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 1.18 2004/12/20 11:03:16 joerg 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/proc.h> 45 #include <sys/file.h> 46 #include <sys/filedesc.h> 47 #include <sys/mbuf.h> 48 #include <sys/nlookup.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/mount.h> 55 #include <sys/sysctl.h> 56 #include <sys/un.h> 57 #include <sys/unpcb.h> 58 #include <sys/vnode.h> 59 #include <sys/file2.h> 60 61 #include <vm/vm_zone.h> 62 63 static struct vm_zone *unp_zone; 64 static unp_gen_t unp_gencnt; 65 static u_int unp_count; 66 67 static struct unp_head unp_shead, unp_dhead; 68 69 /* 70 * Unix communications domain. 71 * 72 * TODO: 73 * SEQPACKET, RDM 74 * rethink name space problems 75 * need a proper out-of-band 76 * lock pushdown 77 */ 78 static struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL }; 79 static ino_t unp_ino; /* prototype for fake inode numbers */ 80 81 static int unp_attach (struct socket *, struct pru_attach_info *); 82 static void unp_detach (struct unpcb *); 83 static int unp_bind (struct unpcb *,struct sockaddr *, struct thread *); 84 static int unp_connect (struct socket *,struct sockaddr *, 85 struct thread *); 86 static void unp_disconnect (struct unpcb *); 87 static void unp_shutdown (struct unpcb *); 88 static void unp_drop (struct unpcb *, int); 89 static void unp_gc (void); 90 static void unp_scan (struct mbuf *, void (*)(struct file *)); 91 static void unp_mark (struct file *); 92 static void unp_discard (struct file *); 93 static int unp_internalize (struct mbuf *, struct thread *); 94 static int unp_listen (struct unpcb *, struct thread *); 95 96 static int 97 uipc_abort(struct socket *so) 98 { 99 struct unpcb *unp = sotounpcb(so); 100 101 if (unp == 0) 102 return EINVAL; 103 unp_drop(unp, ECONNABORTED); 104 unp_detach(unp); 105 sofree(so); 106 return 0; 107 } 108 109 static int 110 uipc_accept(struct socket *so, struct sockaddr **nam) 111 { 112 struct unpcb *unp = sotounpcb(so); 113 114 if (unp == 0) 115 return EINVAL; 116 117 /* 118 * Pass back name of connected socket, 119 * if it was bound and we are still connected 120 * (our peer may have closed already!). 121 */ 122 if (unp->unp_conn && unp->unp_conn->unp_addr) { 123 *nam = dup_sockaddr((struct sockaddr *)unp->unp_conn->unp_addr); 124 } else { 125 *nam = dup_sockaddr((struct sockaddr *)&sun_noname); 126 } 127 return 0; 128 } 129 130 static int 131 uipc_attach(struct socket *so, int proto, struct pru_attach_info *ai) 132 { 133 struct unpcb *unp = sotounpcb(so); 134 135 if (unp != 0) 136 return EISCONN; 137 return unp_attach(so, ai); 138 } 139 140 static int 141 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 142 { 143 struct unpcb *unp = sotounpcb(so); 144 145 if (unp == 0) 146 return EINVAL; 147 return unp_bind(unp, nam, td); 148 } 149 150 static int 151 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 152 { 153 struct unpcb *unp = sotounpcb(so); 154 155 if (unp == 0) 156 return EINVAL; 157 return unp_connect(so, nam, td); 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 thread *td) 198 { 199 struct unpcb *unp = sotounpcb(so); 200 201 if (unp == 0 || unp->unp_vnode == 0) 202 return EINVAL; 203 return unp_listen(unp, td); 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 else { 216 /* 217 * XXX: It seems that this test always fails even when 218 * connection is established. So, this else clause is 219 * added as workaround to return PF_LOCAL sockaddr. 220 */ 221 *nam = dup_sockaddr((struct sockaddr *)&sun_noname); 222 } 223 return 0; 224 } 225 226 static int 227 uipc_rcvd(struct socket *so, int flags) 228 { 229 struct unpcb *unp = sotounpcb(so); 230 struct socket *so2; 231 u_long newhiwat; 232 233 if (unp == 0) 234 return EINVAL; 235 switch (so->so_type) { 236 case SOCK_DGRAM: 237 panic("uipc_rcvd DGRAM?"); 238 /*NOTREACHED*/ 239 240 case SOCK_STREAM: 241 if (unp->unp_conn == 0) 242 break; 243 so2 = unp->unp_conn->unp_socket; 244 /* 245 * Adjust backpressure on sender 246 * and wakeup any waiting to write. 