc( 985 CMSG_SPACE(nfds * sizeof(struct file *)), 986 M_MBUF, M_WAITOK); 987 if (newcm == NULL) 988 return (E2BIG); 989 memcpy(newcm, cm, sizeof(struct cmsghdr)); 990 files = (struct file **)CMSG_DATA(newcm); 991 } else { 992 /* we can convert in-place */ 993 newcm = NULL; 994 files = (struct file **)CMSG_DATA(cm); 995 } 996 997 /* 998 * Transform the file descriptors into struct file pointers, in 999 * reverse order so that if pointers are bigger than ints, the 1000 * int won't get until we're done. 1001 */ 1002 fdp = (int *)CMSG_DATA(cm) + nfds - 1; 1003 rp = files + nfds - 1; 1004 for (i = 0; i < nfds; i++) { 1005 fp = fdescp->fd_ofiles[*fdp--]; 1006 simple_lock(&fp->f_slock); 1007 #ifdef DIAGNOSTIC 1008 if (fp->f_iflags & FIF_WANTCLOSE) 1009 panic("unp_internalize: file already closed"); 1010 #endif 1011 *rp-- = fp; 1012 fp->f_count++; 1013 fp->f_msgcount++; 1014 simple_unlock(&fp->f_slock); 1015 unp_rights++; 1016 } 1017 1018 if (newcm) { 1019 if (control->m_flags & M_EXT) 1020 MEXTREMOVE(control); 1021 MEXTADD(control, newcm, 1022 CMSG_SPACE(nfds * sizeof(struct file *)), 1023 M_MBUF, NULL, NULL); 1024 cm = newcm; 1025 } 1026 1027 /* adjust message & mbuf to note amount of space actually used. */ 1028 cm->cmsg_len = CMSG_LEN(nfds * sizeof(struct file *)); 1029 control->m_len = CMSG_SPACE(nfds * sizeof(struct file *)); 1030 1031 return (0); 1032 } 1033 1034 struct mbuf * 1035 unp_addsockcred(struct proc *p, struct mbuf *control) 1036 { 1037 struct cmsghdr *cmp; 1038 struct sockcred *sc; 1039 struct mbuf *m, *n; 1040 int len, space, i; 1041 1042 len = CMSG_LEN(SOCKCREDSIZE(p->p_ucred->cr_ngroups)); 1043 space = CMSG_SPACE(SOCKCREDSIZE(p->p_ucred->cr_ngroups)); 1044 1045 m = m_get(M_WAIT, MT_CONTROL); 1046 if (space > MLEN) { 1047 if (space > MCLBYTES) 1048 MEXTMALLOC(m, space, M_WAITOK); 1049 else 1050 m_clget(m, M_WAIT); 1051 if ((m->m_flags & M_EXT) == 0) { 1052 m_free(m); 1053 return (control); 1054 } 1055 } 1056 1057 m->m_len = space; 1058 m->m_next = NULL; 1059 cmp = mtod(m, struct cmsghdr *); 1060 sc = (struct sockcred *)CMSG_DATA(cmp); 1061 cmp->cmsg_len = len; 1062 cmp->cmsg_level = SOL_SOCKET; 1063 cmp->cmsg_type = SCM_CREDS; 1064 sc->sc_uid = p->p_cred->p_ruid; 1065 sc->sc_euid = p->p_ucred->cr_uid; 1066 sc->sc_gid = p->p_cred->p_rgid; 1067 sc->sc_egid = p->p_ucred->cr_gid; 1068 sc->sc_ngroups = p->p_ucred->cr_ngroups; 1069 for (i = 0; i < sc->sc_ngroups; i++) 1070 sc->sc_groups[i] = p->p_ucred->cr_groups[i]; 1071 1072 /* 1073 * If a control message already exists, append us to the end. 1074 */ 1075 if (control != NULL) { 1076 for (n = control; n->m_next != NULL; n = n->m_next) 1077 ; 1078 n->m_next = m; 1079 } else 1080 control = m; 1081 1082 return (control); 1083 } 1084 1085 int unp_defer, unp_gcing; 1086 extern struct domain unixdomain; 1087 1088 /* 1089 * Comment added long after the fact explaining what's going on here. 1090 * Do a mark-sweep GC of file descriptors on the system, to free up 1091 * any which are caught in flight to an about-to-be-closed socket. 1092 * 1093 * Traditional mark-sweep gc's start at the "root", and mark 1094 * everything reachable from the root (which, in our case would be the 1095 * process table). The mark bits are cleared during the sweep. 1096 * 1097 * XXX For some inexplicable reason (perhaps because the file 1098 * descriptor tables used to live in the u area which could be swapped 1099 * out and thus hard to reach), we do multiple scans over the set of 1100 * descriptors, using use *two* mark bits per object (DEFER and MARK). 1101 * Whenever we find a descriptor which references other descriptors, 1102 * the ones it references are marked with both bits, and we iterate 1103 * over the whole file table until there are no more DEFER bits set. 1104 * We also make an extra pass *before* the GC to clear the mark bits, 1105 * which could have been cleared at almost no cost during the previous 1106 * sweep. 