1 /* $NetBSD: uipc_usrreq.c,v 1.119 2008/10/11 13:40:57 pooka Exp $ */ 2 3 /*- 4 * Copyright (c) 1998, 2000, 2004, 2008 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility, 9 * NASA Ames Research Center. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 30 * POSSIBILITY OF SUCH DAMAGE. 31 */ 32 33 /* 34 * Copyright (c) 1982, 1986, 1989, 1991, 1993 35 * The Regents of the University of California. All rights reserved. 36 * 37 * Redistribution and use in source and binary forms, with or without 38 * modification, are permitted provided that the following conditions 39 * are met: 40 * 1. Redistributions of source code must retain the above copyright 41 * notice, this list of conditions and the following disclaimer. 42 * 2. Redistributions in binary form must reproduce the above copyright 43 * notice, this list of conditions and the following disclaimer in the 44 * documentation and/or other materials provided with the distribution. 45 * 3. Neither the name of the University nor the names of its contributors 46 * may be used to endorse or promote products derived from this software 47 * without specific prior written permission. 48 * 49 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 50 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 51 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 52 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 53 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 54 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 55 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 56 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 57 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 58 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 59 * SUCH DAMAGE. 60 * 61 * @(#)uipc_usrreq.c 8.9 (Berkeley) 5/14/95 62 */ 63 64 /* 65 * Copyright (c) 1997 Christopher G. Demetriou. All rights reserved. 66 * 67 * Redistribution and use in source and binary forms, with or without 68 * modification, are permitted provided that the following conditions 69 * are met: 70 * 1. Redistributions of source code must retain the above copyright 71 * notice, this list of conditions and the following disclaimer. 72 * 2. Redistributions in binary form must reproduce the above copyright 73 * notice, this list of conditions and the following disclaimer in the 74 * documentation and/or other materials provided with the distribution. 75 * 3. All advertising materials mentioning features or use of this software 76 * must display the following acknowledgement: 77 * This product includes software developed by the University of 78 * California, Berkeley and its contributors. 79 * 4. Neither the name of the University nor the names of its contributors 80 * may be used to endorse or promote products derived from this software 81 * without specific prior written permission. 82 * 83 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 84 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 85 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 86 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 87 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 88 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 89 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 90 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 91 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 92 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 93 * SUCH DAMAGE. 94 * 95 * @(#)uipc_usrreq.c 8.9 (Berkeley) 5/14/95 96 */ 97 98 #include <sys/cdefs.h> 99 __KERNEL_RCSID(0, "$NetBSD: uipc_usrreq.c,v 1.119 2008/10/11 13:40:57 pooka Exp $"); 100 101 #include <sys/param.h> 102 #include <sys/systm.h> 103 #include <sys/proc.h> 104 #include <sys/filedesc.h> 105 #include <sys/domain.h> 106 #include <sys/protosw.h> 107 #include <sys/socket.h> 108 #include <sys/socketvar.h> 109 #include <sys/unpcb.h> 110 #include <sys/un.h> 111 #include <sys/namei.h> 112 #include <sys/vnode.h> 113 #include <sys/file.h> 114 #include <sys/stat.h> 115 #include <sys/mbuf.h> 116 #include <sys/kauth.h> 117 #include <sys/kmem.h> 118 #include <sys/atomic.h> 119 #include <sys/uidinfo.h> 120 121 /* 122 * Unix communications domain. 123 * 124 * TODO: 125 * SEQPACKET, RDM 126 * rethink name space problems 127 * need a proper out-of-band 128 * 129 * Notes on locking: 130 * 131 * The generic rules noted in uipc_socket2.c apply. In addition: 132 * 133 * o We have a global lock, uipc_lock. 134 * 135 * o All datagram sockets are locked by uipc_lock. 136 * 137 * o For stream socketpairs, the two endpoints are created sharing the same 138 * independent lock. Sockets presented to PRU_CONNECT2 must already have 139 * matching locks. 140 * 141 * o Stream sockets created via socket() start life with their own 142 * independent lock. 143 * 144 * o Stream connections to a named endpoint are slightly more complicated. 145 * Sockets that have called listen() have their lock pointer mutated to 146 * the global uipc_lock. When establishing a connection, the connecting 147 * socket also has its lock mutated to uipc_lock, which matches the head 148 * (listening socket). We create a new socket for accept() to return, and 149 * that also shares the head's lock. Until the connection is completely 150 * done on both ends, all three sockets are locked by uipc_lock. Once the 151 * connection is complete, the association with the head's lock is broken. 152 * The connecting socket and the socket returned from accept() have their 153 * lock pointers mutated away from uipc_lock, and back to the connecting 154 * socket's original, independent lock. The head continues to be locked 155 * by uipc_lock. 156 * 157 * o If uipc_lock is determined to be a significant source of contention, 158 * it could easily be hashed out. It is difficult to simply make it an 159 * independent lock because of visibility / garbage collection issues: 160 * if a socket has been associated with a lock at any point, that lock 161 * must remain valid until the socket is no longer visible in the system. 