1 /* $NetBSD: uipc_usrreq.c,v 1.152 2014/05/19 02:51:24 rmind Exp $ */ 2 3 /*- 4 * Copyright (c) 1998, 2000, 2004, 2008, 2009 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, and by Andrew Doran. 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 1.152 2014/05/19 02:51:24 rmind 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 #include <sys/kernel.h> 121 #include <sys/kthread.h> 122 123 /* 124 * Unix communications domain. 125 * 126 * TODO: 127 * RDM 128 * rethink name space problems 129 * need a proper out-of-band 130 * 131 * Notes on locking: 132 * 133 * The generic rules noted in uipc_socket2.c apply. In addition: 134 * 135 * o We have a global lock, uipc_lock. 136 * 137 * o All datagram sockets are locked by uipc_lock. 138 * 139 * o For stream socketpairs, the two endpoints are created sharing the same 140 * independent lock. Sockets presented to PRU_CONNECT2 must already have 141 * matching locks. 142 * 143 * o Stream sockets created via socket() start life with their own 144 * independent lock. 145 * 146 * o Stream connections to a named endpoint are slightly more complicated. 147 * Sockets that have called listen() have their lock pointer mutated to 148 * the global uipc_lock. When establishing a connection, the connecting 149 * socket also has its lock mutated to uipc_lock, which matches the head 150 * (listening socket). We create a new socket for accept() to return, and 151 * that also shares the head's lock. Until the connection is completely 152 * done on both ends, all three sockets are locked by uipc_lock. Once the 153 * connection is complete, the association with the head's lock is broken. 154 * The connecting socket and the socket returned from accept() have their 155 * lock pointers mutated away from uipc_lock, and back to the connecting 156 * socket's original, independent lock. The head continues to be locked 157 * by uipc_lock. 158 * 159 * o If uipc_lock is determined to be a significant source of contention, 160 * it could easily be hashed out. It is difficult to simply make it an 161 * independent lock because of visibility / garbage collection issues: 162 * if a socket has been associated with a lock at any point, that lock 163 * must remain valid until the socket is no longer visible in the system. 164 * The lock must not be freed or otherwise destroyed until any sockets 165 * that had referenced it have also been destroyed. 166 */ 167 const struct sockaddr_un sun_noname = { 168 .sun_len = offsetof(struct sockaddr_un, sun_path), 169 .sun_family = AF_LOCAL, 170 }; 171 ino_t unp_ino; /* prototype for fake inode numbers */ 172 173 static void unp_detach(struct socket *); 174 struct mbuf *unp_addsockcred(struct lwp *, struct mbuf *); 175 static void unp_mark(file_t *); 176 static void unp_scan(struct mbuf *, void (*)(file_t *), int); 177 static void unp_discard_now(file_t *); 178 static void unp_discard_later(file_t *); 179 static void unp_thread(void *); 180 static void unp_thread_kick(void); 181 static kmutex_t *uipc_lock; 182 183 static kcondvar_t unp_thread_cv; 184 static lwp_t *unp_thread_lwp; 185 static SLIST_HEAD(,file) unp_thread_discard; 186 static int unp_defer; 187 188 /* 189 * Initialize Unix protocols. 190 */ 191 void 192 uipc_init(void) 193 { 194 int error; 195 196 uipc_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 197 cv_init(&unp_thread_cv, "unpgc"); 198 199 error = kthread_create(PRI_NONE, KTHREAD_MPSAFE, NULL, unp_thread, 200 NULL, &unp_thread_lwp, "unpgc"); 201 if (error != 0) 202 panic("uipc_init %d", error); 203 } 204 205 /* 206 * A connection succeeded: disassociate both endpoints from the head's 207 * lock, and make them share their own lock. There is a race here: for 208 * a very brief time one endpoint will be locked by a different lock 209 * than the other end. However, since the current thread holds the old 210 * lock (the listening socket's lock, the head) access can still only be 211 * made to one side of the connection. 212 */ 213 static void 214 unp_setpeerlocks(struct socket *so, struct socket *so2) 215 { 216 struct unpcb *unp; 217 kmutex_t *lock; 218 219 KASSERT(solocked2(so, so2)); 220 221 /* 222 * Bail out if either end of the socket is not yet fully 223 * connected or accepted. We only break the lock association 224 * with the head when the pair of sockets stand completely 225 * on their own. 226 */ 227 KASSERT(so->so_head == NULL); 228 if (so2->so_head != NULL) 229 return; 230 231 /* 232 * Drop references to old lock. A third reference (from the 233 * queue head) must be held as we still hold its lock. Bonus: 234 * we don't need to worry about garbage collecting the lock. 235 */ 236 lock = so->so_lock; 237 KASSERT(lock == uipc_lock); 