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 1.44 2008/09/06 05:44:58 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/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 #include <sys/spinlock2.h> 61 62 63 static MALLOC_DEFINE(M_UNPCB, "unpcb", "unpcb struct"); 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 * 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 int unp_gc_clearmarks(struct file *, void *); 91 static int unp_gc_checkmarks(struct file *, void *); 92 static int unp_gc_checkrefs(struct file *, void *); 93 static int unp_revoke_gc_check(struct file *, void *); 94 static void unp_scan (struct mbuf *, void (*)(struct file *, void *), 95 void *data); 96 static void unp_mark (struct file *, void *data); 97 static void unp_discard (struct file *, void *); 98 static int unp_internalize (struct mbuf *, struct thread *); 99 static int unp_listen (struct unpcb *, struct thread *); 100 static void unp_fp_externalize(struct proc *p, struct file *fp, int fd); 101 102 static int 103 uipc_abort(struct socket *so) 104 { 105 struct unpcb *unp = so->so_pcb; 106 107 if (unp == NULL) 108 return EINVAL; 109 unp_drop(unp, ECONNABORTED); 110 unp_detach(unp); 111 sofree(so); 112 return 0; 113 } 114 115 static int 116 uipc_accept(struct socket *so, struct sockaddr **nam) 117 { 118 struct unpcb *unp = so->so_pcb; 119 120 if (unp == NULL) 121 return EINVAL; 122 123 /* 124 * Pass back name of connected socket, 125 * if it was bound and we are still connected 126 * (our peer may have closed already!). 127 */ 128 if (unp->unp_conn && unp->unp_conn->unp_addr) { 129 *nam = dup_sockaddr((struct sockaddr *)unp->unp_conn->unp_addr); 130 } else { 131 *nam = dup_sockaddr((struct sockaddr *)&sun_noname); 132 } 133 return 0; 134 } 135 136 static int 137 uipc_attach(struct socket *so, int proto, struct pru_attach_info *ai) 138 { 139 struct unpcb *unp = so->so_pcb; 140 141 if (unp != NULL) 142 return EISCONN; 143 return unp_attach(so, ai); 144 } 145 146 static int 147 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 148 { 149 struct unpcb *unp = so->so_pcb; 150 151 if (unp == NULL) 152 return EINVAL; 153 return unp_bind(unp, nam, td); 154 } 155 156 static int 157 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 158 { 159 struct unpcb *unp = so->so_pcb; 160 161 if (unp == NULL) 162 return EINVAL; 163 return unp_connect(so, nam, td); 164 } 165 166 static int 167 uipc_connect2(struct socket *so1, struct socket *so2) 168 { 169 struct unpcb *unp = so1->so_pcb; 170 171 if (unp == NULL) 172 return EINVAL; 173 174 return unp_connect2(so1, so2); 175 } 176 177 /* control is EOPNOTSUPP */ 178 179 static int 180 uipc_detach(struct socket *so) 181 { 182 struct unpcb *unp = so->so_pcb; 183 184 if (unp == NULL) 185 return EINVAL; 186 187 unp_detach(unp); 188 return 0; 189 } 190 191 static int 192 uipc_disconnect(struct socket *so) 193 { 194 struct unpcb *unp = so->so_pcb; 195 196 if (unp == NULL) 197 return EINVAL; 198 unp_disconnect(unp); 199 return 0; 200 } 201 202 static int 203 uipc_listen(struct socket *so, struct thread *td) 204 { 205 struct unpcb *unp = so->so_pcb; 206 207 if (unp == NULL || unp->unp_vnode == NULL) 208 return EINVAL; 209 return unp_listen(unp, td); 210 } 211 212 static int 213 uipc_peeraddr(struct socket *so, struct sockaddr **nam) 214 { 215 struct unpcb *unp = so->so_pcb; 216 217 if (unp == NULL) 218 return EINVAL; 219 if (unp->unp_conn && unp->unp_conn->unp_addr) 220 *nam = dup_sockaddr((struct sockaddr *)unp->unp_conn->unp_addr); 221 else { 222 /* 223 * XXX: It seems that this test always fails even when 224 * connection is established. So, this else clause is 225 * added as workaround to return PF_LOCAL sockaddr. 226 */ 227 *nam = dup_sockaddr((struct sockaddr *)&sun_noname); 228 } 229 return 0; 230 } 231 232 static int 233 uipc_rcvd(struct socket *so, int flags) 234 { 235 struct unpcb *unp = so->so_pcb; 236 struct socket *so2; 237 238 if (unp == NULL) 239 return EINVAL; 240 switch (so->so_type) { 241 case SOCK_DGRAM: 242 panic("uipc_rcvd DGRAM?"); 243 /*NOTREACHED*/ 244 245 case SOCK_STREAM: 246 case SOCK_SEQPACKET: 247 if (unp->unp_conn == NULL) 248 break; 249 /* 250 * Because we are transfering mbufs directly to the 251 * peer socket we have to use SSB_STOP on the sender 252 * to prevent it from building up infinite mbufs. 