1 /* 2 * Copyright (c) 1982, 1986, 1989, 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. All advertising materials mentioning features or use of this software 14 * must display the following acknowledgement: 15 * This product includes software developed by the University of 16 * California, Berkeley and its contributors. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94 34 * $FreeBSD: src/sys/kern/uipc_usrreq.c,v 1.54.2.10 2003/03/04 17:28:09 nectar Exp $ 35 * $DragonFly: src/sys/kern/uipc_usrreq.c,v 1.2 2003/06/17 04:28:41 dillon Exp $ 36 */ 37 38 #include <sys/param.h> 39 #include <sys/systm.h> 40 #include <sys/kernel.h> 41 #include <sys/domain.h> 42 #include <sys/fcntl.h> 43 #include <sys/malloc.h> /* XXX must be before <sys/file.h> */ 44 #include <sys/file.h> 45 #include <sys/filedesc.h> 46 #include <sys/mbuf.h> 47 #include <sys/namei.h> 48 #include <sys/proc.h> 49 #include <sys/protosw.h> 50 #include <sys/socket.h> 51 #include <sys/socketvar.h> 52 #include <sys/resourcevar.h> 53 #include <sys/stat.h> 54 #include <sys/sysctl.h> 55 #include <sys/un.h> 56 #include <sys/unpcb.h> 57 #include <sys/vnode.h> 58 59 #include <vm/vm_zone.h> 60 61 static struct vm_zone *unp_zone; 62 static unp_gen_t unp_gencnt; 63 static u_int unp_count; 64 65 static struct unp_head unp_shead, unp_dhead; 66 67 /* 68 * Unix communications domain. 69 * 70 * TODO: 71 * SEQPACKET, RDM 72 * rethink name space problems 73 * need a proper out-of-band 74 * lock pushdown 75 */ 76 static struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL }; 77 static ino_t unp_ino; /* prototype for fake inode numbers */ 78 79 static int unp_attach __P((struct socket *)); 80 static void unp_detach __P((struct unpcb *)); 81 static int unp_bind __P((struct unpcb *,struct sockaddr *, struct proc *)); 82 static int unp_connect __P((struct socket *,struct sockaddr *, 83 struct proc *)); 84 static void unp_disconnect __P((struct unpcb *)); 85 static void unp_shutdown __P((struct unpcb *)); 86 static void unp_drop __P((struct unpcb *, int)); 87 static void unp_gc __P((void)); 88 static void unp_scan __P((struct mbuf *, void (*)(struct file *))); 89 static void unp_mark __P((struct file *)); 90 static void unp_discard __P((struct file *)); 91 static int unp_internalize __P((struct mbuf *, struct proc *)); 92 static int unp_listen __P((struct unpcb *, struct proc *)); 93 94 static int 95 uipc_abort(struct socket *so) 96 { 97 struct unpcb *unp = sotounpcb(so); 98 99 if (unp == 0) 100 return EINVAL; 101 unp_drop(unp, ECONNABORTED); 102 unp_detach(unp); 103 sofree(so); 104 return 0; 105 } 106 107 static int 108 uipc_accept(struct socket *so, struct sockaddr **nam) 109 { 110 struct unpcb *unp = sotounpcb(so); 111 112 if (unp == 0) 113 return EINVAL; 114 115 /* 116 * Pass back name of connected socket, 117 * if it was bound and we are still connected 118 * (our peer may have closed already!). 119 */ 120 if (unp->unp_conn && unp->unp_conn->unp_addr) { 121 *nam = dup_sockaddr((struct sockaddr *)unp->unp_conn->unp_addr, 122 1); 123 } else { 124 *nam = dup_sockaddr((struct sockaddr *)&sun_noname, 1); 125 } 126 return 0; 127 } 128 129 static int 130 uipc_attach(struct socket *so, int proto, struct proc *p) 131 { 132 struct unpcb *unp = sotounpcb(so); 133 134 if (unp != 0) 135 return EISCONN; 136 return unp_attach(so); 137 } 138 139 static int 140 uipc_bind(struct socket *so, struct sockaddr *nam, struct proc *p) 141 { 142 struct unpcb *unp = sotounpcb(so); 143 144 if (unp == 0) 145 return EINVAL; 146 147 return unp_bind(unp, nam, p); 148 } 149 150 static int 151 uipc_connect(struct socket *so, struct sockaddr *nam, struct proc *p) 152 { 153 struct unpcb *unp = sotounpcb(so); 154 155 if (unp == 0) 156 return EINVAL; 157 return unp_connect(so, nam, curproc); 158 } 159 160 static int 161 uipc_connect2(struct socket *so1, struct socket *so2) 162 { 163 struct unpcb *unp = sotounpcb(so1); 164 165 if (unp == 0) 166 return EINVAL; 167 168 return unp_connect2(so1, so2); 169 } 170 171 /* control is EOPNOTSUPP */ 172 173 static int 174 uipc_detach(struct socket *so) 175 { 176 struct unpcb *unp = sotounpcb(so); 177 178 if (unp == 0) 179 return EINVAL; 180 181 unp_detach(unp); 182 return 0; 183 } 184 185 static int 186 uipc_disconnect(struct socket *so) 187 { 188 struct unpcb *unp = sotounpcb(so); 189 190 if (unp == 0) 191 return EINVAL; 192 unp_disconnect(unp); 193 return 0; 194 } 195 196 static int 197 