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 1.15 2004/06/06 19:16:06 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/namei.h> 49 #include <sys/protosw.h> 50 #include <sys/socket.h> 51 #include <sys/socketvar.h> 52 #include <sys/resourcevar.h> 53 #include <sys/stat.h> 54 #include <sys/sysctl.h> 55 #include <sys/un.h> 56 #include <sys/unpcb.h> 57 #include <sys/vnode.h> 58 #include <sys/file2.h> 59 60 #include <vm/vm_zone.h> 61 62 static struct vm_zone *unp_zone; 63 static unp_gen_t unp_gencnt; 64 static u_int unp_count; 65 66 static struct unp_head unp_shead, unp_dhead; 67 68 /* 69 * Unix communications domain. 70 * 71 * TODO: 72 * SEQPACKET, RDM 73 * rethink name space problems 74 * need a proper out-of-band 75 * lock pushdown 76 */ 77 static struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL }; 78 static ino_t unp_ino; /* prototype for fake inode numbers */ 79 80 static int unp_attach (struct socket *, struct pru_attach_info *); 81 static void unp_detach (struct unpcb *); 82 static int unp_bind (struct unpcb *,struct sockaddr *, struct thread *); 83 static int unp_connect (struct socket *,struct sockaddr *, 84 struct thread *); 85 static void unp_disconnect (struct unpcb *); 86 static void unp_shutdown (struct unpcb *); 87 static void unp_drop (struct unpcb *, int); 88 static void unp_gc (void); 89 static void unp_scan (struct mbuf *, void (*)(struct file *)); 90 static void unp_mark (struct file *); 91 static void unp_discard (struct file *); 92 static int unp_internalize (struct mbuf *, struct thread *); 93 static int unp_listen (struct unpcb *, struct thread *); 94 95 static int 96 uipc_abort(struct socket *so) 97 { 98 struct unpcb *unp = sotounpcb(so); 99 100 if (unp == 0) 101 return EINVAL; 102 unp_drop(unp, ECONNABORTED); 103 unp_detach(unp); 104 sofree(so); 105 return 0; 106 } 107 108 static int 109 uipc_accept(struct socket *so, struct sockaddr **nam) 110 { 111 struct unpcb *unp = sotounpcb(so); 112 113 if (unp == 0) 114 return EINVAL; 115 116 /* 117 * Pass back name of connected socket, 118 * if it was bound and we are still connected 119 * (our peer may have closed already!). 120 */ 121 if (unp->unp_conn && unp->unp_conn->unp_addr) { 122 *nam = dup_sockaddr((struct sockaddr *)unp->unp_conn->unp_addr); 123 } else { 124 *nam = dup_sockaddr((struct sockaddr *)&sun_noname); 125 } 126 return 0; 127 } 128 129 static int 130 uipc_attach(struct socket *so, int proto, struct pru_attach_info *ai) 131 { 132 struct unpcb *unp = sotounpcb(so); 133 134 if (unp != 0) 135 return EISCONN; 136 return unp_attach(so, ai); 137 } 138 139 static int 140 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 141 { 142 struct unpcb *unp = sotounpcb(so); 143 144 if (unp == 0) 145 return EINVAL; 146 return unp_bind(unp, nam, td); 147 } 148 149 static int 150 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 151 { 152 struct unpcb *unp = sotounpcb(so); 153 154 if (unp == 0) 155 return EINVAL; 156 return unp_connect(so, nam, td); 157 } 158 159 static int 160 uipc_connect2(struct socket *so1, struct socket *so2) 161 { 162 struct unpcb *unp = sotounpcb(so1); 163 164 if (unp == 0) 165 return EINVAL; 166 167 return unp_connect2(so1, so2); 168 } 169 170 /* control is EOPNOTSUPP */ 171 172 static int 173 uipc_detach(struct socket *so) 174 { 175 struct unpcb *unp = sotounpcb(so); 176 177 if (unp == 0) 178 return EINVAL; 179 180 unp_detach(unp); 181 return 0; 182 } 183 184 static int 185 uipc_disconnect(struct socket *so) 186 { 187 struct unpcb *unp = sotounpcb(so); 188 189 if (unp == 0) 190 return EINVAL; 191 unp_disconnect(unp); 192 return 0; 193 } 194 195 static int 196 uipc_listen(struct socket *so, struct thread *td) 197 { 198 struct unpcb *unp = sotounpcb(so); 199 200 if (unp == 0 || unp->unp_vnode == 0) 201 return EINVAL; 202 return unp_listen(unp, td); 203 } 204 205 static int 206 uipc_peeraddr(struct socket *so, struct sockaddr **nam) 207 { 208 struct unpcb *unp = sotounpcb(so); 209 210 if (unp == 0) 211 return EINVAL; 212 if (unp->unp_conn && unp->unp_conn->unp_addr) 213 *nam = dup_sockaddr((struct sockaddr *)unp->unp_conn->unp_addr); 214 else { 215 /* 216 * XXX: It seems that this test always fails even when 217 * connection is established. So, this else clause is 218 * added as workaround to return PF_LOCAL sockaddr. 