162 * The lock must not be freed or otherwise destroyed until any sockets 163 * that had referenced it have also been destroyed. 164 */ 165 const struct sockaddr_un sun_noname = { 166 .sun_len = sizeof(sun_noname), 167 .sun_family = AF_LOCAL, 168 }; 169 ino_t unp_ino; /* prototype for fake inode numbers */ 170 171 struct mbuf *unp_addsockcred(struct lwp *, struct mbuf *); 172 static kmutex_t *uipc_lock; 173 174 /* 175 * Initialize Unix protocols. 176 */ 177 void 178 uipc_init(void) 179 { 180 181 uipc_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 182 } 183 184 /* 185 * A connection succeeded: disassociate both endpoints from the head's 186 * lock, and make them share their own lock. There is a race here: for 187 * a very brief time one endpoint will be locked by a different lock 188 * than the other end. However, since the current thread holds the old 189 * lock (the listening socket's lock, the head) access can still only be 190 * made to one side of the connection. 191 */ 192 static void 193 unp_setpeerlocks(struct socket *so, struct socket *so2) 194 { 195 struct unpcb *unp; 196 kmutex_t *lock; 197 198 KASSERT(solocked2(so, so2)); 199 200 /* 201 * Bail out if either end of the socket is not yet fully 202 * connected or accepted. We only break the lock association 203 * with the head when the pair of sockets stand completely 204 * on their own. 205 */ 206 if (so->so_head != NULL || so2->so_head != NULL) 207 return; 208 209 /* 210 * Drop references to old lock. A third reference (from the 211 * queue head) must be held as we still hold its lock. Bonus: 212 * we don't need to worry about garbage collecting the lock. 213 */ 214 lock = so->so_lock; 215 KASSERT(lock == uipc_lock); 216 mutex_obj_free(lock); 217 mutex_obj_free(lock); 218 219 /* 220 * Grab stream lock from the initiator and share between the two 221 * endpoints. Issue memory barrier to ensure all modifications 222 * become globally visible before the lock change. so2 is 223 * assumed not to have a stream lock, because it was created 224 * purely for the server side to accept this connection and 225 * started out life using the domain-wide lock. 226 */ 227 unp = sotounpcb(so); 228 KASSERT(unp->unp_streamlock != NULL); 229 KASSERT(sotounpcb(so2)->unp_streamlock == NULL); 230 lock = unp->unp_streamlock; 231 unp->unp_streamlock = NULL; 232 mutex_obj_hold(lock); 233 membar_exit(); 234 solockreset(so, lock); 235 solockreset(so2, lock); 236 } 237 238 /* 239 * Reset a socket's lock back to the domain-wide lock. 240 */ 241 static void 242 unp_resetlock(struct socket *so) 243 { 244 kmutex_t *olock, *nlock; 245 struct unpcb *unp; 246 247 KASSERT(solocked(so)); 248 249 olock = so->so_lock; 250 nlock = uipc_lock; 251 if (olock == nlock) 252 return; 253 unp = sotounpcb(so); 254 KASSERT(unp->unp_streamlock == NULL); 255 unp->unp_streamlock = olock; 256 mutex_obj_hold(nlock); 257 mutex_enter(nlock); 258 solockreset(so, nlock); 259 mutex_exit(olock); 260 } 261 262 static void 263 unp_free(struct unpcb *unp) 264 { 265 266 if (unp->unp_addr) 267 free(unp->unp_addr, M_SONAME); 268 if (unp->unp_streamlock != NULL) 269 mutex_obj_free(unp->unp_streamlock); 270 free(unp, M_PCB); 271 } 272 273 int 274 unp_output(struct mbuf *m, struct mbuf *control, struct unpcb *unp, 275 struct lwp *l) 276 { 277 struct socket *so2; 278 const struct sockaddr_un *sun; 279 280 so2 = unp->unp_conn->unp_socket; 281 282 KASSERT(solocked(so2)); 283 284 if (unp->unp_addr) 285 sun = unp->unp_addr; 286 else 287 sun = &sun_noname; 288 if (unp->unp_conn->unp_flags & UNP_WANTCRED) 289 control = unp_addsockcred(l, control); 290 if (sbappendaddr(&so2->so_rcv, (const struct sockaddr *)sun, m, 291 control) == 0) { 292 so2->so_rcv.sb_overflowed++; 293 sounlock(so2); 294 unp_dispose(control); 295 m_freem(control); 296 m_freem(m); 297 solock(so2); 298 return (ENOBUFS); 299 } else { 300 sorwakeup(so2); 301 return (0); 302 } 303 } 304 305 void 306 unp_setaddr(struct socket *so, struct mbuf *nam, bool peeraddr) 307 { 308 const struct sockaddr_un *sun; 309 struct unpcb *unp; 310 bool ext; 311 312 unp = sotounpcb(so); 313 ext = false; 314 315 for (;;) { 316 sun = NULL; 317 if (peeraddr) { 318 if (unp->unp_conn && unp->unp_conn->unp_addr) 319 sun = unp->unp_conn->unp_addr; 320 } else { 321 if (unp->unp_addr) 322 sun = unp->unp_addr; 323 } 324 if (sun == NULL) 325 sun = &sun_noname; 326 nam->m_len = sun->sun_len; 327 if (nam->m_len > MLEN && !ext) { 328 sounlock(so); 329 MEXTMALLOC(nam, MAXPATHLEN * 2, M_WAITOK); 330 solock(so); 331 ext = true; 332 } else { 333 KASSERT(nam->m_len <= MAXPATHLEN * 2); 334 memcpy(mtod(nam, void *), sun, (size_t)nam->m_len); 335 break; 336 } 337 } 338 } 339 340 /*ARGSUSED*/ 341 int 342 uipc_usrreq(struct socket *so, int req, struct mbuf *m, struct mbuf *nam, 343 struct mbuf *control, struct lwp *l) 344 { 345 struct unpcb *unp = sotounpcb(so); 346 struct socket *so2; 347 struct proc *p; 348 u_int newhiwat; 349 int error = 0; 350 351 if (req == PRU_CONTROL) 352 return (EOPNOTSUPP); 353 354 #ifdef DIAGNOSTIC 355 if (req != PRU_SEND && req != PRU_SENDOOB && control) 356 panic("uipc_usrreq: unexpected control mbuf"); 357 #endif 358 p = l ? l->l_proc : NULL; 359 if (req != PRU_ATTACH) { 360 if (unp == 0) { 361 error = EINVAL; 362 goto release; 363 } 364 KASSERT(solocked(so)); 365 } 366 367 switch (req) { 368 369 case PRU_ATTACH: 370 if (unp != 0) { 371 error = EISCONN; 372 break; 373 } 374 error = unp_attach(so); 375 break; 376 377 case PRU_DETACH: 378 unp_detach(unp); 379 break; 380 381 case PRU_BIND: 382 KASSERT(l != NULL); 383 error = unp_bind(so, nam, l); 384 break; 385 386 case PRU_LISTEN: 387 /* 388 * If the socket can accept a connection, it must be 389 * locked by uipc_lock. 390 */ 391 unp_resetlock(so); 392 if (unp->unp_vnode == 0) 393 error = EINVAL; 394 break; 395 396 case PRU_CONNECT: 397 KASSERT(l != NULL); 398 error = unp_connect(so, nam, l); 399 break; 400 401 case PRU_CONNECT2: 402 error = unp_connect2(so, (struct socket *)nam, PRU_CONNECT2); 403 break; 404 405 case PRU_DISCONNECT: 406 unp_disconnect(unp); 407 break; 408 409 case PRU_ACCEPT: 410 KASSERT(so->so_lock == uipc_lock); 411 /* 412 * Mark the initiating STREAM socket as connected *ONLY* 413 * after it's been accepted. This prevents a client from 414 * overrunning a server and receiving ECONNREFUSED. 