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 * $FreeBSD$ 35 */ 36 37 #include <sys/param.h> 38 #include <sys/systm.h> 39 #include <sys/kernel.h> 40 #include <sys/domain.h> 41 #include <sys/fcntl.h> 42 #include <sys/malloc.h> /* XXX must be before <sys/file.h> */ 43 #include <sys/file.h> 44 #include <sys/filedesc.h> 45 #include <sys/mbuf.h> 46 #include <sys/namei.h> 47 #include <sys/proc.h> 48 #include <sys/protosw.h> 49 #include <sys/socket.h> 50 #include <sys/socketvar.h> 51 #include <sys/stat.h> 52 #include <sys/sysctl.h> 53 #include <sys/un.h> 54 #include <sys/unpcb.h> 55 #include <sys/vnode.h> 56 57 #include <vm/vm_zone.h> 58 59 static struct vm_zone *unp_zone; 60 static unp_gen_t unp_gencnt; 61 static u_int unp_count; 62 63 static struct unp_head unp_shead, unp_dhead; 64 65 /* 66 * Unix communications domain. 67 * 68 * TODO: 69 * SEQPACKET, RDM 70 * rethink name space problems 71 * need a proper out-of-band 72 * lock pushdown 73 */ 74 static struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL }; 75 static ino_t unp_ino; /* prototype for fake inode numbers */ 76 77 static int unp_attach __P((struct socket *)); 78 static void unp_detach __P((struct unpcb *)); 79 static int unp_bind __P((struct unpcb *,struct sockaddr *, struct proc *)); 80 static int unp_connect __P((struct socket *,struct sockaddr *, 81 struct proc *)); 82 static void unp_disconnect __P((struct unpcb *)); 83 static void unp_shutdown __P((struct unpcb *)); 84 static void unp_drop __P((struct unpcb *, int)); 85 static void unp_gc __P((void)); 86 static void unp_scan __P((struct mbuf *, void (*)(struct file *))); 87 static void unp_mark __P((struct file *)); 88 static void unp_discard __P((struct file *)); 89 static int unp_internalize __P((struct mbuf *, struct proc *)); 90 91 static int 92 uipc_abort(struct socket *so) 93 { 94 struct unpcb *unp = sotounpcb(so); 95 96 if (unp == 0) 97 return EINVAL; 98 unp_drop(unp, ECONNABORTED); 99 return 0; 100 } 101 102 static int 103 uipc_accept(struct socket *so, struct sockaddr **nam) 104 { 105 struct unpcb *unp = sotounpcb(so); 106 107 if (unp == 0) 108 return EINVAL; 109 110 /* 111 * Pass back name of connected socket, 112 * if it was bound and we are still connected 113 * (our peer may have closed already!). 114 */ 115 if (unp->unp_conn && unp->unp_conn->unp_addr) { 116 *nam = dup_sockaddr((struct sockaddr *)unp->unp_conn->unp_addr, 117 1); 118 } else { 119 *nam = dup_sockaddr((struct sockaddr *)&sun_noname, 1); 120 } 121 return 0; 122 } 123 124 static int 125 uipc_attach(struct socket *so, int proto, struct proc *p) 126 { 127 struct unpcb *unp = sotounpcb(so); 128 129 if (unp != 0) 130 return EISCONN; 131 return unp_attach(so); 132 } 133 134 static int 135 uipc_bind(struct socket *so, struct sockaddr *nam, struct proc *p) 136 { 137 struct unpcb *unp = sotounpcb(so); 138 139 if (unp == 0) 140 return EINVAL; 141 142 return unp_bind(unp, nam, p); 143 } 144 145 static int 146 uipc_connect(struct socket *so, struct sockaddr *nam, struct proc *p) 147 { 148 struct unpcb *unp = sotounpcb(so); 149 150 if (unp == 0) 151 return EINVAL; 152 return unp_connect(so, nam, curproc); 153 } 154 155 static int 156 uipc_connect2(struct socket *so1, struct socket *so2) 157 { 158 struct unpcb *unp = sotounpcb(so1); 159 160 if (unp == 0) 161 return EINVAL; 162 163 return unp_connect2(so1, so2); 164 } 165 166 /* control is EOPNOTSUPP */ 167 168 static int 169 uipc_detach(struct socket *so) 170 { 171 struct unpcb *unp = sotounpcb(so); 172 173 if (unp == 0) 174 return EINVAL; 175 176 unp_detach(unp); 177 return 0; 178 } 179 180 static int 181 uipc_disconnect(struct socket *so) 182 { 183 struct unpcb *unp = sotounpcb(so); 184 185 if (unp == 0) 186 return EINVAL; 187 unp_disconnect(unp); 188 return 0; 189 } 190 191 static int 192 