1107 * 1108 * XXX MP: this needs to run with locks such that no other thread of 1109 * control can create or destroy references to file descriptors. it 1110 * may be necessary to defer the GC until later (when the locking 1111 * situation is more hospitable); it may be necessary to push this 1112 * into a separate thread. 1113 */ 1114 void 1115 unp_gc(void) 1116 { 1117 struct file *fp, *nextfp; 1118 struct socket *so, *so1; 1119 struct file **extra_ref, **fpp; 1120 int nunref, i; 1121 1122 if (unp_gcing) 1123 return; 1124 unp_gcing = 1; 1125 unp_defer = 0; 1126 1127 /* Clear mark bits */ 1128 LIST_FOREACH(fp, &filehead, f_list) 1129 fp->f_flag &= ~(FMARK|FDEFER); 1130 1131 /* 1132 * Iterate over the set of descriptors, marking ones believed 1133 * (based on refcount) to be referenced from a process, and 1134 * marking for rescan descriptors which are queued on a socket. 1135 */ 1136 do { 1137 LIST_FOREACH(fp, &filehead, f_list) { 1138 if (fp->f_flag & FDEFER) { 1139 fp->f_flag &= ~FDEFER; 1140 unp_defer--; 1141 #ifdef DIAGNOSTIC 1142 if (fp->f_count == 0) 1143 panic("unp_gc: deferred unreferenced socket"); 1144 #endif 1145 } else { 1146 if (fp->f_count == 0) 1147 continue; 1148 if (fp->f_flag & FMARK) 1149 continue; 1150 if (fp->f_count == fp->f_msgcount) 1151 continue; 1152 } 1153 fp->f_flag |= FMARK; 1154 1155 if (fp->f_type != DTYPE_SOCKET || 1156 (so = (struct socket *)fp->f_data) == 0) 1157 continue; 1158 if (so->so_proto->pr_domain != &unixdomain || 1159 (so->so_proto->pr_flags&PR_RIGHTS) == 0) 1160 continue; 1161 #ifdef notdef 1162 if (so->so_rcv.sb_flags & SB_LOCK) { 1163 /* 1164 * This is problematical; it's not clear 1165 * we need to wait for the sockbuf to be 1166 * unlocked (on a uniprocessor, at least), 1167 * and it's also not clear what to do 1168 * if sbwait returns an error due to receipt 1169 * of a signal. If sbwait does return 1170 * an error, we'll go into an infinite 1171 * loop. Delete all of this for now. 1172 */ 1173 (void) sbwait(&so->so_rcv); 1174 goto restart; 1175 } 1176 #endif 1177 unp_scan(so->so_rcv.sb_mb, unp_mark, 0); 1178 /* 1179 * mark descriptors referenced from sockets queued on the accept queue as well. 1180 */ 1181 if (so->so_options & SO_ACCEPTCONN) { 1182 TAILQ_FOREACH(so1, &so->so_q0, so_qe) { 1183 unp_scan(so1->so_rcv.sb_mb, unp_mark, 0); 1184 } 1185 TAILQ_FOREACH(so1, &so->so_q, so_qe) { 1186 unp_scan(so1->so_rcv.sb_mb, unp_mark, 0); 1187 } 1188 } 1189 1190 } 1191 } while (unp_defer); 1192 /* 1193 * Sweep pass. Find unmarked descriptors, and free them. 1194 * 1195 * We grab an extra reference to each of the file table entries 1196 * that are not otherwise accessible and then free the rights 1197 * that are stored in messages on them. 1198 * 1199 * The bug in the original code is a little tricky, so I'll describe 1200 * what's wrong with it here. 1201 * 1202 * It is incorrect to simply unp_discard each entry for f_msgcount 1203 * times -- consider the case of sockets A and B that contain 1204 * references to each other. On a last close of some other socket, 1205 * we trigger a gc since the number of outstanding rights (unp_rights) 1206 * is non-zero. If during the sweep phase the gc code un_discards, 1207 * we end up doing a (full) closef on the descriptor. A closef on A 1208 * results in the following chain. Closef calls soo_close, which 1209 * calls soclose. Soclose calls first (through the switch 1210 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply 1211 * returns because the previous instance had set unp_gcing, and 1212 * we return all the way back to soclose, which marks the socket 1213 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush 1214 * to free up the rights that are queued in messages on the socket A, 1215 * i.e., the reference on B. The sorflush calls via the dom_dispose 1216 * switch unp_dispose, which unp_scans with unp_discard. This second 1217 * instance of unp_discard just calls closef on B. 1218 * 1219 * Well, a similar chain occurs on B, resulting in a sorflush on B, 1220 * which results in another closef on A. Unfortunately, A is already 1221 * being closed, and the descriptor has already been marked with 1222 * SS_NOFDREF, and soclose panics at this point. 