415 */ 416 if (unp->unp_conn == NULL) 417 break; 418 so2 = unp->unp_conn->unp_socket; 419 if (so2->so_state & SS_ISCONNECTING) { 420 KASSERT(solocked2(so, so->so_head)); 421 KASSERT(solocked2(so2, so->so_head)); 422 soisconnected(so2); 423 } 424 /* 425 * If the connection is fully established, break the 426 * association with uipc_lock and give the connected 427 * pair a seperate lock to share. 428 */ 429 unp_setpeerlocks(so2, so); 430 /* 431 * Only now return peer's address, as we may need to 432 * block in order to allocate memory. 433 * 434 * XXX Minor race: connection can be broken while 435 * lock is dropped in unp_setaddr(). We will return 436 * error == 0 and sun_noname as the peer address. 437 */ 438 unp_setaddr(so, nam, true); 439 break; 440 441 case PRU_SHUTDOWN: 442 socantsendmore(so); 443 unp_shutdown(unp); 444 break; 445 446 case PRU_RCVD: 447 switch (so->so_type) { 448 449 case SOCK_DGRAM: 450 panic("uipc 1"); 451 /*NOTREACHED*/ 452 453 case SOCK_STREAM: 454 #define rcv (&so->so_rcv) 455 #define snd (&so2->so_snd) 456 if (unp->unp_conn == 0) 457 break; 458 so2 = unp->unp_conn->unp_socket; 459 KASSERT(solocked2(so, so2)); 460 /* 461 * Adjust backpressure on sender 462 * and wakeup any waiting to write. 463 */ 464 snd->sb_mbmax += unp->unp_mbcnt - rcv->sb_mbcnt; 465 unp->unp_mbcnt = rcv->sb_mbcnt; 466 newhiwat = snd->sb_hiwat + unp->unp_cc - rcv->sb_cc; 467 (void)chgsbsize(so2->so_uidinfo, 468 &snd->sb_hiwat, newhiwat, RLIM_INFINITY); 469 unp->unp_cc = rcv->sb_cc; 470 sowwakeup(so2); 471 #undef snd 472 #undef rcv 473 break; 474 475 default: 476 panic("uipc 2"); 477 } 478 break; 479 480 case PRU_SEND: 481 /* 482 * Note: unp_internalize() rejects any control message 483 * other than SCM_RIGHTS, and only allows one. This 484 * has the side-effect of preventing a caller from 485 * forging SCM_CREDS. 486 */ 487 if (control) { 488 sounlock(so); 489 error = unp_internalize(&control); 490 solock(so); 491 if (error != 0) { 492 m_freem(control); 493 m_freem(m); 494 break; 495 } 496 } 497 switch (so->so_type) { 498 499 case SOCK_DGRAM: { 500 KASSERT(so->so_lock == uipc_lock); 501 if (nam) { 502 if ((so->so_state & SS_ISCONNECTED) != 0) 503 error = EISCONN; 504 else { 505 /* 506 * Note: once connected, the 507 * socket's lock must not be 508 * dropped until we have sent 509 * the message and disconnected. 510 * This is necessary to prevent 511 * intervening control ops, like 512 * another connection. 513 */ 514 error = unp_connect(so, nam, l); 515 } 516 } else { 517 if ((so->so_state & SS_ISCONNECTED) == 0) 518 error = ENOTCONN; 519 } 520 if (error) { 521 sounlock(so); 522 unp_dispose(control); 523 m_freem(control); 524 m_freem(m); 525 solock(so); 526 break; 527 } 528 KASSERT(p != NULL); 529 error = unp_output(m, control, unp, l); 530 if (nam) 531 unp_disconnect(unp); 532 break; 533 } 534 535 case SOCK_STREAM: 536 #define rcv (&so2->so_rcv) 537 #define snd (&so->so_snd) 538 if (unp->unp_conn == NULL) { 539 error = ENOTCONN; 540 break; 541 } 542 so2 = unp->unp_conn->unp_socket; 543 KASSERT(solocked2(so, so2)); 544 if (unp->unp_conn->unp_flags & UNP_WANTCRED) { 545 /* 546 * Credentials are passed only once on 547 * SOCK_STREAM. 548 */ 549 unp->unp_conn->unp_flags &= ~UNP_WANTCRED; 550 control = unp_addsockcred(l, control); 551 } 552 /* 553 * Send to paired receive port, and then reduce 554 * send buffer hiwater marks to maintain backpressure. 555 * Wake up readers. 556 */ 557 if (control) { 558 if (sbappendcontrol(rcv, m, control) != 0) 559 control = NULL; 560 } else 561 sbappend(rcv, m); 562 snd->sb_mbmax -= 563 rcv->sb_mbcnt - unp->unp_conn->unp_mbcnt; 564 unp->unp_conn->unp_mbcnt = rcv->sb_mbcnt; 565 newhiwat = snd->sb_hiwat - 566 (rcv->sb_cc - unp->unp_conn->unp_cc); 567 (void)chgsbsize(so->so_uidinfo, 568 &snd->sb_hiwat, newhiwat, RLIM_INFINITY); 569 unp->unp_conn->unp_cc = rcv->sb_cc; 570 sorwakeup(so2); 571 #undef snd 572 #undef rcv 573 if (control != NULL) { 574 sounlock(so); 575 unp_dispose(control); 576 m_freem(control); 577 solock(so); 578 } 579 break; 580 581 default: 582 panic("uipc 4"); 583 } 584 break; 585 586 case PRU_ABORT: 587 (void)unp_drop(unp, ECONNABORTED); 588 589 KASSERT(so->so_head == NULL); 590 #ifdef DIAGNOSTIC 591 if (so->so_pcb == 0) 592 panic("uipc 5: drop killed pcb"); 593 #endif 594 unp_detach(unp); 595 break; 596 597 case PRU_SENSE: 598 ((struct stat *) m)->st_blksize = so->so_snd.sb_hiwat; 599 if (so->so_type == SOCK_STREAM && unp->unp_conn != 0) { 600 so2 = unp->unp_conn->unp_socket; 601 KASSERT(solocked2(so, so2)); 602 ((struct stat *) m)->st_blksize += so2->so_rcv.sb_cc; 603 } 604 ((struct stat *) m)->st_dev = NODEV; 605 if (unp->unp_ino == 0) 606 unp->unp_ino = unp_ino++; 607 ((struct stat *) m)->st_atimespec = 608 ((struct stat *) m)->st_mtimespec = 609 ((struct stat *) m)->st_ctimespec = unp->unp_ctime; 610 ((struct stat *) m)->st_ino = unp->unp_ino; 611 return (0); 612 613 case PRU_RCVOOB: 614 error = EOPNOTSUPP; 615 break; 616 617 case PRU_SENDOOB: 618 m_freem(control); 619 m_freem(m); 620 error = EOPNOTSUPP; 621 break; 622 623 case PRU_SOCKADDR: 624 unp_setaddr(so, nam, false); 625 break; 626 627 case PRU_PEERADDR: 628 unp_setaddr(so, nam, true); 629 break; 630 631 default: 632 panic("piusrreq"); 633 } 634 635 release: 636 return (error); 637 } 638 639 /* 640 * Unix domain socket option processing. 