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 1.26 2006/05/26 02:26:01 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 #include <vm/vm_zone.h> 63 64 static struct vm_zone *unp_zone; 65 static unp_gen_t unp_gencnt; 66 static u_int unp_count; 67 68 static struct unp_head unp_shead, unp_dhead; 69 70 /* 71 * Unix communications domain. 72 * 73 * TODO: 74 * SEQPACKET, RDM 75 * rethink name space problems 76 * need a proper out-of-band 77 * lock pushdown 78 */ 79 static struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL }; 80 static ino_t unp_ino; /* prototype for fake inode numbers */ 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 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 *); 100 101 static int 102 uipc_abort(struct socket *so) 103 { 104 struct unpcb *unp = so->so_pcb; 105 106 if (unp == NULL) 107 return EINVAL; 108 unp_drop(unp, ECONNABORTED); 109 unp_detach(unp); 110 sofree(so); 111 return 0; 112 } 113 114 static int 115 uipc_accept(struct socket *so, struct sockaddr **nam) 116 { 117 struct unpcb *unp = so->so_pcb; 118 119 if (unp == NULL) 120 return EINVAL; 121 122 /* 123 * Pass back name of connected socket, 124 * if it was bound and we are still connected 125 * (our peer may have closed already!). 126 */ 127 if (unp->unp_conn && unp->unp_conn->unp_addr) { 128 *nam = dup_sockaddr((struct sockaddr *)unp->unp_conn->unp_addr); 129 } else { 130 *nam = dup_sockaddr((struct sockaddr *)&sun_noname); 131 } 132 return 0; 133 } 134 135 static int 136 uipc_attach(struct socket *so, int proto, struct pru_attach_info *ai) 137 { 138 struct unpcb *unp = so->so_pcb; 139 140 if (unp != NULL) 141 return EISCONN; 142 return unp_attach(so, ai); 143 } 144 145 static int 146 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 147 { 148 struct unpcb *unp = so->so_pcb; 149 150 if (unp == NULL) 151 return EINVAL; 152 return unp_bind(unp, nam, td); 153 } 154 155 static int 156 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 157 { 158 struct unpcb *unp = so->so_pcb; 159 160 if (unp == NULL) 161 return EINVAL; 162 return unp_connect(so, nam, td); 163 } 164 165 static int 166 uipc_connect2(struct socket *so1, struct socket *so2) 167 { 168 struct unpcb *unp = so1->so_pcb; 169 170 if (unp == NULL) 171 return EINVAL; 172 173 return unp_connect2(so1, so2); 174 } 175 176 /* control is EOPNOTSUPP */ 177 178 static int 179 uipc_detach(struct socket *so) 180 { 181 struct unpcb *unp = so->so_pcb; 182 183 if (unp == NULL) 184 return EINVAL; 185 186 unp_detach(unp); 187 return 0; 188 } 189 190 static int 191 uipc_disconnect(struct socket *so) 192 { 193 struct unpcb *unp = so->so_pcb; 194 195 if (unp == NULL) 196 return EINVAL; 197 unp_disconnect(unp); 198 return 0; 199 } 200 201 static int 202 uipc_listen(struct socket *so, struct thread *td) 203 { 204 struct unpcb *unp = so->so_pcb; 205 206 if (unp == NULL || unp->unp_vnode == NULL) 207 return EINVAL; 208 return unp_listen(unp, td); 209 } 210 211 static int 212 uipc_peeraddr(struct socket *so, struct sockaddr **nam) 213 { 214 struct unpcb *unp = so->so_pcb; 215 216 if (unp == NULL) 217 return EINVAL; 218 if (unp->unp_conn && unp->unp_conn->unp_addr) 219 *nam = dup_sockaddr((struct sockaddr *)unp->unp_conn->unp_addr); 220 else { 221 /* 222 * XXX: It seems that this test always fails even when 223 * connection is established. So, this else clause is 224 * added as workaround to return PF_LOCAL sockaddr. 225 */ 226 *nam = dup_sockaddr((struct sockaddr *)&sun_noname); 227 } 228 return 0; 229 } 230 231 static int 232 uipc_rcvd(struct socket *so, int flags) 233 { 234 struct unpcb *unp = so->so_pcb; 235 struct socket *so2; 236 u_long newhiwat; 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 if (unp->unp_conn == NULL) 247 break; 248 so2 = unp->unp_conn->unp_socket; 249 /* 250 * Adjust backpressure on sender 251 * and wakeup any waiting to write. 