uipc_listen(struct socket *so, struct proc *p) 193 { 194 struct unpcb *unp = sotounpcb(so); 195 196 if (unp == 0 || unp->unp_vnode == 0) 197 return EINVAL; 198 return 0; 199 } 200 201 static int 202 uipc_peeraddr(struct socket *so, struct sockaddr **nam) 203 { 204 struct unpcb *unp = sotounpcb(so); 205 206 if (unp == 0) 207 return EINVAL; 208 if (unp->unp_conn && unp->unp_conn->unp_addr) 209 *nam = dup_sockaddr((struct sockaddr *)unp->unp_conn->unp_addr, 210 1); 211 return 0; 212 } 213 214 static int 215 uipc_rcvd(struct socket *so, int flags) 216 { 217 struct unpcb *unp = sotounpcb(so); 218 struct socket *so2; 219 220 if (unp == 0) 221 return EINVAL; 222 switch (so->so_type) { 223 case SOCK_DGRAM: 224 panic("uipc_rcvd DGRAM?"); 225 /*NOTREACHED*/ 226 227 case SOCK_STREAM: 228 if (unp->unp_conn == 0) 229 break; 230 so2 = unp->unp_conn->unp_socket; 231 /* 232 * Adjust backpressure on sender 233 * and wakeup any waiting to write. 234 */ 235 so2->so_snd.sb_mbmax += unp->unp_mbcnt - so->so_rcv.sb_mbcnt; 236 unp->unp_mbcnt = so->so_rcv.sb_mbcnt; 237 so2->so_snd.sb_hiwat += unp->unp_cc - so->so_rcv.sb_cc; 238 (void)chgsbsize(so2->so_cred->cr_uid, 239 (rlim_t)unp->unp_cc - so->so_rcv.sb_cc); 240 unp->unp_cc = so->so_rcv.sb_cc; 241 sowwakeup(so2); 242 break; 243 244 default: 245 panic("uipc_rcvd unknown socktype"); 246 } 247 return 0; 248 } 249 250 /* pru_rcvoob is EOPNOTSUPP */ 251 252 static int 253 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 254 struct mbuf *control, struct proc *p) 255 { 256 int error = 0; 257 struct unpcb *unp = sotounpcb(so); 258 struct socket *so2; 259 260 if (unp == 0) { 261 error = EINVAL; 262 goto release; 263 } 264 if (flags & PRUS_OOB) { 265 error = EOPNOTSUPP; 266 goto release; 267 } 268 269 if (control && (error = unp_internalize(control, p))) 270 goto release; 271 272 switch (so->so_type) { 273 case SOCK_DGRAM: 274 { 275 struct sockaddr *from; 276 277 if (nam) { 278 if (unp->unp_conn) { 279 error = EISCONN; 280 break; 281 } 282 error = unp_connect(so, nam, p); 283 if (error) 284 break; 285 } else { 286 if (unp->unp_conn == 0) { 287 error = ENOTCONN; 288 break; 289 } 290 } 291 so2 = unp->unp_conn->unp_socket; 292 if (unp->unp_addr) 293 from = (struct sockaddr *)unp->unp_addr; 294 else 295 from = &sun_noname; 296 if (sbappendaddr(&so2->so_rcv, from, m, control)) { 297 sorwakeup(so2); 298 m = 0; 299 control = 0; 300 } else 301 error = ENOBUFS; 302 if (nam) 303 unp_disconnect(unp); 304 break; 305 } 306 307 case SOCK_STREAM: 308 /* Connect if not connected yet. */ 309 /* 310 * Note: A better implementation would complain 311 * if not equal to the peer's address. 312 */ 313 if ((so->so_state & SS_ISCONNECTED) == 0) { 314 if (nam) { 315 error = unp_connect(so, nam, p); 316 if (error) 317 break; /* XXX */ 318 } else { 319 error = ENOTCONN; 320 break; 321 } 322 } 323 324 if (so->so_state & SS_CANTSENDMORE) { 325 error = EPIPE; 326 break; 327 } 328 if (unp->unp_conn == 0) 329 panic("uipc_send connected but no connection?"); 330 so2 = unp->unp_conn->unp_socket; 331 /* 332 * Send to paired receive port, and then reduce 333 * send buffer hiwater marks to maintain backpressure. 334 * Wake up readers. 335 */ 336 if (control) { 337 if (sbappendcontrol(&so2->so_rcv, m, control)) 338 control = 0; 339 } else 340 sbappend(&so2->so_rcv, m); 341 so->so_snd.sb_mbmax -= 342 so2->so_rcv.sb_mbcnt - unp->unp_conn->unp_mbcnt; 343 unp->unp_conn->unp_mbcnt = so2->so_rcv.sb_mbcnt; 344 so->so_snd.sb_hiwat -= 345 so2->so_rcv.sb_cc - unp->unp_conn->unp_cc; 346 (void)chgsbsize(so->so_cred->cr_uid, 347 (rlim_t)unp->unp_conn->unp_cc - so2->so_rcv.sb_cc); 348 unp->unp_conn->unp_cc = so2->so_rcv.sb_cc; 349 sorwakeup(so2); 350 m = 0; 351 break; 352 353 default: 354 panic("uipc_send unknown socktype"); 355 } 356 357 /* 358 * SEND_EOF is equivalent to a SEND followed by 359 * a SHUTDOWN. 