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 = 1; /* prototype for fake inode numbers */ 80 static struct spinlock unp_ino_spin = SPINLOCK_INITIALIZER(&unp_ino_spin); 81 82 static int unp_attach (struct socket *, struct pru_attach_info *); 83 static void unp_detach (struct unpcb *); 84 static int unp_bind (struct unpcb *,struct sockaddr *, struct thread *); 85 static int unp_connect (struct socket *,struct sockaddr *, 86 struct thread *); 87 static void unp_disconnect (struct unpcb *); 88 static void unp_shutdown (struct unpcb *); 89 static void unp_drop (struct unpcb *, int); 90 static void unp_gc (void); 91 static int unp_gc_clearmarks(struct file *, void *); 92 static int unp_gc_checkmarks(struct file *, void *); 93 static int unp_gc_checkrefs(struct file *, void *); 94 static int unp_revoke_gc_check(struct file *, void *); 95 static void unp_scan (struct mbuf *, void (*)(struct file *, void *), 96 void *data); 97 static void unp_mark (struct file *, void *data); 98 static void unp_discard (struct file *, void *); 99 static int unp_internalize (struct mbuf *, struct thread *); 100 static int unp_listen (struct unpcb *, struct thread *); 101 static void unp_fp_externalize(struct lwp *lp, struct file *fp, int fd); 102 103 static int 104 uipc_abort(struct socket *so) 105 { 106 struct unpcb *unp = so->so_pcb; 107 108 if (unp == NULL) 109 return EINVAL; 110 unp_drop(unp, ECONNABORTED); 111 unp_detach(unp); 112 sofree(so); 113 return 0; 114 } 115 116 static int 117 uipc_accept(struct socket *so, struct sockaddr **nam) 118 { 119 struct unpcb *unp = so->so_pcb; 120 121 if (unp == NULL) 122 return EINVAL; 123 124 /* 125 * Pass back name of connected socket, 126 * if it was bound and we are still connected 127 * (our peer may have closed already!). 128 */ 129 if (unp->unp_conn && unp->unp_conn->unp_addr) { 130 *nam = dup_sockaddr((struct sockaddr *)unp->unp_conn->unp_addr); 131 } else { 132 *nam = dup_sockaddr((struct sockaddr *)&sun_noname); 133 } 134 return 0; 135 } 136 137 static int 138 uipc_attach(struct socket *so, int proto, struct pru_attach_info *ai) 139 { 140 struct unpcb *unp = so->so_pcb; 141 142 if (unp != NULL) 143 return EISCONN; 144 return unp_attach(so, ai); 145 } 146 147 static int 148 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 149 { 150 struct unpcb *unp = so->so_pcb; 151 152 if (unp == NULL) 153 return EINVAL; 154 return unp_bind(unp, nam, td); 155 } 156 157 static int 158 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 159 { 160 struct unpcb *unp = so->so_pcb; 161 162 if (unp == NULL) 163 return EINVAL; 164 return unp_connect(so, nam, td); 165 } 166 167 static int 168 uipc_connect2(struct socket *so1, struct socket *so2) 169 { 170 struct unpcb *unp = so1->so_pcb; 171 172 if (unp == NULL) 173 return EINVAL; 174 175 return unp_connect2(so1, so2); 176 } 177 178 /* control is EOPNOTSUPP */ 179 180 static int 181 uipc_detach(struct socket *so) 182 { 183 struct unpcb *unp = so->so_pcb; 184 185 if (unp == NULL) 186 return EINVAL; 187 188 unp_detach(unp); 189 return 0; 190 } 191 192 static int 193 uipc_disconnect(struct socket *so) 194 { 195 struct unpcb *unp = so->so_pcb; 196 197 if (unp == NULL) 198 return EINVAL; 199 unp_disconnect(unp); 200 return 0; 201 } 202 203 static int 204 uipc_listen(struct socket *so, struct thread *td) 205 { 206 struct unpcb *unp = so->so_pcb; 207 208 if (unp == NULL || unp->unp_vnode == NULL) 209 return EINVAL; 210 return unp_listen(unp, td); 211 } 212 213 static int 214 uipc_peeraddr(struct socket *so, struct sockaddr **nam) 215 { 216 struct unpcb *unp = so->so_pcb; 217 218 if (unp == NULL) 219 return EINVAL; 220 if (unp->unp_conn && unp->unp_conn->unp_addr) 221 *nam = dup_sockaddr((struct sockaddr *)unp->unp_conn->unp_addr); 222 else { 223 /* 224 * XXX: It seems that this test always fails even when 225 * connection is established. So, this else clause is 226 * added as workaround to return PF_LOCAL sockaddr. 227 */ 228 *nam = dup_sockaddr((struct sockaddr *)&sun_noname); 229 } 230 return 0; 231 } 232 233 static int 234 uipc_rcvd(struct socket *so, int flags) 235 { 236 struct unpcb *unp = so->so_pcb; 237 struct socket *so2; 238 239 if (unp == NULL) 240 return EINVAL; 241 switch (so->so_type) { 242 case SOCK_DGRAM: 243 panic("uipc_rcvd DGRAM?"); 244 /*NOTREACHED*/ 245 246 case SOCK_STREAM: 247 case SOCK_SEQPACKET: 248 if (unp->unp_conn == NULL) 249 break; 250 /* 251 * Because we are transfering mbufs directly to the 252 * peer socket we have to use SSB_STOP on the sender 253 * to prevent it from building up infinite mbufs. 