252 */ 253 so2->so_snd.sb_mbmax += unp->unp_mbcnt - so->so_rcv.sb_mbcnt; 254 unp->unp_mbcnt = so->so_rcv.sb_mbcnt; 255 newhiwat = 256 so2->so_snd.sb_hiwat + unp->unp_cc - so->so_rcv.sb_cc; 257 chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat, 258 newhiwat, RLIM_INFINITY); 259 unp->unp_cc = so->so_rcv.sb_cc; 260 sowwakeup(so2); 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 u_long newhiwat; 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 (sbappendaddr(&so2->so_rcv, from, m, control)) { 317 sorwakeup(so2); 318 m = NULL; 319 control = NULL; 320 } else 321 error = ENOBUFS; 322 if (nam) 323 unp_disconnect(unp); 324 break; 325 } 326 327 case SOCK_STREAM: 328 /* Connect if not connected yet. */ 329 /* 330 * Note: A better implementation would complain 331 * if not equal to the peer's address. 332 */ 333 if (!(so->so_state & SS_ISCONNECTED)) { 334 if (nam) { 335 error = unp_connect(so, nam, td); 336 if (error) 337 break; /* XXX */ 338 } else { 339 error = ENOTCONN; 340 break; 341 } 342 } 343 344 if (so->so_state & SS_CANTSENDMORE) { 345 error = EPIPE; 346 break; 347 } 348 if (unp->unp_conn == NULL) 349 panic("uipc_send connected but no connection?"); 350 so2 = unp->unp_conn->unp_socket; 351 /* 352 * Send to paired receive port, and then reduce 353 * send buffer hiwater marks to maintain backpressure. 354 * Wake up readers. 355 */ 356 if (control) { 357 if (sbappendcontrol(&so2->so_rcv, m, control)) 358 control = NULL; 359 } else 360 sbappend(&so2->so_rcv, m); 361 so->so_snd.sb_mbmax -= 362 so2->so_rcv.sb_mbcnt - unp->unp_conn->unp_mbcnt; 363 unp->unp_conn->unp_mbcnt = so2->so_rcv.sb_mbcnt; 364 newhiwat = so->so_snd.sb_hiwat - 365 (so2->so_rcv.sb_cc - unp->unp_conn->unp_cc); 366 chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat, 367 newhiwat, RLIM_INFINITY); 368 unp->unp_conn->unp_cc = so2->so_rcv.sb_cc; 369 sorwakeup(so2); 370 m = NULL; 371 break; 372 373 default: 374 panic("uipc_send unknown socktype"); 375 } 376 377 /* 378 * SEND_EOF is equivalent to a SEND followed by a SHUTDOWN. 379 */ 380 if (flags & PRUS_EOF) { 381 socantsendmore(so); 382 unp_shutdown(unp); 383 } 384 385 if (control && error != 0) 386 unp_dispose(control); 387 388 release: 389 if (control) 390 m_freem(control); 391 if (m) 392 m_freem(m); 393 return error; 394 } 395 396 static int 397 uipc_sense(struct socket *so, struct stat *sb) 398 { 399 struct unpcb *unp = so->so_pcb; 400 struct socket *so2; 401 402 if (unp == NULL) 403 return EINVAL; 404 sb->st_blksize = so->so_snd.sb_hiwat; 405 if (so->so_type == SOCK_STREAM && unp->unp_conn != NULL) { 406 so2 = unp->unp_conn->unp_socket; 407 sb->st_blksize += so2->so_rcv.sb_cc; 408 } 409 sb->st_dev = NOUDEV; 410 if (unp->unp_ino == 0) /* make up a non-zero inode number */ 411 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino; 412 sb->st_ino = unp->unp_ino; 413 return (0); 414 } 415 416 static int 417 uipc_shutdown(struct socket *so) 418 { 419 struct unpcb *unp = so->so_pcb; 420 421 if (unp == NULL) 422 return EINVAL; 423 socantsendmore(so); 424 unp_shutdown(unp); 425 return 0; 426 } 427 428 static int 429 uipc_sockaddr(struct socket *so, struct sockaddr **nam) 430 { 431 struct unpcb *unp = so->so_pcb; 432 433 if (unp == NULL) 434 return EINVAL; 435 if (unp->unp_addr) 436 *nam = dup_sockaddr((struct sockaddr *)unp->unp_addr); 437 return 0; 438 } 439 440 struct pr_usrreqs uipc_usrreqs = { 441 uipc_abort, uipc_accept, uipc_attach, uipc_bind, uipc_connect, 442 uipc_connect2, pru_control_notsupp, uipc_detach, uipc_disconnect, 443 uipc_listen, uipc_peeraddr, uipc_rcvd, pru_rcvoob_notsupp, 444 uipc_send, uipc_sense, uipc_shutdown, uipc_sockaddr, 445 sosend, soreceive, sopoll 446 }; 447 448 int 449 uipc_ctloutput(struct socket *so, struct sockopt *sopt) 450 { 451 struct unpcb *unp = so->so_pcb; 452 int error; 453 454 switch (sopt->sopt_dir) { 455 case SOPT_GET: 456 switch (sopt->sopt_name) { 457 case LOCAL_PEERCRED: 458 if (unp->unp_flags & UNP_HAVEPC) 459 error = sooptcopyout(sopt, &unp->unp_peercred, 460 sizeof(unp->unp_peercred)); 461 else { 462 if (so->so_type == SOCK_STREAM) 463 error = ENOTCONN; 464 else 465 error = EINVAL; 466 } 467 break; 468 default: 469 error = EOPNOTSUPP; 470 break; 471 } 472 break; 473 case SOPT_SET: 474 default: 475 error = EOPNOTSUPP; 476 break; 477 } 478 return (error); 479 } 480 481 /* 482 * Both send and receive buffers are allocated PIPSIZ bytes of buffering 483 * for stream sockets, although the total for sender and receiver is 484 * actually only PIPSIZ. 