360 */ 361 if (flags & PRUS_EOF) { 362 socantsendmore(so); 363 unp_shutdown(unp); 364 } 365 366 if (control && error != 0) 367 unp_dispose(control); 368 369 release: 370 if (control) 371 m_freem(control); 372 if (m) 373 m_freem(m); 374 return error; 375 } 376 377 static int 378 uipc_sense(struct socket *so, struct stat *sb) 379 { 380 struct unpcb *unp = sotounpcb(so); 381 struct socket *so2; 382 383 if (unp == 0) 384 return EINVAL; 385 sb->st_blksize = so->so_snd.sb_hiwat; 386 if (so->so_type == SOCK_STREAM && unp->unp_conn != 0) { 387 so2 = unp->unp_conn->unp_socket; 388 sb->st_blksize += so2->so_rcv.sb_cc; 389 } 390 sb->st_dev = NOUDEV; 391 if (unp->unp_ino == 0) 392 unp->unp_ino = unp_ino++; 393 sb->st_ino = unp->unp_ino; 394 return (0); 395 } 396 397 static int 398 uipc_shutdown(struct socket *so) 399 { 400 struct unpcb *unp = sotounpcb(so); 401 402 if (unp == 0) 403 return EINVAL; 404 socantsendmore(so); 405 unp_shutdown(unp); 406 return 0; 407 } 408 409 static int 410 uipc_sockaddr(struct socket *so, struct sockaddr **nam) 411 { 412 struct unpcb *unp = sotounpcb(so); 413 414 if (unp == 0) 415 return EINVAL; 416 if (unp->unp_addr) 417 *nam = dup_sockaddr((struct sockaddr *)unp->unp_addr, 1); 418 return 0; 419 } 420 421 struct pr_usrreqs uipc_usrreqs = { 422 uipc_abort, uipc_accept, uipc_attach, uipc_bind, uipc_connect, 423 uipc_connect2, pru_control_notsupp, uipc_detach, uipc_disconnect, 424 uipc_listen, uipc_peeraddr, uipc_rcvd, pru_rcvoob_notsupp, 425 uipc_send, uipc_sense, uipc_shutdown, uipc_sockaddr, 426 sosend, soreceive, sopoll 427 }; 428 429 /* 430 * Both send and receive buffers are allocated PIPSIZ bytes of buffering 431 * for stream sockets, although the total for sender and receiver is 432 * actually only PIPSIZ. 433 * Datagram sockets really use the sendspace as the maximum datagram size, 434 * and don't really want to reserve the sendspace. Their recvspace should 435 * be large enough for at least one max-size datagram plus address. 436 */ 437 #ifndef PIPSIZ 438 #define PIPSIZ 8192 439 #endif 440 static u_long unpst_sendspace = PIPSIZ; 441 static u_long unpst_recvspace = PIPSIZ; 442 static u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 443 static u_long unpdg_recvspace = 4*1024; 444 445 static int unp_rights; /* file descriptors in flight */ 446 447 SYSCTL_DECL(_net_local_stream); 448 SYSCTL_INT(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 449 &unpst_sendspace, 0, ""); 450 SYSCTL_INT(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 451 &unpst_recvspace, 0, ""); 452 SYSCTL_DECL(_net_local_dgram); 453 SYSCTL_INT(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 454 &unpdg_sendspace, 0, ""); 455 SYSCTL_INT(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 456 &unpdg_recvspace, 0, ""); 457 SYSCTL_DECL(_net_local); 458 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, ""); 459 460 static int 461 unp_attach(so) 462 struct socket *so; 463 { 464 register struct unpcb *unp; 465 int error; 466 467 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 468 switch (so->so_type) { 469 470 case SOCK_STREAM: 471 error = soreserve(so, unpst_sendspace, unpst_recvspace); 472 break; 473 474 case SOCK_DGRAM: 475 error = soreserve(so, unpdg_sendspace, unpdg_recvspace); 476 break; 477 478 default: 479 panic("unp_attach"); 480 } 481 if (error) 482 return (error); 483 } 484 unp = zalloc(unp_zone); 485 if (unp == NULL) 486 return (ENOBUFS); 487 bzero(unp, sizeof *unp); 488 unp->unp_gencnt = ++unp_gencnt; 489 unp_count++; 490 LIST_INIT(&unp->unp_refs); 491 unp->unp_socket = so; 492 unp->unp_rvnode = curproc->p_fd->fd_rdir; 493 LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead 494 : &unp_shead, unp, unp_link); 495 so->so_pcb = (caddr_t)unp; 496 