253 */ 254 so2 = unp->unp_conn->unp_socket; 255 if (so->so_rcv.ssb_cc < so2->so_snd.ssb_hiwat && 256 so->so_rcv.ssb_mbcnt < so2->so_snd.ssb_mbmax 257 ) { 258 so2->so_snd.ssb_flags &= ~SSB_STOP; 259 sowwakeup(so2); 260 } 261 break; 262 263 default: 264 panic("uipc_rcvd unknown socktype"); 265 } 266 return 0; 267 } 268 269 /* pru_rcvoob is EOPNOTSUPP */ 270 271 static int 272 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 273 struct mbuf *control, struct thread *td) 274 { 275 int error = 0; 276 struct unpcb *unp = so->so_pcb; 277 struct socket *so2; 278 279 if (unp == NULL) { 280 error = EINVAL; 281 goto release; 282 } 283 if (flags & PRUS_OOB) { 284 error = EOPNOTSUPP; 285 goto release; 286 } 287 288 if (control && (error = unp_internalize(control, td))) 289 goto release; 290 291 switch (so->so_type) { 292 case SOCK_DGRAM: 293 { 294 struct sockaddr *from; 295 296 if (nam) { 297 if (unp->unp_conn) { 298 error = EISCONN; 299 break; 300 } 301 error = unp_connect(so, nam, td); 302 if (error) 303 break; 304 } else { 305 if (unp->unp_conn == NULL) { 306 error = ENOTCONN; 307 break; 308 } 309 } 310 so2 = unp->unp_conn->unp_socket; 311 if (unp->unp_addr) 312 from = (struct sockaddr *)unp->unp_addr; 313 else 314 from = &sun_noname; 315 if (ssb_appendaddr(&so2->so_rcv, from, m, control)) { 316 sorwakeup(so2); 317 m = NULL; 318 control = NULL; 319 } else { 320 error = ENOBUFS; 321 } 322 if (nam) 323 unp_disconnect(unp); 324 break; 325 } 326 327 case SOCK_STREAM: 328 case SOCK_SEQPACKET: 329 /* Connect if not connected yet. */ 330 /* 331 * Note: A better implementation would complain 332 * if not equal to the peer's address. 333 */ 334 if (!(so->so_state & SS_ISCONNECTED)) { 335 if (nam) { 336 error = unp_connect(so, nam, td); 337 if (error) 338 break; /* XXX */ 339 } else { 340 error = ENOTCONN; 341 break; 342 } 343 } 344 345 if (so->so_state & SS_CANTSENDMORE) { 346 error = EPIPE; 347 break; 348 } 349 if (unp->unp_conn == NULL) 350 panic("uipc_send connected but no connection?"); 351 so2 = unp->unp_conn->unp_socket; 352 /* 353 * Send to paired receive port, and then reduce 354 * send buffer hiwater marks to maintain backpressure. 355 * Wake up readers. 356 */ 357 if (control) { 358 if (ssb_appendcontrol(&so2->so_rcv, m, control)) { 359 control = NULL; 360 m = NULL; 361 } 362 } else if (so->so_type == SOCK_SEQPACKET) { 363 sbappendrecord(&so2->so_rcv.sb, m); 364 m = NULL; 365 } else { 366 sbappend(&so2->so_rcv.sb, m); 367 m = NULL; 368 } 369 370 /* 371 * Because we are transfering mbufs directly to the 372 * peer socket we have to use SSB_STOP on the sender 373 * to prevent it from building up infinite mbufs. 374 */ 375 if (so2->so_rcv.ssb_cc >= so->so_snd.ssb_hiwat || 376 so2->so_rcv.ssb_mbcnt >= so->so_snd.ssb_mbmax 377 ) { 378 so->so_snd.ssb_flags |= SSB_STOP; 379 } 380 sorwakeup(so2); 381 break; 382 383 default: 384 panic("uipc_send unknown socktype"); 385 } 386 387 /* 388 * SEND_EOF is equivalent to a SEND followed by a SHUTDOWN. 389 */ 390 if (flags & PRUS_EOF) { 391 socantsendmore(so); 392 unp_shutdown(unp); 393 } 394 395 if (control && error != 0) 396 unp_dispose(control); 397 398 release: 399 if (control) 400 m_freem(control); 401 if (m) 402 m_freem(m); 403 return error; 404 } 405 406 static int 407 uipc_sense(struct socket *so, struct stat *sb) 408 { 409 struct unpcb *unp = so->so_pcb; 410 411 if (unp == NULL) 412 return EINVAL; 413 sb->st_blksize = so->so_snd.ssb_hiwat; 414 sb->st_dev = NOUDEV; 415 if (unp->unp_ino == 0) /* make up a non-zero inode number */ 416 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino; 417 sb->st_ino = unp->unp_ino; 418 return (0); 419 } 420 421 static int 422 uipc_shutdown(struct socket *so) 423 { 424 struct unpcb *unp = so->so_pcb; 425 426 if (unp == NULL) 427 return EINVAL; 428 socantsendmore(so); 429 unp_shutdown(unp); 430 return 0; 431 } 432 433 static int 434 uipc_sockaddr(struct socket *so, struct sockaddr **nam) 435 { 436 struct unpcb *unp = so->so_pcb; 437 438 if (unp == NULL) 439 return EINVAL; 440 if (unp->unp_addr) 441 *nam = dup_sockaddr((struct sockaddr *)unp->unp_addr); 442 return 0; 443 } 444 445 struct pr_usrreqs uipc_usrreqs = { 446 .pru_abort = uipc_abort, 447 .pru_accept = uipc_accept, 448 .pru_attach = uipc_attach, 449 .pru_bind = uipc_bind, 450 .pru_connect = uipc_connect, 451 .pru_connect2 = uipc_connect2, 452 .pru_control = pru_control_notsupp, 453 .pru_detach = uipc_detach, 454 .pru_disconnect = uipc_disconnect, 455 .pru_listen = uipc_listen, 456 .pru_peeraddr = uipc_peeraddr, 457 .pru_rcvd = uipc_rcvd, 458 .pru_rcvoob = pru_rcvoob_notsupp, 459 .pru_send = uipc_send, 460 .pru_sense = uipc_sense, 461 .pru_shutdown = uipc_shutdown, 462 .pru_sockaddr = uipc_sockaddr, 463 .pru_sosend = sosend, 464 .pru_soreceive = soreceive, 465 .pru_sopoll = sopoll 466 }; 467 468 int 469 uipc_ctloutput(struct socket *so, struct sockopt *sopt) 470 { 471 struct unpcb *unp = so->so_pcb; 472 int error = 0; 473 474 switch (sopt->sopt_dir) { 475 case SOPT_GET: 476 switch (sopt->sopt_name) { 477 case LOCAL_PEERCRED: 478 if (unp->unp_flags & UNP_HAVEPC) 479 soopt_from_kbuf(sopt, &unp->unp_peercred, 480 sizeof(unp->unp_peercred)); 481 else { 482 if (so->so_type == SOCK_STREAM) 483 error = ENOTCONN; 484 else if (so->so_type == SOCK_SEQPACKET) 485 error = ENOTCONN; 486 else 487 error = EINVAL; 488 } 489 break; 490 default: 491 error = EOPNOTSUPP; 492 break; 493 } 494 break; 495 case SOPT_SET: 496 default: 497 error = EOPNOTSUPP; 498 break; 499 } 500 return (error); 501 } 502 503 /* 504 * Both send and receive buffers are allocated PIPSIZ bytes of buffering 505 * for stream sockets, although the total for sender and receiver is 506 * actually only PIPSIZ. 