485 * Datagram sockets really use the sendspace as the maximum datagram size, 486 * and don't really want to reserve the sendspace. Their recvspace should 487 * be large enough for at least one max-size datagram plus address. 488 */ 489 #ifndef PIPSIZ 490 #define PIPSIZ 8192 491 #endif 492 static u_long unpst_sendspace = PIPSIZ; 493 static u_long unpst_recvspace = PIPSIZ; 494 static u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 495 static u_long unpdg_recvspace = 4*1024; 496 497 static int unp_rights; /* file descriptors in flight */ 498 static struct spinlock unp_spin = SPINLOCK_INITIALIZER(&unp_spin); 499 500 SYSCTL_DECL(_net_local_stream); 501 SYSCTL_INT(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 502 &unpst_sendspace, 0, ""); 503 SYSCTL_INT(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 504 &unpst_recvspace, 0, ""); 505 506 SYSCTL_DECL(_net_local_dgram); 507 SYSCTL_INT(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 508 &unpdg_sendspace, 0, ""); 509 SYSCTL_INT(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 510 &unpdg_recvspace, 0, ""); 511 512 SYSCTL_DECL(_net_local); 513 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, ""); 514 515 static int 516 unp_attach(struct socket *so, struct pru_attach_info *ai) 517 { 518 struct unpcb *unp; 519 int error; 520 521 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 522 switch (so->so_type) { 523 524 case SOCK_STREAM: 525 error = soreserve(so, unpst_sendspace, unpst_recvspace, 526 ai->sb_rlimit); 527 break; 528 529 case SOCK_DGRAM: 530 error = soreserve(so, unpdg_sendspace, unpdg_recvspace, 531 ai->sb_rlimit); 532 break; 533 534 default: 535 panic("unp_attach"); 536 } 537 if (error) 538 return (error); 539 } 540 unp = zalloc(unp_zone); 541 if (unp == NULL) 542 return (ENOBUFS); 543 bzero(unp, sizeof *unp); 544 unp->unp_gencnt = ++unp_gencnt; 545 unp_count++; 546 LIST_INIT(&unp->unp_refs); 547 unp->unp_socket = so; 548 unp->unp_rvnode = ai->fd_rdir; /* jail cruft XXX JH */ 549 LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead 550 : &unp_shead, unp, unp_link); 551 so->so_pcb = (caddr_t)unp; 552 return (0); 553 } 554 555 static void 556 unp_detach(struct unpcb *unp) 557 { 558 LIST_REMOVE(unp, unp_link); 559 unp->unp_gencnt = ++unp_gencnt; 560 --unp_count; 561 if (unp->unp_vnode) { 562 unp->unp_vnode->v_socket = NULL; 563 vrele(unp->unp_vnode); 564 unp->unp_vnode = NULL; 565 } 566 if (unp->unp_conn) 567 unp_disconnect(unp); 568 while (!LIST_EMPTY(&unp->unp_refs)) 569 unp_drop(LIST_FIRST(&unp->unp_refs), ECONNRESET); 570 soisdisconnected(unp->unp_socket); 571 unp->unp_socket->so_pcb = NULL; 572 if (unp_rights) { 573 /* 574 * Normally the receive buffer is flushed later, 575 * in sofree, but if our receive buffer holds references 576 * to descriptors that are now garbage, we will dispose 577 * of those descriptor references after the garbage collector 578 * gets them (resulting in a "panic: closef: count < 0"). 