238 mutex_obj_free(lock); 239 mutex_obj_free(lock); 240 241 /* 242 * Grab stream lock from the initiator and share between the two 243 * endpoints. Issue memory barrier to ensure all modifications 244 * become globally visible before the lock change. so2 is 245 * assumed not to have a stream lock, because it was created 246 * purely for the server side to accept this connection and 247 * started out life using the domain-wide lock. 248 */ 249 unp = sotounpcb(so); 250 KASSERT(unp->unp_streamlock != NULL); 251 KASSERT(sotounpcb(so2)->unp_streamlock == NULL); 252 lock = unp->unp_streamlock; 253 unp->unp_streamlock = NULL; 254 mutex_obj_hold(lock); 255 membar_exit(); 256 /* 257 * possible race if lock is not held - see comment in 258 * uipc_usrreq(PRU_ACCEPT). 259 */ 260 KASSERT(mutex_owned(lock)); 261 solockreset(so, lock); 262 solockreset(so2, lock); 263 } 264 265 /* 266 * Reset a socket's lock back to the domain-wide lock. 267 */ 268 static void 269 unp_resetlock(struct socket *so) 270 { 271 kmutex_t *olock, *nlock; 272 struct unpcb *unp; 273 274 KASSERT(solocked(so)); 275 276 olock = so->so_lock; 277 nlock = uipc_lock; 278 if (olock == nlock) 279 return; 280 unp = sotounpcb(so); 281 KASSERT(unp->unp_streamlock == NULL); 282 unp->unp_streamlock = olock; 283 mutex_obj_hold(nlock); 284 mutex_enter(nlock); 285 solockreset(so, nlock); 286 mutex_exit(olock); 287 } 288 289 static void 290 unp_free(struct unpcb *unp) 291 { 292 if (unp->unp_addr) 293 free(unp->unp_addr, M_SONAME); 294 if (unp->unp_streamlock != NULL) 295 mutex_obj_free(unp->unp_streamlock); 296 kmem_free(unp, sizeof(*unp)); 297 } 298 299 int 300 unp_output(struct mbuf *m, struct mbuf *control, struct unpcb *unp, 301 struct lwp *l) 302 { 303 struct socket *so2; 304 const struct sockaddr_un *sun; 305 306 so2 = unp->unp_conn->unp_socket; 307 308 KASSERT(solocked(so2)); 309 310 if (unp->unp_addr) 311 sun = unp->unp_addr; 312 else 313 sun = &sun_noname; 314 if (unp->unp_conn->unp_flags & UNP_WANTCRED) 315 control = unp_addsockcred(l, control); 316 if (sbappendaddr(&so2->so_rcv, (const struct sockaddr *)sun, m, 317 control) == 0) { 318 so2->so_rcv.sb_overflowed++; 319 unp_dispose(control); 320 m_freem(control); 321 m_freem(m); 322 return (ENOBUFS); 323 } else { 324 sorwakeup(so2); 325 return (0); 326 } 327 } 328 329 void 330 unp_setaddr(struct socket *so, struct mbuf *nam, bool peeraddr) 331 { 332 const struct sockaddr_un *sun; 333 struct unpcb *unp; 334 bool ext; 335 336 KASSERT(solocked(so)); 337 unp = sotounpcb(so); 338 ext = false; 339 340 for (;;) { 341 sun = NULL; 342 if (peeraddr) { 343 if (unp->unp_conn && unp->unp_conn->unp_addr) 344 sun = unp->unp_conn->unp_addr; 345 } else { 346 if (unp->unp_addr) 347 sun = unp->unp_addr; 348 } 349 if (sun == NULL) 350 sun = &sun_noname; 351 nam->m_len = sun->sun_len; 352 if (nam->m_len > MLEN && !ext) { 353 sounlock(so); 354 MEXTMALLOC(nam, MAXPATHLEN * 2, M_WAITOK); 355 solock(so); 356 ext = true; 357 } else { 358 KASSERT(nam->m_len <= MAXPATHLEN * 2); 359 memcpy(mtod(nam, void *), sun, (size_t)nam->m_len); 360 break; 361 } 362 } 363 } 364 365 static int 366 unp_usrreq(struct socket *so, int req, struct mbuf *m, struct mbuf *nam, 367 struct mbuf *control, struct lwp *l) 368 { 369 struct unpcb *unp; 370 struct socket *so2; 371 u_int newhiwat; 372 int error = 0; 373 374 KASSERT(req != PRU_ATTACH); 375 KASSERT(req != PRU_DETACH); 376 377 if (req == PRU_CONTROL) { 378 return EOPNOTSUPP; 379 } 380 KASSERT(solocked(so)); 381 unp = sotounpcb(so); 382 383 KASSERT(!control || (req == PRU_SEND || req == PRU_SENDOOB)); 384 if (unp == NULL) { 385 error = EINVAL; 386 goto release; 387 } 388 389 switch (req) { 390 case PRU_BIND: 391 KASSERT(l != NULL); 392 error = unp_bind(so, nam, l); 393 break; 394 395 case PRU_LISTEN: 396 /* 397 * If the socket can accept a connection, it must be 398 * locked by uipc_lock. 399 */ 400 unp_resetlock(so); 401 if (unp->unp_vnode == NULL) 402 error = EINVAL; 403 break; 404 405 case PRU_CONNECT: 406 KASSERT(l != NULL); 407 error = unp_connect(so, nam, l); 408 break; 409 410 case PRU_CONNECT2: 411 error = unp_connect2(so, (struct socket *)nam, PRU_CONNECT2); 412 break; 413 414 case PRU_DISCONNECT: 415 unp_disconnect(unp); 416 break; 417 418 case PRU_ACCEPT: 419 KASSERT(so->so_lock == uipc_lock); 420 /* 421 * Mark the initiating STREAM socket as connected *ONLY* 422 * after it's been accepted. This prevents a client from 423 * overrunning a server and receiving ECONNREFUSED. 424 */ 425 if (unp->unp_conn == NULL) { 426 /* 427 * This will use the empty socket and will not 428 * allocate. 