507 * 508 * Datagram sockets really use the sendspace as the maximum datagram size, 509 * and don't really want to reserve the sendspace. Their recvspace should 510 * be large enough for at least one max-size datagram plus address. 511 * 512 * We want the local send/recv space to be significant larger then lo0's 513 * mtu of 16384. 514 */ 515 #ifndef PIPSIZ 516 #define PIPSIZ 57344 517 #endif 518 static u_long unpst_sendspace = PIPSIZ; 519 static u_long unpst_recvspace = PIPSIZ; 520 static u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 521 static u_long unpdg_recvspace = 4*1024; 522 523 static int unp_rights; /* file descriptors in flight */ 524 static struct spinlock unp_spin = SPINLOCK_INITIALIZER(&unp_spin); 525 526 SYSCTL_DECL(_net_local_seqpacket); 527 SYSCTL_DECL(_net_local_stream); 528 SYSCTL_INT(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 529 &unpst_sendspace, 0, ""); 530 SYSCTL_INT(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 531 &unpst_recvspace, 0, ""); 532 533 SYSCTL_DECL(_net_local_dgram); 534 SYSCTL_INT(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 535 &unpdg_sendspace, 0, ""); 536 SYSCTL_INT(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 537 &unpdg_recvspace, 0, ""); 538 539 SYSCTL_DECL(_net_local); 540 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, ""); 541 542 static int 543 unp_attach(struct socket *so, struct pru_attach_info *ai) 544 { 545 struct unpcb *unp; 546 int error; 547 548 if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) { 549 switch (so->so_type) { 550 551 case SOCK_STREAM: 552 case SOCK_SEQPACKET: 553 error = soreserve(so, unpst_sendspace, unpst_recvspace, 554 ai->sb_rlimit); 555 break; 556 557 case SOCK_DGRAM: 558 error = soreserve(so, unpdg_sendspace, unpdg_recvspace, 559 ai->sb_rlimit); 560 break; 561 562 default: 563 panic("unp_attach"); 564 } 565 if (error) 566 return (error); 567 } 568 unp = kmalloc(sizeof(*unp), M_UNPCB, M_NOWAIT|M_ZERO); 569 if (unp == NULL) 570 return (ENOBUFS); 571 unp->unp_gencnt = ++unp_gencnt; 572 unp_count++; 573 LIST_INIT(&unp->unp_refs); 574 unp->unp_socket = so; 575 unp->unp_rvnode = ai->fd_rdir; /* jail cruft XXX JH */ 576 LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead 577 : &unp_shead, unp, unp_link); 578 so->so_pcb = (caddr_t)unp; 579 return (0); 580 } 581 582 static void 583 unp_detach(struct unpcb *unp) 584 { 585 LIST_REMOVE(unp, unp_link); 586 unp->unp_gencnt = ++unp_gencnt; 587 --unp_count; 588 if (unp->unp_vnode) { 589 unp->unp_vnode->v_socket = NULL; 590 vrele(unp->unp_vnode); 591 unp->unp_vnode = NULL; 592 } 593 if (unp->unp_conn) 594 unp_disconnect(unp); 595 while (!LIST_EMPTY(&unp->unp_refs)) 596 unp_drop(LIST_FIRST(&unp->unp_refs), ECONNRESET); 597 soisdisconnected(unp->unp_socket); 598 unp->unp_socket->so_pcb = NULL; 599 if (unp_rights) { 600 /* 601 * Normally the receive buffer is flushed later, 602 * in sofree, but if our receive buffer holds references 603 * to descriptors that are now garbage, we will dispose 604 * of those descriptor references after the garbage collector 605 * gets them (resulting in a "panic: closef: count < 0"). 606 */ 607 sorflush(unp->unp_socket); 608 unp_gc(); 609 } 610 if (unp->unp_addr) 611 kfree(unp->unp_addr, M_SONAME); 612 kfree(unp, M_UNPCB); 613 } 614 615 static int 616 unp_bind(struct unpcb *unp, struct sockaddr *nam, struct thread *td) 617 { 618 struct proc *p = td->td_proc; 619 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 620 struct vnode *vp; 621 struct vattr vattr; 622 int error, namelen; 623 struct nlookupdata nd; 624 char buf[SOCK_MAXADDRLEN]; 625 626 if (unp->unp_vnode != NULL) 627 return (EINVAL); 628 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 629 if (namelen <= 0) 630 return (EINVAL); 631 strncpy(buf, soun->sun_path, namelen); 632 buf[namelen] = 0; /* null-terminate the string */ 633 error = nlookup_init(&nd, buf, UIO_SYSSPACE, 634 NLC_LOCKVP | NLC_CREATE | NLC_REFDVP); 635 if (error == 0) 636 error = nlookup(&nd); 637 if (error == 0 && nd.nl_nch.ncp->nc_vp != NULL) 638 error = EADDRINUSE; 639 if (error) 640 goto done; 641 642 VATTR_NULL(&vattr); 643 vattr.va_type = VSOCK; 644 vattr.va_mode = (ACCESSPERMS & ~p->p_fd->fd_cmask); 645 error = VOP_NCREATE(&nd.nl_nch, nd.nl_dvp, &vp, nd.nl_cred, &vattr); 646 if (error == 0) { 647 vp->v_socket = unp->unp_socket; 648 unp->unp_vnode = vp; 649 unp->unp_addr = (struct sockaddr_un *)dup_sockaddr(nam); 650 vn_unlock(vp); 651 } 652 done: 653 nlookup_done(&nd); 654 return (error); 655 } 656 657 static int 658 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 659 { 660 struct proc *p = td->td_proc; 661 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 662 struct vnode *vp; 663 struct socket *so2, *so3; 664 struct unpcb *unp, *unp2, *unp3; 665 int error, len; 666 struct nlookupdata nd; 667 char buf[SOCK_MAXADDRLEN]; 668 669 KKASSERT(p); 670 671 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 672 if (len <= 0) 673 return EINVAL; 674 strncpy(buf, soun->sun_path, len); 675 buf[len] = 0; 676 677 vp = NULL; 678 error = nlookup_init(&nd, buf, UIO_SYSSPACE, NLC_FOLLOW); 679 if (error == 0) 680 error = nlookup(&nd); 681 if (error == 0) 682 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp); 683 nlookup_done(&nd); 684 if (error) 685 return (error); 686 687 if (vp->v_type != VSOCK) { 688 error = ENOTSOCK; 689 goto bad; 690 } 691 error = VOP_ACCESS(vp, VWRITE, p->p_ucred); 692 if (error) 693 goto bad; 694 so2 = vp->v_socket; 695 if (so2 == NULL) { 696 error = ECONNREFUSED; 697 goto bad; 698 } 699 if (so->so_type != so2->so_type) { 700 error = EPROTOTYPE; 701 goto bad; 702 } 703 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 704 if (!(so2->so_options & SO_ACCEPTCONN) || 705 (so3 = sonewconn(so2, 0)) == NULL) { 706 error = ECONNREFUSED; 707 goto bad; 708 } 709 unp = so->so_pcb; 710 unp2 = so2->so_pcb; 711 unp3 = so3->so_pcb; 712 if (unp2->unp_addr) 713 unp3->unp_addr = (struct sockaddr_un *) 714 dup_sockaddr((struct sockaddr *)unp2->unp_addr); 715 716 /* 717 * unp_peercred management: 718 * 719 * The connecter's (client's) credentials are copied 720 * from its process structure at the time of connect() 721 * (which is now). 722 */ 723 cru2x(p->p_ucred, &unp3->unp_peercred); 724 unp3->unp_flags |= UNP_HAVEPC; 725 /* 726 * The receiver's (server's) credentials are copied 727 * from the unp_peercred member of socket on which the 728 * former called listen(); unp_listen() cached that 729 * process's credentials at that time so we can use 730 * them now. 731 */ 732 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED, 733 ("unp_connect: listener without cached peercred")); 734 memcpy(&unp->unp_peercred, &unp2->unp_peercred, 735 sizeof(unp->unp_peercred)); 736 unp->unp_flags |= UNP_HAVEPC; 737 738 so2 = so3; 739 } 740 error = unp_connect2(so, so2); 741 bad: 742 vput(vp); 743 return (error); 744 } 745 746 int 747 unp_connect2(struct socket *so, struct socket *so2) 748 { 749 struct unpcb *unp = so->so_pcb; 750 struct unpcb *unp2; 751 752 if (so2->so_type != so->so_type) 753 return (EPROTOTYPE); 754 unp2 = so2->so_pcb; 755 unp->unp_conn = unp2; 756 switch (so->so_type) { 757 758 case SOCK_DGRAM: 759 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 760 soisconnected(so); 761 break; 762 763 case SOCK_STREAM: 764 case SOCK_SEQPACKET: 765 unp2->unp_conn = unp; 766 soisconnected(so); 767 soisconnected(so2); 768 break; 769 770 default: 771 panic("unp_connect2"); 772 } 773 return (0); 774 } 775 776 static void 777 unp_disconnect(struct unpcb *unp) 778 { 779 struct unpcb *unp2 = unp->unp_conn; 780 781 if (unp2 == NULL) 782 return; 783 784 unp->unp_conn = NULL; 785 786 switch (unp->unp_socket->so_type) { 787 case SOCK_DGRAM: 788 LIST_REMOVE(unp, unp_reflink); 789 unp->unp_socket->so_state &= ~SS_ISCONNECTED; 790 break; 791 case SOCK_STREAM: 792 case SOCK_SEQPACKET: 793 soisdisconnected(unp->unp_socket); 794 unp2->unp_conn = NULL; 795 soisdisconnected(unp2->unp_socket); 796 break; 797 } 798 } 799 800 #ifdef notdef 801 void 802 unp_abort(struct unpcb *unp) 803 { 804 805 unp_detach(unp); 806 } 807 #endif 808 809 static int 810 prison_unpcb(struct thread *td, struct unpcb *unp) 811 { 812 struct proc *p; 813 814 if (td == NULL) 815 return (0); 816 if ((p = td->td_proc) == NULL) 817 return (0); 818 if (!p->p_ucred->cr_prison) 819 return (0); 820 if (p->p_fd->fd_rdir == unp->unp_rvnode) 821 return (0); 822 return (1); 823 } 824 825 static int 826 unp_pcblist(SYSCTL_HANDLER_ARGS) 827 { 828 int error, i, n; 829 struct unpcb *unp, **unp_list; 830 unp_gen_t gencnt; 831 struct unp_head *head; 832 833 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead); 834 835 KKASSERT(curproc != NULL); 836 837 /* 838 * The process of preparing the PCB list is too time-consuming and 839 * resource-intensive to repeat twice on every request. 