uipc_listen(struct socket *so, struct proc *p) 198 { 199 struct unpcb *unp = sotounpcb(so); 200 201 if (unp == 0 || unp->unp_vnode == 0) 202 return EINVAL; 203 return unp_listen(unp, p); 204 } 205 206 static int 207 uipc_peeraddr(struct socket *so, struct sockaddr **nam) 208 { 209 struct unpcb *unp = sotounpcb(so); 210 211 if (unp == 0) 212 return EINVAL; 213 if (unp->unp_conn && unp->unp_conn->unp_addr) 214 *nam = dup_sockaddr((struct sockaddr *)unp->unp_conn->unp_addr, 215 1); 216 else { 217 /* 218 * XXX: It seems that this test always fails even when 219 * connection is established. So, this else clause is 220 * added as workaround to return PF_LOCAL sockaddr. 221 */ 222 *nam = dup_sockaddr((struct sockaddr *)&sun_noname, 1); 223 } 224 return 0; 225 } 226 227 static int 228 uipc_rcvd(struct socket *so, int flags) 229 { 230 struct unpcb *unp = sotounpcb(so); 231 struct socket *so2; 232 u_long newhiwat; 233 234 if (unp == 0) 235 return EINVAL; 236 switch (so->so_type) { 237 case SOCK_DGRAM: 238 panic("uipc_rcvd DGRAM?"); 239 /*NOTREACHED*/ 240 241 case SOCK_STREAM: 242 if (unp->unp_conn == 0) 243 break; 244 so2 = unp->unp_conn->unp_socket; 245 /* 246 * Adjust backpressure on sender 247 * and wakeup any waiting to write. 248 */ 249 so2->so_snd.sb_mbmax += unp->unp_mbcnt - so->so_rcv.sb_mbcnt; 250 unp->unp_mbcnt = so->so_rcv.sb_mbcnt; 251 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - 252 so->so_rcv.sb_cc; 253 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat, 254 newhiwat, RLIM_INFINITY); 255 unp->unp_cc = so->so_rcv.sb_cc; 256 sowwakeup(so2); 257 break; 258 259 default: 260 panic("uipc_rcvd unknown socktype"); 261 } 262 return 0; 263 } 264 265 /* pru_rcvoob is EOPNOTSUPP */ 266 267 static int 268 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 269 struct mbuf *control, struct proc *p) 270 { 271 int error = 0; 272 struct unpcb *unp = sotounpcb(so); 273 struct socket *so2; 274 u_long newhiwat; 275 276 if (unp == 0) { 277 error = EINVAL; 278 goto release; 279 } 280 if (flags & PRUS_OOB) { 281 error = EOPNOTSUPP; 282 goto release; 283 } 284 285 if (control && (error = unp_internalize(control, p))) 286 goto release; 287 288 switch (so->so_type) { 289 case SOCK_DGRAM: 290 { 291 struct sockaddr *from; 292 293 if (nam) { 294 if (unp->unp_conn) { 295 error = EISCONN; 296 break; 297 } 298 error = unp_connect(so, nam, p); 299 if (error) 300 break; 301 } else { 302 if (unp->unp_conn == 0) { 303 error = ENOTCONN; 304 break; 305 } 306 } 307 so2 = unp->unp_conn->unp_socket; 308 if (unp->unp_addr) 309 from = (struct sockaddr *)unp->unp_addr; 310 else 311 from = &sun_noname; 312 if (sbappendaddr(&so2->so_rcv, from, m, control)) { 313 sorwakeup(so2); 314 m = 0; 315 control = 0; 316 } else 317 error = ENOBUFS; 318 if (nam) 319 unp_disconnect(unp); 320 break; 321 } 322 323 case SOCK_STREAM: 324 /* Connect if not connected yet. */ 325 /* 326 * Note: A better implementation would complain 327 * if not equal to the peer's address. 328 */ 329 if ((so->so_state & SS_ISCONNECTED) == 0) { 330 if (nam) { 331 error = unp_connect(so, nam, p); 332 if (error) 333 break; /* XXX */ 334 } else { 335 error = ENOTCONN; 336 break; 337 } 338 } 339 340 if (so->so_state & SS_CANTSENDMORE) { 341 error = EPIPE; 342 break; 343 } 344 if (unp->unp_conn == 0) 345 panic("uipc_send connected but no connection?"); 346 so2 = unp->unp_conn->unp_socket; 347 /* 348 * Send to paired receive port, and then reduce 349 * send buffer hiwater marks to maintain backpressure. 350 * Wake up readers. 351 */ 352 if (control) { 353 if (sbappendcontrol(&so2->so_rcv, m, control)) 354 control = 0; 355 } else 356 sbappend(&so2->so_rcv, m); 357 so->so_snd.sb_mbmax -= 358 so2->so_rcv.sb_mbcnt - unp->unp_conn->unp_mbcnt; 359 unp->unp_conn->unp_mbcnt = so2->so_rcv.sb_mbcnt; 360 newhiwat = so->so_snd.sb_hiwat - 361 (so2->so_rcv.sb_cc - unp->unp_conn->unp_cc); 362 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat, 363 newhiwat, RLIM_INFINITY); 364 unp->unp_conn->unp_cc = so2->so_rcv.sb_cc; 365 sorwakeup(so2); 366 m = 0; 367 break; 368 369 default: 370 panic("uipc_send unknown socktype"); 371 } 372 373 /* 374 * SEND_EOF is equivalent to a SEND followed by 375 * a SHUTDOWN. 