641 */ 642 int 643 uipc_ctloutput(int op, struct socket *so, struct sockopt *sopt) 644 { 645 struct unpcb *unp = sotounpcb(so); 646 int optval = 0, error = 0; 647 648 KASSERT(solocked(so)); 649 650 if (sopt->sopt_level != 0) { 651 error = ENOPROTOOPT; 652 } else switch (op) { 653 654 case PRCO_SETOPT: 655 switch (sopt->sopt_name) { 656 case LOCAL_CREDS: 657 case LOCAL_CONNWAIT: 658 error = sockopt_getint(sopt, &optval); 659 if (error) 660 break; 661 switch (sopt->sopt_name) { 662 #define OPTSET(bit) \ 663 if (optval) \ 664 unp->unp_flags |= (bit); \ 665 else \ 666 unp->unp_flags &= ~(bit); 667 668 case LOCAL_CREDS: 669 OPTSET(UNP_WANTCRED); 670 break; 671 case LOCAL_CONNWAIT: 672 OPTSET(UNP_CONNWAIT); 673 break; 674 } 675 break; 676 #undef OPTSET 677 678 default: 679 error = ENOPROTOOPT; 680 break; 681 } 682 break; 683 684 case PRCO_GETOPT: 685 sounlock(so); 686 switch (sopt->sopt_name) { 687 case LOCAL_PEEREID: 688 if (unp->unp_flags & UNP_EIDSVALID) { 689 error = sockopt_set(sopt, 690 &unp->unp_connid, sizeof(unp->unp_connid)); 691 } else { 692 error = EINVAL; 693 } 694 break; 695 case LOCAL_CREDS: 696 #define OPTBIT(bit) (unp->unp_flags & (bit) ? 1 : 0) 697 698 optval = OPTBIT(UNP_WANTCRED); 699 error = sockopt_setint(sopt, optval); 700 break; 701 #undef OPTBIT 702 703 default: 704 error = ENOPROTOOPT; 705 break; 706 } 707 solock(so); 708 break; 709 } 710 return (error); 711 } 712 713 /* 714 * Both send and receive buffers are allocated PIPSIZ bytes of buffering 715 * for stream sockets, although the total for sender and receiver is 716 * actually only PIPSIZ. 717 * Datagram sockets really use the sendspace as the maximum datagram size, 718 * and don't really want to reserve the sendspace. Their recvspace should 719 * be large enough for at least one max-size datagram plus address. 720 */ 721 #define PIPSIZ 4096 722 u_long unpst_sendspace = PIPSIZ; 723 u_long unpst_recvspace = PIPSIZ; 724 u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 725 u_long unpdg_recvspace = 4*1024; 726 727 u_int unp_rights; /* file descriptors in flight */ 728 729 int 730 unp_attach(struct socket *so) 731 { 732 struct unpcb *unp; 733 int error; 734 735 switch (so->so_type) { 736 case SOCK_STREAM: 737 if (so->so_lock == NULL) { 738 /* 739 * XXX Assuming that no socket locks are held, 740 * as this call may sleep. 741 */ 742 so->so_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 743 solock(so); 744 } 745 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 746 error = soreserve(so, unpst_sendspace, unpst_recvspace); 747 if (error != 0) 748 return (error); 749 } 750 break; 751 752 case SOCK_DGRAM: 753 if (so->so_lock == NULL) { 754 mutex_obj_hold(uipc_lock); 755 so->so_lock = uipc_lock; 756 solock(so); 757 } 758 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 759 error = soreserve(so, unpdg_sendspace, unpdg_recvspace); 760 if (error != 0) 761 return (error); 762 } 763 break; 764 765 default: 766 panic("unp_attach"); 767 } 768 KASSERT(solocked(so)); 769 unp = malloc(sizeof(*unp), M_PCB, M_NOWAIT); 770 if (unp == NULL) 771 return (ENOBUFS); 772 memset((void *)unp, 0, sizeof(*unp)); 773 unp->unp_socket = so; 774 so->so_pcb = unp; 775 nanotime(&unp->unp_ctime); 776 return (0); 777 } 778 779 void 780 unp_detach(struct unpcb *unp) 781 { 782 struct socket *so; 783 vnode_t *vp; 784 785 so = unp->unp_socket; 786 787 retry: 788 if ((vp = unp->unp_vnode) != NULL) { 789 sounlock(so); 790 /* Acquire v_interlock to protect against unp_connect(). */ 791 /* XXXAD racy */ 792 mutex_enter(&vp->v_interlock); 793 vp->v_socket = NULL; 794 vrelel(vp, 0); 795 solock(so); 796 unp->unp_vnode = NULL; 797 } 798 if (unp->unp_conn) 799 unp_disconnect(unp); 800 while (unp->unp_refs) { 801 KASSERT(solocked2(so, unp->unp_refs->unp_socket)); 802 if (unp_drop(unp->unp_refs, ECONNRESET)) { 803 solock(so); 804 goto retry; 805 } 806 } 807 soisdisconnected(so); 808 so->so_pcb = NULL; 809 if (unp_rights) { 810 /* 811 * Normally the receive buffer is flushed later, 812 * in sofree, but if our receive buffer holds references 813 * to descriptors that are now garbage, we will dispose 814 * of those descriptor references after the garbage collector 815 * gets them (resulting in a "panic: closef: count < 0"). 816 */ 817 sorflush(so); 818 unp_free(unp); 819 sounlock(so); 820 unp_gc(); 821 solock(so); 822 } else 823 unp_free(unp); 824 } 825 826 int 827 unp_bind(struct socket *so, struct mbuf *nam, struct lwp *l) 828 { 829 struct sockaddr_un *sun; 830 struct unpcb *unp; 831 vnode_t *vp; 832 struct vattr vattr; 833 size_t addrlen; 834 int error; 835 struct nameidata nd; 836 proc_t *p; 837 838 unp = sotounpcb(so); 839 if (unp->unp_vnode != NULL) 840 return (EINVAL); 841 if ((unp->unp_flags & UNP_BUSY) != 0) { 842 /* 843 * EALREADY may not be strictly accurate, but since this 844 * is a major application error it's hardly a big deal. 845 */ 846 return (EALREADY); 847 } 848 unp->unp_flags |= UNP_BUSY; 849 sounlock(so); 850 851 /* 852 * Allocate the new sockaddr. We have to allocate one 853 * extra byte so that we can ensure that the pathname 854 * is nul-terminated. 