219 */ 220 *nam = dup_sockaddr((struct sockaddr *)&sun_noname); 221 } 222 return 0; 223 } 224 225 static int 226 uipc_rcvd(struct socket *so, int flags) 227 { 228 struct unpcb *unp = sotounpcb(so); 229 struct socket *so2; 230 u_long newhiwat; 231 232 if (unp == 0) 233 return EINVAL; 234 switch (so->so_type) { 235 case SOCK_DGRAM: 236 panic("uipc_rcvd DGRAM?"); 237 /*NOTREACHED*/ 238 239 case SOCK_STREAM: 240 if (unp->unp_conn == 0) 241 break; 242 so2 = unp->unp_conn->unp_socket; 243 /* 244 * Adjust backpressure on sender 245 * and wakeup any waiting to write. 246 */ 247 so2->so_snd.sb_mbmax += unp->unp_mbcnt - so->so_rcv.sb_mbcnt; 248 unp->unp_mbcnt = so->so_rcv.sb_mbcnt; 249 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - 250 so->so_rcv.sb_cc; 251 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat, 252 newhiwat, RLIM_INFINITY); 253 unp->unp_cc = so->so_rcv.sb_cc; 254 sowwakeup(so2); 255 break; 256 257 default: 258 panic("uipc_rcvd unknown socktype"); 259 } 260 return 0; 261 } 262 263 /* pru_rcvoob is EOPNOTSUPP */ 264 265 static int 266 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 267 struct mbuf *control, struct thread *td) 268 { 269 int error = 0; 270 struct unpcb *unp = sotounpcb(so); 271 struct socket *so2; 272 u_long newhiwat; 273 274 if (unp == 0) { 275 error = EINVAL; 276 goto release; 277 } 278 if (flags & PRUS_OOB) { 279 error = EOPNOTSUPP; 280 goto release; 281 } 282 283 if (control && (error = unp_internalize(control, td))) 284 goto release; 285 286 switch (so->so_type) { 287 case SOCK_DGRAM: 288 { 289 struct sockaddr *from; 290 291 if (nam) { 292 if (unp->unp_conn) { 293 error = EISCONN; 294 break; 295 } 296 error = unp_connect(so, nam, td); 297 if (error) 298 break; 299 } else { 300 if (unp->unp_conn == 0) { 301 error = ENOTCONN; 302 break; 303 } 304 } 305 so2 = unp->unp_conn->unp_socket; 306 if (unp->unp_addr) 307 from = (struct sockaddr *)unp->unp_addr; 308 else 309 from = &sun_noname; 310 if (sbappendaddr(&so2->so_rcv, from, m, control)) { 311 sorwakeup(so2); 312 m = 0; 313 control = 0; 314 } else 315 error = ENOBUFS; 316 if (nam) 317 unp_disconnect(unp); 318 break; 319 } 320 321 case SOCK_STREAM: 322 /* Connect if not connected yet. */ 323 /* 324 * Note: A better implementation would complain 325 * if not equal to the peer's address. 326 */ 327 if ((so->so_state & SS_ISCONNECTED) == 0) { 328 if (nam) { 329 error = unp_connect(so, nam, td); 330 if (error) 331 break; /* XXX */ 332 } else { 333 error = ENOTCONN; 334 break; 335 } 336 } 337 338 if (so->so_state & SS_CANTSENDMORE) { 339 error = EPIPE; 340 break; 341 } 342 if (unp->unp_conn == 0) 343 panic("uipc_send connected but no connection?"); 344 so2 = unp->unp_conn->unp_socket; 345 /* 346 * Send to paired receive port, and then reduce 347 * send buffer hiwater marks to maintain backpressure. 348 * Wake up readers. 349 */ 350 if (control) { 351 if (sbappendcontrol(&so2->so_rcv, m, control)) 352 control = 0; 353 } else 354 sbappend(&so2->so_rcv, m); 355 so->so_snd.sb_mbmax -= 356 so2->so_rcv.sb_mbcnt - unp->unp_conn->unp_mbcnt; 357 unp->unp_conn->unp_mbcnt = so2->so_rcv.sb_mbcnt; 358 newhiwat = so->so_snd.sb_hiwat - 359 (so2->so_rcv.sb_cc - unp->unp_conn->unp_cc); 360 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat, 361 newhiwat, RLIM_INFINITY); 362 unp->unp_conn->unp_cc = so2->so_rcv.sb_cc; 363 sorwakeup(so2); 364 m = 0; 365 break; 366 367 default: 368 panic("uipc_send unknown socktype"); 369 } 370 371 /* 372 * SEND_EOF is equivalent to a SEND followed by 373 * a SHUTDOWN. 374 */ 375 if (flags & PRUS_EOF) { 376 socantsendmore(so); 377 unp_shutdown(unp); 378 } 379 380 if (control && error != 0) 381 unp_dispose(control); 382 383 release: 384 if (control) 385 m_freem(control); 386 if (m) 387 m_freem(m); 388 return error; 389 } 390 391 static int 392 uipc_sense(struct socket *so, struct stat *sb) 393 { 394 struct unpcb *unp = sotounpcb(so); 395 struct socket *so2; 396 397 if (unp == 0) 398 return EINVAL; 399 sb->st_blksize = so->so_snd.sb_hiwat; 400 if (so->so_type == SOCK_STREAM && unp->unp_conn != 0) { 401 so2 = unp->unp_conn->unp_socket; 402 sb->st_blksize += so2->so_rcv.sb_cc; 403 } 404 sb->st_dev = NOUDEV; 405 if (unp->unp_ino == 0) /* make up a non-zero inode number */ 406 