579 */ 580 sorflush(unp->unp_socket); 581 unp_gc(); 582 } 583 if (unp->unp_addr) 584 free(unp->unp_addr, M_SONAME); 585 zfree(unp_zone, unp); 586 } 587 588 static int 589 unp_bind(struct unpcb *unp, struct sockaddr *nam, struct thread *td) 590 { 591 struct proc *p = td->td_proc; 592 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 593 struct vnode *vp; 594 struct vattr vattr; 595 int error, namelen; 596 struct nlookupdata nd; 597 char buf[SOCK_MAXADDRLEN]; 598 599 if (unp->unp_vnode != NULL) 600 return (EINVAL); 601 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 602 if (namelen <= 0) 603 return (EINVAL); 604 strncpy(buf, soun->sun_path, namelen); 605 buf[namelen] = 0; /* null-terminate the string */ 606 error = nlookup_init(&nd, buf, UIO_SYSSPACE, NLC_LOCKVP|NLC_CREATE); 607 if (error == 0) 608 error = nlookup(&nd); 609 if (error == 0 && nd.nl_ncp->nc_vp != NULL) 610 error = EADDRINUSE; 611 if (error) 612 goto done; 613 614 VATTR_NULL(&vattr); 615 vattr.va_type = VSOCK; 616 vattr.va_mode = (ACCESSPERMS & ~p->p_fd->fd_cmask); 617 error = VOP_NCREATE(nd.nl_ncp, &vp, nd.nl_cred, &vattr); 618 if (error == 0) { 619 vp->v_socket = unp->unp_socket; 620 unp->unp_vnode = vp; 621 unp->unp_addr = (struct sockaddr_un *)dup_sockaddr(nam); 622 VOP_UNLOCK(vp, 0); 623 } 624 done: 625 nlookup_done(&nd); 626 return (error); 627 } 628 629 static int 630 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 631 { 632 struct proc *p = td->td_proc; 633 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 634 struct vnode *vp; 635 struct socket *so2, *so3; 636 struct unpcb *unp, *unp2, *unp3; 637 int error, len; 638 struct nlookupdata nd; 639 char buf[SOCK_MAXADDRLEN]; 640 641 KKASSERT(p); 642 643 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 644 if (len <= 0) 645 return EINVAL; 646 strncpy(buf, soun->sun_path, len); 647 buf[len] = 0; 648 649 vp = NULL; 650 error = nlookup_init(&nd, buf, UIO_SYSSPACE, NLC_FOLLOW); 651 if (error == 0) 652 error = nlookup(&nd); 653 if (error == 0) 654 error = cache_vget(nd.nl_ncp, nd.nl_cred, LK_EXCLUSIVE, &vp); 655 nlookup_done(&nd); 656 if (error) 657 return (error); 658 659 if (vp->v_type != VSOCK) { 660 error = ENOTSOCK; 661 goto bad; 662 } 663 error = VOP_ACCESS(vp, VWRITE, p->p_ucred); 664 if (error) 665 goto bad; 666 so2 = vp->v_socket; 667 if (so2 == NULL) { 668 error = ECONNREFUSED; 669 goto bad; 670 } 671 if (so->so_type != so2->so_type) { 672 error = EPROTOTYPE; 673 goto bad; 674 } 675 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 676 if (!(so2->so_options & SO_ACCEPTCONN) || 677 (so3 = sonewconn(so2, 0)) == NULL) { 678 error = ECONNREFUSED; 679 goto bad; 680 } 681 unp = so->so_pcb; 682 unp2 = so2->so_pcb; 683 unp3 = so3->so_pcb; 684 if (unp2->unp_addr) 685 unp3->unp_addr = (struct sockaddr_un *) 686 dup_sockaddr((struct sockaddr *)unp2->unp_addr); 687 688 /* 689 * unp_peercred management: 690 * 691 * The connecter's (client's) credentials are copied 692 * from its process structure at the time of connect() 693 * (which is now). 694 */ 695 cru2x(p->p_ucred, &unp3->unp_peercred); 696 unp3->unp_flags |= UNP_HAVEPC; 697 /* 698 * The receiver's (server's) credentials are copied 699 * from the unp_peercred member of socket on which the 700 * former called listen(); unp_listen() cached that 701 * process's credentials at that time so we can use 702 * them now. 703 */ 704 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED, 705 ("unp_connect: listener without cached peercred")); 706 memcpy(&unp->unp_peercred, &unp2->unp_peercred, 707 sizeof(unp->unp_peercred)); 708 unp->unp_flags |= UNP_HAVEPC; 709 710 so2 = so3; 711 } 712 error = unp_connect2(so, so2); 713 bad: 714 vput(vp); 715 return (error); 716 } 717 718 int 719 unp_connect2(struct socket *so, struct socket *so2) 720 { 721 struct unpcb *unp = so->so_pcb; 722 struct unpcb *unp2; 723 724 if (so2->so_type != so->so_type) 725 return (EPROTOTYPE); 726 unp2 = so2->so_pcb; 727 unp->unp_conn = unp2; 728 switch (so->so_type) { 729 730 case SOCK_DGRAM: 731 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 732 soisconnected(so); 733 break; 734 735 case SOCK_STREAM: 736 unp2->unp_conn = unp; 737 soisconnected(so); 738 soisconnected(so2); 739 break; 740 741 default: 742 panic("unp_connect2"); 743 } 744 return (0); 745 } 746 747 static void 748 unp_disconnect(struct unpcb *unp) 749 { 750 struct unpcb *unp2 = unp->unp_conn; 751 752 if (unp2 == NULL) 753 return; 754 755 unp->unp_conn = NULL; 756 757 switch (unp->unp_socket->so_type) { 758 case SOCK_DGRAM: 759 LIST_REMOVE(unp, unp_reflink); 760 unp->unp_socket->so_state &= ~SS_ISCONNECTED; 761 break; 762 case SOCK_STREAM: 763 soisdisconnected(unp->unp_socket); 764 unp2->unp_conn = NULL; 765 soisdisconnected(unp2->unp_socket); 766 break; 767 } 768 } 769 770 #ifdef notdef 771 void 772 unp_abort(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 == NULL) { 812 n = unp_count; 813 req->oldidx = (n + n/8) * sizeof(struct xunpcb); 814 return 0; 815 } 816 817 if (req->newptr != NULL) 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 == NULL) 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(struct unpcb *unp) 873 { 874 struct socket *so; 875 876 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn != NULL && 877 (so = unp->unp_conn->unp_socket)) 878 socantrcvmore(so); 879 } 880 881 static void 882 unp_drop(struct unpcb *unp, int errno) 883 { 884 struct socket *so = unp->unp_socket; 885 886 so->so_error = errno; 887 unp_disconnect(unp); 888 } 889 890 #ifdef notdef 891 void 892 unp_drain() 893 { 894 895 } 896 #endif 897 898 int 899 unp_externalize(struct mbuf *rights) 900 { 901 struct proc *p = curproc; /* XXX */ 902 int i; 903 struct cmsghdr *cm = mtod(rights, struct cmsghdr *); 904 int *fdp; 905 struct file **rp; 906 struct file *fp; 907 int newfds = (cm->cmsg_len - (CMSG_DATA(cm) - (u_char *)cm)) 908 / sizeof (struct file *); 909 int f; 910 911 /* 912 * if the new FD's will not fit, then we free them all 913 */ 914 if (!fdavail(p, newfds)) { 915 rp = (struct file **)CMSG_DATA(cm); 916 for (i = 0; i < newfds; i++) { 917 fp = *rp; 918 /* 919 * zero the pointer before calling unp_discard, 920 * since it may end up in unp_gc().. 921 */ 922 *rp++ = 0; 923 unp_discard(fp, NULL); 924 } 925 return (EMSGSIZE); 926 } 927 /* 928 * now change each pointer to an fd in the global table to 929 * an integer that is the index to the local fd table entry 930 * that we set up to point to the global one we are transferring. 931 * If sizeof (struct file *) is bigger than or equal to sizeof int, 932 * then do it in forward order. In that case, an integer will 933 * always come in the same place or before its corresponding 934 * struct file pointer. 935 * If sizeof (struct file *) is smaller than sizeof int, then 936 * do it in reverse order. 937 */ 938 if (sizeof (struct file *) >= sizeof (int)) { 939 fdp = (int *)(cm + 1); 940 rp = (struct file **)CMSG_DATA(cm); 941 for (i = 0; i < newfds; i++) { 942 if (fdalloc(p, 0, &f)) 943 panic("unp_externalize"); 944 fp = *rp++; 945 fsetfd(p, fp, f); 946 fdrop(fp); 947 spin_lock_wr(&unp_spin); 948 fp->f_msgcount--; 949 unp_rights--; 950 spin_unlock_wr(&unp_spin); 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 fsetfd(p, fp, f); 961 fdrop(fp); 962 spin_lock_wr(&unp_spin); 963 fp->f_msgcount--; 964 unp_rights--; 965 spin_unlock_wr(&unp_spin); 966 *fdp-- = f; 967 } 968 } 969 970 /* 971 * Adjust length, in case sizeof(struct file *) and sizeof(int) 972 * differs. 973 */ 974 cm->cmsg_len = CMSG_LEN(newfds * sizeof(int)); 975 rights->m_len = cm->cmsg_len; 976 return (0); 977 } 978 979 void 980 unp_init(void) 981 { 982 unp_zone = zinit("unpcb", sizeof(struct unpcb), nmbclusters, 0, 0); 983 if (unp_zone == NULL) 984 panic("unp_init"); 985 LIST_INIT(&unp_dhead); 986 LIST_INIT(&unp_shead); 987 spin_init(&unp_spin); 988 } 989 990 static int 991 unp_internalize(struct mbuf *control, struct thread *td) 992 { 993 struct proc *p = td->td_proc; 994 struct filedesc *fdescp; 995 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 996 struct file **rp; 997 struct file *fp; 998 int i, fd, *fdp; 999 struct cmsgcred *cmcred; 1000 int oldfds; 1001 u_int newlen; 1002 1003 KKASSERT(p); 1004 fdescp = p->p_fd; 1005 if ((cm->cmsg_type != SCM_RIGHTS && cm->cmsg_type != SCM_CREDS) || 1006 cm->cmsg_level != SOL_SOCKET || cm->cmsg_len != control->m_len) 1007 return (EINVAL); 1008 1009 /* 1010 * Fill in credential information. 