return (0); 497 } 498 499 static void 500 unp_detach(unp) 501 register struct unpcb *unp; 502 { 503 LIST_REMOVE(unp, unp_link); 504 unp->unp_gencnt = ++unp_gencnt; 505 --unp_count; 506 if (unp->unp_vnode) { 507 unp->unp_vnode->v_socket = 0; 508 vrele(unp->unp_vnode); 509 unp->unp_vnode = 0; 510 } 511 if (unp->unp_conn) 512 unp_disconnect(unp); 513 while (unp->unp_refs.lh_first) 514 unp_drop(unp->unp_refs.lh_first, ECONNRESET); 515 soisdisconnected(unp->unp_socket); 516 unp->unp_socket->so_pcb = 0; 517 if (unp_rights) { 518 /* 519 * Normally the receive buffer is flushed later, 520 * in sofree, but if our receive buffer holds references 521 * to descriptors that are now garbage, we will dispose 522 * of those descriptor references after the garbage collector 523 * gets them (resulting in a "panic: closef: count < 0"). 524 */ 525 sorflush(unp->unp_socket); 526 unp_gc(); 527 } 528 if (unp->unp_addr) 529 FREE(unp->unp_addr, M_SONAME); 530 zfree(unp_zone, unp); 531 } 532 533 static int 534 unp_bind(unp, nam, p) 535 struct unpcb *unp; 536 struct sockaddr *nam; 537 struct proc *p; 538 { 539 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 540 register struct vnode *vp; 541 struct vattr vattr; 542 int error, namelen; 543 struct nameidata nd; 544 char buf[SOCK_MAXADDRLEN]; 545 546 if (unp->unp_vnode != NULL) 547 return (EINVAL); 548 #define offsetof(s, e) ((char *)&((s *)0)->e - (char *)((s *)0)) 549 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 550 if (namelen <= 0) 551 return EINVAL; 552 strncpy(buf, soun->sun_path, namelen); 553 buf[namelen] = 0; /* null-terminate the string */ 554 NDINIT(&nd, CREATE, NOFOLLOW | LOCKPARENT, UIO_SYSSPACE, 555 buf, p); 556 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 557 error = namei(&nd); 558 if (error) 559 return (error); 560 vp = nd.ni_vp; 561 if (vp != NULL) { 562 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 563 if (nd.ni_dvp == vp) 564 vrele(nd.ni_dvp); 565 else 566 vput(nd.ni_dvp); 567 vrele(vp); 568 return (EADDRINUSE); 569 } 570 VATTR_NULL(&vattr); 571 vattr.va_type = VSOCK; 572 vattr.va_mode = (ACCESSPERMS & ~p->p_fd->fd_cmask); 573 VOP_LEASE(nd.ni_dvp, p, p->p_ucred, LEASE_WRITE); 574 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 575 vput(nd.ni_dvp); 576 if (error) 577 return (error); 578 vp = nd.ni_vp; 579 vp->v_socket = unp->unp_socket; 580 unp->unp_vnode = vp; 581 unp->unp_addr = (struct sockaddr_un *)dup_sockaddr(nam, 1); 582 VOP_UNLOCK(vp, 0, p); 583 return (0); 584 } 585 586 static int 587 unp_connect(so, nam, p) 588 struct socket *so; 589 struct sockaddr *nam; 590 struct proc *p; 591 { 592 register struct sockaddr_un *soun = (struct sockaddr_un *)nam; 593 register struct vnode *vp; 594 register struct socket *so2, *so3; 595 struct unpcb *unp2, *unp3; 596 int error, len; 597 struct nameidata nd; 598 char buf[SOCK_MAXADDRLEN]; 599 600 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 601 if (len <= 0) 602 return EINVAL; 603 strncpy(buf, soun->sun_path, len); 604 buf[len] = 0; 605 606 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, p); 607 error = namei(&nd); 608 if (error) 609 return (error); 610 vp = nd.ni_vp; 611 if (vp->v_type != VSOCK) { 612 error = ENOTSOCK; 613 goto bad; 614 } 615 error = VOP_ACCESS(vp, VWRITE, p->p_ucred, p); 616 if (error) 617 goto bad; 618 so2 = vp->v_socket; 619 if (so2 == 0) { 620 error = ECONNREFUSED; 621 goto bad; 622 } 623 if (so->so_type != so2->so_type) { 624 error = EPROTOTYPE; 625 goto bad; 626 } 627 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 628 if ((so2->so_options & SO_ACCEPTCONN) == 0 || 629 (so3 = sonewconn3(so2, 0, p)) == 0) { 630 error = ECONNREFUSED; 631 goto bad; 632 } 633 unp2 = sotounpcb(so2); 634 unp3 = sotounpcb(so3); 635 if (unp2->unp_addr) 636 unp3->unp_addr = (struct sockaddr_un *) 637 dup_sockaddr((struct sockaddr *) 638 unp2->unp_addr, 1); 639 so2 = so3; 640 } 641 error = unp_connect2(so, so2); 642 bad: 643 vput(vp); 644 return (error); 645 } 646 647 int 648 unp_connect2(so, so2) 649 register struct socket *so; 650 register struct socket *so2; 651 { 652 register struct unpcb *unp = sotounpcb(so); 653 register struct unpcb *unp2; 654 655 if (so2->so_type != so->so_type) 656 return (EPROTOTYPE); 657 unp2 = sotounpcb(so2); 658 unp->unp_conn = unp2; 659 switch (so->so_type) { 660 661 case SOCK_DGRAM: 662 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 663 soisconnected(so); 664 break; 665 666 case SOCK_STREAM: 667 unp2->unp_conn = unp; 668 soisconnected(so); 669 soisconnected(so2); 670 break; 671 672 default: 673 panic("unp_connect2"); 674 } 675 return (0); 676 } 677 678 static void 679 unp_disconnect(unp) 680 struct unpcb *unp; 681 { 682 register struct unpcb *unp2 = unp->unp_conn; 683 684 if (unp2 == 0) 685 return; 686 unp->unp_conn = 0; 687 switch (unp->unp_socket->so_type) { 688 689 case SOCK_DGRAM: 690 LIST_REMOVE(unp, unp_reflink); 691 unp->unp_socket->so_state &= ~SS_ISCONNECTED; 692 break; 693 694 case SOCK_STREAM: 695 soisdisconnected(unp->unp_socket); 696 unp2->unp_conn = 0; 697 soisdisconnected(unp2->unp_socket); 698 break; 699 } 700 } 701 702 #ifdef notdef 703 void 704 unp_abort(unp) 705 struct unpcb *unp; 706 { 707 708 unp_detach(unp); 709 } 710 #endif 711 712 static int 713 prison_unpcb(struct proc *p, struct unpcb *unp) 714 { 715 if (!p->p_prison) 716 return (0); 717 if (p->p_fd->fd_rdir == unp->unp_rvnode) 718 return (0); 719 return (1); 720 } 721 722 static int 723 unp_pcblist SYSCTL_HANDLER_ARGS 724 { 725 int error, i, n; 726 struct unpcb *unp, **unp_list; 727 unp_gen_t gencnt; 728 struct xunpgen xug; 729 struct unp_head *head; 730 731 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead); 732 733 /* 734 * The process of preparing the PCB list is too time-consuming and 735 * resource-intensive to repeat twice on every request. 736 */ 737 if (req->oldptr == 0) { 738 n = unp_count; 739 req->oldidx = 2 * (sizeof xug) 740 + (n + n/8) * sizeof(struct xunpcb); 741 return 0; 742 } 743 744 if (req->newptr != 0) 745 return EPERM; 746 747 /* 748 * OK, now we're committed to doing something. 749 */ 750 gencnt = unp_gencnt; 751 n = unp_count; 752 753 xug.xug_len = sizeof xug; 754 xug.xug_count = n; 755 xug.xug_gen = gencnt; 756 xug.xug_sogen = so_gencnt; 757 error = SYSCTL_OUT(req, &xug, sizeof xug); 758 if (error) 759 return error; 760 761 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 762 if (unp_list == 0) 763 return ENOMEM; 764 765 for (unp = head->lh_first, i = 0; unp && i < n; 766 unp = unp->unp_link.le_next) { 767 if (unp->unp_gencnt <= gencnt && !prison_unpcb(req->p, unp)) 768 unp_list[i++] = unp; 769 } 770 n = i; /* in case we lost some during malloc */ 771 772 error = 0; 773 for (i = 0; i < n; i++) { 774 unp = unp_list[i]; 775 if (unp->unp_gencnt <= gencnt) { 776 struct xunpcb xu; 777 xu.xu_len = sizeof xu; 778 xu.xu_unpp = unp; 779 /* 780 * XXX - need more locking here to protect against 781 * connect/disconnect races for SMP. 782 */ 783 if (unp->unp_addr) 784 bcopy(unp->unp_addr, &xu.xu_addr, 785 unp->unp_addr->sun_len); 786 if (unp->unp_conn && unp->unp_conn->unp_addr) 787 bcopy(unp->unp_conn->unp_addr, 788 &xu.xu_caddr, 789 unp->unp_conn->unp_addr->sun_len); 790 bcopy(unp, &xu.xu_unp, sizeof *unp); 791 sotoxsocket(unp->unp_socket, &xu.xu_socket); 792 error = SYSCTL_OUT(req, &xu, sizeof xu); 793 } 794 } 795 if (!error) { 796 /* 797 * Give the user an updated idea of our state. 798 * If the generation differs from what we told 799 * her before, she knows that something happened 800 * while we were processing this request, and it 801 * might be necessary to retry. 