376 */ 377 if (flags & PRUS_EOF) { 378 socantsendmore(so); 379 unp_shutdown(unp); 380 } 381 382 if (control && error != 0) 383 unp_dispose(control); 384 385 release: 386 if (control) 387 m_freem(control); 388 if (m) 389 m_freem(m); 390 return error; 391 } 392 393 static int 394 uipc_sense(struct socket *so, struct stat *sb) 395 { 396 struct unpcb *unp = sotounpcb(so); 397 struct socket *so2; 398 399 if (unp == 0) 400 return EINVAL; 401 sb->st_blksize = so->so_snd.sb_hiwat; 402 if (so->so_type == SOCK_STREAM && unp->unp_conn != 0) { 403 so2 = unp->unp_conn->unp_socket; 404 sb->st_blksize += so2->so_rcv.sb_cc; 405 } 406 sb->st_dev = NOUDEV; 407 if (unp->unp_ino == 0) /* make up a non-zero inode number */ 408 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino; 409 sb->st_ino = unp->unp_ino; 410 return (0); 411 } 412 413 static int 414 uipc_shutdown(struct socket *so) 415 { 416 struct unpcb *unp = sotounpcb(so); 417 418 if (unp == 0) 419 return EINVAL; 420 socantsendmore(so); 421 unp_shutdown(unp); 422 return 0; 423 } 424 425 static int 426 uipc_sockaddr(struct socket *so, struct sockaddr **nam) 427 { 428 struct unpcb *unp = sotounpcb(so); 429 430 if (unp == 0) 431 return EINVAL; 432 if (unp->unp_addr) 433 *nam = dup_sockaddr((struct sockaddr *)unp->unp_addr, 1); 434 return 0; 435 } 436 437 struct pr_usrreqs uipc_usrreqs = { 438 uipc_abort, uipc_accept, uipc_attach, uipc_bind, uipc_connect, 439 uipc_connect2, pru_control_notsupp, uipc_detach, uipc_disconnect, 440 uipc_listen, uipc_peeraddr, uipc_rcvd, pru_rcvoob_notsupp, 441 uipc_send, uipc_sense, uipc_shutdown, uipc_sockaddr, 442 sosend, soreceive, sopoll 443 }; 444 445 int 446 uipc_ctloutput(so, sopt) 447 struct socket *so; 448 struct sockopt *sopt; 449 { 450 struct unpcb *unp = sotounpcb(so); 451 int error; 452 453 switch (sopt->sopt_dir) { 454 case SOPT_GET: 455 switch (sopt->sopt_name) { 456 case LOCAL_PEERCRED: 457 if (unp->unp_flags & UNP_HAVEPC) 458 error = sooptcopyout(sopt, &unp->unp_peercred, 459 sizeof(unp->unp_peercred)); 460 else { 461 if (so->so_type == SOCK_STREAM) 462 error = ENOTCONN; 463 else 464 error = EINVAL; 465 } 466 break; 467 default: 468 error = EOPNOTSUPP; 469 break; 470 } 471 break; 472 case SOPT_SET: 473 default: 474 error = EOPNOTSUPP; 475 break; 476 } 477 return (error); 478 } 479 480 /* 481 * Both send and receive buffers are allocated PIPSIZ bytes of buffering 482 * for stream sockets, although the total for sender and receiver is 483 * actually only PIPSIZ. 484 * Datagram sockets really use the sendspace as the maximum datagram size, 485 * and don't really want to reserve the sendspace. Their recvspace should 486 * be large enough for at least one max-size datagram plus address. 487 */ 488 #ifndef PIPSIZ 489 #define PIPSIZ 8192 490 #endif 491 static u_long unpst_sendspace = PIPSIZ; 492 static u_long unpst_recvspace = PIPSIZ; 493 static u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 494 static u_long unpdg_recvspace = 4*1024; 495 496 static int unp_rights; /* file descriptors in flight */ 497 498 SYSCTL_DECL(_net_local_stream); 499 SYSCTL_INT(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 500 &unpst_sendspace, 0, ""); 501 SYSCTL_INT(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 502 &unpst_recvspace, 0, ""); 503 SYSCTL_DECL(_net_local_dgram); 504 SYSCTL_INT(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 505 &unpdg_sendspace, 0, ""); 506 SYSCTL_INT(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 507 &unpdg_recvspace, 0, ""); 508 SYSCTL_DECL(_net_local); 509 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, ""); 510 511 static int 512 unp_attach(so) 513 struct socket *so; 514 { 515 register struct unpcb *unp; 516 int error; 517 518 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 519 switch (so->so_type) { 520 521 case SOCK_STREAM: 522 error = soreserve(so, unpst_sendspace, unpst_recvspace); 523 break; 524 525 case SOCK_DGRAM: 526 error = soreserve(so, unpdg_sendspace, unpdg_recvspace); 527 break; 528 529 default: 530 panic("unp_attach"); 531 } 532 if (error) 533 return (error); 534 } 535 unp = zalloc(unp_zone); 536 if (unp == NULL) 537 return (ENOBUFS); 538 bzero(unp, sizeof *unp); 539 unp->unp_gencnt = ++unp_gencnt; 540 unp_count++; 541 LIST_INIT(&unp->unp_refs); 542 unp->unp_socket = so; 543 unp->unp_rvnode = curproc->p_fd->fd_rdir; 544 LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead 545 : &unp_shead, unp, unp_link); 546 so->so_pcb = (caddr_t)unp; 547 return (0); 548 } 549 550 static void 551 unp_detach(unp) 552 register struct unpcb *unp; 553 { 554 LIST_REMOVE(unp, unp_link); 555 unp->unp_gencnt = ++unp_gencnt; 556 --unp_count; 557 if (unp->unp_vnode) { 558 unp->unp_vnode->v_socket = 0; 559 vrele(unp->unp_vnode); 560 unp->unp_vnode = 0; 561 } 562 if (unp->unp_conn) 563 unp_disconnect(unp); 564 while (!LIST_EMPTY(&unp->unp_refs)) 565 unp_drop(LIST_FIRST(&unp->unp_refs), ECONNRESET); 566 soisdisconnected(unp->unp_socket); 567 unp->unp_socket->so_pcb = 0; 568 if (unp_rights) { 569 /* 570 * Normally the receive buffer is flushed later, 571 * in sofree, but if our receive buffer holds references 572 * to descriptors that are now garbage, we will dispose 573 * of those descriptor references after the garbage collector 574 * gets them (resulting in a "panic: closef: count < 0"). 