855 */ 856 p = l->l_proc; 857 addrlen = nam->m_len + 1; 858 sun = malloc(addrlen, M_SONAME, M_WAITOK); 859 m_copydata(nam, 0, nam->m_len, (void *)sun); 860 *(((char *)sun) + nam->m_len) = '\0'; 861 862 NDINIT(&nd, CREATE, FOLLOW | LOCKPARENT | TRYEMULROOT, UIO_SYSSPACE, 863 sun->sun_path); 864 865 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 866 if ((error = namei(&nd)) != 0) 867 goto bad; 868 vp = nd.ni_vp; 869 if (vp != NULL) { 870 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 871 if (nd.ni_dvp == vp) 872 vrele(nd.ni_dvp); 873 else 874 vput(nd.ni_dvp); 875 vrele(vp); 876 error = EADDRINUSE; 877 goto bad; 878 } 879 VATTR_NULL(&vattr); 880 vattr.va_type = VSOCK; 881 vattr.va_mode = ACCESSPERMS & ~(p->p_cwdi->cwdi_cmask); 882 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 883 if (error) 884 goto bad; 885 vp = nd.ni_vp; 886 solock(so); 887 vp->v_socket = unp->unp_socket; 888 unp->unp_vnode = vp; 889 unp->unp_addrlen = addrlen; 890 unp->unp_addr = sun; 891 unp->unp_connid.unp_pid = p->p_pid; 892 unp->unp_connid.unp_euid = kauth_cred_geteuid(l->l_cred); 893 unp->unp_connid.unp_egid = kauth_cred_getegid(l->l_cred); 894 unp->unp_flags |= UNP_EIDSBIND; 895 VOP_UNLOCK(vp, 0); 896 unp->unp_flags &= ~UNP_BUSY; 897 return (0); 898 899 bad: 900 free(sun, M_SONAME); 901 solock(so); 902 unp->unp_flags &= ~UNP_BUSY; 903 return (error); 904 } 905 906 int 907 unp_connect(struct socket *so, struct mbuf *nam, struct lwp *l) 908 { 909 struct sockaddr_un *sun; 910 vnode_t *vp; 911 struct socket *so2, *so3; 912 struct unpcb *unp, *unp2, *unp3; 913 size_t addrlen; 914 int error; 915 struct nameidata nd; 916 917 unp = sotounpcb(so); 918 if ((unp->unp_flags & UNP_BUSY) != 0) { 919 /* 920 * EALREADY may not be strictly accurate, but since this 921 * is a major application error it's hardly a big deal. 922 */ 923 return (EALREADY); 924 } 925 unp->unp_flags |= UNP_BUSY; 926 sounlock(so); 927 928 /* 929 * Allocate a temporary sockaddr. We have to allocate one extra 930 * byte so that we can ensure that the pathname is nul-terminated. 931 * When we establish the connection, we copy the other PCB's 932 * sockaddr to our own. 933 */ 934 addrlen = nam->m_len + 1; 935 sun = malloc(addrlen, M_SONAME, M_WAITOK); 936 m_copydata(nam, 0, nam->m_len, (void *)sun); 937 *(((char *)sun) + nam->m_len) = '\0'; 938 939 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, UIO_SYSSPACE, 940 sun->sun_path); 941 942 if ((error = namei(&nd)) != 0) 943 goto bad2; 944 vp = nd.ni_vp; 945 if (vp->v_type != VSOCK) { 946 error = ENOTSOCK; 947 goto bad; 948 } 949 if ((error = VOP_ACCESS(vp, VWRITE, l->l_cred)) != 0) 950 goto bad; 951 /* Acquire v_interlock to protect against unp_detach(). */ 952 mutex_enter(&vp->v_interlock); 953 so2 = vp->v_socket; 954 if (so2 == NULL) { 955 mutex_exit(&vp->v_interlock); 956 error = ECONNREFUSED; 957 goto bad; 958 } 959 if (so->so_type != so2->so_type) { 960 mutex_exit(&vp->v_interlock); 961 error = EPROTOTYPE; 962 goto bad; 963 } 964 solock(so); 965 unp_resetlock(so); 966 mutex_exit(&vp->v_interlock); 967 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) { 968 /* 969 * This may seem somewhat fragile but is OK: if we can 970 * see SO_ACCEPTCONN set on the endpoint, then it must 971 * be locked by the domain-wide uipc_lock. 972 */ 973 KASSERT((so->so_options & SO_ACCEPTCONN) == 0 || 974 so2->so_lock == uipc_lock); 975 if ((so2->so_options & SO_ACCEPTCONN) == 0 || 976 (so3 = sonewconn(so2, 0)) == 0) { 977 error = ECONNREFUSED; 978 sounlock(so); 979 goto bad; 980 } 981 unp2 = sotounpcb(so2); 982 unp3 = sotounpcb(so3); 983 if (unp2->unp_addr) { 984 unp3->unp_addr = malloc(unp2->unp_addrlen, 985 M_SONAME, M_WAITOK); 986 memcpy(unp3->unp_addr, unp2->unp_addr, 987 unp2->unp_addrlen); 988 unp3->unp_addrlen = unp2->unp_addrlen; 989 } 990 unp3->unp_flags = unp2->unp_flags; 991 unp3->unp_connid.unp_pid = l->l_proc->p_pid; 992 unp3->unp_connid.unp_euid = kauth_cred_geteuid(l->l_cred); 993 unp3->unp_connid.unp_egid = kauth_cred_getegid(l->l_cred); 994 unp3->unp_flags |= UNP_EIDSVALID; 995 if (unp2->unp_flags & UNP_EIDSBIND) { 996 unp->unp_connid = unp2->unp_connid; 997 unp->unp_flags |= UNP_EIDSVALID; 998 } 999 so2 = so3; 1000 } 1001 error = unp_connect2(so, so2, PRU_CONNECT); 1002 sounlock(so); 1003 bad: 1004 vput(vp); 1005 bad2: 1006 free(sun, M_SONAME); 1007 solock(so); 1008 unp->unp_flags &= ~UNP_BUSY; 1009 return (error); 1010 } 1011 1012 int 1013 unp_connect2(struct socket *so, struct socket *so2, int req) 1014 { 1015 struct unpcb *unp = sotounpcb(so); 1016 struct unpcb *unp2; 1017 1018 if (so2->so_type != so->so_type) 1019 return (EPROTOTYPE); 1020 1021 /* 1022 * All three sockets involved must be locked by same lock: 1023 * 1024 * local endpoint (so) 1025 * remote endpoint (so2) 1026 * queue head (so->so_head, only if PR_CONNREQUIRED) 1027 */ 1028 KASSERT(solocked2(so, so2)); 1029 if (so->so_head != NULL) { 1030 KASSERT(so->so_lock == uipc_lock); 1031 KASSERT(solocked2(so, so->so_head)); 1032 } 1033 1034 unp2 = sotounpcb(so2); 1035 unp->unp_conn = unp2; 1036 switch (so->so_type) { 1037 1038 case SOCK_DGRAM: 1039 unp->unp_nextref = unp2->unp_refs; 1040 unp2->unp_refs = unp; 1041 soisconnected(so); 1042 break; 1043 1044 case SOCK_STREAM: 1045 unp2->unp_conn = unp; 1046 if (req == PRU_CONNECT && 1047 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 1048 soisconnecting(so); 1049 else 1050 soisconnected(so); 1051 soisconnected(so2); 1052 /* 1053 * If the connection is fully established, break the 1054 * association with uipc_lock and give the connected 1055 * pair a seperate lock to share. For CONNECT2, we 1056 * require that the locks already match (the sockets 1057 * are created that way). 