802 */ 803 xug.xug_gen = unp_gencnt; 804 xug.xug_sogen = so_gencnt; 805 xug.xug_count = unp_count; 806 error = SYSCTL_OUT(req, &xug, sizeof xug); 807 } 808 free(unp_list, M_TEMP); 809 return error; 810 } 811 812 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD, 813 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 814 "List of active local datagram sockets"); 815 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD, 816 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 817 "List of active local stream sockets"); 818 819 static void 820 unp_shutdown(unp) 821 struct unpcb *unp; 822 { 823 struct socket *so; 824 825 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn && 826 (so = unp->unp_conn->unp_socket)) 827 socantrcvmore(so); 828 } 829 830 static void 831 unp_drop(unp, errno) 832 struct unpcb *unp; 833 int errno; 834 { 835 struct socket *so = unp->unp_socket; 836 837 so->so_error = errno; 838 unp_disconnect(unp); 839 if (so->so_head) { 840 LIST_REMOVE(unp, unp_link); 841 unp->unp_gencnt = ++unp_gencnt; 842 unp_count--; 843 so->so_pcb = (caddr_t) 0; 844 if (unp->unp_addr) 845 FREE(unp->unp_addr, M_SONAME); 846 zfree(unp_zone, unp); 847 sofree(so); 848 } 849 } 850 851 #ifdef notdef 852 void 853 unp_drain() 854 { 855 856 } 857 #endif 858 859 int 860 unp_externalize(rights) 861 struct mbuf *rights; 862 { 863 struct proc *p = curproc; /* XXX */ 864 register int i; 865 register struct cmsghdr *cm = mtod(rights, struct cmsghdr *); 866 register struct file **rp = (struct file **)(cm + 1); 867 register struct file *fp; 868 int newfds = (cm->cmsg_len - sizeof(*cm)) / sizeof (int); 869 int f; 870 871 /* 872 * if the new FD's will not fit, then we free them all 873 */ 874 if (!fdavail(p, newfds)) { 875 for (i = 0; i < newfds; i++) { 876 fp = *rp; 877 unp_discard(fp); 878 *rp++ = 0; 879 } 880 return (EMSGSIZE); 881 } 882 /* 883 * now change each pointer to an fd in the global table to 884 * an integer that is the index to the local fd table entry 885 * that we set up to point to the global one we are transferring. 886 * XXX this assumes a pointer and int are the same size...! 887 */ 888 for (i = 0; i < newfds; i++) { 889 if (fdalloc(p, 0, &f)) 890 panic("unp_externalize"); 891 fp = *rp; 892 p->p_fd->fd_ofiles[f] = fp; 893 fp->f_msgcount--; 894 unp_rights--; 895 *(int *)rp++ = f; 896 } 897 return (0); 898 } 899 900 void 901 unp_init(void) 902 { 903 unp_zone = zinit("unpcb", sizeof(struct unpcb), nmbclusters, 0, 0); 904 if (unp_zone == 0) 905 panic("unp_init"); 906 LIST_INIT(&unp_dhead); 907 LIST_INIT(&unp_shead); 908 } 909 910 #ifndef MIN 911 #define MIN(a,b) (((a)<(b))?(a):(b)) 912 #endif 913 914 static int 915 unp_internalize(control, p) 916 struct mbuf *control; 917 struct proc *p; 918 { 919 struct filedesc *fdp = p->p_fd; 920 register struct cmsghdr *cm = mtod(control, struct cmsghdr *); 921 register struct file **rp; 922 register struct file *fp; 923 register int i, fd; 924 register struct cmsgcred *cmcred; 925 int oldfds; 926 927 if ((cm->cmsg_type != SCM_RIGHTS && cm->cmsg_type != SCM_CREDS) || 928 cm->cmsg_level != SOL_SOCKET || cm->cmsg_len != control->m_len) 929 return (EINVAL); 930 931 /* 932 * Fill in credential information. 933 */ 934 if (cm->cmsg_type == SCM_CREDS) { 935 cmcred = (struct cmsgcred *)(cm + 1); 936 cmcred->cmcred_pid = p->p_pid; 937 cmcred->cmcred_uid = p->p_cred->p_ruid; 938 cmcred->cmcred_gid = p->p_cred->p_rgid; 939 cmcred->cmcred_euid = p->p_ucred->cr_uid; 940 cmcred->cmcred_ngroups = MIN(p->p_ucred->cr_ngroups, 941 CMGROUP_MAX); 942 for (i = 0; i < cmcred->cmcred_ngroups; i++) 943 cmcred->cmcred_groups[i] = p->p_ucred->cr_groups[i]; 944 return(0); 945 } 946 947 oldfds = (cm->cmsg_len - sizeof (*cm)) / sizeof (int); 948 /* 949 * check that all the FDs passed in refer to legal OPEN files 950 * If not, reject the entire operation. 