247 */ 248 so2->so_snd.sb_mbmax += unp->unp_mbcnt - so->so_rcv.sb_mbcnt; 249 unp->unp_mbcnt = so->so_rcv.sb_mbcnt; 250 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - 251 so->so_rcv.sb_cc; 252 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat, 253 newhiwat, RLIM_INFINITY); 254 unp->unp_cc = so->so_rcv.sb_cc; 255 sowwakeup(so2); 256 break; 257 258 default: 259 panic("uipc_rcvd unknown socktype"); 260 } 261 return 0; 262 } 263 264 /* pru_rcvoob is EOPNOTSUPP */ 265 266 static int 267 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 268 struct mbuf *control, struct thread *td) 269 { 270 int error = 0; 271 struct unpcb *unp = sotounpcb(so); 272 struct socket *so2; 273 u_long newhiwat; 274 275 if (unp == 0) { 276 error = EINVAL; 277 goto release; 278 } 279 if (flags & PRUS_OOB) { 280 error = EOPNOTSUPP; 281 goto release; 282 } 283 284 if (control && (error = unp_internalize(control, td))) 285 goto release; 286 287 switch (so->so_type) { 288 case SOCK_DGRAM: 289 { 290 struct sockaddr *from; 291 292 if (nam) { 293 if (unp->unp_conn) { 294 error = EISCONN; 295 break; 296 } 297 error = unp_connect(so, nam, td); 298 if (error) 299 break; 300 } else { 301 if (unp->unp_conn == 0) { 302 error = ENOTCONN; 303 break; 304 } 305 } 306 so2 = unp->unp_conn->unp_socket; 307 if (unp->unp_addr) 308 from = (struct sockaddr *)unp->unp_addr; 309 else 310 from = &sun_noname; 311 if (sbappendaddr(&so2->so_rcv, from, m, control)) { 312 sorwakeup(so2); 313 m = 0; 314 control = 0; 315 } else 316 error = ENOBUFS; 317 if (nam) 318 unp_disconnect(unp); 319 break; 320 } 321 322 case SOCK_STREAM: 323 /* Connect if not connected yet. */ 324 /* 325 * Note: A better implementation would complain 326 * if not equal to the peer's address. 327 */ 328 if ((so->so_state & SS_ISCONNECTED) == 0) { 329 if (nam) { 330 error = unp_connect(so, nam, td); 331 if (error) 332 break; /* XXX */ 333 } else { 334 error = ENOTCONN; 335 break; 336 } 337 } 338 339 if (so->so_state & SS_CANTSENDMORE) { 340 error = EPIPE; 341 break; 342 } 343 if (unp->unp_conn == 0) 344 panic("uipc_send connected but no connection?"); 345 so2 = unp->unp_conn->unp_socket; 346 /* 347 * Send to paired receive port, and then reduce 348 * send buffer hiwater marks to maintain backpressure. 349 * Wake up readers. 350 */ 351 if (control) { 352 if (sbappendcontrol(&so2->so_rcv, m, control)) 353 control = 0; 354 } else 355 sbappend(&so2->so_rcv, m); 356 so->so_snd.sb_mbmax -= 357 so2->so_rcv.sb_mbcnt - unp->unp_conn->unp_mbcnt; 358 unp->unp_conn->unp_mbcnt = so2->so_rcv.sb_mbcnt; 359 newhiwat = so->so_snd.sb_hiwat - 360 (so2->so_rcv.sb_cc - unp->unp_conn->unp_cc); 361 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat, 362 newhiwat, RLIM_INFINITY); 363 unp->unp_conn->unp_cc = so2->so_rcv.sb_cc; 364 sorwakeup(so2); 365 m = 0; 366 break; 367 368 default: 369 panic("uipc_send unknown socktype"); 370 } 371 372 /* 373 * SEND_EOF is equivalent to a SEND followed by 374 * a SHUTDOWN. 375 */ 376 if (flags & PRUS_EOF) { 377 socantsendmore(so); 378 unp_shutdown(unp); 379 } 380 381 if (control && error != 0) 382 unp_dispose(control); 383 384 release: 385 if (control) 386 m_freem(control); 387 if (m) 388 m_freem(m); 389 return error; 390 } 391 392 static int 393 uipc_sense(struct socket *so, struct stat *sb) 394 { 395 struct unpcb *unp = sotounpcb(so); 396 struct socket *so2; 397 398 if (unp == 0) 399 return EINVAL; 400 sb->st_blksize = so->so_snd.sb_hiwat; 401 if (so->so_type == SOCK_STREAM && unp->unp_conn != 0) { 402 so2 = unp->unp_conn->unp_socket; 403 sb->st_blksize += so2->so_rcv.sb_cc; 404 } 405 sb->st_dev = NOUDEV; 406 if (unp->unp_ino == 0) /* make up a non-zero inode number */ 407 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino; 408 sb->st_ino = unp->unp_ino; 409 return (0); 410 } 411 412 static int 413 uipc_shutdown(struct socket *so) 414 { 415 struct