1058 */ 1059 if (req == PRU_CONNECT) 1060 unp_setpeerlocks(so, so2); 1061 break; 1062 1063 default: 1064 panic("unp_connect2"); 1065 } 1066 return (0); 1067 } 1068 1069 void 1070 unp_disconnect(struct unpcb *unp) 1071 { 1072 struct unpcb *unp2 = unp->unp_conn; 1073 struct socket *so; 1074 1075 if (unp2 == 0) 1076 return; 1077 unp->unp_conn = 0; 1078 so = unp->unp_socket; 1079 switch (so->so_type) { 1080 case SOCK_DGRAM: 1081 if (unp2->unp_refs == unp) 1082 unp2->unp_refs = unp->unp_nextref; 1083 else { 1084 unp2 = unp2->unp_refs; 1085 for (;;) { 1086 KASSERT(solocked2(so, unp2->unp_socket)); 1087 if (unp2 == 0) 1088 panic("unp_disconnect"); 1089 if (unp2->unp_nextref == unp) 1090 break; 1091 unp2 = unp2->unp_nextref; 1092 } 1093 unp2->unp_nextref = unp->unp_nextref; 1094 } 1095 unp->unp_nextref = 0; 1096 so->so_state &= ~SS_ISCONNECTED; 1097 break; 1098 1099 case SOCK_STREAM: 1100 KASSERT(solocked2(so, unp2->unp_socket)); 1101 soisdisconnected(so); 1102 unp2->unp_conn = 0; 1103 soisdisconnected(unp2->unp_socket); 1104 break; 1105 } 1106 } 1107 1108 #ifdef notdef 1109 unp_abort(struct unpcb *unp) 1110 { 1111 unp_detach(unp); 1112 } 1113 #endif 1114 1115 void 1116 unp_shutdown(struct unpcb *unp) 1117 { 1118 struct socket *so; 1119 1120 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn && 1121 (so = unp->unp_conn->unp_socket)) 1122 socantrcvmore(so); 1123 } 1124 1125 bool 1126 unp_drop(struct unpcb *unp, int errno) 1127 { 1128 struct socket *so = unp->unp_socket; 1129 1130 KASSERT(solocked(so)); 1131 1132 so->so_error = errno; 1133 unp_disconnect(unp); 1134 if (so->so_head) { 1135 so->so_pcb = NULL; 1136 /* sofree() drops the socket lock */ 1137 sofree(so); 1138 unp_free(unp); 1139 return true; 1140 } 1141 return false; 1142 } 1143 1144 #ifdef notdef 1145 unp_drain(void) 1146 { 1147 1148 } 1149 #endif 1150 1151 int 1152 unp_externalize(struct mbuf *rights, struct lwp *l) 1153 { 1154 struct cmsghdr *cm = mtod(rights, struct cmsghdr *); 1155 struct proc *p = l->l_proc; 1156 int i, *fdp; 1157 file_t **rp; 1158 file_t *fp; 1159 int nfds, error = 0; 1160 1161 nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) / 1162 sizeof(file_t *); 1163 rp = (file_t **)CMSG_DATA(cm); 1164 1165 fdp = malloc(nfds * sizeof(int), M_TEMP, M_WAITOK); 1166 rw_enter(&p->p_cwdi->cwdi_lock, RW_READER); 1167 1168 /* Make sure the recipient should be able to see the descriptors.. */ 1169 if (p->p_cwdi->cwdi_rdir != NULL) { 1170 rp = (file_t **)CMSG_DATA(cm); 1171 for (i = 0; i < nfds; i++) { 1172 fp = *rp++; 1173 /* 1174 * If we are in a chroot'ed directory, and 1175 * someone wants to pass us a directory, make 1176 * sure it's inside the subtree we're allowed 1177 * to access. 1178 */ 1179 if (fp->f_type == DTYPE_VNODE) { 1180 vnode_t *vp = (vnode_t *)fp->f_data; 1181 if ((vp->v_type == VDIR) && 1182 !vn_isunder(vp, p->p_cwdi->cwdi_rdir, l)) { 1183 error = EPERM; 1184 break; 1185 } 1186 } 1187 } 1188 } 1189 1190 restart: 1191 rp = (file_t **)CMSG_DATA(cm); 1192 if (error != 0) { 1193 for (i = 0; i < nfds; i++) { 1194 fp = *rp; 1195 /* 1196 * zero the pointer before calling unp_discard, 1197 * since it may end up in unp_gc().. 1198 */ 1199 *rp++ = 0; 1200 unp_discard(fp); 1201 } 1202 goto out; 1203 } 1204 1205 /* 1206 * First loop -- allocate file descriptor table slots for the 1207 * new descriptors. 1208 */ 1209 for (i = 0; i < nfds; i++) { 1210 fp = *rp++; 1211 if ((error = fd_alloc(p, 0, &fdp[i])) != 0) { 1212 /* 1213 * Back out what we've done so far. 1214 */ 1215 for (--i; i >= 0; i--) { 1216 fd_abort(p, NULL, fdp[i]); 1217 } 1218 if (error == ENOSPC) { 1219 fd_tryexpand(p); 1220 error = 0; 1221 } else { 1222 /* 1223 * This is the error that has historically 1224 * been returned, and some callers may 1225 * expect it. 1226 */ 1227 error = EMSGSIZE; 1228 } 1229 goto restart; 1230 } 1231 } 1232 1233 /* 1234 * Now that adding them has succeeded, update all of the 1235 * descriptor passing state. 1236 */ 1237 rp = (file_t **)CMSG_DATA(cm); 1238 for (i = 0; i < nfds; i++) { 1239 fp = *rp++; 1240 atomic_dec_uint(&unp_rights); 1241 fd_affix(p, fp, fdp[i]); 1242 mutex_enter(&fp->f_lock); 1243 fp->f_msgcount--; 1244 mutex_exit(&fp->f_lock); 1245 /* 1246 * Note that fd_affix() adds a reference to the file. 1247 * The file may already have been closed by another 1248 * LWP in the process, so we must drop the reference 1249 * added by unp_internalize() with closef(). 1250 */ 1251 closef(fp); 1252 } 1253 1254 /* 1255 * Copy temporary array to message and adjust length, in case of 1256 * transition from large file_t pointers to ints. 1257 */ 1258 memcpy(CMSG_DATA(cm), fdp, nfds * sizeof(int)); 1259 cm->cmsg_len = CMSG_LEN(nfds * sizeof(int)); 1260 rights->m_len = CMSG_SPACE(nfds * sizeof(int)); 1261 out: 1262 rw_exit(&p->p_cwdi->cwdi_lock); 1263 free(fdp, M_TEMP); 1264 return (error); 1265 } 1266 1267 int 1268 unp_internalize(struct mbuf **controlp) 1269 { 1270 struct filedesc *fdescp = curlwp->l_fd; 1271 struct mbuf *control = *controlp; 1272 struct cmsghdr *newcm, *cm = mtod(control, struct cmsghdr *); 1273 file_t **rp, **files; 1274 file_t *fp; 1275 int i, fd, *fdp; 1276 int nfds, error; 1277 1278 error = 0; 1279 newcm = NULL; 1280 1281 /* Sanity check the control message header. */ 1282 if (cm->cmsg_type != SCM_RIGHTS || cm->cmsg_level != SOL_SOCKET || 1283 cm->cmsg_len > control->m_len || 1284 cm->cmsg_len < CMSG_ALIGN(sizeof(*cm))) 1285 return (EINVAL); 1286 1287 /* 1288 * Verify that the file descriptors are valid, and acquire 1289 * a reference to each. 1290 */ 1291 nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) / sizeof(int); 1292 fdp = (int *)CMSG_DATA(cm); 1293 for (i = 0; i < nfds; i++) { 1294 fd = *fdp++; 1295 if ((fp = fd_getfile(fd)) == NULL) { 1296 nfds = i + 1; 1297 error = EBADF; 1298 goto out; 1299 } 1300 } 1301 1302 /* Allocate new space and copy header into it. */ 1303 newcm = malloc(CMSG_SPACE(nfds * sizeof(file_t *)), M_MBUF, M_WAITOK); 1304 if (newcm == NULL) { 1305 error = E2BIG; 1306 goto out; 1307 } 1308 memcpy(newcm, cm, sizeof(struct cmsghdr)); 1309 files = (file_t **)CMSG_DATA(newcm); 1310 1311 /* 1312 * Transform the file descriptors into file_t pointers, in 1313 * reverse order so that if pointers are bigger than ints, the 1314 * int won't get until we're done. No need to lock, as we have 1315 * already validated the descriptors with fd_getfile(). 