575 */ 576 sorflush(unp->unp_socket); 577 unp_gc(); 578 } 579 if (unp->unp_addr) 580 FREE(unp->unp_addr, M_SONAME); 581 zfree(unp_zone, unp); 582 } 583 584 static int 585 unp_bind(unp, nam, p) 586 struct unpcb *unp; 587 struct sockaddr *nam; 588 struct proc *p; 589 { 590 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 591 register struct vnode *vp; 592 struct vattr vattr; 593 int error, namelen; 594 struct nameidata nd; 595 char buf[SOCK_MAXADDRLEN]; 596 597 if (unp->unp_vnode != NULL) 598 return (EINVAL); 599 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 600 if (namelen <= 0) 601 return EINVAL; 602 strncpy(buf, soun->sun_path, namelen); 603 buf[namelen] = 0; /* null-terminate the string */ 604 NDINIT(&nd, CREATE, NOFOLLOW | LOCKPARENT, UIO_SYSSPACE, 605 buf, p); 606 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 607 error = namei(&nd); 608 if (error) 609 return (error); 610 vp = nd.ni_vp; 611 if (vp != NULL) { 612 NDFREE(&nd, NDF_ONLY_PNBUF); 613 if (nd.ni_dvp == vp) 614 vrele(nd.ni_dvp); 615 else 616 vput(nd.ni_dvp); 617 vrele(vp); 618 return (EADDRINUSE); 619 } 620 VATTR_NULL(&vattr); 621 vattr.va_type = VSOCK; 622 vattr.va_mode = (ACCESSPERMS & ~p->p_fd->fd_cmask); 623 VOP_LEASE(nd.ni_dvp, p, p->p_ucred, LEASE_WRITE); 624 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 625 NDFREE(&nd, NDF_ONLY_PNBUF); 626 vput(nd.ni_dvp); 627 if (error) 628 return (error); 629 vp = nd.ni_vp; 630 vp->v_socket = unp->unp_socket; 631 unp->unp_vnode = vp; 632 unp->unp_addr = (struct sockaddr_un *)dup_sockaddr(nam, 1); 633 VOP_UNLOCK(vp, 0, p); 634 return (0); 635 } 636 637 static int 638 unp_connect(so, nam, p) 639 struct socket *so; 640 struct sockaddr *nam; 641 struct proc *p; 642 { 643 register struct sockaddr_un *soun = (struct sockaddr_un *)nam; 644 register struct vnode *vp; 645 register struct socket *so2, *so3; 646 struct unpcb *unp, *unp2, *unp3; 647 int error, len; 648 struct nameidata nd; 649 char buf[SOCK_MAXADDRLEN]; 650 651 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 652 if (len <= 0) 653 return EINVAL; 654 strncpy(buf, soun->sun_path, len); 655 buf[len] = 0; 656 657 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, p); 658 error = namei(&nd); 659 if (error) 660 return (error); 661 vp = nd.ni_vp; 662 NDFREE(&nd, NDF_ONLY_PNBUF); 663 if (vp->v_type != VSOCK) { 664 error = ENOTSOCK; 665 goto bad; 666 } 667 error = VOP_ACCESS(vp, VWRITE, p->p_ucred, p); 668 if (error) 669 goto bad; 670 so2 = vp->v_socket; 671 if (so2 == 0) { 672 error = ECONNREFUSED; 673 goto bad; 674 } 675 if (so->so_type != so2->so_type) { 676 error = EPROTOTYPE; 677 goto bad; 678 } 679 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 680 if ((so2->so_options & SO_ACCEPTCONN) == 0 || 681 (so3 = sonewconn3(so2, 0, p)) == 0) { 682 error = ECONNREFUSED; 683 goto bad; 684 } 685 unp = sotounpcb(so); 686 unp2 = sotounpcb(so2); 687 unp3 = sotounpcb(so3); 688 if (unp2->unp_addr) 689 unp3->unp_addr = (struct sockaddr_un *) 690 dup_sockaddr((struct sockaddr *) 691 unp2->unp_addr, 1); 692 693 /* 694 * unp_peercred management: 695 * 696 * The connecter's (client's) credentials are copied 697 * from its process structure at the time of connect() 698 * (which is now). 