1011 */ 1012 if (cm->cmsg_type == SCM_CREDS) { 1013 cmcred = (struct cmsgcred *)(cm + 1); 1014 cmcred->cmcred_pid = p->p_pid; 1015 cmcred->cmcred_uid = p->p_ucred->cr_ruid; 1016 cmcred->cmcred_gid = p->p_ucred->cr_rgid; 1017 cmcred->cmcred_euid = p->p_ucred->cr_uid; 1018 cmcred->cmcred_ngroups = MIN(p->p_ucred->cr_ngroups, 1019 CMGROUP_MAX); 1020 for (i = 0; i < cmcred->cmcred_ngroups; i++) 1021 cmcred->cmcred_groups[i] = p->p_ucred->cr_groups[i]; 1022 return(0); 1023 } 1024 1025 oldfds = (cm->cmsg_len - sizeof (*cm)) / sizeof (int); 1026 /* 1027 * check that all the FDs passed in refer to legal OPEN files 1028 * If not, reject the entire operation. 1029 */ 1030 fdp = (int *)(cm + 1); 1031 for (i = 0; i < oldfds; i++) { 1032 fd = *fdp++; 1033 if ((unsigned)fd >= fdescp->fd_nfiles || 1034 fdescp->fd_files[fd].fp == NULL) 1035 return (EBADF); 1036 if (fdescp->fd_files[fd].fp->f_type == DTYPE_KQUEUE) 1037 return (EOPNOTSUPP); 1038 } 1039 /* 1040 * Now replace the integer FDs with pointers to 1041 * the associated global file table entry.. 1042 * Allocate a bigger buffer as necessary. But if an cluster is not 1043 * enough, return E2BIG. 1044 */ 1045 newlen = CMSG_LEN(oldfds * sizeof(struct file *)); 1046 if (newlen > MCLBYTES) 1047 return (E2BIG); 1048 if (newlen - control->m_len > M_TRAILINGSPACE(control)) { 1049 if (control->m_flags & M_EXT) 1050 return (E2BIG); 1051 MCLGET(control, MB_WAIT); 1052 if (!(control->m_flags & M_EXT)) 1053 return (ENOBUFS); 1054 1055 /* copy the data to the cluster */ 1056 memcpy(mtod(control, char *), cm, cm->cmsg_len); 1057 cm = mtod(control, struct cmsghdr *); 1058 } 1059 1060 /* 1061 * Adjust length, in case sizeof(struct file *) and sizeof(int) 1062 * differs. 1063 */ 1064 control->m_len = cm->cmsg_len = newlen; 1065 1066 /* 1067 * Transform the file descriptors into struct file pointers. 1068 * If sizeof (struct file *) is bigger than or equal to sizeof int, 1069 * then do it in reverse order so that the int won't get until 1070 * we're done. 1071 * If sizeof (struct file *) is smaller than sizeof int, then 1072 * do it in forward order. 1073 */ 1074 if (sizeof (struct file *) >= sizeof (int)) { 1075 fdp = (int *)(cm + 1) + oldfds - 1; 1076 rp = (struct file **)CMSG_DATA(cm) + oldfds - 1; 1077 for (i = 0; i < oldfds; i++) { 1078 fp = fdescp->fd_files[*fdp--].fp; 1079 *rp-- = fp; 1080 fhold(fp); 1081 spin_lock_wr(&unp_spin); 1082 fp->f_msgcount++; 1083 unp_rights++; 1084 spin_unlock_wr(&unp_spin); 1085 } 1086 } else { 1087 fdp = (int *)(cm + 1); 1088 rp = (struct file **)CMSG_DATA(cm); 1089 for (i = 0; i < oldfds; i++) { 1090 fp = fdescp->fd_files[*fdp++].fp; 1091 *rp++ = fp; 1092 fhold(fp); 1093 spin_lock_wr(&unp_spin); 1094 fp->f_msgcount++; 1095 unp_rights++; 1096 spin_unlock_wr(&unp_spin); 1097 } 1098 } 1099 return (0); 1100 } 1101 1102 /* 1103 * Garbage collect in-transit file descriptors that get lost due to 1104 * loops (i.e. when a socket is sent to another process over itself, 1105 * and more complex situations). 1106 * 1107 * NOT MPSAFE - TODO socket flush code and maybe closef. Rest is MPSAFE. 1108 */ 1109 1110 struct unp_gc_info { 1111 struct file **extra_ref; 1112 struct file *locked_fp; 1113 int defer; 1114 int index; 1115 int maxindex; 1116 }; 1117 1118 static void 1119 unp_gc() 1120 { 1121 struct unp_gc_info info; 1122 static boolean_t unp_gcing; 1123 struct file **fpp; 1124 int i; 1125 1126 spin_lock_wr(&unp_spin); 1127 if (unp_gcing) { 1128 spin_unlock_wr(&unp_spin); 1129 return; 1130 } 1131 unp_gcing = TRUE; 1132 spin_unlock_wr(&unp_spin); 1133 1134 /* 1135 * before going through all this, set all FDs to 1136 * be NOT defered and NOT externally accessible 1137 */ 1138 info.defer = 0; 1139 allfiles_scan_exclusive(unp_gc_clearmarks, NULL); 1140 do { 1141 allfiles_scan_exclusive(unp_gc_checkmarks, &info); 