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino; 407 sb->st_ino = unp->unp_ino; 408 return (0); 409 } 410 411 static int 412 uipc_shutdown(struct socket *so) 413 { 414 struct unpcb *unp = sotounpcb(so); 415 416 if (unp == 0) 417 return EINVAL; 418 socantsendmore(so); 419 unp_shutdown(unp); 420 return 0; 421 } 422 423 static int 424 uipc_sockaddr(struct socket *so, struct sockaddr **nam) 425 { 426 struct unpcb *unp = sotounpcb(so); 427 428 if (unp == 0) 429 return EINVAL; 430 if (unp->unp_addr) 431 *nam = dup_sockaddr((struct sockaddr *)unp->unp_addr); 432 return 0; 433 } 434 435 struct pr_usrreqs uipc_usrreqs = { 436 uipc_abort, uipc_accept, uipc_attach, uipc_bind, uipc_connect, 437 uipc_connect2, pru_control_notsupp, uipc_detach, uipc_disconnect, 438 uipc_listen, uipc_peeraddr, uipc_rcvd, pru_rcvoob_notsupp, 439 uipc_send, uipc_sense, uipc_shutdown, uipc_sockaddr, 440 sosend, soreceive, sopoll 441 }; 442 443 int 444 uipc_ctloutput(so, sopt) 445 struct socket *so; 446 struct sockopt *sopt; 447 { 448 struct unpcb *unp = sotounpcb(so); 449 int error; 450 451 switch (sopt->sopt_dir) { 452 case SOPT_GET: 453 switch (sopt->sopt_name) { 454 case LOCAL_PEERCRED: 455 if (unp->unp_flags & UNP_HAVEPC) 456 error = sooptcopyout(sopt, &unp->unp_peercred, 457 sizeof(unp->unp_peercred)); 458 else { 459 if (so->so_type == SOCK_STREAM) 460 error = ENOTCONN; 461 else 462 error = EINVAL; 463 } 464 break; 465 default: 466 error = EOPNOTSUPP; 467 break; 468 } 469 break; 470 case SOPT_SET: 471 default: 472 error = EOPNOTSUPP; 473 break; 474 } 475 return (error); 476 } 477 478 /* 479 * Both send and receive buffers are allocated PIPSIZ bytes of buffering 480 * for stream sockets, although the total for sender and receiver is 481 * actually only PIPSIZ. 482 * Datagram sockets really use the sendspace as the maximum datagram size, 483 * and don't really want to reserve the sendspace. Their recvspace should 484 * be large enough for at least one max-size datagram plus address. 485 */ 486 #ifndef PIPSIZ 487 #define PIPSIZ 8192 488 #endif 489 static u_long unpst_sendspace = PIPSIZ; 490 static u_long unpst_recvspace = PIPSIZ; 491 static u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 492 static u_long unpdg_recvspace = 4*1024; 493 494 static int unp_rights; /* file descriptors in flight */ 495 496 SYSCTL_DECL(_net_local_stream); 497 SYSCTL_INT(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 498 &unpst_sendspace, 0, ""); 499 SYSCTL_INT(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 500 &unpst_recvspace, 0, ""); 501 SYSCTL_DECL(_net_local_dgram); 502 SYSCTL_INT(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 503 &unpdg_sendspace, 0, ""); 504 SYSCTL_INT(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 505 &unpdg_recvspace, 0, ""); 506 SYSCTL_DECL(_net_local); 507 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, ""); 508 509 static int 510 unp_attach(struct socket *so, struct pru_attach_info *ai) 511 { 512 struct unpcb *unp; 513 int error; 514 515 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 516 switch (so->so_type) { 517 518 case SOCK_STREAM: 519 error = soreserve(so, unpst_sendspace, unpst_recvspace, 520 ai->sb_rlimit); 521 break; 522 523 case SOCK_DGRAM: 524 error = soreserve(so, unpdg_sendspace, unpdg_recvspace, 525 ai->sb_rlimit); 526 break; 527 528 default: 529 panic("unp_attach"); 530 } 531 if (error) 532 return (error); 533 } 534 unp = zalloc(unp_zone); 535 if (unp == NULL) 536 return (ENOBUFS); 537 bzero(unp, sizeof *unp); 538 unp->unp_gencnt = ++unp_gencnt; 539 unp_count++; 540 LIST_INIT(&unp->unp_refs); 541 unp->unp_socket = so; 542 unp->unp_rvnode = ai->fd_rdir; /* jail cruft XXX JH */ 543 LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead 544 : &unp_shead, unp, unp_link); 545 so->so_pcb = (caddr_t)unp; 546 return (0); 547 } 548 549 static void 550 unp_detach(unp) 551 struct unpcb *unp; 552 { 553 LIST_REMOVE(unp, unp_link); 554 unp->unp_gencnt = ++unp_gencnt; 555 --unp_count; 556 if (unp->unp_vnode) { 557 unp->unp_vnode->v_socket = 0; 558 vrele(unp->unp_vnode); 559 unp->unp_vnode = 0; 560 } 561 if (unp->unp_conn) 562 unp_disconnect(unp); 563 while (!LIST_EMPTY(&unp->unp_refs)) 564 unp_drop(LIST_FIRST(&unp->unp_refs), ECONNRESET); 565 soisdisconnected(unp->unp_socket); 566 unp->unp_socket->so_pcb = 0; 567 if (unp_rights) { 568 /* 569 * Normally the receive buffer is flushed later, 570 * in sofree, but if our receive buffer holds references 571 * to descriptors that are now garbage, we will dispose 572 * of those descriptor references after the garbage collector 573 * gets them (resulting in a "panic: closef: count < 0"). 