254 */ 255 so2 = unp->unp_conn->unp_socket; 256 if (so->so_rcv.ssb_cc < so2->so_snd.ssb_hiwat && 257 so->so_rcv.ssb_mbcnt < so2->so_snd.ssb_mbmax 258 ) { 259 atomic_clear_int(&so2->so_snd.ssb_flags, SSB_STOP); 260 sowwakeup(so2); 261 } 262 break; 263 264 default: 265 panic("uipc_rcvd unknown socktype"); 266 } 267 return 0; 268 } 269 270 /* pru_rcvoob is EOPNOTSUPP */ 271 272 static int 273 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 274 struct mbuf *control, struct thread *td) 275 { 276 int error = 0; 277 struct unpcb *unp = so->so_pcb; 278 struct socket *so2; 279 280 if (unp == NULL) { 281 error = EINVAL; 282 goto release; 283 } 284 if (flags & PRUS_OOB) { 285 error = EOPNOTSUPP; 286 goto release; 287 } 288 289 if (control && (error = unp_internalize(control, td))) 290 goto release; 291 292 switch (so->so_type) { 293 case SOCK_DGRAM: 294 { 295 struct sockaddr *from; 296 297 if (nam) { 298 if (unp->unp_conn) { 299 error = EISCONN; 300 break; 301 } 302 error = unp_connect(so, nam, td); 303 if (error) 304 break; 305 } else { 306 if (unp->unp_conn == NULL) { 307 error = ENOTCONN; 308 break; 309 } 310 } 311 so2 = unp->unp_conn->unp_socket; 312 if (unp->unp_addr) 313 from = (struct sockaddr *)unp->unp_addr; 314 else 315 from = &sun_noname; 316 if (ssb_appendaddr(&so2->so_rcv, from, m, control)) { 317 sorwakeup(so2); 318 m = NULL; 319 control = NULL; 320 } else { 321 error = ENOBUFS; 322 } 323 if (nam) 324 unp_disconnect(unp); 325 break; 326 } 327 328 case SOCK_STREAM: 329 case SOCK_SEQPACKET: 330 /* Connect if not connected yet. */ 331 /* 332 * Note: A better implementation would complain 333 * if not equal to the peer's address. 334 */ 335 if (!(so->so_state & SS_ISCONNECTED)) { 336 if (nam) { 337 error = unp_connect(so, nam, td); 338 if (error) 339 break; /* XXX */ 340 } else { 341 error = ENOTCONN; 342 break; 343 } 344 } 345 346 if (so->so_state & SS_CANTSENDMORE) { 347 error = EPIPE; 348 break; 349 } 350 if (unp->unp_conn == NULL) 351 panic("uipc_send connected but no connection?"); 352 so2 = unp->unp_conn->unp_socket; 353 /* 354 * Send to paired receive port, and then reduce 355 * send buffer hiwater marks to maintain backpressure. 356 * Wake up readers. 357 */ 358 if (control) { 359 if (ssb_appendcontrol(&so2->so_rcv, m, control)) { 360 control = NULL; 361 m = NULL; 362 } 363 } else if (so->so_type == SOCK_SEQPACKET) { 364 sbappendrecord(&so2->so_rcv.sb, m); 365 m = NULL; 366 } else { 367 sbappend(&so2->so_rcv.sb, m); 368 m = NULL; 369 } 370 371 /* 372 * Because we are transfering mbufs directly to the 373 * peer socket we have to use SSB_STOP on the sender 374 * to prevent it from building up infinite mbufs. 375 */ 376 if (so2->so_rcv.ssb_cc >= so->so_snd.ssb_hiwat || 377 so2->so_rcv.ssb_mbcnt >= so->so_snd.ssb_mbmax 378 ) { 379 atomic_set_int(&so->so_snd.ssb_flags, SSB_STOP); 380 } 381 sorwakeup(so2); 382 break; 383 384 default: 385 panic("uipc_send unknown socktype"); 386 } 387 388 /* 389 * SEND_EOF is equivalent to a SEND followed by a SHUTDOWN. 390 */ 391 if (flags & PRUS_EOF) { 392 socantsendmore(so); 393 unp_shutdown(unp); 394 } 395 396 if (control && error != 0) 397 unp_dispose(control); 398 399 release: 400 if (control) 401 m_freem(control); 402 if (m) 403 m_freem(m); 404 return error; 405 } 406 407 /* 408 * MPSAFE 409 */ 410 static int 411 uipc_sense(struct socket *so, struct stat *sb) 412 { 413 struct unpcb *unp = so->so_pcb; 414 415 if (unp == NULL) 416 return EINVAL; 417 sb->st_blksize = so->so_snd.ssb_hiwat; 418 sb->st_dev = NOUDEV; 419 if (unp->unp_ino == 0) { /* make up a non-zero inode number */ 420 spin_lock(&unp_ino_spin); 421 unp->unp_ino = unp_ino++; 422 spin_unlock(&unp_ino_spin); 423 } 424 sb->st_ino = unp->unp_ino; 425 return (0); 426 } 427 428 static int 429 uipc_shutdown(struct socket *so) 430 { 431 struct unpcb *unp = so->so_pcb; 432 433 if (unp == NULL) 434 return EINVAL; 435 socantsendmore(so); 436 unp_shutdown(unp); 437 return 0; 438 } 439 440 static int 441 uipc_sockaddr(struct socket *so, struct sockaddr **nam) 442 { 443 struct unpcb *unp = so->so_pcb; 444 445 if (unp == NULL) 446 return EINVAL; 447 if (unp->unp_addr) 448 *nam = dup_sockaddr((struct sockaddr *)unp->unp_addr); 449 return 0; 450 } 451 452 struct pr_usrreqs uipc_usrreqs = { 453 .pru_abort = uipc_abort, 454 .pru_accept = uipc_accept, 455 .pru_attach = uipc_attach, 456 .pru_bind = uipc_bind, 457 .pru_connect = uipc_connect, 458 .pru_connect2 = uipc_connect2, 459 .pru_control = pru_control_notsupp, 460 .pru_detach = uipc_detach, 461 .pru_disconnect = uipc_disconnect, 462 .pru_listen = uipc_listen, 463 .pru_peeraddr = uipc_peeraddr, 464 .pru_rcvd = uipc_rcvd, 465 .pru_rcvoob = pru_rcvoob_notsupp, 466 .pru_send = uipc_send, 467 .pru_sense = uipc_sense, 468 .pru_shutdown = uipc_shutdown, 469 .pru_sockaddr = uipc_sockaddr, 470 .pru_sosend = sosend, 471 .pru_soreceive = soreceive 472 }; 473 474 int 475 uipc_ctloutput(struct socket *so, struct sockopt *sopt) 476 { 477 struct unpcb *unp = so->so_pcb; 478 int error = 0; 479 480 switch (sopt->sopt_dir) { 481 case SOPT_GET: 482 switch (sopt->sopt_name) { 483 case LOCAL_PEERCRED: 484 if (unp->unp_flags & UNP_HAVEPC) 485 soopt_from_kbuf(sopt, &unp->unp_peercred, 486 sizeof(unp->unp_peercred)); 487 else { 488 if (so->so_type == SOCK_STREAM) 489 error = ENOTCONN; 490 else if (so->so_type == SOCK_SEQPACKET) 491 error = ENOTCONN; 492 else 493 error = EINVAL; 494 } 495 break; 496 default: 497 error = EOPNOTSUPP; 498 break; 499 } 500 break; 501 case SOPT_SET: 502 default: 503 error = EOPNOTSUPP; 504 break; 505 } 506 return (error); 507 } 508 509 /* 510 * Both send and receive buffers are allocated PIPSIZ bytes of buffering 511 * for stream sockets, although the total for sender and receiver is 512 * actually only PIPSIZ. 513 * 514 * Datagram sockets really use the sendspace as the maximum datagram size, 515 * and don't really want to reserve the sendspace. Their recvspace should 516 * be large enough for at least one max-size datagram plus address. 517 * 518 * We want the local send/recv space to be significant larger then lo0's 519 * mtu of 16384. 520 */ 521 #ifndef PIPSIZ 522 #define PIPSIZ 57344 523 #endif 524 static u_long unpst_sendspace = PIPSIZ; 525 static u_long unpst_recvspace = PIPSIZ; 526 static u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 527 static u_long unpdg_recvspace = 4*1024; 528 529 static int unp_rights; /* file descriptors in flight */ 530 static struct spinlock unp_spin = SPINLOCK_INITIALIZER(&unp_spin); 531 532 SYSCTL_DECL(_net_local_seqpacket); 533 SYSCTL_DECL(_net_local_stream); 534 SYSCTL_INT(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 535 &unpst_sendspace, 0, ""); 536 SYSCTL_INT(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 537 &unpst_recvspace, 0, ""); 538 539 SYSCTL_DECL(_net_local_dgram); 540 SYSCTL_INT(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 541 &unpdg_sendspace, 0, ""); 542 SYSCTL_INT(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 543 &unpdg_recvspace, 0, ""); 544 545 SYSCTL_DECL(_net_local); 546 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, ""); 547 548 static int 549 unp_attach(struct socket *so, struct pru_attach_info *ai) 550 { 551 struct unpcb *unp; 552 int error; 553 554 if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) { 555 switch (so->so_type) { 556 557 case SOCK_STREAM: 558 case SOCK_SEQPACKET: 559 error = soreserve(so, unpst_sendspace, unpst_recvspace, 560 ai->sb_rlimit); 561 break; 562 563 case SOCK_DGRAM: 564 error = soreserve(so, unpdg_sendspace, unpdg_recvspace, 565 ai->sb_rlimit); 566 break; 567 568 default: 569 panic("unp_attach"); 570 } 571 if (error) 572 return (error); 573 } 574 unp = kmalloc(sizeof(*unp), M_UNPCB, M_NOWAIT|M_ZERO); 575 if (unp == NULL) 576 return (ENOBUFS); 577 unp->unp_gencnt = ++unp_gencnt; 578 unp_count++; 579 LIST_INIT(&unp->unp_refs); 580 unp->unp_socket = so; 581 unp->unp_rvnode = ai->fd_rdir; /* jail cruft XXX JH */ 582 LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead 583 : &unp_shead, unp, unp_link); 584 so->so_pcb = (caddr_t)unp; 585 so->so_port = sync_soport(so, NULL, NULL); 586 return (0); 587 } 588 589 static void 590 unp_detach(struct unpcb *unp) 591 { 592 LIST_REMOVE(unp, unp_link); 593 unp->unp_gencnt = ++unp_gencnt; 594 --unp_count; 595 if (unp->unp_vnode) { 596 unp->unp_vnode->v_socket = NULL; 597 vrele(unp->unp_vnode); 598 unp->unp_vnode = NULL; 599 } 600 if (unp->unp_conn) 601 unp_disconnect(unp); 602 while (!LIST_EMPTY(&unp->unp_refs)) 603 unp_drop(LIST_FIRST(&unp->unp_refs), ECONNRESET); 604 soisdisconnected(unp->unp_socket); 605 unp->unp_socket->so_pcb = NULL; 606 if (unp_rights) { 607 /* 608 * Normally the receive buffer is flushed later, 609 * in sofree, but if our receive buffer holds references 610 * to descriptors that are now garbage, we will dispose 611 * of those descriptor references after the garbage collector 612 * gets them (resulting in a "panic: closef: count < 0"). 