1316 */ 1317 fdp = (int *)CMSG_DATA(cm) + nfds; 1318 rp = files + nfds; 1319 for (i = 0; i < nfds; i++) { 1320 fp = fdescp->fd_ofiles[*--fdp]->ff_file; 1321 KASSERT(fp != NULL); 1322 mutex_enter(&fp->f_lock); 1323 *--rp = fp; 1324 fp->f_count++; 1325 fp->f_msgcount++; 1326 mutex_exit(&fp->f_lock); 1327 atomic_inc_uint(&unp_rights); 1328 } 1329 1330 out: 1331 /* Release descriptor references. */ 1332 fdp = (int *)CMSG_DATA(cm); 1333 for (i = 0; i < nfds; i++) { 1334 fd_putfile(*fdp++); 1335 } 1336 1337 if (error == 0) { 1338 if (control->m_flags & M_EXT) { 1339 m_freem(control); 1340 *controlp = control = m_get(M_WAIT, MT_CONTROL); 1341 } 1342 MEXTADD(control, newcm, CMSG_SPACE(nfds * sizeof(file_t *)), 1343 M_MBUF, NULL, NULL); 1344 cm = newcm; 1345 /* 1346 * Adjust message & mbuf to note amount of space 1347 * actually used. 1348 */ 1349 cm->cmsg_len = CMSG_LEN(nfds * sizeof(file_t *)); 1350 control->m_len = CMSG_SPACE(nfds * sizeof(file_t *)); 1351 } 1352 1353 return error; 1354 } 1355 1356 struct mbuf * 1357 unp_addsockcred(struct lwp *l, struct mbuf *control) 1358 { 1359 struct cmsghdr *cmp; 1360 struct sockcred *sc; 1361 struct mbuf *m, *n; 1362 int len, space, i; 1363 1364 len = CMSG_LEN(SOCKCREDSIZE(kauth_cred_ngroups(l->l_cred))); 1365 space = CMSG_SPACE(SOCKCREDSIZE(kauth_cred_ngroups(l->l_cred))); 1366 1367 m = m_get(M_WAIT, MT_CONTROL); 1368 if (space > MLEN) { 1369 if (space > MCLBYTES) 1370 MEXTMALLOC(m, space, M_WAITOK); 1371 else 1372 m_clget(m, M_WAIT); 1373 if ((m->m_flags & M_EXT) == 0) { 1374 m_free(m); 1375 return (control); 1376 } 1377 } 1378 1379 m->m_len = space; 1380 m->m_next = NULL; 1381 cmp = mtod(m, struct cmsghdr *); 1382 sc = (struct sockcred *)CMSG_DATA(cmp); 1383 cmp->cmsg_len = len; 1384 cmp->cmsg_level = SOL_SOCKET; 1385 cmp->cmsg_type = SCM_CREDS; 1386 sc->sc_uid = kauth_cred_getuid(l->l_cred); 1387 sc->sc_euid = kauth_cred_geteuid(l->l_cred); 1388 sc->sc_gid = kauth_cred_getgid(l->l_cred); 1389 sc->sc_egid = kauth_cred_getegid(l->l_cred); 1390 sc->sc_ngroups = kauth_cred_ngroups(l->l_cred); 1391 for (i = 0; i < sc->sc_ngroups; i++) 1392 sc->sc_groups[i] = kauth_cred_group(l->l_cred, i); 1393 1394 /* 1395 * If a control message already exists, append us to the end. 1396 */ 1397 if (control != NULL) { 1398 for (n = control; n->m_next != NULL; n = n->m_next) 1399 ; 1400 n->m_next = m; 1401 } else 1402 control = m; 1403 1404 return (control); 1405 } 1406 1407 int unp_defer, unp_gcing; 1408 extern struct domain unixdomain; 1409 1410 /* 1411 * Comment added long after the fact explaining what's going on here. 1412 * Do a mark-sweep GC of file descriptors on the system, to free up 1413 * any which are caught in flight to an about-to-be-closed socket. 1414 * 1415 * Traditional mark-sweep gc's start at the "root", and mark 1416 * everything reachable from the root (which, in our case would be the 1417 * process table). The mark bits are cleared during the sweep. 1418 * 1419 * XXX For some inexplicable reason (perhaps because the file 1420 * descriptor tables used to live in the u area which could be swapped 1421 * out and thus hard to reach), we do multiple scans over the set of 1422 * descriptors, using use *two* mark bits per object (DEFER and MARK). 1423 * Whenever we find a descriptor which references other descriptors, 1424 * the ones it references are marked with both bits, and we iterate 1425 * over the whole file table until there are no more DEFER bits set. 1426 * We also make an extra pass *before* the GC to clear the mark bits, 1427 * which could have been cleared at almost no cost during the previous 1428 * sweep. 1429 */ 1430 void 1431 unp_gc(void) 1432 { 1433 file_t *fp, *nextfp; 1434 struct socket *so, *so1; 1435 file_t **extra_ref, **fpp; 1436 int nunref, nslots, i; 1437 1438 if (atomic_swap_uint(&unp_gcing, 1) == 1) 1439 return; 1440 1441 restart: 1442 nslots = nfiles * 2; 1443 extra_ref = kmem_alloc(nslots * sizeof(file_t *), KM_SLEEP); 1444 1445 mutex_enter(&filelist_lock); 1446 unp_defer = 0; 1447 1448 /* Clear mark bits */ 1449 LIST_FOREACH(fp, &filehead, f_list) { 1450 atomic_and_uint(&fp->f_flag, ~(FMARK|FDEFER)); 1451 } 1452 1453 /* 1454 * Iterate over the set of descriptors, marking ones believed 1455 * (based on refcount) to be referenced from a process, and 1456 * marking for rescan descriptors which are queued on a socket. 1457 */ 1458 do { 1459 LIST_FOREACH(fp, &filehead, f_list) { 1460 mutex_enter(&fp->f_lock); 1461 if (fp->f_flag & FDEFER) { 1462 atomic_and_uint(&fp->f_flag, ~FDEFER); 1463 unp_defer--; 1464 KASSERT(fp->f_count != 0); 1465 } else { 1466 if (fp->f_count == 0 || 1467 (fp->f_flag & FMARK) || 1468 fp->f_count == fp->f_msgcount) { 1469 mutex_exit(&fp->f_lock); 1470 continue; 1471 } 1472 } 1473 atomic_or_uint(&fp->f_flag, FMARK); 1474 1475 if (fp->f_type != DTYPE_SOCKET || 1476 (so = fp->f_data) == NULL || 1477 so->so_proto->pr_domain != &unixdomain || 1478 (so->so_proto->pr_flags&PR_RIGHTS) == 0) { 1479 mutex_exit(&fp->f_lock); 1480 continue; 1481 } 1482 #ifdef notdef 1483 if (so->so_rcv.sb_flags & SB_LOCK) { 1484 mutex_exit(&fp->f_lock); 1485 mutex_exit(&filelist_lock); 1486 kmem_free(extra_ref, nslots * sizeof(file_t *)); 1487 /* 1488 * This is problematical; it's not clear 1489 * we need to wait for the sockbuf to be 1490 * unlocked (on a uniprocessor, at least), 1491 * and it's also not clear what to do 1492 * if sbwait returns an error due to receipt 1493 * of a signal. If sbwait does return 1494 * an error, we'll go into an infinite 1495 * loop. Delete all of this for now. 1496 */ 1497 (void) sbwait(&so->so_rcv); 1498 goto restart; 1499 } 1500 #endif 1501 mutex_exit(&fp->f_lock); 1502 1503 /* 1504 * XXX Locking a socket with filelist_lock held 1505 * is ugly. filelist_lock can be taken by the 1506 * pagedaemon when reclaiming items from file_cache. 