699 */ 700 cru2x(p->p_ucred, &unp3->unp_peercred); 701 unp3->unp_flags |= UNP_HAVEPC; 702 /* 703 * The receiver's (server's) credentials are copied 704 * from the unp_peercred member of socket on which the 705 * former called listen(); unp_listen() cached that 706 * process's credentials at that time so we can use 707 * them now. 708 */ 709 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED, 710 ("unp_connect: listener without cached peercred")); 711 memcpy(&unp->unp_peercred, &unp2->unp_peercred, 712 sizeof(unp->unp_peercred)); 713 unp->unp_flags |= UNP_HAVEPC; 714 715 so2 = so3; 716 } 717 error = unp_connect2(so, so2); 718 bad: 719 vput(vp); 720 return (error); 721 } 722 723 int 724 unp_connect2(so, so2) 725 register struct socket *so; 726 register struct socket *so2; 727 { 728 register struct unpcb *unp = sotounpcb(so); 729 register struct unpcb *unp2; 730 731 if (so2->so_type != so->so_type) 732 return (EPROTOTYPE); 733 unp2 = sotounpcb(so2); 734 unp->unp_conn = unp2; 735 switch (so->so_type) { 736 737 case SOCK_DGRAM: 738 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 739 soisconnected(so); 740 break; 741 742 case SOCK_STREAM: 743 unp2->unp_conn = unp; 744 soisconnected(so); 745 soisconnected(so2); 746 break; 747 748 default: 749 panic("unp_connect2"); 750 } 751 return (0); 752 } 753 754 static void 755 unp_disconnect(unp) 756 struct unpcb *unp; 757 { 758 register struct unpcb *unp2 = unp->unp_conn; 759 760 if (unp2 == 0) 761 return; 762 unp->unp_conn = 0; 763 switch (unp->unp_socket->so_type) { 764 765 case SOCK_DGRAM: 766 LIST_REMOVE(unp, unp_reflink); 767 unp->unp_socket->so_state &= ~SS_ISCONNECTED; 768 break; 769 770 case SOCK_STREAM: 771 soisdisconnected(unp->unp_socket); 772 unp2->unp_conn = 0; 773 soisdisconnected(unp2->unp_socket); 774 break; 775 } 776 } 777 778 #ifdef notdef 779 void 780 unp_abort(unp) 781 struct unpcb *unp; 782 { 783 784 unp_detach(unp); 785 } 786 #endif 787 788 static int 789 prison_unpcb(struct proc *p, struct unpcb *unp) 790 { 791 if (!p->p_prison) 792 return (0); 793 if (p->p_fd->fd_rdir == unp->unp_rvnode) 794 return (0); 795 return (1); 796 } 797 798 static int 799 unp_pcblist(SYSCTL_HANDLER_ARGS) 800 { 801 int error, i, n; 802 struct unpcb *unp, **unp_list; 803 unp_gen_t gencnt; 804 struct xunpgen xug; 805 struct unp_head *head; 806 807 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead); 808 809 /* 810 * The process of preparing the PCB list is too time-consuming and 811 * resource-intensive to repeat twice on every request. 812 */ 813 if (req->oldptr == 0) { 814 n = unp_count; 815 req->oldidx = 2 * (sizeof xug) 816 + (n + n/8) * sizeof(struct xunpcb); 817 return 0; 818 } 819 820 if (req->newptr != 0) 821 return EPERM; 822 823 /* 824 * OK, now we're committed to doing something. 825 */ 826 gencnt = unp_gencnt; 827 n = unp_count; 828 829 xug.xug_len = sizeof xug; 830 xug.xug_count = n; 831 xug.xug_gen = gencnt; 832 xug.xug_sogen = so_gencnt; 833 error = SYSCTL_OUT(req, &xug, sizeof xug); 834 if (error) 835 return error; 836 837 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 838 if (unp_list == 0) 839 return ENOMEM; 840 841 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 842 unp = LIST_NEXT(unp, unp_link)) { 843 if (unp->unp_gencnt <= gencnt && !prison_unpcb(req->p, unp)) 844 unp_list[i++] = unp; 845 } 846 n = i; /* in case we lost some during malloc */ 847 848 error = 0; 849 for (i = 0; i < n; i++) { 850 unp = unp_list[i]; 851 if (unp->unp_gencnt <= gencnt) { 852 struct xunpcb xu; 853 xu.xu_len = sizeof xu; 854 xu.xu_unpp = unp; 855 /* 856 * XXX - need more locking here to protect against 857 * connect/disconnect races for SMP. 858 */ 859 if (unp->unp_addr) 860 bcopy(unp->unp_addr, &xu.xu_addr, 861 unp->unp_addr->sun_len); 862 if (unp->unp_conn && unp->unp_conn->unp_addr) 863 bcopy(unp->unp_conn->unp_addr, 864 &xu.xu_caddr, 865 unp->unp_conn->unp_addr->sun_len); 866 bcopy(unp, &xu.xu_unp, sizeof *unp); 867 sotoxsocket(unp->unp_socket, &xu.xu_socket); 868 error = SYSCTL_OUT(req, &xu, sizeof xu); 869 } 870 } 871 if (!error) { 872 /* 873 * Give the user an updated idea of our state. 874 * If the generation differs from what we told 875 * her before, she knows that something happened 876 * while we were processing this request, and it 877 * might be necessary to retry. 878 */ 879 xug.xug_gen = unp_gencnt; 880 xug.xug_sogen = so_gencnt; 881 xug.xug_count = unp_count; 882 error = SYSCTL_OUT(req, &xug, sizeof xug); 883 } 884 free(unp_list, M_TEMP); 885 return error; 886 } 887 888 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD, 889 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 890 "List of active local datagram sockets"); 891 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD, 892 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 893 "List of active local stream sockets"); 894 895 static void 896 unp_shutdown(unp) 897 struct unpcb *unp; 898 { 899 struct socket *so; 900 901 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn && 902 (so = unp->unp_conn->unp_socket)) 903 socantrcvmore(so); 904 } 905 906 static void 907 unp_drop(unp, errno) 908 struct unpcb *unp; 909 int errno; 910 { 911 struct socket *so = unp->unp_socket; 912 913 so->so_error = errno; 914 unp_disconnect(unp); 915 } 916 917 #ifdef notdef 918 void 919 unp_drain() 920 { 921 922 } 923 #endif 924 925 int 926 unp_externalize(rights) 927 struct mbuf *rights; 928 { 929 struct proc *p = curproc; /* XXX */ 930 register int i; 931 register struct cmsghdr *cm = mtod(rights, struct cmsghdr *); 932 register int *fdp; 933 register struct file **rp; 934 register struct file *fp; 935 int newfds = (cm->cmsg_len - (CMSG_DATA(cm) - (u_char *)cm)) 936 / sizeof (struct file *); 937 int f; 938 939 /* 940 * if the new FD's will not fit, then we free them all 941 */ 942 if (!fdavail(p, newfds)) { 943 rp = (struct file **)CMSG_DATA(cm); 944 for (i = 0; i < newfds; i++) { 945 fp = *rp; 946 /* 947 * zero the pointer before calling unp_discard, 948 * since it may end up in unp_gc().. 949 */ 950 *rp++ = 0; 951 unp_discard(fp); 952 } 953 return (EMSGSIZE); 954 } 955 /* 956 * now change each pointer to an fd in the global table to 957 * an integer that is the index to the local fd table entry 958 * that we set up to point to the global one we are transferring. 959 * If sizeof (struct file *) is bigger than or equal to sizeof int, 960 * then do it in forward order. In that case, an integer will 961 * always come in the same place or before its corresponding 962 * struct file pointer. 963 * If sizeof (struct file *) is smaller than sizeof int, then 964 * do it in reverse order. 965 */ 966 if (sizeof (struct file *) >= sizeof (int)) { 967 fdp = (int *)(cm + 1); 968 rp = (struct file **)CMSG_DATA(cm); 969 for (i = 0; i < newfds; i++) { 970 if (fdalloc(p, 0, &f)) 971 panic("unp_externalize"); 972 fp = *rp++; 973 p->p_fd->fd_ofiles[f] = fp; 974 fp->f_msgcount--; 975 unp_rights--; 976 *fdp++ = f; 977 } 978 } else { 979 fdp = (int *)(cm + 1) + newfds - 1; 980 rp = (struct file **)CMSG_DATA(cm) + newfds - 1; 981 for (i = 0; i < newfds; i++) { 982 if (fdalloc(p, 0, &f)) 983 panic("unp_externalize"); 984 fp = *rp--; 985 p->p_fd->fd_ofiles[f] = fp; 986 fp->f_msgcount--; 987 unp_rights--; 988 *fdp-- = f; 989 } 990 } 991 992 /* 993 * Adjust length, in case sizeof(struct file *) and sizeof(int) 994 * differs. 995 */ 996 cm->cmsg_len = CMSG_LEN(newfds * sizeof(int)); 997 rights->m_len = cm->cmsg_len; 998 return (0); 999 } 1000 1001 void 1002 unp_init(void) 1003 { 1004 unp_zone = zinit("unpcb", sizeof(struct unpcb), nmbclusters, 0, 0); 1005 if (unp_zone == 0) 1006 panic("unp_init"); 1007 LIST_INIT(&unp_dhead); 1008 LIST_INIT(&unp_shead); 1009 } 1010 1011 #ifndef MIN 1012 #define MIN(a,b) (((a)<(b))?(a):(b)) 1013 #endif 1014 1015 static int 1016 unp_internalize(control, p) 1017 struct mbuf *control; 1018 struct proc *p; 1019 { 1020 struct filedesc *fdescp = p->p_fd; 1021 register struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1022 register struct file **rp; 1023 register struct file *fp; 1024 register int i, fd, *fdp; 1025 register struct cmsgcred *cmcred; 1026 int oldfds; 1027 u_int newlen; 1028 1029 if ((cm->cmsg_type != SCM_RIGHTS && cm->cmsg_type != SCM_CREDS) || 1030 cm->cmsg_level != SOL_SOCKET || cm->cmsg_len != control->m_len) 1031 return (EINVAL); 1032 1033 /* 1034 * Fill in credential information. 1035 */ 1036 if (cm->cmsg_type == SCM_CREDS) { 1037 cmcred = (struct cmsgcred *)(cm + 1); 1038 cmcred->cmcred_pid = p->p_pid; 1039 cmcred->cmcred_uid = p->p_cred->p_ruid; 1040 cmcred->cmcred_gid = p->p_cred->p_rgid; 1041 cmcred->cmcred_euid = p->p_ucred->cr_uid; 1042 cmcred->cmcred_ngroups = MIN(p->p_ucred->cr_ngroups, 1043 CMGROUP_MAX); 1044 for (i = 0; i < cmcred->cmcred_ngroups; i++) 1045 cmcred->cmcred_groups[i] = p->p_ucred->cr_groups[i]; 1046 return(0); 1047 } 1048 1049 oldfds = (cm->cmsg_len - sizeof (*cm)) / sizeof (int); 1050 /* 1051 * check that all the FDs passed in refer to legal OPEN files 1052 * If not, reject the entire operation. 1053 */ 1054 fdp = (int *)(cm + 1); 1055 for (i = 0; i < oldfds; i++) { 1056 fd = *fdp++; 1057 if ((unsigned)fd >= fdescp->fd_nfiles || 1058 fdescp->fd_ofiles[fd] == NULL) 1059 return (EBADF); 1060 if (fdescp->fd_ofiles[fd]->f_type == DTYPE_KQUEUE) 1061 return (EOPNOTSUPP); 1062 } 1063 /* 1064 * Now replace the integer FDs with pointers to 1065 * the associated global file table entry.. 