613 */ 614 sorflush(unp->unp_socket); 615 unp_gc(); 616 } 617 if (unp->unp_addr) 618 kfree(unp->unp_addr, M_SONAME); 619 kfree(unp, M_UNPCB); 620 } 621 622 static int 623 unp_bind(struct unpcb *unp, struct sockaddr *nam, struct thread *td) 624 { 625 struct proc *p = td->td_proc; 626 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 627 struct vnode *vp; 628 struct vattr vattr; 629 int error, namelen; 630 struct nlookupdata nd; 631 char buf[SOCK_MAXADDRLEN]; 632 633 if (unp->unp_vnode != NULL) 634 return (EINVAL); 635 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 636 if (namelen <= 0) 637 return (EINVAL); 638 strncpy(buf, soun->sun_path, namelen); 639 buf[namelen] = 0; /* null-terminate the string */ 640 error = nlookup_init(&nd, buf, UIO_SYSSPACE, 641 NLC_LOCKVP | NLC_CREATE | NLC_REFDVP); 642 if (error == 0) 643 error = nlookup(&nd); 644 if (error == 0 && nd.nl_nch.ncp->nc_vp != NULL) 645 error = EADDRINUSE; 646 if (error) 647 goto done; 648 649 VATTR_NULL(&vattr); 650 vattr.va_type = VSOCK; 651 vattr.va_mode = (ACCESSPERMS & ~p->p_fd->fd_cmask); 652 error = VOP_NCREATE(&nd.nl_nch, nd.nl_dvp, &vp, nd.nl_cred, &vattr); 653 if (error == 0) { 654 vp->v_socket = unp->unp_socket; 655 unp->unp_vnode = vp; 656 unp->unp_addr = (struct sockaddr_un *)dup_sockaddr(nam); 657 vn_unlock(vp); 658 } 659 done: 660 nlookup_done(&nd); 661 return (error); 662 } 663 664 static int 665 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 666 { 667 struct proc *p = td->td_proc; 668 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 669 struct vnode *vp; 670 struct socket *so2, *so3; 671 struct unpcb *unp, *unp2, *unp3; 672 int error, len; 673 struct nlookupdata nd; 674 char buf[SOCK_MAXADDRLEN]; 675 676 KKASSERT(p); 677 678 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 679 if (len <= 0) 680 return EINVAL; 681 strncpy(buf, soun->sun_path, len); 682 buf[len] = 0; 683 684 vp = NULL; 685 error = nlookup_init(&nd, buf, UIO_SYSSPACE, NLC_FOLLOW); 686 if (error == 0) 687 error = nlookup(&nd); 688 if (error == 0) 689 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp); 690 nlookup_done(&nd); 691 if (error) 692 return (error); 693 694 if (vp->v_type != VSOCK) { 695 error = ENOTSOCK; 696 goto bad; 697 } 698 error = VOP_ACCESS(vp, VWRITE, p->p_ucred); 699 if (error) 700 goto bad; 701 so2 = vp->v_socket; 702 if (so2 == NULL) { 703 error = ECONNREFUSED; 704 goto bad; 705 } 706 if (so->so_type != so2->so_type) { 707 error = EPROTOTYPE; 708 goto bad; 709 } 710 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 711 if (!(so2->so_options & SO_ACCEPTCONN) || 712 (so3 = sonewconn(so2, 0)) == NULL) { 713 error = ECONNREFUSED; 714 goto bad; 715 } 716 unp = so->so_pcb; 717 unp2 = so2->so_pcb; 718 unp3 = so3->so_pcb; 719 if (unp2->unp_addr) 720 unp3->unp_addr = (struct sockaddr_un *) 721 dup_sockaddr((struct sockaddr *)unp2->unp_addr); 722 723 /* 724 * unp_peercred management: 725 * 726 * The connecter's (client's) credentials are copied 727 * from its process structure at the time of connect() 728 * (which is now). 729 */ 730 cru2x(p->p_ucred, &unp3->unp_peercred); 731 unp3->unp_flags |= UNP_HAVEPC; 732 /* 733 * The receiver's (server's) credentials are copied 734 * from the unp_peercred member of socket on which the 735 * former called listen(); unp_listen() cached that 736 * process's credentials at that time so we can use 737 * them now. 738 */ 739 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED, 740 ("unp_connect: listener without cached peercred")); 741 memcpy(&unp->unp_peercred, &unp2->unp_peercred, 742 sizeof(unp->unp_peercred)); 743 unp->unp_flags |= UNP_HAVEPC; 744 745 so2 = so3; 746 } 747 error = unp_connect2(so, so2); 748 bad: 749 vput(vp); 750 return (error); 751 } 752 753 int 754 unp_connect2(struct socket *so, struct socket *so2) 755 { 756 struct unpcb *unp = so->so_pcb; 757 struct unpcb *unp2; 758 759 if (so2->so_type != so->so_type) 760 return (EPROTOTYPE); 761 unp2 = so2->so_pcb; 762 unp->unp_conn = unp2; 763 switch (so->so_type) { 764 765 case SOCK_DGRAM: 766 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 767 soisconnected(so); 768 break; 769 770 case SOCK_STREAM: 771 case SOCK_SEQPACKET: 772 unp2->unp_conn = unp; 773 soisconnected(so); 774 soisconnected(so2); 775 break; 776 777 default: 778 panic("unp_connect2"); 779 } 780 return (0); 781 } 782 783 static void 784 unp_disconnect(struct unpcb *unp) 785 { 786 struct unpcb *unp2 = unp->unp_conn; 787 788 if (unp2 == NULL) 789 return; 790 791 unp->unp_conn = NULL; 792 793 switch (unp->unp_socket->so_type) { 794 case SOCK_DGRAM: 795 LIST_REMOVE(unp, unp_reflink); 796 unp->unp_socket->so_state &= ~SS_ISCONNECTED; 797 break; 798 case SOCK_STREAM: 799 case SOCK_SEQPACKET: 800 soisdisconnected(unp->unp_socket); 801 unp2->unp_conn = NULL; 802 soisdisconnected(unp2->unp_socket); 803 break; 804 } 805 } 806 807 #ifdef notdef 808 void 809 unp_abort(struct unpcb *unp) 810 { 811 812 unp_detach(unp); 813 } 814 #endif 815 816 static int 817 prison_unpcb(struct thread *td, struct unpcb *unp) 818 { 819 struct proc *p; 820 821 if (td == NULL) 822 return (0); 823 if ((p = td->td_proc) == NULL) 824 return (0); 825 if (!p->p_ucred->cr_prison) 826 return (0); 827 if (p->p_fd->fd_rdir == unp->unp_rvnode) 828 return (0); 829 return (1); 830 } 831 832 static int 833 unp_pcblist(SYSCTL_HANDLER_ARGS) 834 { 835 int error, i, n; 836 struct unpcb *unp, **unp_list; 837 unp_gen_t gencnt; 838 struct unp_head *head; 839 840 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead); 841 842 KKASSERT(curproc != NULL); 843 844 /* 845 * The process of preparing the PCB list is too time-consuming and 846 * resource-intensive to repeat twice on every request. 