1223 * 1224 * Here, we first take an extra reference to each inaccessible 1225 * descriptor. Then, if the inaccessible descriptor is a 1226 * socket, we call sorflush in case it is a Unix domain 1227 * socket. After we destroy all the rights carried in 1228 * messages, we do a last closef to get rid of our extra 1229 * reference. This is the last close, and the unp_detach etc 1230 * will shut down the socket. 1231 * 1232 * 91/09/19, bsy@cs.cmu.edu 1233 */ 1234 extra_ref = malloc(nfiles * sizeof(struct file *), M_FILE, M_WAITOK); 1235 for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref; fp != 0; 1236 fp = nextfp) { 1237 nextfp = LIST_NEXT(fp, f_list); 1238 simple_lock(&fp->f_slock); 1239 if (fp->f_count != 0 && 1240 fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) { 1241 *fpp++ = fp; 1242 nunref++; 1243 fp->f_count++; 1244 } 1245 simple_unlock(&fp->f_slock); 1246 } 1247 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1248 fp = *fpp; 1249 simple_lock(&fp->f_slock); 1250 FILE_USE(fp); 1251 if (fp->f_type == DTYPE_SOCKET) 1252 sorflush((struct socket *)fp->f_data); 1253 FILE_UNUSE(fp, NULL); 1254 } 1255 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1256 fp = *fpp; 1257 simple_lock(&fp->f_slock); 1258 FILE_USE(fp); 1259 (void) closef(fp, (struct lwp *)0); 1260 } 1261 free((caddr_t)extra_ref, M_FILE); 1262 unp_gcing = 0; 1263 } 1264 1265 void 1266 unp_dispose(struct mbuf *m) 1267 { 1268 1269 if (m) 1270 unp_scan(m, unp_discard, 1); 1271 } 1272 1273 void 1274 unp_scan(struct mbuf *m0, void (*op)(struct file *), int discard) 1275 { 1276 struct mbuf *m; 1277 struct file **rp; 1278 struct cmsghdr *cm; 1279 int i; 1280 int qfds; 1281 1282 while (m0) { 1283 for (m = m0; m; m = m->m_next) { 1284 if (m->m_type == MT_CONTROL && 1285 m->m_len >= sizeof(*cm)) { 1286 cm = mtod(m, struct cmsghdr *); 1287 if (cm->cmsg_level != SOL_SOCKET || 1288 cm->cmsg_type != SCM_RIGHTS) 1289 continue; 1290 qfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) 1291 / sizeof(struct file *); 1292 rp = (struct file **)CMSG_DATA(cm); 1293 for (i = 0; i < qfds; i++) { 1294 struct file *fp = *rp; 1295 if (discard) 1296 *rp = 0; 1297 (*op)(fp); 1298 rp++; 1299 } 1300 break; /* XXX, but saves time */ 1301 } 1302 } 1303 m0 = m0->m_nextpkt; 1304 } 1305 } 1306 1307 void 1308 unp_mark(struct file *fp) 1309 { 1310 if (fp == NULL) 1311 return; 1312 1313 if (fp->f_flag & FMARK) 1314 return; 1315 1316 /* If we're already deferred, don't screw up the defer count */ 1317 if (fp->f_flag & FDEFER) 1318 return; 1319 1320 /* 1321 * Minimize the number of deferrals... Sockets are the only 1322 * type of descriptor which can hold references to another 1323 * descriptor, so just mark other descriptors, and defer 1324 * unmarked sockets for the next pass. 1325 */ 1326 if (fp->f_type == DTYPE_SOCKET) { 1327 unp_defer++; 1328 if (fp->f_count == 0) 1329 panic("unp_mark: queued unref"); 1330 fp->f_flag |= FDEFER; 1331 } else { 1332 fp->f_flag |= FMARK; 1333 } 1334 return; 1335 } 1336 1337 void 1338 unp_discard(struct file *fp) 1339 { 1340 if (fp == NULL) 1341 return; 1342 simple_lock(&fp->f_slock); 1343 fp->f_usecount++; /* i.e. FILE_USE(fp) sans locking */ 1344 fp->f_msgcount--; 1345 simple_unlock(&fp->f_slock); 1346 unp_rights--; 1347 (void) closef(fp, (struct lwp *)0); 1348 } 1349