951 */ 952 rp = (struct file **)(cm + 1); 953 for (i = 0; i < oldfds; i++) { 954 fd = *(int *)rp++; 955 if ((unsigned)fd >= fdp->fd_nfiles || 956 fdp->fd_ofiles[fd] == NULL) 957 return (EBADF); 958 } 959 /* 960 * Now replace the integer FDs with pointers to 961 * the associated global file table entry.. 962 * XXX this assumes a pointer and an int are the same size! 963 */ 964 rp = (struct file **)(cm + 1); 965 for (i = 0; i < oldfds; i++) { 966 fp = fdp->fd_ofiles[*(int *)rp]; 967 *rp++ = fp; 968 fp->f_count++; 969 fp->f_msgcount++; 970 unp_rights++; 971 } 972 return (0); 973 } 974 975 static int unp_defer, unp_gcing; 976 977 static void 978 unp_gc() 979 { 980 register struct file *fp, *nextfp; 981 register struct socket *so; 982 struct file **extra_ref, **fpp; 983 int nunref, i; 984 985 if (unp_gcing) 986 return; 987 unp_gcing = 1; 988 unp_defer = 0; 989 /* 990 * before going through all this, set all FDs to 991 * be NOT defered and NOT externally accessible 992 */ 993 for (fp = filehead.lh_first; fp != 0; fp = fp->f_list.le_next) 994 fp->f_flag &= ~(FMARK|FDEFER); 995 do { 996 for (fp = filehead.lh_first; fp != 0; fp = fp->f_list.le_next) { 997 /* 998 * If the file is not open, skip it 999 */ 1000 if (fp->f_count == 0) 1001 continue; 1002 /* 1003 * If we already marked it as 'defer' in a 1004 * previous pass, then try process it this time 1005 * and un-mark it 1006 */ 1007 if (fp->f_flag & FDEFER) { 1008 fp->f_flag &= ~FDEFER; 1009 unp_defer--; 1010 } else { 1011 /* 1012 * if it's not defered, then check if it's 1013 * already marked.. if so skip it 1014 */ 1015 if (fp->f_flag & FMARK) 1016 continue; 1017 /* 1018 * If all references are from messages 1019 * in transit, then skip it. it's not 1020 * externally accessible. 1021 */ 1022 if (fp->f_count == fp->f_msgcount) 1023 continue; 1024 /* 1025 * If it got this far then it must be 1026 * externally accessible. 1027 */ 1028 fp->f_flag |= FMARK; 1029 } 1030 /* 1031 * either it was defered, or it is externally 1032 * accessible and not already marked so. 1033 * Now check if it is possibly one of OUR sockets. 1034 */ 1035 if (fp->f_type != DTYPE_SOCKET || 1036 (so = (struct socket *)fp->f_data) == 0) 1037 continue; 1038 if (so->so_proto->pr_domain != &localdomain || 1039 (so->so_proto->pr_flags&PR_RIGHTS) == 0) 1040 continue; 1041 #ifdef notdef 1042 if (so->so_rcv.sb_flags & SB_LOCK) { 1043 /* 1044 * This is problematical; it's not clear 1045 * we need to wait for the sockbuf to be 1046 * unlocked (on a uniprocessor, at least), 1047 * and it's also not clear what to do 1048 * if sbwait returns an error due to receipt 1049 * of a signal. If sbwait does return 1050 * an error, we'll go into an infinite 1051 * loop. Delete all of this for now. 1052 */ 1053 (void) sbwait(&so->so_rcv); 1054 goto restart; 1055 } 1056 #endif 1057 /* 1058 * So, Ok, it's one of our sockets and it IS externally 1059 * accessible (or was defered). Now we look 1060 * to see if we hold any file descriptors in its 1061 * message buffers. Follow those links and mark them 1062 * as accessible too. 1063 */ 1064 unp_scan(so->so_rcv.sb_mb, unp_mark); 1065 } 1066 } while (unp_defer); 1067 /* 1068 * We grab an extra reference to each of the file table entries 1069 * that are not otherwise accessible and then free the rights 1070 * that are stored in messages on them. 1071 * 1072 * The bug in the orginal code is a little tricky, so I'll describe 1073 * what's wrong with it here. 1074 * 1075 * It is incorrect to simply unp_discard each entry for f_msgcount 1076 * times -- consider the case of sockets A and B that contain 1077 * references to each other. On a last close of some other socket, 1078 * we trigger a gc since the number of outstanding rights (unp_rights) 1079 * is non-zero. If