429 */ 430 unp_setaddr(so, nam, true); 431 break; 432 } 433 so2 = unp->unp_conn->unp_socket; 434 if (so2->so_state & SS_ISCONNECTING) { 435 KASSERT(solocked2(so, so->so_head)); 436 KASSERT(solocked2(so2, so->so_head)); 437 soisconnected(so2); 438 } 439 /* 440 * If the connection is fully established, break the 441 * association with uipc_lock and give the connected 442 * pair a separate lock to share. 443 * There is a race here: sotounpcb(so2)->unp_streamlock 444 * is not locked, so when changing so2->so_lock 445 * another thread can grab it while so->so_lock is still 446 * pointing to the (locked) uipc_lock. 447 * this should be harmless, except that this makes 448 * solocked2() and solocked() unreliable. 449 * Another problem is that unp_setaddr() expects the 450 * the socket locked. Grabing sotounpcb(so2)->unp_streamlock 451 * fixes both issues. 452 */ 453 mutex_enter(sotounpcb(so2)->unp_streamlock); 454 unp_setpeerlocks(so2, so); 455 /* 456 * Only now return peer's address, as we may need to 457 * block in order to allocate memory. 458 * 459 * XXX Minor race: connection can be broken while 460 * lock is dropped in unp_setaddr(). We will return 461 * error == 0 and sun_noname as the peer address. 462 */ 463 unp_setaddr(so, nam, true); 464 /* so_lock now points to unp_streamlock */ 465 mutex_exit(so2->so_lock); 466 break; 467 468 case PRU_SHUTDOWN: 469 socantsendmore(so); 470 unp_shutdown(unp); 471 break; 472 473 case PRU_RCVD: 474 switch (so->so_type) { 475 476 case SOCK_DGRAM: 477 panic("uipc 1"); 478 /*NOTREACHED*/ 479 480 case SOCK_SEQPACKET: /* FALLTHROUGH */ 481 case SOCK_STREAM: 482 #define rcv (&so->so_rcv) 483 #define snd (&so2->so_snd) 484 if (unp->unp_conn == 0) 485 break; 486 so2 = unp->unp_conn->unp_socket; 487 KASSERT(solocked2(so, so2)); 488 /* 489 * Adjust backpressure on sender 490 * and wakeup any waiting to write. 491 */ 492 snd->sb_mbmax += unp->unp_mbcnt - rcv->sb_mbcnt; 493 unp->unp_mbcnt = rcv->sb_mbcnt; 494 newhiwat = snd->sb_hiwat + unp->unp_cc - rcv->sb_cc; 495 (void)chgsbsize(so2->so_uidinfo, 496 &snd->sb_hiwat, newhiwat, RLIM_INFINITY); 497 unp->unp_cc = rcv->sb_cc; 498 sowwakeup(so2); 499 #undef snd 500 #undef rcv 501 break; 502 503 default: 504 panic("uipc 2"); 505 } 506 break; 507 508 case PRU_SEND: 509 /* 510 * Note: unp_internalize() rejects any control message 511 * other than SCM_RIGHTS, and only allows one. This 512 * has the side-effect of preventing a caller from 513 * forging SCM_CREDS. 514 */ 515 if (control) { 516 sounlock(so); 517 error = unp_internalize(&control); 518 solock(so); 519 if (error != 0) { 520 m_freem(control); 521 m_freem(m); 522 break; 523 } 524 } 525 switch (so->so_type) { 526 527 case SOCK_DGRAM: { 528 KASSERT(so->so_lock == uipc_lock); 529 if (nam) { 530 if ((so->so_state & SS_ISCONNECTED) != 0) 531 error = EISCONN; 532 else { 533 /* 534 * Note: once connected, the 535 * socket's lock must not be 536 * dropped until we have sent 537 * the message and disconnected. 538 * This is necessary to prevent 539 * intervening control ops, like 540 * another connection. 541 */ 542 error = unp_connect(so, nam, l); 543 } 544 } else { 545 if ((so->so_state & SS_ISCONNECTED) == 0) 546 error = ENOTCONN; 547 } 548 if (error) { 549 unp_dispose(control); 550 m_freem(control); 551 m_freem(m); 552 break; 553 } 554 KASSERT(l != NULL); 555 error = unp_output(m, control, unp, l); 556 if (nam) 557 unp_disconnect(unp); 558 break; 559 } 560 561 case SOCK_SEQPACKET: /* FALLTHROUGH */ 562 case SOCK_STREAM: 563 #define rcv (&so2->so_rcv) 564 #define snd (&so->so_snd) 565 if (unp->unp_conn == NULL) { 566 error = ENOTCONN; 567 break; 568 } 569 so2 = unp->unp_conn->unp_socket; 570 KASSERT(solocked2(so, so2)); 571 if (unp->unp_conn->unp_flags & UNP_WANTCRED) { 572 /* 573 * Credentials are passed only once on 574 * SOCK_STREAM and SOCK_SEQPACKET. 575 */ 576 unp->unp_conn->unp_flags &= ~UNP_WANTCRED; 577 control = unp_addsockcred(l, control); 578 } 579 /* 580 * Send to paired receive port, and then reduce 581 * send buffer hiwater marks to maintain backpressure. 582 * Wake up readers. 583 */ 584 if (control) { 585 if (sbappendcontrol(rcv, m, control) != 0) 586 control = NULL; 587 } else { 588 switch(so->so_type) { 589 case SOCK_SEQPACKET: 590 sbappendrecord(rcv, m); 591 break; 592 case SOCK_STREAM: 593 sbappend(rcv, m); 594 break; 595 default: 596 panic("uipc_usrreq"); 597 break; 598 } 599 } 600 snd->sb_mbmax -= 601 rcv->sb_mbcnt - unp->unp_conn->unp_mbcnt; 602 unp->unp_conn->unp_mbcnt = rcv->sb_mbcnt; 603 newhiwat = snd->sb_hiwat - 604 (rcv->sb_cc - unp->unp_conn->unp_cc); 605 (void)chgsbsize(so->so_uidinfo, 606 &snd->sb_hiwat, newhiwat, RLIM_INFINITY); 607 unp->unp_conn->unp_cc = rcv->sb_cc; 608 sorwakeup(so2); 609 #undef snd 610 #undef rcv 611 if (control != NULL) { 612 unp_dispose(control); 