unpcb *unp = sotounpcb(so); 416 417 if (unp == 0) 418 return EINVAL; 419 socantsendmore(so); 420 unp_shutdown(unp); 421 return 0; 422 } 423 424 static int 425 uipc_sockaddr(struct socket *so, struct sockaddr **nam) 426 { 427 struct unpcb *unp = sotounpcb(so); 428 429 if (unp == 0) 430 return EINVAL; 431 if (unp->unp_addr) 432 *nam = dup_sockaddr((struct sockaddr *)unp->unp_addr); 433 return 0; 434 } 435 436 struct pr_usrreqs uipc_usrreqs = { 437 uipc_abort, uipc_accept, uipc_attach, uipc_bind, uipc_connect, 438 uipc_connect2, pru_control_notsupp, uipc_detach, uipc_disconnect, 439 uipc_listen, uipc_peeraddr, uipc_rcvd, pru_rcvoob_notsupp, 440 uipc_send, uipc_sense, uipc_shutdown, uipc_sockaddr, 441 sosend, soreceive, sopoll 442 }; 443 444 int 445 uipc_ctloutput(so, sopt) 446 struct socket *so; 447 struct sockopt *sopt; 448 { 449 struct unpcb *unp = sotounpcb(so); 450 int error; 451 452 switch (sopt->sopt_dir) { 453 case SOPT_GET: 454 switch (sopt->sopt_name) { 455 case LOCAL_PEERCRED: 456 if (unp->unp_flags & UNP_HAVEPC) 457 error = sooptcopyout(sopt, &unp->unp_peercred, 458 sizeof(unp->unp_peercred)); 459 else { 460 if (so->so_type == SOCK_STREAM) 461 error = ENOTCONN; 462 else 463 error = EINVAL; 464 } 465 break; 466 default: 467 error = EOPNOTSUPP; 468 break; 469 } 470 break; 471 case SOPT_SET: 472 default: 473 error = EOPNOTSUPP; 474 break; 475 } 476 return (error); 477 } 478 479 /* 480 * Both send and receive buffers are allocated PIPSIZ bytes of buffering 481 * for stream sockets, although the total for sender and receiver is 482 * actually only PIPSIZ. 483 * Datagram sockets really use the sendspace as the maximum datagram size, 484 * and don't really want to reserve the sendspace. Their recvspace should 485 * be large enough for at least one max-size datagram plus address. 486 */ 487 #ifndef PIPSIZ 488 #define PIPSIZ 8192 489 #endif 490 static u_long unpst_sendspace = PIPSIZ; 491 static u_long unpst_recvspace = PIPSIZ; 492 static u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 493 static u_long unpdg_recvspace = 4*1024; 494 495 static int unp_rights; /* file descriptors in flight */ 496 497 SYSCTL_DECL(_net_local_stream); 498 SYSCTL_INT(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 499 &unpst_sendspace, 0, ""); 500 SYSCTL_INT(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 501 &unpst_recvspace, 0, ""); 502 SYSCTL_DECL(_net_local_dgram); 503 SYSCTL_INT(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 504 &unpdg_sendspace, 0, ""); 505 SYSCTL_INT(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 506 &unpdg_recvspace, 0, ""); 507 SYSCTL_DECL(_net_local); 508 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, ""); 509 510 static int 511 unp_attach(struct socket *so, struct pru_attach_info *ai) 512 { 513 struct unpcb *unp; 514 int error; 515 516 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 517 switch (so->so_type) { 518 519 case SOCK_STREAM: 520 error = soreserve(so, unpst_sendspace, unpst_recvspace, 521 ai->sb_rlimit); 522 break; 523 524 case SOCK_DGRAM: 525 error = soreserve(so, unpdg_sendspace, unpdg_recvspace, 526 ai->sb_rlimit); 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 = ai->fd_rdir; /* jail cruft XXX JH */ 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 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(struct unpcb *unp, struct sockaddr *nam, struct thread *td) 586 { 587 struct proc *p = td->td_proc; 588 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 589 struct vnode *vp; 590 struct vattr vattr; 591 int error, namelen; 592 struct nlookupdata nd; 593 char buf[SOCK_MAXADDRLEN]; 594 595 if (unp->unp_vnode != NULL) 596 return (EINVAL); 597 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 598 if (namelen <= 0) 599 return (EINVAL); 600 strncpy(buf, soun->sun_path, namelen); 601 buf[namelen] = 0; /* null-terminate the string */ 602 error = nlookup_init(&nd, buf, UIO_SYSSPACE, NLC_LOCKVP|NLC_CREATE); 603 