574 */ 575 sorflush(unp->unp_socket); 576 unp_gc(); 577 } 578 if (unp->unp_addr) 579 FREE(unp->unp_addr, M_SONAME); 580 zfree(unp_zone, unp); 581 } 582 583 static int 584 unp_bind(struct unpcb *unp, struct sockaddr *nam, struct thread *td) 585 { 586 struct proc *p = td->td_proc; 587 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 588 struct vnode *vp; 589 struct vattr vattr; 590 int error, namelen; 591 struct nameidata nd; 592 char buf[SOCK_MAXADDRLEN]; 593 594 if (unp->unp_vnode != NULL) 595 return (EINVAL); 596 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 597 if (namelen <= 0) 598 return EINVAL; 599 strncpy(buf, soun->sun_path, namelen); 600 buf[namelen] = 0; /* null-terminate the string */ 601 NDINIT(&nd, NAMEI_CREATE, CNP_LOCKPARENT, UIO_SYSSPACE, buf, td); 602 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 603 error = namei(&nd); 604 if (error) 605 return (error); 606 vp = nd.ni_vp; 607 if (vp != NULL) { 608 NDFREE(&nd, NDF_ONLY_PNBUF); 609 if (nd.ni_dvp == vp) 610 vrele(nd.ni_dvp); 611 else 612 vput(nd.ni_dvp); 613 vrele(vp); 614 return (EADDRINUSE); 615 } 616 VATTR_NULL(&vattr); 617 vattr.va_type = VSOCK; 618 vattr.va_mode = (ACCESSPERMS & ~p->p_fd->fd_cmask); 619 VOP_LEASE(nd.ni_dvp, td, p->p_ucred, LEASE_WRITE); 620 error = VOP_CREATE(nd.ni_dvp, NCPNULL, &nd.ni_vp, &nd.ni_cnd, &vattr); 621 NDFREE(&nd, NDF_ONLY_PNBUF); 622 vput(nd.ni_dvp); 623 if (error) 624 return (error); 625 vp = nd.ni_vp; 626 vp->v_socket = unp->unp_socket; 627 unp->unp_vnode = vp; 628 unp->unp_addr = (struct sockaddr_un *)dup_sockaddr(nam); 629 VOP_UNLOCK(vp, NULL, 0, td); 630 return (0); 631 } 632 633 static int 634 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 635 { 636 struct proc *p = td->td_proc; 637 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 638 struct vnode *vp; 639 struct socket *so2, *so3; 640 struct unpcb *unp, *unp2, *unp3; 641 int error, len; 642 struct nameidata nd; 643 char buf[SOCK_MAXADDRLEN]; 644 645 KKASSERT(p); 646 647 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 648 if (len <= 0) 649 return EINVAL; 650 strncpy(buf, soun->sun_path, len); 651 buf[len] = 0; 652 653 NDINIT(&nd, NAMEI_LOOKUP, CNP_FOLLOW | CNP_LOCKLEAF, 654 UIO_SYSSPACE, buf, td); 655 error = namei(&nd); 656 if (error) 657 return (error); 658 vp = nd.ni_vp; 659 NDFREE(&nd, NDF_ONLY_PNBUF); 660 if (vp->v_type != VSOCK) { 661 error = ENOTSOCK; 662 goto bad; 663 } 664 error = VOP_ACCESS(vp, VWRITE, p->p_ucred, td); 665 if (error) 666 goto bad; 667 so2 = vp->v_socket; 668 if (so2 == 0) { 669 error = ECONNREFUSED; 670 goto bad; 671 } 672 if (so->so_type != so2->so_type) { 673 error = EPROTOTYPE; 674 goto bad; 675 } 676 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 677 if ((so2->so_options & SO_ACCEPTCONN) == 0 || 678 (so3 = sonewconn(so2, 0)) == 0) { 679 error = ECONNREFUSED; 680 goto bad; 681 } 682 unp = sotounpcb(so); 683 unp2 = sotounpcb(so2); 684 unp3 = sotounpcb(so3); 685 if (unp2->unp_addr) 686 unp3->unp_addr = (struct sockaddr_un *) 687 dup_sockaddr((struct sockaddr *)unp2->unp_addr); 688 689 /* 690 * unp_peercred management: 691 * 692 * The connecter's (client's) credentials are copied 693 * from its process structure at the time of connect() 694 * (which is now). 695 */ 696 cru2x(p->p_ucred, &unp3->unp_peercred); 697 unp3->unp_flags |= UNP_HAVEPC; 698 /* 699 * The receiver's (server's) credentials are copied 700 * from the unp_peercred member of socket on which the 701 * former called listen(); unp_listen() cached that 702 * process's credentials at that time so we can use 703 * them now. 704 */ 705 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED, 706 ("unp_connect: listener without cached peercred")); 707 memcpy(&unp->unp_peercred, &unp2->unp_peercred, 708 sizeof(unp->unp_peercred)); 709 unp->unp_flags |= UNP_HAVEPC; 710 711 so2 = so3; 712 } 713 error = unp_connect2(so, so2); 714 bad: 715 vput(vp); 716 return (error); 717 } 718 719 int 720 unp_connect2(so, so2) 