613 m_freem(control); 614 } 615 break; 616 617 default: 618 panic("uipc 4"); 619 } 620 break; 621 622 case PRU_ABORT: 623 (void)unp_drop(unp, ECONNABORTED); 624 KASSERT(so->so_head == NULL); 625 KASSERT(so->so_pcb != NULL); 626 unp_detach(so); 627 break; 628 629 case PRU_SENSE: 630 ((struct stat *) m)->st_blksize = so->so_snd.sb_hiwat; 631 switch (so->so_type) { 632 case SOCK_SEQPACKET: /* FALLTHROUGH */ 633 case SOCK_STREAM: 634 if (unp->unp_conn == 0) 635 break; 636 637 so2 = unp->unp_conn->unp_socket; 638 KASSERT(solocked2(so, so2)); 639 ((struct stat *) m)->st_blksize += so2->so_rcv.sb_cc; 640 break; 641 default: 642 break; 643 } 644 ((struct stat *) m)->st_dev = NODEV; 645 if (unp->unp_ino == 0) 646 unp->unp_ino = unp_ino++; 647 ((struct stat *) m)->st_atimespec = 648 ((struct stat *) m)->st_mtimespec = 649 ((struct stat *) m)->st_ctimespec = unp->unp_ctime; 650 ((struct stat *) m)->st_ino = unp->unp_ino; 651 return (0); 652 653 case PRU_RCVOOB: 654 error = EOPNOTSUPP; 655 break; 656 657 case PRU_SENDOOB: 658 m_freem(control); 659 m_freem(m); 660 error = EOPNOTSUPP; 661 break; 662 663 case PRU_SOCKADDR: 664 unp_setaddr(so, nam, false); 665 break; 666 667 case PRU_PEERADDR: 668 unp_setaddr(so, nam, true); 669 break; 670 671 default: 672 panic("piusrreq"); 673 } 674 675 release: 676 return (error); 677 } 678 679 /* 680 * Unix domain socket option processing. 681 */ 682 int 683 uipc_ctloutput(int op, struct socket *so, struct sockopt *sopt) 684 { 685 struct unpcb *unp = sotounpcb(so); 686 int optval = 0, error = 0; 687 688 KASSERT(solocked(so)); 689 690 if (sopt->sopt_level != 0) { 691 error = ENOPROTOOPT; 692 } else switch (op) { 693 694 case PRCO_SETOPT: 695 switch (sopt->sopt_name) { 696 case LOCAL_CREDS: 697 case LOCAL_CONNWAIT: 698 error = sockopt_getint(sopt, &optval); 699 if (error) 700 break; 701 switch (sopt->sopt_name) { 702 #define OPTSET(bit) \ 703 if (optval) \ 704 unp->unp_flags |= (bit); \ 705 else \ 706 unp->unp_flags &= ~(bit); 707 708 case LOCAL_CREDS: 709 OPTSET(UNP_WANTCRED); 710 break; 711 case LOCAL_CONNWAIT: 712 OPTSET(UNP_CONNWAIT); 713 break; 714 } 715 break; 716 #undef OPTSET 717 718 default: 719 error = ENOPROTOOPT; 720 break; 721 } 722 break; 723 724 case PRCO_GETOPT: 725 sounlock(so); 726 switch (sopt->sopt_name) { 727 case LOCAL_PEEREID: 728 if (unp->unp_flags & UNP_EIDSVALID) { 729 error = sockopt_set(sopt, 730 &unp->unp_connid, sizeof(unp->unp_connid)); 731 } else { 732 error = EINVAL; 733 } 734 break; 735 case LOCAL_CREDS: 736 #define OPTBIT(bit) (unp->unp_flags & (bit) ? 1 : 0) 737 738 optval = OPTBIT(UNP_WANTCRED); 739 error = sockopt_setint(sopt, optval); 740 break; 741 #undef OPTBIT 742 743 default: 744 error = ENOPROTOOPT; 745 break; 746 } 747 solock(so); 748 break; 749 } 750 return (error); 751 } 752 753 /* 754 * Both send and receive buffers are allocated PIPSIZ bytes of buffering 755 * for stream sockets, although the total for sender and receiver is 756 * actually only PIPSIZ. 757 * Datagram sockets really use the sendspace as the maximum datagram size, 758 * and don't really want to reserve the sendspace. Their recvspace should 759 * be large enough for at least one max-size datagram plus address. 760 */ 761 #define PIPSIZ 4096 762 u_long unpst_sendspace = PIPSIZ; 763 u_long unpst_recvspace = PIPSIZ; 764 u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 765 u_long unpdg_recvspace = 4*1024; 766 767 u_int unp_rights; /* files in flight */ 768 u_int unp_rights_ratio = 2; /* limit, fraction of maxfiles */ 769 770 static int 771 unp_attach(struct socket *so, int proto) 772 { 773 struct unpcb *unp = sotounpcb(so); 774 u_long sndspc, rcvspc; 775 int error; 776 777 KASSERT(unp == NULL); 778 779 switch (so->so_type) { 780 case SOCK_SEQPACKET: 781 /* FALLTHROUGH */ 782 case SOCK_STREAM: 783 if (so->so_lock == NULL) { 784 so->so_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 785 solock(so); 786 } 787 sndspc = unpst_sendspace; 788 rcvspc = unpst_recvspace; 789 break; 790 791 case SOCK_DGRAM: 792 if (so->so_lock == NULL) { 793 mutex_obj_hold(uipc_lock); 794 so->so_lock = uipc_lock; 795 solock(so); 796 } 797 sndspc = unpdg_sendspace; 798 rcvspc = unpdg_recvspace; 799 break; 800 801 default: 802 panic("unp_attach"); 803 } 804 805 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 806 error = soreserve(so, sndspc, rcvspc); 807 if (error) { 808 return error; 809 } 810 } 811 812 unp = kmem_zalloc(sizeof(*unp), KM_SLEEP); 813 nanotime(&unp->unp_ctime); 814 unp->unp_socket = so; 815 so->so_pcb = unp; 816 817 KASSERT(solocked(so)); 818 return 0; 819 } 820 821 static void 822 unp_detach(struct socket *so) 823 { 824 struct unpcb *unp; 825 vnode_t *vp; 826 827 unp = sotounpcb(so); 