if (error == 0) 604 error = nlookup(&nd); 605 if (error == 0 && nd.nl_ncp->nc_vp != NULL) 606 error = EADDRINUSE; 607 if (error) 608 goto done; 609 610 VATTR_NULL(&vattr); 611 vattr.va_type = VSOCK; 612 vattr.va_mode = (ACCESSPERMS & ~p->p_fd->fd_cmask); 613 error = VOP_NCREATE(nd.nl_ncp, &vp, nd.nl_cred, &vattr); 614 if (error == 0) { 615 vp->v_socket = unp->unp_socket; 616 unp->unp_vnode = vp; 617 unp->unp_addr = (struct sockaddr_un *)dup_sockaddr(nam); 618 VOP_UNLOCK(vp, 0, td); 619 } 620 done: 621 nlookup_done(&nd); 622 return (error); 623 } 624 625 static int 626 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 627 { 628 struct proc *p = td->td_proc; 629 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 630 struct vnode *vp; 631 struct socket *so2, *so3; 632 struct unpcb *unp, *unp2, *unp3; 633 int error, len; 634 struct nlookupdata nd; 635 char buf[SOCK_MAXADDRLEN]; 636 637 KKASSERT(p); 638 639 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 640 if (len <= 0) 641 return EINVAL; 642 strncpy(buf, soun->sun_path, len); 643 buf[len] = 0; 644 645 vp = NULL; 646 error = nlookup_init(&nd, buf, UIO_SYSSPACE, NLC_FOLLOW); 647 if (error == 0) 648 error = nlookup(&nd); 649 if (error == 0) 650 error = cache_vget(nd.nl_ncp, nd.nl_cred, LK_EXCLUSIVE, &vp); 651 nlookup_done(&nd); 652 if (error) 653 return (error); 654 655 if (vp->v_type != VSOCK) { 656 error = ENOTSOCK; 657 goto bad; 658 } 659 error = VOP_ACCESS(vp, VWRITE, p->p_ucred, td); 660 if (error) 661 goto bad; 662 so2 = vp->v_socket; 663 if (so2 == 0) { 664 error = ECONNREFUSED; 665 goto bad; 666 } 667 if (so->so_type != so2->so_type) { 668 error = EPROTOTYPE; 669 goto bad; 670 } 671 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 672 if ((so2->so_options & SO_ACCEPTCONN) == 0 || 673 (so3 = sonewconn(so2, 0)) == 0) { 674 error = ECONNREFUSED; 675 goto bad; 676 } 677 unp = sotounpcb(so); 678 unp2 = sotounpcb(so2); 679 unp3 = sotounpcb(so3); 680 if (unp2->unp_addr) 681 unp3->unp_addr = (struct sockaddr_un *) 682 dup_sockaddr((struct sockaddr *)unp2->unp_addr); 683 684 /* 685 * unp_peercred management: 686 * 687 * The connecter's (client's) credentials are copied 688 * from its process structure at the time of connect() 689 * (which is now). 690 */ 691 cru2x(p->p_ucred, &unp3->unp_peercred); 692 unp3->unp_flags |= UNP_HAVEPC; 693 /* 694 * The receiver's (server's) credentials are copied 695 * from the unp_peercred member of socket on which the 696 * former called listen(); unp_listen() cached that 697 * process's credentials at that time so we can use 698 * them now. 699 */ 700 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED, 701 ("unp_connect: listener without cached peercred")); 702 memcpy(&unp->unp_peercred, &unp2->unp_peercred, 703 sizeof(unp->unp_peercred)); 704 unp->unp_flags |= UNP_HAVEPC; 705 706 so2 = so3; 707 } 708 error = unp_connect2(so, so2); 709 bad: 710 vput(vp); 711 return (error); 712 } 713 714 int 715 unp_connect2(so, so2) 716 struct socket *so; 717 struct socket *so2; 718 { 719 struct unpcb *unp = sotounpcb(so); 720 struct unpcb *unp2; 721 722 if (so2->so_type != so->so_type) 723 return (EPROTOTYPE); 724 unp2 = sotounpcb(so2); 725 unp->unp_conn = unp2; 726 switch (so->so_type) { 727 728 case SOCK_DGRAM: 729 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 730 soisconnected(so); 731 break; 732 733 case SOCK_STREAM: 734 unp2->unp_conn = unp; 735 soisconnected(so); 736 soisconnected(so2); 737 break; 738 739 default: 740 panic("unp_connect2"); 741 } 742 return (0); 743 } 744 745 static void 746 unp_disconnect(unp) 747 struct unpcb *unp; 748 { 749 struct unpcb *unp2 = unp->unp_conn; 750 751 if (unp2 == 0) 752 return; 753 unp->unp_conn = 0; 754 switch (unp->unp_socket->so_type) { 755 756 case SOCK_DGRAM: 757 LIST_REMOVE(unp, unp_reflink); 758 unp->unp_socket->so_state &= ~SS_ISCONNECTED; 759 break; 760 761 case SOCK_STREAM: 762 soisdisconnected(unp->unp_socket); 763 unp2->unp_conn = 