721 struct socket *so; 722 struct socket *so2; 723 { 724 struct unpcb *unp = sotounpcb(so); 725 struct unpcb *unp2; 726 727 if (so2->so_type != so->so_type) 728 return (EPROTOTYPE); 729 unp2 = sotounpcb(so2); 730 unp->unp_conn = unp2; 731 switch (so->so_type) { 732 733 case SOCK_DGRAM: 734 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 735 soisconnected(so); 736 break; 737 738 case SOCK_STREAM: 739 unp2->unp_conn = unp; 740 soisconnected(so); 741 soisconnected(so2); 742 break; 743 744 default: 745 panic("unp_connect2"); 746 } 747 return (0); 748 } 749 750 static void 751 unp_disconnect(unp) 752 struct unpcb *unp; 753 { 754 struct unpcb *unp2 = unp->unp_conn; 755 756 if (unp2 == 0) 757 return; 758 unp->unp_conn = 0; 759 switch (unp->unp_socket->so_type) { 760 761 case SOCK_DGRAM: 762 LIST_REMOVE(unp, unp_reflink); 763 unp->unp_socket->so_state &= ~SS_ISCONNECTED; 764 break; 765 766 case SOCK_STREAM: 767 soisdisconnected(unp->unp_socket); 768 unp2->unp_conn = 0; 769 soisdisconnected(unp2->unp_socket); 770 break; 771 } 772 } 773 774 #ifdef notdef 775 void 776 unp_abort(unp) 777 struct unpcb *unp; 778 { 779 780 unp_detach(unp); 781 } 782 #endif 783 784 static int 785 prison_unpcb(struct thread *td, struct unpcb *unp) 786 { 787 struct proc *p; 788 789 if (td == NULL) 790 return (0); 791 if ((p = td->td_proc) == NULL) 792 return (0); 793 if (!p->p_ucred->cr_prison) 794 return (0); 795 if (p->p_fd->fd_rdir == unp->unp_rvnode) 796 return (0); 797 return (1); 798 } 799 800 static int 801 unp_pcblist(SYSCTL_HANDLER_ARGS) 802 { 803 int error, i, n; 804 struct unpcb *unp, **unp_list; 805 unp_gen_t gencnt; 806 struct xunpgen xug; 807 struct unp_head *head; 808 809 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead); 810 811 KKASSERT(curproc != NULL); 812 813 /* 814 * The process of preparing the PCB list is too time-consuming and 815 * resource-intensive to repeat twice on every request. 816 */ 817 if (req->oldptr == 0) { 818 n = unp_count; 819 req->oldidx = 2 * (sizeof xug) 820 + (n + n/8) * sizeof(struct xunpcb); 821 return 0; 822 } 823 824 if (req->newptr != 0) 825 return EPERM; 826 827 /* 828 * OK, now we're committed to doing something. 829 */ 830 gencnt = unp_gencnt; 831 n = unp_count; 832 833 xug.xug_len = sizeof xug; 834 xug.xug_count = n; 835 xug.xug_gen = gencnt; 836 xug.xug_sogen = so_gencnt; 837 error = SYSCTL_OUT(req, &xug, sizeof xug); 838 if (error) 839 return error; 840 841 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 842 if (unp_list == 0) 843 return ENOMEM; 844 845 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 846 unp = LIST_NEXT(unp, unp_link)) { 847 if (unp->unp_gencnt <= gencnt && !prison_unpcb(req->td, unp)) 848 unp_list[i++] = unp; 849 } 850 n = i; /* in case we lost some during malloc */ 851 852 error = 0; 853 for (i = 0; i < n; i++) { 854 unp = unp_list[i]; 855 if (unp->unp_gencnt <= gencnt) { 856 struct xunpcb xu; 857 xu.xu_len = sizeof xu; 858 xu.xu_unpp = unp; 859 /* 860 * XXX - need more locking here to protect against 861 * connect/disconnect races for SMP. 862 */ 863 if (unp->unp_addr) 864 bcopy(unp->unp_addr, &xu.xu_addr, 865 unp->unp_addr->sun_len); 866 if (unp->unp_conn && unp->unp_conn->unp_addr) 867 bcopy(unp->unp_conn->unp_addr, 868 &xu.xu_caddr, 869 unp->unp_conn->unp_addr->sun_len); 870 bcopy(unp, &xu.xu_unp, sizeof *unp); 871 sotoxsocket(unp->unp_socket, &xu.xu_socket); 872 error = SYSCTL_OUT(req, &xu, sizeof xu); 873 } 874 } 875 if (!error) { 876 /* 877 * Give the user an updated idea of our state. 878 * If the generation differs from what we told 879 * her before, she knows that something happened 880 * while we were processing this request, and it 881 * might be necessary to retry. 