828 KASSERT(unp != NULL); 829 KASSERT(solocked(so)); 830 retry: 831 if ((vp = unp->unp_vnode) != NULL) { 832 sounlock(so); 833 /* Acquire v_interlock to protect against unp_connect(). */ 834 /* XXXAD racy */ 835 mutex_enter(vp->v_interlock); 836 vp->v_socket = NULL; 837 mutex_exit(vp->v_interlock); 838 vrele(vp); 839 solock(so); 840 unp->unp_vnode = NULL; 841 } 842 if (unp->unp_conn) 843 unp_disconnect(unp); 844 while (unp->unp_refs) { 845 KASSERT(solocked2(so, unp->unp_refs->unp_socket)); 846 if (unp_drop(unp->unp_refs, ECONNRESET)) { 847 solock(so); 848 goto retry; 849 } 850 } 851 soisdisconnected(so); 852 so->so_pcb = NULL; 853 if (unp_rights) { 854 /* 855 * Normally the receive buffer is flushed later, in sofree, 856 * but if our receive buffer holds references to files that 857 * are now garbage, we will enqueue those file references to 858 * the garbage collector and kick it into action. 859 */ 860 sorflush(so); 861 unp_free(unp); 862 unp_thread_kick(); 863 } else 864 unp_free(unp); 865 } 866 867 /* 868 * Allocate the new sockaddr. We have to allocate one 869 * extra byte so that we can ensure that the pathname 870 * is nul-terminated. Note that unlike linux, we don't 871 * include in the address length the NUL in the path 872 * component, because doing so, would exceed sizeof(sockaddr_un) 873 * for fully occupied pathnames. Linux is also inconsistent, 874 * because it does not include the NUL in the length of 875 * what it calls "abstract" unix sockets. 876 */ 877 static struct sockaddr_un * 878 makeun(struct mbuf *nam, size_t *addrlen) { 879 struct sockaddr_un *sun; 880 881 *addrlen = nam->m_len + 1; 882 sun = malloc(*addrlen, M_SONAME, M_WAITOK); 883 m_copydata(nam, 0, nam->m_len, (void *)sun); 884 *(((char *)sun) + nam->m_len) = '\0'; 885 sun->sun_len = strlen(sun->sun_path) + 886 offsetof(struct sockaddr_un, sun_path); 887 return sun; 888 } 889 890 int 891 unp_bind(struct socket *so, struct mbuf *nam, struct lwp *l) 892 { 893 struct sockaddr_un *sun; 894 struct unpcb *unp; 895 vnode_t *vp; 896 struct vattr vattr; 897 size_t addrlen; 898 int error; 899 struct pathbuf *pb; 900 struct nameidata nd; 901 proc_t *p; 902 903 unp = sotounpcb(so); 904 if (unp->unp_vnode != NULL) 905 return (EINVAL); 906 if ((unp->unp_flags & UNP_BUSY) != 0) { 907 /* 908 * EALREADY may not be strictly accurate, but since this 909 * is a major application error it's hardly a big deal. 910 */ 911 return (EALREADY); 912 } 913 unp->unp_flags |= UNP_BUSY; 914 sounlock(so); 915 916 p = l->l_proc; 917 sun = makeun(nam, &addrlen); 918 919 pb = pathbuf_create(sun->sun_path); 920 if (pb == NULL) { 921 error = ENOMEM; 922 goto bad; 923 } 924 NDINIT(&nd, CREATE, FOLLOW | LOCKPARENT | TRYEMULROOT, pb); 925 926 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 927 if ((error = namei(&nd)) != 0) { 928 pathbuf_destroy(pb); 929 goto bad; 930 } 931 vp = nd.ni_vp; 932 if (vp != NULL) { 933 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 934 if (nd.ni_dvp == vp) 935 vrele(nd.ni_dvp); 936 else 937 vput(nd.ni_dvp); 938 vrele(vp); 939 pathbuf_destroy(pb); 940 error = EADDRINUSE; 941 goto bad; 942 } 943 vattr_null(&vattr); 944 vattr.va_type = VSOCK; 945 vattr.va_mode = ACCESSPERMS & ~(p->p_cwdi->cwdi_cmask); 946 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 947 if (error) { 948 vput(nd.ni_dvp); 949 pathbuf_destroy(pb); 950 goto bad; 951 } 952 vp = nd.ni_vp; 953 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 954 solock(so); 955 vp->v_socket = unp->unp_socket; 956 unp->unp_vnode = vp; 957 unp->unp_addrlen = addrlen; 958 unp->unp_addr = sun; 959 unp->unp_connid.unp_pid = p->p_pid; 960 unp->unp_connid.unp_euid = kauth_cred_geteuid(l->l_cred); 961 unp->unp_connid.unp_egid = kauth_cred_getegid(l->l_cred); 962 unp->unp_flags |= UNP_EIDSBIND; 963 VOP_UNLOCK(vp); 964 vput(nd.ni_dvp); 965 unp->unp_flags &= ~UNP_BUSY; 966 pathbuf_destroy(pb); 967 return (0); 968 969 bad: 970 free(sun, M_SONAME); 971 solock(so); 972 unp->unp_flags &= ~UNP_BUSY; 973 return (error); 974 } 975 976 int 977 unp_connect(struct socket *so, struct mbuf *nam, struct lwp *l) 978 { 979 struct sockaddr_un *sun; 980 vnode_t *vp; 981 struct socket *so2, *so3; 982 struct unpcb *unp, *unp2, *unp3; 983 size_t addrlen; 984 int error; 985 struct pathbuf *pb; 986 struct nameidata nd; 987 988 unp = sotounpcb(so); 989 if ((unp->unp_flags & UNP_BUSY) != 0) { 990 /* 991 * EALREADY may not be strictly accurate, but since this 992 * is a major application error it's hardly a big deal. 993 */ 994 return (EALREADY); 995 } 996 unp->unp_flags |= UNP_BUSY; 997 sounlock(so); 998 999 sun = makeun(nam, &addrlen); 1000 pb = pathbuf_create(sun->sun_path); 1001 if (pb == NULL) { 1002 