840 */ 841 if (req->oldptr == NULL) { 842 n = unp_count; 843 req->oldidx = (n + n/8) * sizeof(struct xunpcb); 844 return 0; 845 } 846 847 if (req->newptr != NULL) 848 return EPERM; 849 850 /* 851 * OK, now we're committed to doing something. 852 */ 853 gencnt = unp_gencnt; 854 n = unp_count; 855 856 unp_list = kmalloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 857 858 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 859 unp = LIST_NEXT(unp, unp_link)) { 860 if (unp->unp_gencnt <= gencnt && !prison_unpcb(req->td, unp)) 861 unp_list[i++] = unp; 862 } 863 n = i; /* in case we lost some during malloc */ 864 865 error = 0; 866 for (i = 0; i < n; i++) { 867 unp = unp_list[i]; 868 if (unp->unp_gencnt <= gencnt) { 869 struct xunpcb xu; 870 xu.xu_len = sizeof xu; 871 xu.xu_unpp = unp; 872 /* 873 * XXX - need more locking here to protect against 874 * connect/disconnect races for SMP. 875 */ 876 if (unp->unp_addr) 877 bcopy(unp->unp_addr, &xu.xu_addr, 878 unp->unp_addr->sun_len); 879 if (unp->unp_conn && unp->unp_conn->unp_addr) 880 bcopy(unp->unp_conn->unp_addr, 881 &xu.xu_caddr, 882 unp->unp_conn->unp_addr->sun_len); 883 bcopy(unp, &xu.xu_unp, sizeof *unp); 884 sotoxsocket(unp->unp_socket, &xu.xu_socket); 885 error = SYSCTL_OUT(req, &xu, sizeof xu); 886 } 887 } 888 kfree(unp_list, M_TEMP); 889 return error; 890 } 891 892 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD, 893 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 894 "List of active local datagram sockets"); 895 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD, 896 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 897 "List of active local stream sockets"); 898 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist, CTLFLAG_RD, 899 (caddr_t)(long)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb", 900 "List of active local seqpacket stream sockets"); 901 902 static void 903 unp_shutdown(struct unpcb *unp) 904 { 905 struct socket *so; 906 907 if ((unp->unp_socket->so_type == SOCK_STREAM || 908 unp->unp_socket->so_type == SOCK_SEQPACKET) && 909 unp->unp_conn != NULL && (so = unp->unp_conn->unp_socket)) { 910 socantrcvmore(so); 911 } 912 } 913 914 static void 915 unp_drop(struct unpcb *unp, int err) 916 { 917 struct socket *so = unp->unp_socket; 918 919 so->so_error = err; 920 unp_disconnect(unp); 921 } 922 923 #ifdef notdef 924 void 925 unp_drain(void) 926 { 927 928 } 929 #endif 930 931 int 932 unp_externalize(struct mbuf *rights) 933 { 934 struct proc *p = curproc; /* XXX */ 935 int i; 936 struct cmsghdr *cm = mtod(rights, struct cmsghdr *); 937 int *fdp; 938 struct file **rp; 939 struct file *fp; 940 int newfds = (cm->cmsg_len - (CMSG_DATA(cm) - (u_char *)cm)) 941 / sizeof (struct file *); 942 int f; 943 944 /* 945 * if the new FD's will not fit, then we free them all 946 */ 947 if (!fdavail(p, newfds)) { 948 rp = (struct file **)CMSG_DATA(cm); 949 for (i = 0; i < newfds; i++) { 950 fp = *rp; 951 /* 952 * zero the pointer before calling unp_discard, 953 * since it may end up in unp_gc().. 954 */ 955 *rp++ = 0; 956 unp_discard(fp, NULL); 957 } 958 return (EMSGSIZE); 959 } 960 961 /* 962 * now change each pointer to an fd in the global table to 963 * an integer that is the index to the local fd table entry 964 * that we set up to point to the global one we are transferring. 965 * If sizeof (struct file *) is bigger than or equal to sizeof int, 966 * then do it in forward order. In that case, an integer will 967 * always come in the same place or before its corresponding 968 * struct file pointer. 969 * If sizeof (struct file *) is smaller than sizeof int, then 970 * do it in reverse order. 971 */ 972 if (sizeof (struct file *) >= sizeof (int)) { 973 fdp = (int *)(cm + 1); 974 rp = (struct file **)CMSG_DATA(cm); 975 for (i = 0; i < newfds; i++) { 976 if (fdalloc(p, 0, &f)) 977 panic("unp_externalize"); 978 fp = *rp++; 979 unp_fp_externalize(p, fp, f); 980 *fdp++ = f; 981 } 982 } else { 983 fdp = (int *)(cm + 1) + newfds - 1; 984 rp = (struct file **)CMSG_DATA(cm) + newfds - 1; 985 for (i = 0; i < newfds; i++) { 986 if (fdalloc(p, 0, &f)) 987 panic("unp_externalize"); 988 fp = *rp--; 989 unp_fp_externalize(p, fp, f); 990 *fdp-- = f; 991 } 992 } 993 994 /* 995 * Adjust length, in case sizeof(struct file *) and sizeof(int) 996 * differs. 