0; 764 soisdisconnected(unp2->unp_socket); 765 break; 766 } 767 } 768 769 #ifdef notdef 770 void 771 unp_abort(unp) 772 struct unpcb *unp; 773 { 774 775 unp_detach(unp); 776 } 777 #endif 778 779 static int 780 prison_unpcb(struct thread *td, struct unpcb *unp) 781 { 782 struct proc *p; 783 784 if (td == NULL) 785 return (0); 786 if ((p = td->td_proc) == NULL) 787 return (0); 788 if (!p->p_ucred->cr_prison) 789 return (0); 790 if (p->p_fd->fd_rdir == unp->unp_rvnode) 791 return (0); 792 return (1); 793 } 794 795 static int 796 unp_pcblist(SYSCTL_HANDLER_ARGS) 797 { 798 int error, i, n; 799 struct unpcb *unp, **unp_list; 800 unp_gen_t gencnt; 801 struct unp_head *head; 802 803 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead); 804 805 KKASSERT(curproc != NULL); 806 807 /* 808 * The process of preparing the PCB list is too time-consuming and 809 * resource-intensive to repeat twice on every request. 810 */ 811 if (req->oldptr == 0) { 812 n = unp_count; 813 req->oldidx = (n + n/8) * sizeof(struct xunpcb); 814 return 0; 815 } 816 817 if (req->newptr != 0) 818 return EPERM; 819 820 /* 821 * OK, now we're committed to doing something. 822 */ 823 gencnt = unp_gencnt; 824 n = unp_count; 825 826 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 827 if (unp_list == 0) 828 return ENOMEM; 829 830 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 831 unp = LIST_NEXT(unp, unp_link)) { 832 if (unp->unp_gencnt <= gencnt && !prison_unpcb(req->td, unp)) 833 unp_list[i++] = unp; 834 } 835 n = i; /* in case we lost some during malloc */ 836 837 error = 0; 838 for (i = 0; i < n; i++) { 839 unp = unp_list[i]; 840 if (unp->unp_gencnt <= gencnt) { 841 struct xunpcb xu; 842 xu.xu_len = sizeof xu; 843 xu.xu_unpp = unp; 844 /* 845 * XXX - need more locking here to protect against 846 * connect/disconnect races for SMP. 847 */ 848 if (unp->unp_addr) 849 bcopy(unp->unp_addr, &xu.xu_addr, 850 unp->unp_addr->sun_len); 851 if (unp->unp_conn && unp->unp_conn->unp_addr) 852 bcopy(unp->unp_conn->unp_addr, 853 &xu.xu_caddr, 854 unp->unp_conn->unp_addr->sun_len); 855 bcopy(unp, &xu.xu_unp, sizeof *unp); 856 sotoxsocket(unp->unp_socket, &xu.xu_socket); 857 error = SYSCTL_OUT(req, &xu, sizeof xu); 858 } 859 } 860 free(unp_list, M_TEMP); 861 return error; 862 } 863 864 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD, 865 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 866 "List of active local datagram sockets"); 867 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD, 868 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 869 "List of active local stream sockets"); 870 871 static void 872 unp_shutdown(unp) 873 struct unpcb *unp; 874 { 875 struct socket *so; 876 877 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn && 878 (so = unp->unp_conn->unp_socket)) 879 socantrcvmore(so); 880 } 881 882 static void 883 unp_drop(unp, errno) 884 struct unpcb *unp; 885 int errno; 886 { 887 struct socket *so = unp->unp_socket; 888 889 so->so_error = errno; 890 unp_disconnect(unp); 891 } 892 893 #ifdef notdef 894 void 895 unp_drain() 896 { 897 898 } 899 #endif 900 901 int 902 unp_externalize(struct mbuf *rights) 903 { 904 struct proc *p = curproc; /* XXX */ 905 int i; 906 struct cmsghdr *cm = mtod(rights, struct cmsghdr *); 907 int *fdp; 908 struct file **rp; 909 struct file *fp; 910 int newfds = (cm->cmsg_len - (CMSG_DATA(cm) - (u_char *)cm)) 911 / sizeof (struct file *); 912 int f; 913 914 /* 915 * if the new FD's will not fit, then we free them all 916 */ 917 if (!fdavail(p, newfds)) { 918 rp = (struct file **)CMSG_DATA(cm); 919 for (i = 0; i < newfds; i++) { 920 fp = *rp; 921 /* 922 * zero the pointer before calling unp_discard, 923 * since it may end up in unp_gc().. 924 */ 925 *rp++ = 0; 926 unp_discard(fp); 927 } 928 return (EMSGSIZE); 929 } 930 /* 931 * now change each pointer to an fd in the global table to 932 * an integer that is the index to the local fd table entry 933 * that we set up to point to the global one we are transferring. 