error = ENOMEM; 1003 goto bad2; 1004 } 1005 1006 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb); 1007 1008 if ((error = namei(&nd)) != 0) { 1009 pathbuf_destroy(pb); 1010 goto bad2; 1011 } 1012 vp = nd.ni_vp; 1013 if (vp->v_type != VSOCK) { 1014 error = ENOTSOCK; 1015 goto bad; 1016 } 1017 pathbuf_destroy(pb); 1018 if ((error = VOP_ACCESS(vp, VWRITE, l->l_cred)) != 0) 1019 goto bad; 1020 /* Acquire v_interlock to protect against unp_detach(). */ 1021 mutex_enter(vp->v_interlock); 1022 so2 = vp->v_socket; 1023 if (so2 == NULL) { 1024 mutex_exit(vp->v_interlock); 1025 error = ECONNREFUSED; 1026 goto bad; 1027 } 1028 if (so->so_type != so2->so_type) { 1029 mutex_exit(vp->v_interlock); 1030 error = EPROTOTYPE; 1031 goto bad; 1032 } 1033 solock(so); 1034 unp_resetlock(so); 1035 mutex_exit(vp->v_interlock); 1036 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) { 1037 /* 1038 * This may seem somewhat fragile but is OK: if we can 1039 * see SO_ACCEPTCONN set on the endpoint, then it must 1040 * be locked by the domain-wide uipc_lock. 1041 */ 1042 KASSERT((so2->so_options & SO_ACCEPTCONN) == 0 || 1043 so2->so_lock == uipc_lock); 1044 if ((so2->so_options & SO_ACCEPTCONN) == 0 || 1045 (so3 = sonewconn(so2, false)) == NULL) { 1046 error = ECONNREFUSED; 1047 sounlock(so); 1048 goto bad; 1049 } 1050 unp2 = sotounpcb(so2); 1051 unp3 = sotounpcb(so3); 1052 if (unp2->unp_addr) { 1053 unp3->unp_addr = malloc(unp2->unp_addrlen, 1054 M_SONAME, M_WAITOK); 1055 memcpy(unp3->unp_addr, unp2->unp_addr, 1056 unp2->unp_addrlen); 1057 unp3->unp_addrlen = unp2->unp_addrlen; 1058 } 1059 unp3->unp_flags = unp2->unp_flags; 1060 unp3->unp_connid.unp_pid = l->l_proc->p_pid; 1061 unp3->unp_connid.unp_euid = kauth_cred_geteuid(l->l_cred); 1062 unp3->unp_connid.unp_egid = kauth_cred_getegid(l->l_cred); 1063 unp3->unp_flags |= UNP_EIDSVALID; 1064 if (unp2->unp_flags & UNP_EIDSBIND) { 1065 unp->unp_connid = unp2->unp_connid; 1066 unp->unp_flags |= UNP_EIDSVALID; 1067 } 1068 so2 = so3; 1069 } 1070 error = unp_connect2(so, so2, PRU_CONNECT); 1071 sounlock(so); 1072 bad: 1073 vput(vp); 1074 bad2: 1075 free(sun, M_SONAME); 1076 solock(so); 1077 unp->unp_flags &= ~UNP_BUSY; 1078 return (error); 1079 } 1080 1081 int 1082 unp_connect2(struct socket *so, struct socket *so2, int req) 1083 { 1084 struct unpcb *unp = sotounpcb(so); 1085 struct unpcb *unp2; 1086 1087 if (so2->so_type != so->so_type) 1088 return (EPROTOTYPE); 1089 1090 /* 1091 * All three sockets involved must be locked by same lock: 1092 * 1093 * local endpoint (so) 1094 * remote endpoint (so2) 1095 * queue head (so2->so_head, only if PR_CONNREQUIRED) 1096 */ 1097 KASSERT(solocked2(so, so2)); 1098 KASSERT(so->so_head == NULL); 1099 if (so2->so_head != NULL) { 1100 KASSERT(so2->so_lock == uipc_lock); 1101 KASSERT(solocked2(so2, so2->so_head)); 1102 } 1103 1104 unp2 = sotounpcb(so2); 1105 unp->unp_conn = unp2; 1106 switch (so->so_type) { 1107 1108 case SOCK_DGRAM: 1109 unp->unp_nextref = unp2->unp_refs; 1110 unp2->unp_refs = unp; 1111 soisconnected(so); 1112 break; 1113 1114 case SOCK_SEQPACKET: /* FALLTHROUGH */ 1115 case SOCK_STREAM: 1116 unp2->unp_conn = unp; 1117 if (req == PRU_CONNECT && 1118 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 1119 soisconnecting(so); 1120 else 1121 soisconnected(so); 1122 soisconnected(so2); 1123 /* 1124 * If the connection is fully established, break the 1125 * association with uipc_lock and give the connected 1126 * pair a seperate lock to share. For CONNECT2, we 1127 * require that the locks already match (the sockets 1128 * are created that way). 1129 */ 1130 if (req == PRU_CONNECT) { 1131 KASSERT(so2->so_head != NULL); 1132 unp_setpeerlocks(so, so2); 1133 } 1134 break; 1135 1136 default: 1137 panic("unp_connect2"); 1138 } 1139 return (0); 1140 } 1141 1142 void 1143 unp_disconnect(struct unpcb *unp) 1144 { 1145 struct unpcb *unp2 = unp->unp_conn; 1146 struct socket *so; 1147 1148 if (unp2 == 0) 1149 return; 1150 unp->unp_conn = 0; 1151 so = unp->unp_socket; 1152 switch (so->so_type) { 1153 case SOCK_DGRAM: 1154 if (unp2->unp_refs == unp) 1155 unp2->unp_refs = unp->unp_nextref; 1156 else { 1157 unp2 = unp2->unp_refs; 1158 for (;;) { 1159 KASSERT(solocked2(so, unp2->unp_socket)); 1160 if (unp2 == 0) 1161 panic("unp_disconnect"); 1162 if (unp2->unp_nextref == unp) 1163 break; 1164 unp2 = unp2->unp_nextref; 1165 } 1166 unp2->unp_nextref = unp->unp_nextref; 1167 } 1168 unp->unp_nextref = 0; 1169 so->so_state &= ~SS_ISCONNECTED; 1170 break; 1171 1172 case SOCK_SEQPACKET: /* FALLTHROUGH */ 1173 case SOCK_STREAM: 1174 KASSERT(solocked2(so, unp2->unp_socket)); 1175 soisdisconnected(so); 1176 unp2->unp_conn = 0; 1177 soisdisconnected(unp2->unp_socket); 