997 */ 998 cm->cmsg_len = CMSG_LEN(newfds * sizeof(int)); 999 rights->m_len = cm->cmsg_len; 1000 return (0); 1001 } 1002 1003 static void 1004 unp_fp_externalize(struct proc *p, struct file *fp, int fd) 1005 { 1006 struct file *fx; 1007 int error; 1008 1009 if (p) { 1010 KKASSERT(fd >= 0); 1011 if (fp->f_flag & FREVOKED) { 1012 kprintf("Warning: revoked fp exiting unix socket\n"); 1013 fx = NULL; 1014 error = falloc(p, &fx, NULL); 1015 if (error == 0) 1016 fsetfd(p, fx, fd); 1017 else 1018 fsetfd(p, NULL, fd); 1019 fdrop(fx); 1020 } else { 1021 fsetfd(p, fp, fd); 1022 } 1023 } 1024 spin_lock_wr(&unp_spin); 1025 fp->f_msgcount--; 1026 unp_rights--; 1027 spin_unlock_wr(&unp_spin); 1028 fdrop(fp); 1029 } 1030 1031 1032 void 1033 unp_init(void) 1034 { 1035 LIST_INIT(&unp_dhead); 1036 LIST_INIT(&unp_shead); 1037 spin_init(&unp_spin); 1038 } 1039 1040 static int 1041 unp_internalize(struct mbuf *control, struct thread *td) 1042 { 1043 struct proc *p = td->td_proc; 1044 struct filedesc *fdescp; 1045 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1046 struct file **rp; 1047 struct file *fp; 1048 int i, fd, *fdp; 1049 struct cmsgcred *cmcred; 1050 int oldfds; 1051 u_int newlen; 1052 1053 KKASSERT(p); 1054 fdescp = p->p_fd; 1055 if ((cm->cmsg_type != SCM_RIGHTS && cm->cmsg_type != SCM_CREDS) || 1056 cm->cmsg_level != SOL_SOCKET || cm->cmsg_len != control->m_len) 1057 return (EINVAL); 1058 1059 /* 1060 * Fill in credential information. 1061 */ 1062 if (cm->cmsg_type == SCM_CREDS) { 1063 cmcred = (struct cmsgcred *)(cm + 1); 1064 cmcred->cmcred_pid = p->p_pid; 1065 cmcred->cmcred_uid = p->p_ucred->cr_ruid; 1066 cmcred->cmcred_gid = p->p_ucred->cr_rgid; 1067 cmcred->cmcred_euid = p->p_ucred->cr_uid; 1068 cmcred->cmcred_ngroups = MIN(p->p_ucred->cr_ngroups, 1069 CMGROUP_MAX); 1070 for (i = 0; i < cmcred->cmcred_ngroups; i++) 1071 cmcred->cmcred_groups[i] = p->p_ucred->cr_groups[i]; 1072 return(0); 1073 } 1074 1075 oldfds = (cm->cmsg_len - sizeof (*cm)) / sizeof (int); 1076 /* 1077 * check that all the FDs passed in refer to legal OPEN files 1078 * If not, reject the entire operation. 1079 */ 1080 fdp = (int *)(cm + 1); 1081 for (i = 0; i < oldfds; i++) { 1082 fd = *fdp++; 1083 if ((unsigned)fd >= fdescp->fd_nfiles || 1084 fdescp->fd_files[fd].fp == NULL) 1085 return (EBADF); 1086 if (fdescp->fd_files[fd].fp->f_type == DTYPE_KQUEUE) 1087 return (EOPNOTSUPP); 1088 } 1089 /* 1090 * Now replace the integer FDs with pointers to 1091 * the associated global file table entry.. 1092 * Allocate a bigger buffer as necessary. But if an cluster is not 1093 * enough, return E2BIG. 1094 */ 1095 newlen = CMSG_LEN(oldfds * sizeof(struct file *)); 1096 if (newlen > MCLBYTES) 1097 return (E2BIG); 1098 if (newlen - control->m_len > M_TRAILINGSPACE(control)) { 1099 if (control->m_flags & M_EXT) 1100 return (E2BIG); 1101 MCLGET(control, MB_WAIT); 1102 if (!(control->m_flags & M_EXT)) 1103 return (ENOBUFS); 1104 1105 /* copy the data to the cluster */ 1106 memcpy(mtod(control, char *), cm, cm->cmsg_len); 1107 cm = mtod(control, struct cmsghdr *); 1108 } 1109 1110 /* 1111 * Adjust length, in case sizeof(struct file *) and sizeof(int) 1112 * differs. 1113 */ 1114 control->m_len = cm->cmsg_len = newlen; 1115 1116 /* 1117 * Transform the file descriptors into struct file pointers. 1118 * If sizeof (struct file *) is bigger than or equal to sizeof int, 1119 * then do it in reverse order so that the int won't get until 1120 * we're done. 1121 * If sizeof (struct file *) is smaller than sizeof int, then 1122 * do it in forward order. 1123 */ 1124 if (sizeof (struct file *) >= sizeof (int)) { 1125 fdp = (int *)(cm + 1) + oldfds - 1; 1126 rp = (struct file **)CMSG_DATA(cm) + oldfds - 1; 1127 for (i = 0; i < oldfds; i++) { 1128 fp = fdescp->fd_files[*fdp--].fp; 1129 *rp-- = fp; 1130 fhold(fp); 1131 spin_lock_wr(&unp_spin); 1132 fp->f_msgcount++; 1133 unp_rights++; 1134 spin_unlock_wr(&unp_spin); 1135 } 1136 } else { 1137 fdp = (int *)(cm + 1); 1138 rp = (struct file **)CMSG_DATA(cm); 1139 for (i = 0; i < oldfds; i++) { 1140 fp = fdescp->fd_files[*fdp++].fp; 1141 *rp++ = fp; 1142 fhold(fp); 1143 spin_lock_wr(&unp_spin); 1144 fp->f_msgcount++; 1145 unp_rights++; 1146 spin_unlock_wr(&unp_spin); 1147 } 1148 } 1149 return (0); 1150 } 1151 1152 /* 1153 * Garbage collect in-transit file descriptors that get lost due to 1154 * loops (i.e. when a socket is sent to another process over itself, 1155 * and more complex situations). 1156 * 1157 * NOT MPSAFE - TODO socket flush code and maybe closef. Rest is MPSAFE. 