934 * If sizeof (struct file *) is bigger than or equal to sizeof int, 935 * then do it in forward order. In that case, an integer will 936 * always come in the same place or before its corresponding 937 * struct file pointer. 938 * If sizeof (struct file *) is smaller than sizeof int, then 939 * do it in reverse order. 940 */ 941 if (sizeof (struct file *) >= sizeof (int)) { 942 fdp = (int *)(cm + 1); 943 rp = (struct file **)CMSG_DATA(cm); 944 for (i = 0; i < newfds; i++) { 945 if (fdalloc(p, 0, &f)) 946 panic("unp_externalize"); 947 fp = *rp++; 948 p->p_fd->fd_ofiles[f] = fp; 949 fp->f_msgcount--; 950 unp_rights--; 951 *fdp++ = f; 952 } 953 } else { 954 fdp = (int *)(cm + 1) + newfds - 1; 955 rp = (struct file **)CMSG_DATA(cm) + newfds - 1; 956 for (i = 0; i < newfds; i++) { 957 if (fdalloc(p, 0, &f)) 958 panic("unp_externalize"); 959 fp = *rp--; 960 p->p_fd->fd_ofiles[f] = fp; 961 fp->f_msgcount--; 962 unp_rights--; 963 *fdp-- = f; 964 } 965 } 966 967 /* 968 * Adjust length, in case sizeof(struct file *) and sizeof(int) 969 * differs. 970 */ 971 cm->cmsg_len = CMSG_LEN(newfds * sizeof(int)); 972 rights->m_len = cm->cmsg_len; 973 return (0); 974 } 975 976 void 977 unp_init(void) 978 { 979 unp_zone = zinit("unpcb", sizeof(struct unpcb), nmbclusters, 0, 0); 980 if (unp_zone == 0) 981 panic("unp_init"); 982 LIST_INIT(&unp_dhead); 983 LIST_INIT(&unp_shead); 984 } 985 986 static int 987 unp_internalize(struct mbuf *control, struct thread *td) 988 { 989 struct proc *p = td->td_proc; 990 struct filedesc *fdescp; 991 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 992 struct file **rp; 993 struct file *fp; 994 int i, fd, *fdp; 995 struct cmsgcred *cmcred; 996 int oldfds; 997 u_int newlen; 998 999 KKASSERT(p); 1000 fdescp = p->p_fd; 1001 if ((cm->cmsg_type != SCM_RIGHTS && cm->cmsg_type != SCM_CREDS) || 1002 cm->cmsg_level != SOL_SOCKET || cm->cmsg_len != control->m_len) 1003 return (EINVAL); 1004 1005 /* 1006 * Fill in credential information. 1007 */ 1008 if (cm->cmsg_type == SCM_CREDS) { 1009 cmcred = (struct cmsgcred *)(cm + 1); 1010 cmcred->cmcred_pid = p->p_pid; 1011 cmcred->cmcred_uid = p->p_ucred->cr_ruid; 1012 cmcred->cmcred_gid = p->p_ucred->cr_rgid; 1013 cmcred->cmcred_euid = p->p_ucred->cr_uid; 1014 cmcred->cmcred_ngroups = MIN(p->p_ucred->cr_ngroups, 1015 CMGROUP_MAX); 1016 for (i = 0; i < cmcred->cmcred_ngroups; i++) 1017 cmcred->cmcred_groups[i] = p->p_ucred->cr_groups[i]; 1018 return(0); 1019 } 1020 1021 oldfds = (cm->cmsg_len - sizeof (*cm)) / sizeof (int); 1022 /* 1023 * check that all the FDs passed in refer to legal OPEN files 1024 * If not, reject the entire operation. 1025 */ 1026 fdp = (int *)(cm + 1); 1027 for (i = 0; i < oldfds; i++) { 1028 fd = *fdp++; 1029 if ((unsigned)fd >= fdescp->fd_nfiles || 1030 fdescp->fd_ofiles[fd] == NULL) 1031 return (EBADF); 1032 if (fdescp->fd_ofiles[fd]->f_type == DTYPE_KQUEUE) 1033 return (EOPNOTSUPP); 1034 } 1035 /* 1036 * Now replace the integer FDs with pointers to 1037 * the associated global file table entry.. 1038 * Allocate a bigger buffer as necessary. But if an cluster is not 1039 * enough, return E2BIG. 1040 */ 1041 newlen = CMSG_LEN(oldfds * sizeof(struct file *)); 1042 if (newlen > MCLBYTES) 1043 return (E2BIG); 1044 if (newlen - control->m_len > M_TRAILINGSPACE(control)) { 1045 if (control->m_flags & M_EXT) 1046 return (E2BIG); 1047 MCLGET(control, MB_WAIT); 1048 if ((control->m_flags & M_EXT) == 0) 1049 return (ENOBUFS); 1050 1051 /* copy the data to the cluster */ 1052 memcpy(mtod(control, char *), cm, cm->cmsg_len); 1053 cm = mtod(control, struct cmsghdr *); 1054 } 1055 1056 /* 1057 * Adjust length, in case sizeof(struct file *) and sizeof(int) 1058 * differs. 1059 */ 1060 control->m_len = cm->cmsg_len = newlen; 1061 1062 /* 1063 * Transform the file descriptors into struct file pointers. 