1066 * Allocate a bigger buffer as necessary. But if an cluster is not 1067 * enough, return E2BIG. 1068 */ 1069 newlen = CMSG_LEN(oldfds * sizeof(struct file *)); 1070 if (newlen > MCLBYTES) 1071 return (E2BIG); 1072 if (newlen - control->m_len > M_TRAILINGSPACE(control)) { 1073 if (control->m_flags & M_EXT) 1074 return (E2BIG); 1075 MCLGET(control, M_WAIT); 1076 if ((control->m_flags & M_EXT) == 0) 1077 return (ENOBUFS); 1078 1079 /* copy the data to the cluster */ 1080 memcpy(mtod(control, char *), cm, cm->cmsg_len); 1081 cm = mtod(control, struct cmsghdr *); 1082 } 1083 1084 /* 1085 * Adjust length, in case sizeof(struct file *) and sizeof(int) 1086 * differs. 1087 */ 1088 control->m_len = cm->cmsg_len = newlen; 1089 1090 /* 1091 * Transform the file descriptors into struct file pointers. 1092 * If sizeof (struct file *) is bigger than or equal to sizeof int, 1093 * then do it in reverse order so that the int won't get until 1094 * we're done. 1095 * If sizeof (struct file *) is smaller than sizeof int, then 1096 * do it in forward order. 1097 */ 1098 if (sizeof (struct file *) >= sizeof (int)) { 1099 fdp = (int *)(cm + 1) + oldfds - 1; 1100 rp = (struct file **)CMSG_DATA(cm) + oldfds - 1; 1101 for (i = 0; i < oldfds; i++) { 1102 fp = fdescp->fd_ofiles[*fdp--]; 1103 *rp-- = fp; 1104 fp->f_count++; 1105 fp->f_msgcount++; 1106 unp_rights++; 1107 } 1108 } else { 1109 fdp = (int *)(cm + 1); 1110 rp = (struct file **)CMSG_DATA(cm); 1111 for (i = 0; i < oldfds; i++) { 1112 fp = fdescp->fd_ofiles[*fdp++]; 1113 *rp++ = fp; 1114 fp->f_count++; 1115 fp->f_msgcount++; 1116 unp_rights++; 1117 } 1118 } 1119 return (0); 1120 } 1121 1122 static int unp_defer, unp_gcing; 1123 1124 static void 1125 unp_gc() 1126 { 1127 register struct file *fp, *nextfp; 1128 register struct socket *so; 1129 struct file **extra_ref, **fpp; 1130 int nunref, i; 1131 1132 if (unp_gcing) 1133 return; 1134 unp_gcing = 1; 1135 unp_defer = 0; 1136 /* 1137 * before going through all this, set all FDs to 1138 * be NOT defered and NOT externally accessible 1139 */ 1140 LIST_FOREACH(fp, &filehead, f_list) 1141 fp->f_flag &= ~(FMARK|FDEFER); 1142 do { 1143 LIST_FOREACH(fp, &filehead, f_list) { 1144 /* 1145 * If the file is not open, skip it 1146 */ 1147 if (fp->f_count == 0) 1148 continue; 1149 /* 1150 * If we already marked it as 'defer' in a 1151 * previous pass, then try process it this time 1152 * and un-mark it 1153 */ 1154 if (fp->f_flag & FDEFER) { 1155 fp->f_flag &= ~FDEFER; 1156 unp_defer--; 1157 } else { 1158 /* 1159 * if it's not defered, then check if it's 1160 * already marked.. if so skip it 1161 */ 1162 if (fp->f_flag & FMARK) 1163 continue; 1164 /* 1165 * If all references are from messages 1166 * in transit, then skip it. it's not 1167 * externally accessible. 1168 */ 1169 if (fp->f_count == fp->f_msgcount) 1170 continue; 1171 /* 1172 * If it got this far then it must be 1173 * externally accessible. 1174 */ 1175 fp->f_flag |= FMARK; 1176 } 1177 /* 1178 * either it was defered, or it is externally 1179 * accessible and not already marked so. 1180 * Now check if it is possibly one of OUR sockets. 1181 */ 1182 if (fp->f_type != DTYPE_SOCKET || 1183 (so = (struct socket *)fp->f_data) == 0) 1184 continue; 1185 if (so->so_proto->pr_domain != &localdomain || 1186 (so->so_proto->pr_flags&PR_RIGHTS) == 0) 1187 continue; 1188 #ifdef notdef 1189 if (so->so_rcv.sb_flags & SB_LOCK) { 1190 /* 1191 * This is problematical; it's not clear 1192 * we need to wait for the sockbuf to be 1193 * unlocked (on a uniprocessor, at least), 1194 * and it's also not clear what to do 1195 * if sbwait returns an error due to receipt 1196 * of a signal. If sbwait does return 1197 * an error, we'll go into an infinite 1198 * loop. Delete all of this for now. 1199 */ 1200 (void) sbwait(&so->so_rcv); 1201 goto restart; 1202 } 1203 #endif 1204 /* 1205 * So, Ok, it's one of our sockets and it IS externally 1206 * accessible (or was defered). Now we look 1207 * to see if we hold any file descriptors in its 1208 * message buffers. Follow those links and mark them 1209 * as accessible too. 1210 */ 1211 unp_scan(so->so_rcv.sb_mb, unp_mark); 1212 } 1213 } while (unp_defer); 1214 /* 1215 * We grab an extra reference to each of the file table entries 1216 * that are not otherwise accessible and then free the rights 1217 * that are stored in messages on them. 1218 * 1219 * The bug in the orginal code is