882 */ 883 xug.xug_gen = unp_gencnt; 884 xug.xug_sogen = so_gencnt; 885 xug.xug_count = unp_count; 886 error = SYSCTL_OUT(req, &xug, sizeof xug); 887 } 888 free(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"); 898 899 static void 900 unp_shutdown(unp) 901 struct unpcb *unp; 902 { 903 struct socket *so; 904 905 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn && 906 (so = unp->unp_conn->unp_socket)) 907 socantrcvmore(so); 908 } 909 910 static void 911 unp_drop(unp, errno) 912 struct unpcb *unp; 913 int errno; 914 { 915 struct socket *so = unp->unp_socket; 916 917 so->so_error = errno; 918 unp_disconnect(unp); 919 } 920 921 #ifdef notdef 922 void 923 unp_drain() 924 { 925 926 } 927 #endif 928 929 int 930 unp_externalize(struct mbuf *rights) 931 { 932 struct proc *p = curproc; /* XXX */ 933 int i; 934 struct cmsghdr *cm = mtod(rights, struct cmsghdr *); 935 int *fdp; 936 struct file **rp; 937 struct file *fp; 938 int newfds = (cm->cmsg_len - (CMSG_DATA(cm) - (u_char *)cm)) 939 / sizeof (struct file *); 940 int f; 941 942 /* 943 * if the new FD's will not fit, then we free them all 944 */ 945 if (!fdavail(p, newfds)) { 946 rp = (struct file **)CMSG_DATA(cm); 947 for (i = 0; i < newfds; i++) { 948 fp = *rp; 949 /* 950 * zero the pointer before calling unp_discard, 951 * since it may end up in unp_gc().. 952 */ 953 *rp++ = 0; 954 unp_discard(fp); 955 } 956 return (EMSGSIZE); 957 } 958 /* 959 * now change each pointer to an fd in the global table to 960 * an integer that is the index to the local fd table entry 961 * that we set up to point to the global one we are transferring. 962 * If sizeof (struct file *) is bigger than or equal to sizeof int, 963 * then do it in forward order. In that case, an integer will 964 * always come in the same place or before its corresponding 965 * struct file pointer. 966 * If sizeof (struct file *) is smaller than sizeof int, then 967 * do it in reverse order. 968 */ 969 if (sizeof (struct file *) >= sizeof (int)) { 970 fdp = (int *)(cm + 1); 971 rp = (struct file **)CMSG_DATA(cm); 972 for (i = 0; i < newfds; i++) { 973 if (fdalloc(p, 0, &f)) 974 panic("unp_externalize"); 975 fp = *rp++; 976 p->p_fd->fd_ofiles[f] = fp; 977 fp->f_msgcount--; 978 unp_rights--; 979 *fdp++ = f; 980 } 981 } else { 982 fdp = (int *)(cm + 1) + newfds - 1; 983 rp = (struct file **)CMSG_DATA(cm) + newfds - 1; 984 for (i = 0; i < newfds; i++) { 985 if (fdalloc(p, 0, &f)) 986 panic("unp_externalize"); 987 fp = *rp--; 988 p->p_fd->fd_ofiles[f] = fp; 989 fp->f_msgcount--; 990 unp_rights--; 991 *fdp-- = f; 992 } 993 } 994 995 /* 996 * Adjust length, in case sizeof(struct file *) and sizeof(int) 997 * differs. 998 */ 999 cm->cmsg_len = CMSG_LEN(newfds * sizeof(int)); 1000 rights->m_len = cm->cmsg_len; 1001 return (0); 1002 } 1003 1004 void 1005 unp_init(void) 1006 { 1007 unp_zone = zinit("unpcb", sizeof(struct unpcb), nmbclusters, 0, 0); 1008 if (unp_zone == 0) 1009 panic("unp_init"); 1010 LIST_INIT(&unp_dhead); 1011 LIST_INIT(&unp_shead); 1012 } 1013 1014 static int 1015 unp_internalize(struct mbuf *control, struct thread *td) 1016 { 1017 struct proc *p = td->td_proc; 1018 struct filedesc *fdescp; 1019 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1020 struct file **rp; 1021 struct file *fp; 1022 int i, fd, *fdp; 1023 struct cmsgcred *cmcred; 1024 int oldfds; 1025 u_int newlen; 1026 1027 KKASSERT(p); 1028 fdescp = p->p_fd; 1029 if ((cm->cmsg_type != SCM_RIGHTS && cm->cmsg_type != SCM_CREDS) || 1030 cm->cmsg_level != SOL_SOCKET || cm->cmsg_len != control->m_len) 1031 return (EINVAL); 1032 1033 /* 1034 * Fill in credential information. 1035 */ 1036 if (cm->cmsg_type == SCM_CREDS) { 1037 cmcred = (struct cmsgcred *)(cm + 1); 1038 cmcred->cmcred_pid = p->p_pid; 1039 cmcred->cmcred_uid = p->p_ucred->cr_ruid; 1040 cmcred->cmcred_gid = p->p_ucred->cr_rgid; 1041 cmcred->cmcred_euid = p->p_ucred->cr_uid; 1042 cmcred->cmcred_ngroups = MIN(p->p_ucred->cr_ngroups, 1043 CMGROUP_MAX); 1044 for (i = 0; i < cmcred->cmcred_ngroups; i++) 1045 cmcred->cmcred_groups[i] = p->p_ucred->cr_groups[i]; 1046 return(0); 1047 } 1048 1049 oldfds = (cm->cmsg_len - sizeof (*cm)) / sizeof (int); 1050 /* 1051 * check that all the FDs passed in refer to legal OPEN files 1052 * If not, reject the entire operation. 1053 */ 1054 fdp = (int *)(cm + 1); 1055 for (i = 0; i < oldfds; i++) { 1056 fd = *fdp++; 1057 if ((unsigned)fd >= fdescp->fd_nfiles || 1058 fdescp->fd_ofiles[fd] == NULL) 1059 return (EBADF); 1060 if (fdescp->fd_ofiles[fd]->f_type == DTYPE_KQUEUE) 1061 return (EOPNOTSUPP); 1062 } 1063 /* 1064 * Now replace the integer FDs with pointers to 1065 * the associated global file table entry.. 