1064 * If sizeof (struct file *) is bigger than or equal to sizeof int, 1065 * then do it in reverse order so that the int won't get until 1066 * we're done. 1067 * If sizeof (struct file *) is smaller than sizeof int, then 1068 * do it in forward order. 1069 */ 1070 if (sizeof (struct file *) >= sizeof (int)) { 1071 fdp = (int *)(cm + 1) + oldfds - 1; 1072 rp = (struct file **)CMSG_DATA(cm) + oldfds - 1; 1073 for (i = 0; i < oldfds; i++) { 1074 fp = fdescp->fd_ofiles[*fdp--]; 1075 *rp-- = fp; 1076 fp->f_count++; 1077 fp->f_msgcount++; 1078 unp_rights++; 1079 } 1080 } else { 1081 fdp = (int *)(cm + 1); 1082 rp = (struct file **)CMSG_DATA(cm); 1083 for (i = 0; i < oldfds; i++) { 1084 fp = fdescp->fd_ofiles[*fdp++]; 1085 *rp++ = fp; 1086 fp->f_count++; 1087 fp->f_msgcount++; 1088 unp_rights++; 1089 } 1090 } 1091 return (0); 1092 } 1093 1094 static int unp_defer, unp_gcing; 1095 1096 static void 1097 unp_gc() 1098 { 1099 struct file *fp, *nextfp; 1100 struct socket *so; 1101 struct file **extra_ref, **fpp; 1102 int nunref, i; 1103 1104 if (unp_gcing) 1105 return; 1106 unp_gcing = 1; 1107 unp_defer = 0; 1108 /* 1109 * before going through all this, set all FDs to 1110 * be NOT defered and NOT externally accessible 1111 */ 1112 LIST_FOREACH(fp, &filehead, f_list) 1113 fp->f_flag &= ~(FMARK|FDEFER); 1114 do { 1115 LIST_FOREACH(fp, &filehead, f_list) { 1116 /* 1117 * If the file is not open, skip it 1118 */ 1119 if (fp->f_count == 0) 1120 continue; 1121 /* 1122 * If we already marked it as 'defer' in a 1123 * previous pass, then try process it this time 1124 * and un-mark it 1125 */ 1126 if (fp->f_flag & FDEFER) { 1127 fp->f_flag &= ~FDEFER; 1128 unp_defer--; 1129 } else { 1130 /* 1131 * if it's not defered, then check if it's 1132 * already marked.. if so skip it 1133 */ 1134 if (fp->f_flag & FMARK) 1135 continue; 1136 /* 1137 * If all references are from messages 1138 * in transit, then skip it. it's not 1139 * externally accessible. 1140 */ 1141 if (fp->f_count == fp->f_msgcount) 1142 continue; 1143 /* 1144 * If it got this far then it must be 1145 * externally accessible. 1146 */ 1147 fp->f_flag |= FMARK; 1148 } 1149 /* 1150 * either it was defered, or it is externally 1151 * accessible and not already marked so. 1152 * Now check if it is possibly one of OUR sockets. 1153 */ 1154 if (fp->f_type != DTYPE_SOCKET || 1155 (so = (struct socket *)fp->f_data) == 0) 1156 continue; 1157 if (so->so_proto->pr_domain != &localdomain || 1158 (so->so_proto->pr_flags&PR_RIGHTS) == 0) 1159 continue; 1160 #ifdef notdef 1161 if (so->so_rcv.sb_flags & SB_LOCK) { 1162 /* 1163 * This is problematical; it's not clear 1164 * we need to wait for the sockbuf to be 1165 * unlocked (on a uniprocessor, at least), 1166 * and it's also not clear what to do 1167 * if sbwait returns an error due to receipt 1168 * of a signal. If sbwait does return 1169 * an error, we'll go into an infinite 1170 * loop. Delete all of this for now. 1171 */ 1172 (void) sbwait(&so->so_rcv); 1173 goto restart; 1174 } 1175 #endif 1176 /* 1177 * So, Ok, it's one of our sockets and it IS externally 1178 * accessible (or was defered). Now we look 1179 * to see if we hold any file descriptors in its 1180 * message buffers. Follow those links and mark them 1181 * as accessible too. 1182 */ 1183 unp_scan(so->so_rcv.sb_mb, unp_mark); 1184 } 1185 } while (unp_defer); 1186 /* 1187 * We grab an extra reference to each of the file table entries 1188 * that are not otherwise accessible and then free the rights 1189 * that are stored in messages on them. 1190 * 1191 * The bug in the orginal code is a little tricky, so I'll describe 1192 * what's wrong with it here. 1193 * 1194 * It is incorrect to simply unp_discard each entry for f_msgcount 1195 * times -- consider the case of sockets A and B that contain 1196 * references to each other. On a last close of some other socket, 1197 * we trigger a gc since the number of outstanding rights (unp_rights) 1198 * is non-zero. If during the sweep phase the gc code un_discards, 