847 */ 848 if (req->oldptr == NULL) { 849 n = unp_count; 850 req->oldidx = (n + n/8) * sizeof(struct xunpcb); 851 return 0; 852 } 853 854 if (req->newptr != NULL) 855 return EPERM; 856 857 /* 858 * OK, now we're committed to doing something. 859 */ 860 gencnt = unp_gencnt; 861 n = unp_count; 862 863 unp_list = kmalloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 864 865 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 866 unp = LIST_NEXT(unp, unp_link)) { 867 if (unp->unp_gencnt <= gencnt && !prison_unpcb(req->td, unp)) 868 unp_list[i++] = unp; 869 } 870 n = i; /* in case we lost some during malloc */ 871 872 error = 0; 873 for (i = 0; i < n; i++) { 874 unp = unp_list[i]; 875 if (unp->unp_gencnt <= gencnt) { 876 struct xunpcb xu; 877 xu.xu_len = sizeof xu; 878 xu.xu_unpp = unp; 879 /* 880 * XXX - need more locking here to protect against 881 * connect/disconnect races for SMP. 882 */ 883 if (unp->unp_addr) 884 bcopy(unp->unp_addr, &xu.xu_addr, 885 unp->unp_addr->sun_len); 886 if (unp->unp_conn && unp->unp_conn->unp_addr) 887 bcopy(unp->unp_conn->unp_addr, 888 &xu.xu_caddr, 889 unp->unp_conn->unp_addr->sun_len); 890 bcopy(unp, &xu.xu_unp, sizeof *unp); 891 sotoxsocket(unp->unp_socket, &xu.xu_socket); 892 error = SYSCTL_OUT(req, &xu, sizeof xu); 893 } 894 } 895 kfree(unp_list, M_TEMP); 896 return error; 897 } 898 899 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD, 900 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 901 "List of active local datagram sockets"); 902 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD, 903 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 904 "List of active local stream sockets"); 905 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist, CTLFLAG_RD, 906 (caddr_t)(long)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb", 907 "List of active local seqpacket stream sockets"); 908 909 static void 910 unp_shutdown(struct unpcb *unp) 911 { 912 struct socket *so; 913 914 if ((unp->unp_socket->so_type == SOCK_STREAM || 915 unp->unp_socket->so_type == SOCK_SEQPACKET) && 916 unp->unp_conn != NULL && (so = unp->unp_conn->unp_socket)) { 917 socantrcvmore(so); 918 } 919 } 920 921 static void 922 unp_drop(struct unpcb *unp, int err) 923 { 924 struct socket *so = unp->unp_socket; 925 926 so->so_error = err; 927 unp_disconnect(unp); 928 } 929 930 #ifdef notdef 931 void 932 unp_drain(void) 933 { 934 935 } 936 #endif 937 938 int 939 unp_externalize(struct mbuf *rights) 940 { 941 struct thread *td = curthread; 942 struct proc *p = td->td_proc; /* XXX */ 943 struct lwp *lp = td->td_lwp; 944 struct cmsghdr *cm = mtod(rights, struct cmsghdr *); 945 int *fdp; 946 int i; 947 struct file **rp; 948 struct file *fp; 949 int newfds = (cm->cmsg_len - (CMSG_DATA(cm) - (u_char *)cm)) 950 / sizeof (struct file *); 951 int f; 952 953 /* 954 * if the new FD's will not fit, then we free them all 955 */ 956 if (!fdavail(p, newfds)) { 957 rp = (struct file **)CMSG_DATA(cm); 958 for (i = 0; i < newfds; i++) { 959 fp = *rp; 960 /* 961 * zero the pointer before calling unp_discard, 962 * since it may end up in unp_gc().. 963 */ 964 *rp++ = 0; 965 unp_discard(fp, NULL); 966 } 967 return (EMSGSIZE); 968 } 969 970 /* 971 * now change each pointer to an fd in the global table to 972 * an integer that is the index to the local fd table entry 973 * that we set up to point to the global one we are transferring. 974 * If sizeof (struct file *) is bigger than or equal to sizeof int, 975 * then do it in forward order. In that case, an integer will 976 * always come in the same place or before its corresponding 977 * struct file pointer. 978 * If sizeof (struct file *) is smaller than sizeof int, then 979 * do it in reverse order. 980 */ 981 if (sizeof (struct file *) >= sizeof (int)) { 982 fdp = (int *)CMSG_DATA(cm); 983 rp = (struct file **)CMSG_DATA(cm); 984 for (i = 0; i < newfds; i++) { 985 if (fdalloc(p, 0, &f)) 986 panic("unp_externalize"); 987 fp = *rp++; 988 unp_fp_externalize(lp, fp, f); 989 *fdp++ = f; 990 } 991 } else { 992 fdp = (int *)CMSG_DATA(cm) + newfds - 1; 993 rp = (struct file **)CMSG_DATA(cm) + newfds - 1; 994 for (i = 0; i < newfds; i++) { 995 if (fdalloc(p, 0, &f)) 996 panic("unp_externalize"); 997 fp = *rp--; 998 unp_fp_externalize(lp, fp, f); 999 *fdp-- = f; 1000 } 1001 } 1002 1003 /* 1004 * Adjust length, in case sizeof(struct file *) and sizeof(int) 1005 * differs. 