1158 */ 1159 1160 struct unp_gc_info { 1161 struct file **extra_ref; 1162 struct file *locked_fp; 1163 int defer; 1164 int index; 1165 int maxindex; 1166 }; 1167 1168 static void 1169 unp_gc(void) 1170 { 1171 struct unp_gc_info info; 1172 static boolean_t unp_gcing; 1173 struct file **fpp; 1174 int i; 1175 1176 spin_lock_wr(&unp_spin); 1177 if (unp_gcing) { 1178 spin_unlock_wr(&unp_spin); 1179 return; 1180 } 1181 unp_gcing = TRUE; 1182 spin_unlock_wr(&unp_spin); 1183 1184 /* 1185 * before going through all this, set all FDs to 1186 * be NOT defered and NOT externally accessible 1187 */ 1188 info.defer = 0; 1189 allfiles_scan_exclusive(unp_gc_clearmarks, NULL); 1190 do { 1191 allfiles_scan_exclusive(unp_gc_checkmarks, &info); 1192 } while (info.defer); 1193 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 orginal 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, we call sorflush ourself, since we know 1226 * it is a Unix domain socket anyhow. After we destroy all the 1227 * rights carried in messages, we do a last closef to get rid 1228 * of our extra reference. This is the last close, and the 1229 * unp_detach etc will shut down the socket. 1230 * 1231 * 91/09/19, bsy@cs.cmu.edu 1232 */ 1233 info.extra_ref = kmalloc(256 * sizeof(struct file *), M_FILE, M_WAITOK); 1234 info.maxindex = 256; 1235 1236 do { 1237 /* 1238 * Look for matches 1239 */ 1240 info.index = 0; 1241 allfiles_scan_exclusive(unp_gc_checkrefs, &info); 1242 1243 /* 1244 * For each FD on our hit list, do the following two things 1245 */ 1246 for (i = info.index, fpp = info.extra_ref; --i >= 0; ++fpp) { 1247 struct file *tfp = *fpp; 1248 if (tfp->f_type == DTYPE_SOCKET && tfp->f_data != NULL) 1249 sorflush((struct socket *)(tfp->f_data)); 1250 } 1251 for (i = info.index, fpp = info.extra_ref; --i >= 0; ++fpp) 1252 closef(*fpp, NULL); 1253 } while (info.index == info.maxindex); 1254 kfree((caddr_t)info.extra_ref, M_FILE); 1255 unp_gcing = FALSE; 1256 } 1257 1258 /* 1259 * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry 1260 */ 1261 static int 1262 unp_gc_checkrefs(struct file *fp, void *data) 1263 { 1264 struct unp_gc_info *info = data; 1265 1266 if (fp->f_count == 0) 1267 return(0); 1268 if (info->index == info->maxindex) 1269 return(-1); 1270 1271 /* 1272 * If all refs are from msgs, and it's not marked accessible 1273 * then it must be referenced from some unreachable cycle 1274 * of (shut-down) FDs, so include it in our 1275 * list of FDs to remove 1276 */ 1277 if (fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) { 1278 info->extra_ref[info->index++] = fp; 1279 fhold(fp); 1280 } 1281 return(0); 1282 } 1283 1284 /* 1285 * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry 1286 */ 1287 static int 1288 unp_gc_clearmarks(struct file *fp, void *data __unused) 1289 { 1290 atomic_clear_int(&fp->f_flag, FMARK | FDEFER); 1291 return(0); 1292 } 1293 1294 /* 1295 * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry 1296 */ 1297 static int 1298 unp_gc_checkmarks(struct file *fp, void *data) 1299 { 1300 struct unp_gc_info *info = data; 1301 struct socket *so; 1302 1303 /* 1304 * If the file is not open, skip it 1305 */ 1306 if (fp->f_count == 0) 1307 return(0); 1308 /* 1309 * If we already marked it as 'defer' in a 1310 * previous pass, then try process it this time 1311 * and un-mark it 1312 */ 1313 if (fp->f_flag & FDEFER) { 1314 atomic_clear_int(&fp->f_flag, FDEFER); 1315 --info->defer; 1316 } else { 1317 /* 1318 * if it's not defered, then check if it's 1319 * already marked.. if so skip it 1320 */ 1321 if (fp->f_flag & FMARK) 1322 return(0); 1323 /* 1324 * If all references are from messages 1325 * in transit, then skip it. it's not 1326 * externally accessible. 1327 */ 1328 if (fp->f_count == fp->f_msgcount) 1329 return(0); 1330 /* 1331 * If it got this far then it must be 1332 * externally accessible. 1333 */ 1334 atomic_set_int(&fp->f_flag, FMARK); 1335 } 1336 1337 /* 1338 * either it was defered, or it is externally 1339 * accessible and not already marked so. 1340 * Now check if it is possibly one of OUR sockets. 1341 */ 1342 if (fp->f_type != DTYPE_SOCKET || 1343 (so = (struct socket *)fp->f_data) == NULL) 1344 return(0); 1345 if (so->so_proto->pr_domain != &localdomain || 1346 !(so->so_proto->pr_flags & PR_RIGHTS)) 1347 return(0); 1348 #ifdef notdef 1349 if (so->so_rcv.sb_flags & SB_LOCK) { 1350 /* 1351 * This is problematical; it's not clear 1352 * we need to wait for the sockbuf to be 1353 * unlocked (on a uniprocessor, at least), 1354 * and it's also not clear what to do 1355 * if sbwait returns an error due to receipt 1356 * of a signal. If sbwait does return 1357 * an error, we'll go into an infinite 1358 * loop. Delete all of this for now. 1359 */ 1360 sbwait(&so->so_rcv); 1361 goto restart; 1362 } 1363 #endif 1364 /* 1365 * So, Ok, it's one of our sockets and it IS externally 1366 * accessible (or was defered). Now we look 1367 * to see if we hold any file descriptors in