1199 * we end up doing a (full) closef on the descriptor. A closef on A 1200 * results in the following chain. Closef calls soo_close, which 1201 * calls soclose. Soclose calls first (through the switch 1202 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply 1203 * returns because the previous instance had set unp_gcing, and 1204 * we return all the way back to soclose, which marks the socket 1205 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush 1206 * to free up the rights that are queued in messages on the socket A, 1207 * i.e., the reference on B. The sorflush calls via the dom_dispose 1208 * switch unp_dispose, which unp_scans with unp_discard. This second 1209 * instance of unp_discard just calls closef on B. 1210 * 1211 * Well, a similar chain occurs on B, resulting in a sorflush on B, 1212 * which results in another closef on A. Unfortunately, A is already 1213 * being closed, and the descriptor has already been marked with 1214 * SS_NOFDREF, and soclose panics at this point. 1215 * 1216 * Here, we first take an extra reference to each inaccessible 1217 * descriptor. Then, we call sorflush ourself, since we know 1218 * it is a Unix domain socket anyhow. After we destroy all the 1219 * rights carried in messages, we do a last closef to get rid 1220 * of our extra reference. This is the last close, and the 1221 * unp_detach etc will shut down the socket. 1222 * 1223 * 91/09/19, bsy@cs.cmu.edu 1224 */ 1225 extra_ref = malloc(nfiles * sizeof(struct file *), M_FILE, M_WAITOK); 1226 for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref; fp != 0; 1227 fp = nextfp) { 1228 nextfp = LIST_NEXT(fp, f_list); 1229 /* 1230 * If it's not open, skip it 1231 */ 1232 if (fp->f_count == 0) 1233 continue; 1234 /* 1235 * If all refs are from msgs, and it's not marked accessible 1236 * then it must be referenced from some unreachable cycle 1237 * of (shut-down) FDs, so include it in our 1238 * list of FDs to remove 1239 */ 1240 if (fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) { 1241 *fpp++ = fp; 1242 nunref++; 1243 fp->f_count++; 1244 } 1245 } 1246 /* 1247 * for each FD on our hit list, do the following two things 1248 */ 1249 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1250 struct file *tfp = *fpp; 1251 if (tfp->f_type == DTYPE_SOCKET && tfp->f_data != NULL) 1252 sorflush((struct socket *)(tfp->f_data)); 1253 } 1254 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) 1255 closef(*fpp, NULL); 1256 free((caddr_t)extra_ref, M_FILE); 1257 unp_gcing = 0; 1258 } 1259 1260 void 1261 unp_dispose(struct mbuf *m) 1262 { 1263 if (m) 1264 unp_scan(m, unp_discard); 1265 } 1266 1267 static int 1268 unp_listen(struct unpcb *unp, struct thread *td) 1269 { 1270 struct proc *p = td->td_proc; 1271 1272 KKASSERT(p); 1273 cru2x(p->p_ucred, &unp->unp_peercred); 1274 unp->unp_flags |= UNP_HAVEPCCACHED; 1275 return (0); 1276 } 1277 1278 static void 1279 unp_scan(m0, op) 1280 struct mbuf *m0; 1281 void (*op) (struct file *); 1282 { 1283 struct mbuf *m; 1284 struct file **rp; 1285 struct cmsghdr *cm; 1286 int i; 1287 int qfds; 1288 1289 while (m0) { 1290 for (m = m0; m; m = m->m_next) 1291 if (m->m_type == MT_CONTROL && 1292 m->m_len >= sizeof(*cm)) { 1293 cm = mtod(m, struct cmsghdr *); 1294 if (cm->cmsg_level != SOL_SOCKET || 1295 cm->cmsg_type != SCM_RIGHTS) 1296 continue; 1297 qfds = (cm->cmsg_len - 1298 (CMSG_DATA(cm) - (u_char *)cm)) 1299 / sizeof (struct file *); 1300 rp = (struct file **)CMSG_DATA(cm); 1301 for (i = 0; i < qfds; i++) 1302 (*op)(*rp++); 1303 break; /* XXX, but saves time */ 1304 } 1305 m0 = m0->m_nextpkt; 1306 } 1307 } 1308 1309 static void 1310 unp_mark(fp) 1311 struct file *fp; 1312 { 1313 1314 if (fp->f_flag & FMARK) 1315 return; 1316 unp_defer++; 1317 fp->f_flag |= (FMARK|FDEFER); 1318 } 1319 1320 static void 1321 unp_discard(fp) 1322 struct file *fp; 1323 { 1324 1325 fp->f_msgcount--; 1326 unp_rights--; 1327 (void) closef(fp, NULL); 1328 } 1329