1178 break; 1179 } 1180 } 1181 1182 void 1183 unp_shutdown(struct unpcb *unp) 1184 { 1185 struct socket *so; 1186 1187 switch(unp->unp_socket->so_type) { 1188 case SOCK_SEQPACKET: /* FALLTHROUGH */ 1189 case SOCK_STREAM: 1190 if (unp->unp_conn && (so = unp->unp_conn->unp_socket)) 1191 socantrcvmore(so); 1192 break; 1193 default: 1194 break; 1195 } 1196 } 1197 1198 bool 1199 unp_drop(struct unpcb *unp, int errno) 1200 { 1201 struct socket *so = unp->unp_socket; 1202 1203 KASSERT(solocked(so)); 1204 1205 so->so_error = errno; 1206 unp_disconnect(unp); 1207 if (so->so_head) { 1208 so->so_pcb = NULL; 1209 /* sofree() drops the socket lock */ 1210 sofree(so); 1211 unp_free(unp); 1212 return true; 1213 } 1214 return false; 1215 } 1216 1217 #ifdef notdef 1218 unp_drain(void) 1219 { 1220 1221 } 1222 #endif 1223 1224 int 1225 unp_externalize(struct mbuf *rights, struct lwp *l, int flags) 1226 { 1227 struct cmsghdr * const cm = mtod(rights, struct cmsghdr *); 1228 struct proc * const p = l->l_proc; 1229 file_t **rp; 1230 int error = 0; 1231 1232 const size_t nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) / 1233 sizeof(file_t *); 1234 if (nfds == 0) 1235 goto noop; 1236 1237 int * const fdp = kmem_alloc(nfds * sizeof(int), KM_SLEEP); 1238 rw_enter(&p->p_cwdi->cwdi_lock, RW_READER); 1239 1240 /* Make sure the recipient should be able to see the files.. */ 1241 rp = (file_t **)CMSG_DATA(cm); 1242 for (size_t i = 0; i < nfds; i++) { 1243 file_t * const fp = *rp++; 1244 if (fp == NULL) { 1245 error = EINVAL; 1246 goto out; 1247 } 1248 /* 1249 * If we are in a chroot'ed directory, and 1250 * someone wants to pass us a directory, make 1251 * sure it's inside the subtree we're allowed 1252 * to access. 1253 */ 1254 if (p->p_cwdi->cwdi_rdir != NULL && fp->f_type == DTYPE_VNODE) { 1255 vnode_t *vp = (vnode_t *)fp->f_data; 1256 if ((vp->v_type == VDIR) && 1257 !vn_isunder(vp, p->p_cwdi->cwdi_rdir, l)) { 1258 error = EPERM; 1259 goto out; 1260 } 1261 } 1262 } 1263 1264 restart: 1265 /* 1266 * First loop -- allocate file descriptor table slots for the 1267 * new files. 1268 */ 1269 for (size_t i = 0; i < nfds; i++) { 1270 if ((error = fd_alloc(p, 0, &fdp[i])) != 0) { 1271 /* 1272 * Back out what we've done so far. 1273 */ 1274 while (i-- > 0) { 1275 fd_abort(p, NULL, fdp[i]); 1276 } 1277 if (error == ENOSPC) { 1278 fd_tryexpand(p); 1279 error = 0; 1280 goto restart; 1281 } 1282 /* 1283 * This is the error that has historically 1284 * been returned, and some callers may 1285 * expect it. 1286 */ 1287 error = EMSGSIZE; 1288 goto out; 1289 } 1290 } 1291 1292 /* 1293 * Now that adding them has succeeded, update all of the 1294 * file passing state and affix the descriptors. 1295 */ 1296 rp = (file_t **)CMSG_DATA(cm); 1297 int *ofdp = (int *)CMSG_DATA(cm); 1298 for (size_t i = 0; i < nfds; i++) { 1299 file_t * const fp = *rp++; 1300 const int fd = fdp[i]; 1301 atomic_dec_uint(&unp_rights); 1302 fd_set_exclose(l, fd, (flags & O_CLOEXEC) != 0); 1303 fd_affix(p, fp, fd); 1304 /* 1305 * Done with this file pointer, replace it with a fd; 1306 */ 1307 *ofdp++ = fd; 1308 mutex_enter(&fp->f_lock); 1309 fp->f_msgcount--; 1310 mutex_exit(&fp->f_lock); 1311 /* 1312 * Note that fd_affix() adds a reference to the file. 1313 * The file may already have been closed by another 1314 * LWP in the process, so we must drop the reference 1315 * added by unp_internalize() with closef(). 1316 */ 1317 closef(fp); 1318 } 1319 1320 /* 1321 * Adjust length, in case of transition from large file_t 1322 * pointers to ints. 1323 */ 1324 if (sizeof(file_t *) != sizeof(int)) { 1325 cm->cmsg_len = CMSG_LEN(nfds * sizeof(int)); 1326 rights->m_len = CMSG_SPACE(nfds * sizeof(int)); 1327 } 1328 out: 1329 if (__predict_false(error != 0)) { 1330 file_t **const fpp = (file_t **)CMSG_DATA(cm); 1331 for (size_t i = 0; i < nfds; i++) 1332 unp_discard_now(fpp[i]); 1333 /* 1334 * Truncate the array so that nobody will try to interpret 1335 * what is now garbage in it. 1336 */ 1337 cm->cmsg_len = CMSG_LEN(0); 1338 rights->m_len = CMSG_SPACE(0); 1339 } 1340 rw_exit(&p->p_cwdi->cwdi_lock); 1341 kmem_free(fdp, nfds * sizeof(int)); 1342 1343 noop: 1344 /* 1345 * Don't disclose kernel memory in the alignment space. 