1006 */ 1007 cm->cmsg_len = CMSG_LEN(newfds * sizeof(int)); 1008 rights->m_len = cm->cmsg_len; 1009 return (0); 1010 } 1011 1012 static void 1013 unp_fp_externalize(struct lwp *lp, struct file *fp, int fd) 1014 { 1015 struct file *fx; 1016 int error; 1017 1018 if (lp) { 1019 KKASSERT(fd >= 0); 1020 if (fp->f_flag & FREVOKED) { 1021 kprintf("Warning: revoked fp exiting unix socket\n"); 1022 fx = NULL; 1023 error = falloc(lp, &fx, NULL); 1024 if (error == 0) 1025 fsetfd(lp->lwp_proc->p_fd, fx, fd); 1026 else 1027 fsetfd(lp->lwp_proc->p_fd, NULL, fd); 1028 fdrop(fx); 1029 } else { 1030 fsetfd(lp->lwp_proc->p_fd, fp, fd); 1031 } 1032 } 1033 spin_lock(&unp_spin); 1034 fp->f_msgcount--; 1035 unp_rights--; 1036 spin_unlock(&unp_spin); 1037 fdrop(fp); 1038 } 1039 1040 1041 void 1042 unp_init(void) 1043 { 1044 LIST_INIT(&unp_dhead); 1045 LIST_INIT(&unp_shead); 1046 spin_init(&unp_spin); 1047 } 1048 1049 static int 1050 unp_internalize(struct mbuf *control, struct thread *td) 1051 { 1052 struct proc *p = td->td_proc; 1053 struct filedesc *fdescp; 1054 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1055 struct file **rp; 1056 struct file *fp; 1057 int i, fd, *fdp; 1058 struct cmsgcred *cmcred; 1059 int oldfds; 1060 u_int newlen; 1061 1062 KKASSERT(p); 1063 fdescp = p->p_fd; 1064 if ((cm->cmsg_type != SCM_RIGHTS && cm->cmsg_type != SCM_CREDS) || 1065 cm->cmsg_level != SOL_SOCKET || 1066 CMSG_ALIGN(cm->cmsg_len) != control->m_len) { 1067 return (EINVAL); 1068 } 1069 1070 /* 1071 * Fill in credential information. 1072 */ 1073 if (cm->cmsg_type == SCM_CREDS) { 1074 cmcred = (struct cmsgcred *)CMSG_DATA(cm); 1075 cmcred->cmcred_pid = p->p_pid; 1076 cmcred->cmcred_uid = p->p_ucred->cr_ruid; 1077 cmcred->cmcred_gid = p->p_ucred->cr_rgid; 1078 cmcred->cmcred_euid = p->p_ucred->cr_uid; 1079 cmcred->cmcred_ngroups = MIN(p->p_ucred->cr_ngroups, 1080 CMGROUP_MAX); 1081 for (i = 0; i < cmcred->cmcred_ngroups; i++) 1082 cmcred->cmcred_groups[i] = p->p_ucred->cr_groups[i]; 1083 return(0); 1084 } 1085 1086 /* 1087 * cmsghdr may not be aligned, do not allow calculation(s) to 1088 * go negative. 1089 */ 1090 if (cm->cmsg_len < CMSG_LEN(0)) 1091 return(EINVAL); 1092 1093 oldfds = (cm->cmsg_len - CMSG_LEN(0)) / sizeof (int); 1094 1095 /* 1096 * check that all the FDs passed in refer to legal OPEN files 1097 * If not, reject the entire operation. 1098 */ 1099 fdp = (int *)CMSG_DATA(cm); 1100 for (i = 0; i < oldfds; i++) { 1101 fd = *fdp++; 1102 if ((unsigned)fd >= fdescp->fd_nfiles || 1103 fdescp->fd_files[fd].fp == NULL) 1104 return (EBADF); 1105 if (fdescp->fd_files[fd].fp->f_type == DTYPE_KQUEUE) 1106 return (EOPNOTSUPP); 1107 } 1108 /* 1109 * Now replace the integer FDs with pointers to 1110 * the associated global file table entry.. 1111 * Allocate a bigger buffer as necessary. But if an cluster is not 1112 * enough, return E2BIG. 1113 */ 1114 newlen = CMSG_LEN(oldfds * sizeof(struct file *)); 1115 if (newlen > MCLBYTES) 1116 return (E2BIG); 1117 if (newlen - control->m_len > M_TRAILINGSPACE(control)) { 1118 if (control->m_flags & M_EXT) 1119 return (E2BIG); 1120 MCLGET(control, MB_WAIT); 1121 if (!(control->m_flags & M_EXT)) 1122 return (ENOBUFS); 1123 1124 /* copy the data to the cluster */ 1125 memcpy(mtod(control, char *), cm, cm->cmsg_len); 1126 cm = mtod(control, struct cmsghdr *); 1127 } 1128 1129 /* 1130 * Adjust length, in case sizeof(struct file *) and sizeof(int) 1131 * differs. 1132 */ 1133 cm->cmsg_len = newlen; 1134 control->m_len = CMSG_ALIGN(newlen); 1135 1136 /* 1137 * Transform the file descriptors into struct file pointers. 1138 * If sizeof (struct file *) is bigger than or equal to sizeof int, 1139 * then do it in reverse order so that the int won't get until 1140 * we're done. 1141 * If sizeof (struct file *) is smaller than sizeof int, then 1142 * do it in forward order. 1143 */ 1144 if (sizeof (struct file *) >= sizeof (int)) { 1145 fdp = (int *)CMSG_DATA(cm) + oldfds - 1; 1146 rp = (struct file **)CMSG_DATA(cm) + oldfds - 1; 1147 for (i = 0; i < oldfds; i++) { 1148 fp = fdescp->fd_files[*fdp--].fp; 1149 *rp-- = fp; 1150 fhold(fp); 1151 spin_lock(&unp_spin); 1152 fp->f_msgcount++; 1153 unp_rights++; 1154 spin_unlock(&unp_spin); 1155 } 1156 } else { 1157 fdp = (int *)CMSG_DATA(cm); 1158 rp = (struct file **)CMSG_DATA(cm); 1159 for (i = 0; i < oldfds; i++) { 1160 fp = fdescp->fd_files[*fdp++].fp; 1161 *rp++ = fp; 1162 fhold(fp); 1163 spin_lock(&unp_spin); 1164 fp->f_msgcount++; 1165 unp_rights++; 1166 spin_unlock(&unp_spin); 1167 } 1168 } 1169 return (0); 1170 } 1171 1172 /* 1173 * Garbage collect in-transit file descriptors that get lost due to 1174 * loops (i.e. when a socket is sent to another process over itself, 1175 * and more complex situations). 