1507 * Socket activity could delay the pagedaemon. 1508 */ 1509 solock(so); 1510 unp_scan(so->so_rcv.sb_mb, unp_mark, 0); 1511 /* 1512 * Mark descriptors referenced from sockets queued 1513 * on the accept queue as well. 1514 */ 1515 if (so->so_options & SO_ACCEPTCONN) { 1516 TAILQ_FOREACH(so1, &so->so_q0, so_qe) { 1517 unp_scan(so1->so_rcv.sb_mb, unp_mark, 0); 1518 } 1519 TAILQ_FOREACH(so1, &so->so_q, so_qe) { 1520 unp_scan(so1->so_rcv.sb_mb, unp_mark, 0); 1521 } 1522 } 1523 sounlock(so); 1524 } 1525 } while (unp_defer); 1526 1527 /* 1528 * Sweep pass. Find unmarked descriptors, and free them. 1529 * 1530 * We grab an extra reference to each of the file table entries 1531 * that are not otherwise accessible and then free the rights 1532 * that are stored in messages on them. 1533 * 1534 * The bug in the original code is a little tricky, so I'll describe 1535 * what's wrong with it here. 1536 * 1537 * It is incorrect to simply unp_discard each entry for f_msgcount 1538 * times -- consider the case of sockets A and B that contain 1539 * references to each other. On a last close of some other socket, 1540 * we trigger a gc since the number of outstanding rights (unp_rights) 1541 * is non-zero. If during the sweep phase the gc code un_discards, 1542 * we end up doing a (full) closef on the descriptor. A closef on A 1543 * results in the following chain. Closef calls soo_close, which 1544 * calls soclose. Soclose calls first (through the switch 1545 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply 1546 * returns because the previous instance had set unp_gcing, and 1547 * we return all the way back to soclose, which marks the socket 1548 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush 1549 * to free up the rights that are queued in messages on the socket A, 1550 * i.e., the reference on B. The sorflush calls via the dom_dispose 1551 * switch unp_dispose, which unp_scans with unp_discard. This second 1552 * instance of unp_discard just calls closef on B. 1553 * 1554 * Well, a similar chain occurs on B, resulting in a sorflush on B, 1555 * which results in another closef on A. Unfortunately, A is already 1556 * being closed, and the descriptor has already been marked with 1557 * SS_NOFDREF, and soclose panics at this point. 1558 * 1559 * Here, we first take an extra reference to each inaccessible 1560 * descriptor. Then, if the inaccessible descriptor is a 1561 * socket, we call sorflush in case it is a Unix domain 1562 * socket. After we destroy all the rights carried in 1563 * messages, we do a last closef to get rid of our extra 1564 * reference. This is the last close, and the unp_detach etc 1565 * will shut down the socket. 1566 * 1567 * 91/09/19, bsy@cs.cmu.edu 1568 */ 1569 if (nslots < nfiles) { 1570 mutex_exit(&filelist_lock); 1571 kmem_free(extra_ref, nslots * sizeof(file_t *)); 1572 goto restart; 1573 } 1574 for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref; fp != 0; 1575 fp = nextfp) { 1576 nextfp = LIST_NEXT(fp, f_list); 1577 mutex_enter(&fp->f_lock); 1578 if (fp->f_count != 0 && 1579 fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) { 1580 *fpp++ = fp; 1581 nunref++; 1582 fp->f_count++; 1583 } 1584 mutex_exit(&fp->f_lock); 1585 } 1586 mutex_exit(&filelist_lock); 1587 1588 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1589 fp = *fpp; 1590 if (fp->f_type == DTYPE_SOCKET) { 1591 so = fp->f_data; 1592 solock(so); 1593 sorflush(fp->f_data); 1594 sounlock(so); 1595 } 1596 } 1597 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1598 closef(*fpp); 1599 } 1600 kmem_free(extra_ref, nslots * sizeof(file_t *)); 1601 atomic_swap_uint(&unp_gcing, 0); 1602 } 1603 1604 void 1605 unp_dispose(struct mbuf *m) 1606 { 1607 1608 if (m) 1609 unp_scan(m, unp_discard, 1); 1610 } 1611 1612 void 1613 unp_scan(struct mbuf *m0, void (*op)(file_t *), int discard) 1614 { 1615 struct mbuf *m; 1616 file_t **rp; 1617 struct cmsghdr *cm; 1618 int i; 1619 int qfds; 1620 1621 while (m0) { 1622 for (m = m0; m; m = m->m_next) { 1623 if (m->m_type == MT_CONTROL && 1624 m->m_len >= sizeof(*cm)) { 1625 cm = mtod(m, struct cmsghdr *); 1626 if (cm->cmsg_level != SOL_SOCKET || 1627 cm->cmsg_type != SCM_RIGHTS) 1628 continue; 1629 qfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) 1630 / sizeof(file_t *); 1631 rp = (file_t **)CMSG_DATA(cm); 1632 for (i = 0; i < qfds; i++) { 1633 file_t *fp = *rp; 1634 if (discard) 1635 *rp = 0; 1636 (*op)(fp); 1637 rp++; 1638 } 1639 break; /* XXX, but saves time */ 1640 } 1641 } 1642 m0 = m0->m_nextpkt; 1643 } 1644 } 1645 1646 void 1647 unp_mark(file_t *fp) 1648 { 1649 1650 if (fp == NULL) 1651 return; 1652 1653 /* If we're already deferred, don't screw up the defer count */ 1654 mutex_enter(&fp->f_lock); 1655 if (fp->f_flag & (FMARK | FDEFER)) { 1656 mutex_exit(&fp->f_lock); 1657 return; 1658 } 1659 1660 /* 1661 * Minimize the number of deferrals... Sockets are the only 1662 * type of descriptor which can hold references to another 1663 * descriptor, so just mark other descriptors, and defer 1664 * unmarked sockets for the next pass. 1665 */ 1666 if (fp->f_type == DTYPE_SOCKET) { 1667 unp_defer++; 1668 KASSERT(fp->f_count != 0); 1669 atomic_or_uint(&fp->f_flag, FDEFER); 1670 } else { 1671 atomic_or_uint(&fp->f_flag, FMARK); 1672 } 1673 mutex_exit(&fp->f_lock); 1674 return; 1675 } 1676 1677 void 1678 unp_discard(file_t *fp) 1679 { 1680 1681 if (fp == NULL) 1682 return; 1683 1684 mutex_enter(&fp->f_lock); 1685 KASSERT(fp->f_count > 0); 1686 fp->f_msgcount--; 1687 mutex_exit(&fp->f_lock); 1688 atomic_dec_uint(&unp_rights); 1689 (void)closef(fp); 1690 } 1691