1066 * Allocate a bigger buffer as necessary. But if an cluster is not 1067 * enough, return E2BIG. 1068 */ 1069 newlen = CMSG_LEN(oldfds * sizeof(struct file *)); 1070 if (newlen > MCLBYTES) 1071 return (E2BIG); 1072 if (newlen - control->m_len > M_TRAILINGSPACE(control)) { 1073 if (control->m_flags & M_EXT) 1074 return (E2BIG); 1075 MCLGET(control, MB_WAIT); 1076 if ((control->m_flags & M_EXT) == 0) 1077 return (ENOBUFS); 1078 1079 /* copy the data to the cluster */ 1080 memcpy(mtod(control, char *), cm, cm->cmsg_len); 1081 cm = mtod(control, struct cmsghdr *); 1082 } 1083 1084 /* 1085 * Adjust length, in case sizeof(struct file *) and sizeof(int) 1086 * differs. 1087 */ 1088 control->m_len = cm->cmsg_len = newlen; 1089 1090 /* 1091 * Transform the file descriptors into struct file pointers. 1092 * If sizeof (struct file *) is bigger than or equal to sizeof int, 1093 * then do it in reverse order so that the int won't get until 1094 * we're done. 1095 * If sizeof (struct file *) is smaller than sizeof int, then 1096 * do it in forward order. 1097 */ 1098 if (sizeof (struct file *) >= sizeof (int)) { 1099 fdp = (int *)(cm + 1) + oldfds - 1; 1100 rp = (struct file **)CMSG_DATA(cm) + oldfds - 1; 1101 for (i = 0; i < oldfds; i++) { 1102 fp = fdescp->fd_ofiles[*fdp--]; 1103 *rp-- = fp; 1104 fp->f_count++; 1105 fp->f_msgcount++; 1106 unp_rights++; 1107 } 1108 } else { 1109 fdp = (int *)(cm + 1); 1110 rp = (struct file **)CMSG_DATA(cm); 1111 for (i = 0; i < oldfds; i++) { 1112 fp = fdescp->fd_ofiles[*fdp++]; 1113 *rp++ = fp; 1114 fp->f_count++; 1115 fp->f_msgcount++; 1116 unp_rights++; 1117 } 1118 } 1119 return (0); 1120 } 1121 1122 static int unp_defer, unp_gcing; 1123 1124 static void 1125 unp_gc() 1126 { 1127 struct file *fp, *nextfp; 1128 struct socket *so; 1129 struct file **extra_ref, **fpp; 1130 int nunref, i; 1131 1132 if (unp_gcing) 1133 return; 1134 unp_gcing = 1; 1135 unp_defer = 0; 1136 /* 1137 * before going through all this, set all FDs to 1138 * be NOT defered and NOT externally accessible 1139 */ 1140 LIST_FOREACH(fp, &filehead, f_list) 1141 fp->f_flag &= ~(FMARK|FDEFER); 1142 do { 1143 LIST_FOREACH(fp, &filehead, f_list) { 1144 /* 1145 * If the file is not open, skip it 1146 */ 1147 if (fp->f_count == 0) 1148 continue; 1149 /* 1150 * If we already marked it as 'defer' in a 1151 * previous pass, then try process it this time 1152 * and un-mark it 1153 */ 1154 if (fp->f_flag & FDEFER) { 1155 fp->f_flag &= ~FDEFER; 1156 unp_defer--; 1157 } else { 1158 /* 1159 * if it's not defered, then check if it's 1160 * already marked.. if so skip it 1161 */ 1162 if (fp->f_flag & FMARK) 1163 continue; 1164 /* 1165 * If all references are from messages 1166 * in transit, then skip it. it's not 1167 * externally accessible. 1168 */ 1169 if (fp->f_count == fp->f_msgcount) 1170 continue; 1171 /* 1172 * If it got this far then it must be 1173 * externally accessible. 1174 */ 1175 fp->f_flag |= FMARK; 1176 } 1177 /* 1178 * either it was defered, or it is externally 1179 * accessible and not already marked so. 1180 * Now check if it is possibly one of OUR sockets. 1181 */ 1182 if (fp->f_type != DTYPE_SOCKET || 1183 (so = (struct socket *)fp->f_data) == 0) 1184 continue; 1185 if (so->so_proto->pr_domain != &localdomain || 1186 (so->so_proto->pr_flags&PR_RIGHTS) == 0) 1187 continue; 1188 #ifdef notdef 1189 if (so->so_rcv.sb_flags & SB_LOCK) { 1190 /* 1191 * This is problematical; it's not clear 1192 * we need to wait for the sockbuf to be 1193 * unlocked (on a uniprocessor, at least), 1194 * and it's also not clear what to do 1195 * if sbwait returns an error due to receipt 1196 * of a signal. If sbwait does return 1197 * an error, we'll go into an infinite 1198 * loop. Delete all of this for now. 1199 */ 1200 (void) sbwait(&so->so_rcv); 1201 goto restart; 1202 } 1203 #endif 1204 /* 1205 * So, Ok, it's one of our sockets and it IS externally 1206 * accessible (or was defered). Now we look 1207 * to see if we hold any file descriptors in its 1208 * message buffers. Follow those links and mark them 1209 * as accessible too. 1210 */ 1211 unp_scan(so->so_rcv.sb_mb, unp_mark); 1212 } 1213 } while (unp_defer); 1214 /* 1215 * We grab an extra reference to each of the file table entries 1216 * that are not otherwise accessible and then free the rights 1217 * that are stored in messages on them. 1218 * 1219 * The bug in the orginal code is a little tricky, so I'll describe 1220 * what's wrong with it here. 1221 * 1222 * It