1176 * 1177 * NOT MPSAFE - TODO socket flush code and maybe closef. Rest is MPSAFE. 1178 */ 1179 1180 struct unp_gc_info { 1181 struct file **extra_ref; 1182 struct file *locked_fp; 1183 int defer; 1184 int index; 1185 int maxindex; 1186 }; 1187 1188 static void 1189 unp_gc(void) 1190 { 1191 struct unp_gc_info info; 1192 static boolean_t unp_gcing; 1193 struct file **fpp; 1194 int i; 1195 1196 spin_lock(&unp_spin); 1197 if (unp_gcing) { 1198 spin_unlock(&unp_spin); 1199 return; 1200 } 1201 unp_gcing = TRUE; 1202 spin_unlock(&unp_spin); 1203 1204 /* 1205 * before going through all this, set all FDs to 1206 * be NOT defered and NOT externally accessible 1207 */ 1208 info.defer = 0; 1209 allfiles_scan_exclusive(unp_gc_clearmarks, NULL); 1210 do { 1211 allfiles_scan_exclusive(unp_gc_checkmarks, &info); 1212 } while (info.defer); 1213 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 info.extra_ref = kmalloc(256 * sizeof(struct file *), M_FILE, M_WAITOK); 1254 info.maxindex = 256; 1255 1256 do { 1257 /* 1258 * Look for matches 1259 */ 1260 info.index = 0; 1261 allfiles_scan_exclusive(unp_gc_checkrefs, &info); 1262 1263 /* 1264 * For each FD on our hit list, do the following two things 1265 */ 1266 for (i = info.index, fpp = info.extra_ref; --i >= 0; ++fpp) { 1267 struct file *tfp = *fpp; 1268 if (tfp->f_type == DTYPE_SOCKET && tfp->f_data != NULL) 1269 sorflush((struct socket *)(tfp->f_data)); 1270 } 1271 for (i = info.index, fpp = info.extra_ref; --i >= 0; ++fpp) 1272 closef(*fpp, NULL); 1273 } while (info.index == info.maxindex); 1274 kfree((caddr_t)info.extra_ref, M_FILE); 1275 unp_gcing = FALSE; 1276 } 1277 1278 /* 1279 * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry 1280 */ 1281 static int 1282 unp_gc_checkrefs(struct file *fp, void *data) 1283 { 1284 struct unp_gc_info *info = data; 1285 1286 if (fp->f_count == 0) 1287 return(0); 1288 if (info->index == info->maxindex) 1289 return(-1); 1290 1291 /* 1292 * If all refs are from msgs, and it's not marked accessible 1293 * then it must be referenced from some unreachable cycle 1294 * of (shut-down) FDs, so include it in our 1295 * list of FDs to remove 1296 */ 1297 if (fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) { 1298 info->extra_ref[info->index++] = fp; 1299 fhold(fp); 1300 } 1301 return(0); 1302 } 1303 1304 /* 1305 * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry 1306 */ 1307 static int 1308 unp_gc_clearmarks(struct file *fp, void *data __unused) 1309 { 1310 atomic_clear_int(&fp->f_flag, FMARK | FDEFER); 1311 return(0); 1312 } 1313 1314 /* 1315 * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry 1316 */ 1317 static int 1318 unp_gc_checkmarks(struct file *fp, void *data) 1319 { 1320 struct unp_gc_info *info = data; 1321 struct socket *so; 1322 1323 /* 1324 * If the file is not open, skip it 1325 */ 1326 if (fp->f_count == 0) 1327 return(0); 1328 /* 1329 * If we already marked it as 'defer' in a 1330 * previous pass, then try process it this time 1331 * and un-mark it 1332 */ 1333 if (fp->f_flag & FDEFER) { 1334 atomic_clear_int(&fp->f_flag, FDEFER); 1335 --info->defer; 1336 } else { 1337 /* 1338 * if it's not defered, then check if it's 1339 * already marked.. if so skip it 1340 */ 1341 if (fp->f_flag & FMARK) 1342 return(0); 1343 /* 1344 * If all references are from messages 1345 * in transit, then skip it. it's not 1346 * externally accessible. 1347 */ 1348 if (fp->f_count == fp->f_msgcount) 1349 return(0); 1350 /* 1351 * If it got this far then it must be 1352 * externally accessible. 1353 */ 1354 atomic_set_int(&fp->f_flag, FMARK); 1355 } 1356 1357 /* 1358 * either it was defered, or it is externally 1359 * accessible and not already marked so. 1360 * Now check if it is possibly one of OUR sockets. 1361 */ 1362 if (fp->f_type != DTYPE_SOCKET || 1363 (so = (struct socket *)fp->f_data) == NULL) 1364 return(0); 1365 if (so->so_proto->pr_domain != &localdomain || 1366 !