a little tricky, so I'll describe 1220 * what's wrong with it here. 1221 * 1222 * It is incorrect to simply unp_discard each entry for f_msgcount 1223 * times -- consider the case of sockets A and B that contain 1224 * references to each other. On a last close of some other socket, 1225 * we trigger a gc since the number of outstanding rights (unp_rights) 1226 * is non-zero. If during the sweep phase the gc code un_discards, 1227 * we end up doing a (full) closef on the descriptor. A closef on A 1228 * results in the following chain. Closef calls soo_close, which 1229 * calls soclose. Soclose calls first (through the switch 1230 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply 1231 * returns because the previous instance had set unp_gcing, and 1232 * we return all the way back to soclose, which marks the socket 1233 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush 1234 * to free up the rights that are queued in messages on the socket A, 1235 * i.e., the reference on B. The sorflush calls via the dom_dispose 1236 * switch unp_dispose, which unp_scans with unp_discard. This second 1237 * instance of unp_discard just calls closef on B. 1238 * 1239 * Well, a similar chain occurs on B, resulting in a sorflush on B, 1240 * which results in another closef on A. Unfortunately, A is already 1241 * being closed, and the descriptor has already been marked with 1242 * SS_NOFDREF, and soclose panics at this point. 1243 * 1244 * Here, we first take an extra reference to each inaccessible 1245 * descriptor. Then, we call sorflush ourself, since we know 1246 * it is a Unix domain socket anyhow. After we destroy all the 1247 * rights carried in messages, we do a last closef to get rid 1248 * of our extra reference. This is the last close, and the 1249 * unp_detach etc will shut down the socket. 1250 * 1251 * 91/09/19, bsy@cs.cmu.edu 1252 */ 1253 extra_ref = malloc(nfiles * sizeof(struct file *), M_FILE, M_WAITOK); 1254 for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref; fp != 0; 1255 fp = nextfp) { 1256 nextfp = LIST_NEXT(fp, f_list); 1257 /* 1258 * If it's not open, skip it 1259 */ 1260 if (fp->f_count == 0) 1261 continue; 1262 /* 1263 * If all refs are from msgs, and it's not marked accessible 1264 * then it must be referenced from some unreachable cycle 1265 * of (shut-down) FDs, so include it in our 1266 * list of FDs to remove 1267 */ 1268 if (fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) { 1269 *fpp++ = fp; 1270 nunref++; 1271 fp->f_count++; 1272 } 1273 } 1274 /* 1275 * for each FD on our hit list, do the following two things 1276 */ 1277 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1278 struct file *tfp = *fpp; 1279 if (tfp->f_type == DTYPE_SOCKET && tfp->f_data != NULL) 1280 sorflush((struct socket *)(tfp->f_data)); 1281 } 1282 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) 1283 closef(*fpp, (struct proc *) NULL); 1284 free((caddr_t)extra_ref, M_FILE); 1285 unp_gcing = 0; 1286 } 1287 1288 void 1289 unp_dispose(m) 1290 struct mbuf *m; 1291 { 1292 1293 if (m) 1294 unp_scan(m, unp_discard); 1295 } 1296 1297 static int 1298 unp_listen(unp, p) 1299 struct unpcb *unp; 1300 struct proc *p; 1301 { 1302 1303 cru2x(p->p_ucred, &unp->unp_peercred); 1304 unp->unp_flags |= UNP_HAVEPCCACHED; 1305 return (0); 1306 } 1307 1308 static void 1309 unp_scan(m0, op) 1310 register struct mbuf *m0; 1311 void (*op) __P((struct file *)); 1312 { 1313 register struct mbuf *m; 1314 register struct file **rp; 1315 register struct cmsghdr *cm; 1316 register int i; 1317 int qfds; 1318 1319 while (m0) { 1320 for (m = m0; m; m = m->m_next) 1321 if (m->m_type == MT_CONTROL && 1322 m->m_len >= sizeof(*cm)) { 1323 cm = mtod(m, struct cmsghdr *); 1324 if (cm->cmsg_level != SOL_SOCKET || 1325 cm->cmsg_type != SCM_RIGHTS) 1326 continue; 1327 qfds = (cm->cmsg_len - 1328 (CMSG_DATA(cm) - (u_char *)cm)) 1329 / sizeof (struct file *); 1330 rp = (struct file **)CMSG_DATA(cm); 1331 for (i = 0; i < qfds; i++) 1332 (*op)(*rp++); 1333 break; /* XXX, but saves time */ 1334 } 1335 m0 = m0->m_act; 1336 } 1337 } 1338 1339 static void 1340 unp_mark(fp) 1341 struct file *fp; 1342 { 1343 1344 if (fp->f_flag & FMARK) 1345 return; 1346 unp_defer++; 1347 fp->f_flag |= (FMARK|FDEFER); 1348 } 1349 1350 static void 1351 unp_discard(fp) 1352 struct file *fp; 1353 { 1354 1355 fp->f_msgcount--; 1356 unp_rights--; 1357 (void) closef(fp, (struct proc *)NULL); 1358 } 1359