is incorrect to simply unp_discard each entry for f_msgcount 1223 * times -- consider the case of sockets A and B that contain 1224 * references to each other. On a last close of some other socket, 1225 * we trigger a gc since the number of outstanding rights (unp_rights) 1226 * is non-zero. If during the sweep phase the gc code un_discards, 1227 * we end up doing a (full) closef on the descriptor. A closef on A 1228 * results in the following chain. Closef calls soo_close, which 1229 * calls soclose. Soclose calls first (through the switch 1230 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply 1231 * returns because the previous instance had set unp_gcing, and 1232 * we return all the way back to soclose, which marks the socket 1233 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush 1234 * to free up the rights that are queued in messages on the socket A, 1235 * i.e., the reference on B. The sorflush calls via the dom_dispose 1236 * switch unp_dispose, which unp_scans with unp_discard. This second 1237 * instance of unp_discard just calls closef on B. 1238 * 1239 * Well, a similar chain occurs on B, resulting in a sorflush on B, 1240 * which results in another closef on A. Unfortunately, A is already 1241 * being closed, and the descriptor has already been marked with 1242 * SS_NOFDREF, and soclose panics at this point. 1243 * 1244 * Here, we first take an extra reference to each inaccessible 1245 * descriptor. Then, we call sorflush ourself, since we know 1246 * it is a Unix domain socket anyhow. After we destroy all the 1247 * rights carried in messages, we do a last closef to get rid 1248 * of our extra reference. This is the last close, and the 1249 * unp_detach etc will shut down the socket. 1250 * 1251 * 91/09/19, bsy@cs.cmu.edu 1252 */ 1253 extra_ref = malloc(nfiles * sizeof(struct file *), M_FILE, M_WAITOK); 1254 for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref; fp != 0; 1255 fp = nextfp) { 1256 nextfp = LIST_NEXT(fp, f_list); 1257 /* 1258 * If it's not open, skip it 1259 */ 1260 if (fp->f_count == 0) 1261 continue; 1262 /* 1263 * If all refs are from msgs, and it's not marked accessible 1264 * then it must be referenced from some unreachable cycle 1265 * of (shut-down) FDs, so include it in our 1266 * list of FDs to remove 1267 */ 1268 if (fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) { 1269 *fpp++ = fp; 1270 nunref++; 1271 fp->f_count++; 1272 } 1273 } 1274 /* 1275 * for each FD on our hit list, do the following two things 1276 */ 1277 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1278 struct file *tfp = *fpp; 1279 if (tfp->f_type == DTYPE_SOCKET && tfp->f_data != NULL) 1280 sorflush((struct socket *)(tfp->f_data)); 1281 } 1282 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) 1283 closef(*fpp, NULL); 1284 free((caddr_t)extra_ref, M_FILE); 1285 unp_gcing = 0; 1286 } 1287 1288 void 1289 unp_dispose(struct mbuf *m) 1290 { 1291 if (m) 1292 unp_scan(m, unp_discard); 1293 } 1294 1295 static int 1296 unp_listen(struct unpcb *unp, struct thread *td) 1297 { 1298 struct proc *p = td->td_proc; 1299 1300 KKASSERT(p); 1301 cru2x(p->p_ucred, &unp->unp_peercred); 1302 unp->unp_flags |= UNP_HAVEPCCACHED; 1303 return (0); 1304 } 1305 1306 static void 1307 unp_scan(m0, op) 1308 struct mbuf *m0; 1309 void (*op) (struct file *); 1310 { 1311 struct mbuf *m; 1312 struct file **rp; 1313 struct cmsghdr *cm; 1314 int i; 1315 int qfds; 1316 1317 while (m0) { 1318 for (m = m0; m; m = m->m_next) 1319 if (m->m_type == MT_CONTROL && 1320 m->m_len >= sizeof(*cm)) { 1321 cm = mtod(m, struct cmsghdr *); 1322 if (cm->cmsg_level != SOL_SOCKET || 1323 cm->cmsg_type != SCM_RIGHTS) 1324 continue; 1325 qfds = (cm->cmsg_len - 1326 (CMSG_DATA(cm) - (u_char *)cm)) 1327 / sizeof (struct file *); 1328 rp = (struct file **)CMSG_DATA(cm); 1329 for (i = 0; i < qfds; i++) 1330 (*op)(*rp++); 1331 break; /* XXX, but saves time */ 1332 } 1333 m0 = m0->m_nextpkt; 1334 } 1335 } 1336 1337 static void 1338 unp_mark(fp) 1339 struct file *fp; 1340 { 1341 1342 if (fp->f_flag & FMARK) 1343 return; 1344 unp_defer++; 1345 fp->f_flag |= (FMARK|FDEFER); 1346 } 1347 1348 static void 1349 unp_discard(fp) 1350 struct file *fp; 1351 { 1352 1353 fp->f_msgcount--; 1354 unp_rights--; 1355 (void) closef(fp, NULL); 1356 } 1357