during the sweep phase the gc code un_discards, 1080 * we end up doing a (full) closef on the descriptor. A closef on A 1081 * results in the following chain. Closef calls soo_close, which 1082 * calls soclose. Soclose calls first (through the switch 1083 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply 1084 * returns because the previous instance had set unp_gcing, and 1085 * we return all the way back to soclose, which marks the socket 1086 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush 1087 * to free up the rights that are queued in messages on the socket A, 1088 * i.e., the reference on B. The sorflush calls via the dom_dispose 1089 * switch unp_dispose, which unp_scans with unp_discard. This second 1090 * instance of unp_discard just calls closef on B. 1091 * 1092 * Well, a similar chain occurs on B, resulting in a sorflush on B, 1093 * which results in another closef on A. Unfortunately, A is already 1094 * being closed, and the descriptor has already been marked with 1095 * SS_NOFDREF, and soclose panics at this point. 1096 * 1097 * Here, we first take an extra reference to each inaccessible 1098 * descriptor. Then, we call sorflush ourself, since we know 1099 * it is a Unix domain socket anyhow. After we destroy all the 1100 * rights carried in messages, we do a last closef to get rid 1101 * of our extra reference. This is the last close, and the 1102 * unp_detach etc will shut down the socket. 1103 * 1104 * 91/09/19, bsy@cs.cmu.edu 1105 */ 1106 extra_ref = malloc(nfiles * sizeof(struct file *), M_FILE, M_WAITOK); 1107 for (nunref = 0, fp = filehead.lh_first, fpp = extra_ref; fp != 0; 1108 fp = nextfp) { 1109 nextfp = fp->f_list.le_next; 1110 /* 1111 * If it's not open, skip it 1112 */ 1113 if (fp->f_count == 0) 1114 continue; 1115 /* 1116 * If all refs are from msgs, and it's not marked accessible 1117 * then it must be referenced from some unreachable cycle 1118 * of (shut-down) FDs, so include it in our 1119 * list of FDs to remove 1120 */ 1121 if (fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) { 1122 *fpp++ = fp; 1123 nunref++; 1124 fp->f_count++; 1125 } 1126 } 1127 /* 1128 * for each FD on our hit list, do the following two things 1129 */ 1130 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1131 struct file *tfp = *fpp; 1132 if (tfp->f_type == DTYPE_SOCKET && tfp->f_data != NULL) 1133 sorflush((struct socket *)(tfp->f_data)); 1134 } 1135 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) 1136 closef(*fpp, (struct proc *) NULL); 1137 free((caddr_t)extra_ref, M_FILE); 1138 unp_gcing = 0; 1139 } 1140 1141 void 1142 unp_dispose(m) 1143 struct mbuf *m; 1144 { 1145 1146 if (m) 1147 unp_scan(m, unp_discard); 1148 } 1149 1150 static void 1151 unp_scan(m0, op) 1152 register struct mbuf *m0; 1153 void (*op) __P((struct file *)); 1154 { 1155 register struct mbuf *m; 1156 register struct file **rp; 1157 register struct cmsghdr *cm; 1158 register int i; 1159 int qfds; 1160 1161 while (m0) { 1162 for (m = m0; m; m = m->m_next) 1163 if (m->m_type == MT_CONTROL && 1164 m->m_len >= sizeof(*cm)) { 1165 cm = mtod(m, struct cmsghdr *); 1166 if (cm->cmsg_level != SOL_SOCKET || 1167 cm->cmsg_type != SCM_RIGHTS) 1168 continue; 1169 qfds = (cm->cmsg_len - sizeof *cm) 1170 / sizeof (struct file *); 1171 rp = (struct file **)(cm + 1); 1172 for (i = 0; i < qfds; i++) 1173 (*op)(*rp++); 1174 break; /* XXX, but saves time */ 1175 } 1176 m0 = m0->m_act; 1177 } 1178 } 1179 1180 static void 1181 unp_mark(fp) 1182 struct file *fp; 1183 { 1184 1185 if (fp->f_flag & FMARK) 1186 return; 1187 unp_defer++; 1188 fp->f_flag |= (FMARK|FDEFER); 1189 } 1190 1191 static void 1192 unp_discard(fp) 1193 struct file *fp; 1194 { 1195 1196 fp->f_msgcount--; 1197 unp_rights--; 1198 (void) closef(fp, (struct proc *)NULL); 1199 } 1200