1142 } while (info.defer); 1143 1144 /* 1145 * We grab an extra reference to each of the file table entries 1146 * that are not otherwise accessible and then free the rights 1147 * that are stored in messages on them. 1148 * 1149 * The bug in the orginal code is a little tricky, so I'll describe 1150 * what's wrong with it here. 1151 * 1152 * It is incorrect to simply unp_discard each entry for f_msgcount 1153 * times -- consider the case of sockets A and B that contain 1154 * references to each other. On a last close of some other socket, 1155 * we trigger a gc since the number of outstanding rights (unp_rights) 1156 * is non-zero. If during the sweep phase the gc code un_discards, 1157 * we end up doing a (full) closef on the descriptor. A closef on A 1158 * results in the following chain. Closef calls soo_close, which 1159 * calls soclose. Soclose calls first (through the switch 1160 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply 1161 * returns because the previous instance had set unp_gcing, and 1162 * we return all the way back to soclose, which marks the socket 1163 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush 1164 * to free up the rights that are queued in messages on the socket A, 1165 * i.e., the reference on B. The sorflush calls via the dom_dispose 1166 * switch unp_dispose, which unp_scans with unp_discard. This second 1167 * instance of unp_discard just calls closef on B. 1168 * 1169 * Well, a similar chain occurs on B, resulting in a sorflush on B, 1170 * which results in another closef on A. Unfortunately, A is already 1171 * being closed, and the descriptor has already been marked with 1172 * SS_NOFDREF, and soclose panics at this point. 1173 * 1174 * Here, we first take an extra reference to each inaccessible 1175 * descriptor. Then, we call sorflush ourself, since we know 1176 * it is a Unix domain socket anyhow. After we destroy all the 1177 * rights carried in messages, we do a last closef to get rid 1178 * of our extra reference. This is the last close, and the 1179 * unp_detach etc will shut down the socket. 1180 * 1181 * 91/09/19, bsy@cs.cmu.edu 1182 */ 1183 info.extra_ref = malloc(256 * sizeof(struct file *), M_FILE, M_WAITOK); 1184 info.maxindex = 256; 1185 1186 do { 1187 /* 1188 * Look for matches 1189 */ 1190 info.index = 0; 1191 allfiles_scan_exclusive(unp_gc_checkrefs, &info); 1192 1193 /* 1194 * For each FD on our hit list, do the following two things 1195 */ 1196 for (i = info.index, fpp = info.extra_ref; --i >= 0; ++fpp) { 1197 struct file *tfp = *fpp; 1198 if (tfp->f_type == DTYPE_SOCKET && tfp->f_data != NULL) 1199 sorflush((struct socket *)(tfp->f_data)); 1200 } 1201 for (i = info.index, fpp = info.extra_ref; --i >= 0; ++fpp) 1202 closef(*fpp, NULL); 1203 } while (info.index == info.maxindex); 1204 free((caddr_t)info.extra_ref, M_FILE); 1205 unp_gcing = FALSE; 1206 } 1207 1208 /* 1209 * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry 1210 */ 1211 static int 1212 unp_gc_checkrefs(struct file *fp, void *data) 1213 { 1214 struct unp_gc_info *info = data; 1215 1216 if (fp->f_count == 0) 1217 return(0); 1218 if (info->index == info->maxindex) 1219 return(-1); 1220 1221 /* 1222 * If all refs are from msgs, and it's not marked accessible 1223 * then it must be referenced from some unreachable cycle 1224 * of (shut-down) FDs, so include it in our 1225 * list of FDs to remove 1226 */ 1227 if (fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) { 1228 info->extra_ref[info->index++] = fp; 1229 fhold(fp); 1230 } 1231 return(0); 1232 } 1233 1234 /* 1235 * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry 1236 */ 1237 static int 1238 unp_gc_clearmarks(struct file *fp, void *data __unused) 1239 { 1240 fp->f_flag &= ~(FMARK|FDEFER); 1241 return(0); 1242 } 1243 1244 /* 1245 * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry 1246 */ 1247 static int 1248 unp_gc_checkmarks(struct file *fp, void *data) 1249 { 1250 struct unp_gc_info *info = data; 1251 struct socket *so; 1252 1253 /* 1254 * If the file is not open, skip it 1255 */ 1256 if (fp->f_count == 0) 1257 return(0); 1258 /* 1259 * If we already marked it as 'defer' in a 1260 * previous pass, then try process it this time 1261 * and un-mark it 1262 */ 1263 if (fp->f_flag & FDEFER) { 1264 fp->f_flag &= ~FDEFER; 1265 --info->defer; 1266 } else { 1267 /* 1268 * if it's not defered, then check if it's 1269 * already marked.. if so skip it 1270 */ 1271 if (fp->f_flag & FMARK) 1272 return(0); 1273 /* 1274 * If all references are from messages 1275 * in transit, then skip it. it's not 1276 * externally accessible. 1277 */ 1278 if (fp->f_count == fp->f_msgcount) 1279 return(0); 1280 /* 1281 * If it got this far then it must be 1282 * externally accessible. 1283 */ 1284 fp->f_flag |= FMARK; 1285 } 1286 /* 1287 * either it was defered, or it is externally 1288 * accessible and not already marked so. 1289 * Now check if it is possibly one of OUR sockets. 1290 */ 1291 if (fp->f_type != DTYPE_SOCKET || 1292 (so = (struct socket *)fp->f_data) == NULL) 1293 return(0); 1294 if (so->so_proto->pr_domain != &localdomain || 1295 !(so->so_proto->pr_flags & PR_RIGHTS)) 1296 return(0); 1297 #ifdef notdef 1298 XXX note: exclusive fp->f_spin lock held 1299 if (so->so_rcv.sb_flags & SB_LOCK) { 1300 /* 1301 * This is problematical; it's not clear 1302 * we need to wait for the sockbuf to be 1303 * unlocked (on a uniprocessor, at least), 1304 * and it's also not clear what to do 1305 * if sbwait returns an error due to receipt 1306 * of a signal. If sbwait does return 1307 * an error, we'll go into an infinite 1308 * loop. Delete all of this for now. 1309 */ 1310 sbwait(&so->so_rcv); 1311 goto restart; 1312 } 1313 #endif 1314 /* 1315 * So, Ok, it's one of our sockets and it IS externally 1316 * accessible (or was defered). Now we look 1317 * to see if we hold any file descriptors in its 1318 * message buffers. Follow those links and mark them 1319 * as accessible too. 1320 */ 1321 info->locked_fp = fp; 1322 /* spin_lock_wr(&so->so_rcv.sb_spin); */ 1323 unp_scan(so->so_rcv.sb_mb, unp_mark, info); 1324 /* spin_unlock_wr(&so->so_rcv.sb_spin);*/ 1325 return (0); 1326 } 1327 1328 void 1329 unp_dispose(struct mbuf *m) 1330 { 1331 if (m) 1332 unp_scan(m, unp_discard, NULL); 1333 } 1334 1335 static int 1336 unp_listen(struct unpcb *unp, struct thread *td) 1337 { 1338 struct proc *p = td->td_proc; 1339 1340 KKASSERT(p); 1341 cru2x(p->p_ucred, &unp->unp_peercred); 1342 unp->unp_flags |= UNP_HAVEPCCACHED; 1343 return (0); 1344 } 1345 1346 static void 1347 unp_scan(struct mbuf *m0, void (*op)(struct file *, void *), void *data) 1348 { 1349 struct mbuf *m; 1350 struct file **rp; 1351 struct cmsghdr *cm; 1352 int i; 1353 int qfds; 1354 1355 while (m0) { 1356 for (m = m0; m; m = m->m_next) { 1357 if (m->m_type == MT_CONTROL && 1358 m->m_len >= sizeof(*cm)) { 1359 cm = mtod(m, struct cmsghdr *); 1360 if (cm->cmsg_level != SOL_SOCKET || 1361 cm->cmsg_type != SCM_RIGHTS) 1362 continue; 1363 qfds = (cm->cmsg_len - 1364 (CMSG_DATA(cm) - (u_char *)cm)) 1365 / sizeof (struct file *); 1366 rp = (struct file **)CMSG_DATA(cm); 1367 for (i = 0; i < qfds; i++) 1368 (*op)(*rp++, data); 1369 break; /* XXX, but saves time */ 1370 } 1371 } 1372 m0 = m0->m_nextpkt; 1373 } 1374 } 1375 1376 static void 1377 unp_mark(struct file *fp, void *data) 1378 { 1379 struct unp_gc_info *info = data; 1380 1381 if (info->locked_fp != fp) 1382 spin_lock_wr(&fp->f_spin); 1383 if (fp->f_flag & FMARK) 1384 return; 1385 ++info->defer; 1386 fp->f_flag |= (FMARK|FDEFER); 1387 if (info->locked_fp != fp) 1388 spin_unlock_wr(&fp->f_spin); 1389 } 1390 1391 static void 1392 unp_discard(struct file *fp, void *data __unused) 1393 { 1394 spin_lock_wr(&unp_spin); 1395 fp->f_msgcount--; 1396 unp_rights--; 1397 spin_unlock_wr(&unp_spin); 1398 closef(fp, NULL); 1399 } 1400 1401