(so->so_proto->pr_flags & PR_RIGHTS)) 1367 return(0); 1368 #ifdef notdef 1369 if (so->so_rcv.ssb_flags & SSB_LOCK) { 1370 /* 1371 * This is problematical; it's not clear 1372 * we need to wait for the sockbuf to be 1373 * unlocked (on a uniprocessor, at least), 1374 * and it's also not clear what to do 1375 * if sbwait returns an error due to receipt 1376 * of a signal. If sbwait does return 1377 * an error, we'll go into an infinite 1378 * loop. Delete all of this for now. 1379 */ 1380 sbwait(&so->so_rcv); 1381 goto restart; 1382 } 1383 #endif 1384 /* 1385 * So, Ok, it's one of our sockets and it IS externally 1386 * accessible (or was defered). Now we look 1387 * to see if we hold any file descriptors in its 1388 * message buffers. Follow those links and mark them 1389 * as accessible too. 1390 */ 1391 info->locked_fp = fp; 1392 /* spin_lock_wr(&so->so_rcv.sb_spin); */ 1393 unp_scan(so->so_rcv.ssb_mb, unp_mark, info); 1394 /* spin_unlock_wr(&so->so_rcv.sb_spin);*/ 1395 return (0); 1396 } 1397 1398 /* 1399 * Scan all unix domain sockets and replace any revoked file pointers 1400 * found with the dummy file pointer fx. We don't worry about races 1401 * against file pointers being read out as those are handled in the 1402 * externalize code. 1403 */ 1404 1405 #define REVOKE_GC_MAXFILES 32 1406 1407 struct unp_revoke_gc_info { 1408 struct file *fx; 1409 struct file *fary[REVOKE_GC_MAXFILES]; 1410 int fcount; 1411 }; 1412 1413 void 1414 unp_revoke_gc(struct file *fx) 1415 { 1416 struct unp_revoke_gc_info info; 1417 int i; 1418 1419 info.fx = fx; 1420 do { 1421 info.fcount = 0; 1422 allfiles_scan_exclusive(unp_revoke_gc_check, &info); 1423 for (i = 0; i < info.fcount; ++i) 1424 unp_fp_externalize(NULL, info.fary[i], -1); 1425 } while (info.fcount == REVOKE_GC_MAXFILES); 1426 } 1427 1428 /* 1429 * Check for and replace revoked descriptors. 1430 * 1431 * WARNING: This routine is not allowed to block. 1432 */ 1433 static int 1434 unp_revoke_gc_check(struct file *fps, void *vinfo) 1435 { 1436 struct unp_revoke_gc_info *info = vinfo; 1437 struct file *fp; 1438 struct socket *so; 1439 struct mbuf *m0; 1440 struct mbuf *m; 1441 struct file **rp; 1442 struct cmsghdr *cm; 1443 int i; 1444 int qfds; 1445 1446 /* 1447 * Is this a unix domain socket with rights-passing abilities? 1448 */ 1449 if (fps->f_type != DTYPE_SOCKET) 1450 return (0); 1451 if ((so = (struct socket *)fps->f_data) == NULL) 1452 return(0); 1453 if (so->so_proto->pr_domain != &localdomain) 1454 return(0); 1455 if ((so->so_proto->pr_flags & PR_RIGHTS) == 0) 1456 return(0); 1457 1458 /* 1459 * Scan the mbufs for control messages and replace any revoked 1460 * descriptors we find. 1461 */ 1462 m0 = so->so_rcv.ssb_mb; 1463 while (m0) { 1464 for (m = m0; m; m = m->m_next) { 1465 if (m->m_type != MT_CONTROL) 1466 continue; 1467 if (m->m_len < sizeof(*cm)) 1468 continue; 1469 cm = mtod(m, struct cmsghdr *); 1470 if (cm->cmsg_level != SOL_SOCKET || 1471 cm->cmsg_type != SCM_RIGHTS) { 1472 continue; 1473 } 1474 qfds = (cm->cmsg_len - CMSG_LEN(0)) / sizeof(void *); 1475 rp = (struct file **)CMSG_DATA(cm); 1476 for (i = 0; i < qfds; i++) { 1477 fp = rp[i]; 1478 if (fp->f_flag & FREVOKED) { 1479 kprintf("Warning: Removing revoked fp from unix domain socket queue\n"); 1480 fhold(info->fx); 1481 info->fx->f_msgcount++; 1482 unp_rights++; 1483 rp[i] = info->fx; 1484 info->fary[info->fcount++] = fp; 1485 } 1486 if (info->fcount == REVOKE_GC_MAXFILES) 1487 break; 1488 } 1489 if (info->fcount == REVOKE_GC_MAXFILES) 1490 break; 1491 } 1492 m0 = m0->m_nextpkt; 1493 if (info->fcount == REVOKE_GC_MAXFILES) 1494 break; 1495 } 1496 1497 /* 1498 * Stop the scan if we filled up our array. 1499 */ 1500 if (info->fcount == REVOKE_GC_MAXFILES) 1501 return(-1); 1502 return(0); 1503 } 1504 1505 void 1506 unp_dispose(struct mbuf *m) 1507 { 1508 if (m) 1509 unp_scan(m, unp_discard, NULL); 1510 } 1511 1512 static int 1513 unp_listen(struct unpcb *unp, struct thread *td) 1514 { 1515 struct proc *p = td->td_proc; 1516 1517 KKASSERT(p); 1518 cru2x(p->p_ucred, &unp->unp_peercred); 1519 unp->unp_flags |= UNP_HAVEPCCACHED; 1520 return (0); 1521 } 1522 1523 static void 1524 unp_scan(struct mbuf *m0, void (*op)(struct file *, void *), void *data) 1525 { 1526 struct mbuf *m; 1527 struct file **rp; 1528 struct cmsghdr *cm; 1529 int i; 1530 int qfds; 1531 1532 while (m0) { 1533 for (m = m0; m; m = m->m_next) { 1534 if (m->m_type == MT_CONTROL && 1535 m->m_len >= sizeof(*cm)) { 1536 cm = mtod(m, struct cmsghdr *); 1537 if (cm->cmsg_level != SOL_SOCKET || 1538 cm->cmsg_type != SCM_RIGHTS) 1539 continue; 1540 qfds = (cm->cmsg_len - CMSG_LEN(0)) / 1541 sizeof(void *); 1542 rp = (struct file **)CMSG_DATA(cm); 1543 for (i = 0; i < qfds; i++) 1544 (*op)(*rp++, data); 1545 break; /* XXX, but saves time */ 1546 } 1547 } 1548 m0 = m0->m_nextpkt; 1549 } 1550 } 1551 1552 static void 1553 unp_mark(struct file *fp, void *data) 1554 { 1555 struct unp_gc_info *info = data; 1556 1557 if ((fp->f_flag & FMARK) == 0) { 1558 ++info->defer; 1559 atomic_set_int(&fp->f_flag, FMARK | FDEFER); 1560 } 1561 } 1562 1563 static void 1564 unp_discard(struct file *fp, void *data __unused) 1565 { 1566 spin_lock(&unp_spin); 1567 fp->f_msgcount--; 1568 unp_rights--; 1569 spin_unlock(&unp_spin); 1570 closef(fp, NULL); 1571 } 1572 1573