its 1368 * message buffers. Follow those links and mark them 1369 * as accessible too. 1370 */ 1371 info->locked_fp = fp; 1372 /* spin_lock_wr(&so->so_rcv.sb_spin); */ 1373 unp_scan(so->so_rcv.ssb_mb, unp_mark, info); 1374 /* spin_unlock_wr(&so->so_rcv.sb_spin);*/ 1375 return (0); 1376 } 1377 1378 /* 1379 * Scan all unix domain sockets and replace any revoked file pointers 1380 * found with the dummy file pointer fx. We don't worry about races 1381 * against file pointers being read out as those are handled in the 1382 * externalize code. 1383 */ 1384 1385 #define REVOKE_GC_MAXFILES 32 1386 1387 struct unp_revoke_gc_info { 1388 struct file *fx; 1389 struct file *fary[REVOKE_GC_MAXFILES]; 1390 int fcount; 1391 }; 1392 1393 void 1394 unp_revoke_gc(struct file *fx) 1395 { 1396 struct unp_revoke_gc_info info; 1397 int i; 1398 1399 info.fx = fx; 1400 do { 1401 info.fcount = 0; 1402 allfiles_scan_exclusive(unp_revoke_gc_check, &info); 1403 for (i = 0; i < info.fcount; ++i) 1404 unp_fp_externalize(NULL, info.fary[i], -1); 1405 } while (info.fcount == REVOKE_GC_MAXFILES); 1406 } 1407 1408 /* 1409 * Check for and replace revoked descriptors. 1410 * 1411 * WARNING: This routine is not allowed to block. 1412 */ 1413 static int 1414 unp_revoke_gc_check(struct file *fps, void *vinfo) 1415 { 1416 struct unp_revoke_gc_info *info = vinfo; 1417 struct file *fp; 1418 struct socket *so; 1419 struct mbuf *m0; 1420 struct mbuf *m; 1421 struct file **rp; 1422 struct cmsghdr *cm; 1423 int i; 1424 int qfds; 1425 1426 /* 1427 * Is this a unix domain socket with rights-passing abilities? 1428 */ 1429 if (fps->f_type != DTYPE_SOCKET) 1430 return (0); 1431 if ((so = (struct socket *)fps->f_data) == NULL) 1432 return(0); 1433 if (so->so_proto->pr_domain != &localdomain) 1434 return(0); 1435 if ((so->so_proto->pr_flags & PR_RIGHTS) == 0) 1436 return(0); 1437 1438 /* 1439 * Scan the mbufs for control messages and replace any revoked 1440 * descriptors we find. 1441 */ 1442 m0 = so->so_rcv.ssb_mb; 1443 while (m0) { 1444 for (m = m0; m; m = m->m_next) { 1445 if (m->m_type != MT_CONTROL) 1446 continue; 1447 if (m->m_len < sizeof(*cm)) 1448 continue; 1449 cm = mtod(m, struct cmsghdr *); 1450 if (cm->cmsg_level != SOL_SOCKET || 1451 cm->cmsg_type != SCM_RIGHTS) { 1452 continue; 1453 } 1454 qfds = (cm->cmsg_len - 1455 (CMSG_DATA(cm) - (u_char *)cm)) 1456 / sizeof (struct file *); 1457 rp = (struct file **)CMSG_DATA(cm); 1458 for (i = 0; i < qfds; i++) { 1459 fp = rp[i]; 1460 if (fp->f_flag & FREVOKED) { 1461 kprintf("Warning: Removing revoked fp from unix domain socket queue\n"); 1462 fhold(info->fx); 1463 info->fx->f_msgcount++; 1464 unp_rights++; 1465 rp[i] = info->fx; 1466 info->fary[info->fcount++] = fp; 1467 } 1468 if (info->fcount == REVOKE_GC_MAXFILES) 1469 break; 1470 } 1471 if (info->fcount == REVOKE_GC_MAXFILES) 1472 break; 1473 } 1474 m0 = m0->m_nextpkt; 1475 if (info->fcount == REVOKE_GC_MAXFILES) 1476 break; 1477 } 1478 1479 /* 1480 * Stop the scan if we filled up our array. 1481 */ 1482 if (info->fcount == REVOKE_GC_MAXFILES) 1483 return(-1); 1484 return(0); 1485 } 1486 1487 void 1488 unp_dispose(struct mbuf *m) 1489 { 1490 if (m) 1491 unp_scan(m, unp_discard, NULL); 1492 } 1493 1494 static int 1495 unp_listen(struct unpcb *unp, struct thread *td) 1496 { 1497 struct proc *p = td->td_proc; 1498 1499 KKASSERT(p); 1500 cru2x(p->p_ucred, &unp->unp_peercred); 1501 unp->unp_flags |= UNP_HAVEPCCACHED; 1502 return (0); 1503 } 1504 1505 static void 1506 unp_scan(struct mbuf *m0, void (*op)(struct file *, void *), void *data) 1507 { 1508 struct mbuf *m; 1509 struct file **rp; 1510 struct cmsghdr *cm; 1511 int i; 1512 int qfds; 1513 1514 while (m0) { 1515 for (m = m0; m; m = m->m_next) { 1516 if (m->m_type == MT_CONTROL && 1517 m->m_len >= sizeof(*cm)) { 1518 cm = mtod(m, struct cmsghdr *); 1519 if (cm->cmsg_level != SOL_SOCKET || 1520 cm->cmsg_type != SCM_RIGHTS) 1521 continue; 1522 qfds = (cm->cmsg_len - 1523 (CMSG_DATA(cm) - (u_char *)cm)) 1524 / sizeof (struct file *); 1525 rp = (struct file **)CMSG_DATA(cm); 1526 for (i = 0; i < qfds; i++) 1527 (*op)(*rp++, data); 1528 break; /* XXX, but saves time */ 1529 } 1530 } 1531 m0 = m0->m_nextpkt; 1532 } 1533 } 1534 1535 static void 1536 unp_mark(struct file *fp, void *data) 1537 { 1538 struct unp_gc_info *info = data; 1539 1540 if ((fp->f_flag & FMARK) == 0) { 1541 ++info->defer; 1542 atomic_set_int(&fp->f_flag, FMARK | FDEFER); 1543 } 1544 } 1545 1546 static void 1547 unp_discard(struct file *fp, void *data __unused) 1548 { 1549 spin_lock_wr(&unp_spin); 1550 fp->f_msgcount--; 1551 unp_rights--; 1552 spin_unlock_wr(&unp_spin); 1553 closef(fp, NULL); 1554 } 1555 1556