1346 */ 1347 KASSERT(cm->cmsg_len <= rights->m_len); 1348 memset(&mtod(rights, char *)[cm->cmsg_len], 0, rights->m_len - 1349 cm->cmsg_len); 1350 return error; 1351 } 1352 1353 int 1354 unp_internalize(struct mbuf **controlp) 1355 { 1356 filedesc_t *fdescp = curlwp->l_fd; 1357 struct mbuf *control = *controlp; 1358 struct cmsghdr *newcm, *cm = mtod(control, struct cmsghdr *); 1359 file_t **rp, **files; 1360 file_t *fp; 1361 int i, fd, *fdp; 1362 int nfds, error; 1363 u_int maxmsg; 1364 1365 error = 0; 1366 newcm = NULL; 1367 1368 /* Sanity check the control message header. */ 1369 if (cm->cmsg_type != SCM_RIGHTS || cm->cmsg_level != SOL_SOCKET || 1370 cm->cmsg_len > control->m_len || 1371 cm->cmsg_len < CMSG_ALIGN(sizeof(*cm))) 1372 return (EINVAL); 1373 1374 /* 1375 * Verify that the file descriptors are valid, and acquire 1376 * a reference to each. 1377 */ 1378 nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) / sizeof(int); 1379 fdp = (int *)CMSG_DATA(cm); 1380 maxmsg = maxfiles / unp_rights_ratio; 1381 for (i = 0; i < nfds; i++) { 1382 fd = *fdp++; 1383 if (atomic_inc_uint_nv(&unp_rights) > maxmsg) { 1384 atomic_dec_uint(&unp_rights); 1385 nfds = i; 1386 error = EAGAIN; 1387 goto out; 1388 } 1389 if ((fp = fd_getfile(fd)) == NULL 1390 || fp->f_type == DTYPE_KQUEUE) { 1391 if (fp) 1392 fd_putfile(fd); 1393 atomic_dec_uint(&unp_rights); 1394 nfds = i; 1395 error = EBADF; 1396 goto out; 1397 } 1398 } 1399 1400 /* Allocate new space and copy header into it. */ 1401 newcm = malloc(CMSG_SPACE(nfds * sizeof(file_t *)), M_MBUF, M_WAITOK); 1402 if (newcm == NULL) { 1403 error = E2BIG; 1404 goto out; 1405 } 1406 memcpy(newcm, cm, sizeof(struct cmsghdr)); 1407 files = (file_t **)CMSG_DATA(newcm); 1408 1409 /* 1410 * Transform the file descriptors into file_t pointers, in 1411 * reverse order so that if pointers are bigger than ints, the 1412 * int won't get until we're done. No need to lock, as we have 1413 * already validated the descriptors with fd_getfile(). 1414 */ 1415 fdp = (int *)CMSG_DATA(cm) + nfds; 1416 rp = files + nfds; 1417 for (i = 0; i < nfds; i++) { 1418 fp = fdescp->fd_dt->dt_ff[*--fdp]->ff_file; 1419 KASSERT(fp != NULL); 1420 mutex_enter(&fp->f_lock); 1421 *--rp = fp; 1422 fp->f_count++; 1423 fp->f_msgcount++; 1424 mutex_exit(&fp->f_lock); 1425 } 1426 1427 out: 1428 /* Release descriptor references. */ 1429 fdp = (int *)CMSG_DATA(cm); 1430 for (i = 0; i < nfds; i++) { 1431 fd_putfile(*fdp++); 1432 if (error != 0) { 1433 atomic_dec_uint(&unp_rights); 1434 } 1435 } 1436 1437 if (error == 0) { 1438 if (control->m_flags & M_EXT) { 1439 m_freem(control); 1440 *controlp = control = m_get(M_WAIT, MT_CONTROL); 1441 } 1442 MEXTADD(control, newcm, CMSG_SPACE(nfds * sizeof(file_t *)), 1443 M_MBUF, NULL, NULL); 1444 cm = newcm; 1445 /* 1446 * Adjust message & mbuf to note amount of space 1447 * actually used. 1448 */ 1449 cm->cmsg_len = CMSG_LEN(nfds * sizeof(file_t *)); 1450 control->m_len = CMSG_SPACE(nfds * sizeof(file_t *)); 1451 } 1452 1453 return error; 1454 } 1455 1456 struct mbuf * 1457 unp_addsockcred(struct lwp *l, struct mbuf *control) 1458 { 1459 struct sockcred *sc; 1460 struct mbuf *m; 1461 void *p; 1462 1463 m = sbcreatecontrol1(&p, SOCKCREDSIZE(kauth_cred_ngroups(l->l_cred)), 1464 SCM_CREDS, SOL_SOCKET, M_WAITOK); 1465 if (m == NULL) 1466 return control; 1467 1468 sc = p; 1469 sc->sc_uid = kauth_cred_getuid(l->l_cred); 1470 sc->sc_euid = kauth_cred_geteuid(l->l_cred); 1471 sc->sc_gid = kauth_cred_getgid(l->l_cred); 1472 sc->sc_egid = kauth_cred_getegid(l->l_cred); 1473 sc->sc_ngroups = kauth_cred_ngroups(l->l_cred); 1474 1475 for (int i = 0; i < sc->sc_ngroups; i++) 1476 sc->sc_groups[i] = kauth_cred_group(l->l_cred, i); 1477 1478 return m_add(control, m); 1479 } 1480 1481 /* 1482 * Do a mark-sweep GC of files in the system, to free up any which are 1483 * caught in flight to an about-to-be-closed socket. Additionally, 1484 * process deferred file closures. 1485 */ 1486 static void 1487 unp_gc(file_t *dp) 1488 { 1489 extern struct domain unixdomain; 1490 file_t *fp, *np; 1491 struct socket *so, *so1; 1492 u_int i, old, new; 1493 bool didwork; 1494 1495 KASSERT(curlwp == unp_thread_lwp); 1496 KASSERT(mutex_owned(&filelist_lock)); 1497 1498 /* 1499 * First, process deferred file closures. 1500 */ 1501 while (!SLIST_EMPTY(&unp_thread_discard)) { 1502 fp = SLIST_FIRST(&unp_thread_discard); 1503 KASSERT(fp->f_unpcount > 0); 1504 KASSERT(fp->f_count > 0); 1505 KASSERT(fp->f_msgcount > 0); 1506 KASSERT(fp->f_count >= fp->f_unpcount); 1507 KASSERT(fp->f_count >= fp->f_msgcount); 1508 KASSERT(fp->f_msgcount >= fp->f_unpcount); 1509 SLIST_REMOVE_HEAD(&unp_thread_discard, f_unplist); 1510 i = fp->f_unpcount; 1511 fp->f_unpcount = 0; 1512 mutex_exit(&filelist_lock); 1513 for (; i != 0; i--) { 1514 unp_discard_now(fp); 1515 } 1516 mutex_enter(&filelist_lock); 1517 } 1518 1519 /* 1520 * Clear mark bits. Ensure that we don't consider new files 1521 * entering the file table during this loop (they will not have 1522 * FSCAN set). 1523 */ 1524 unp_defer = 0; 1525 LIST_FOREACH(fp, &filehead, f_list) { 1526 for (old = fp->f_flag;; old = new) { 1527 new = atomic_cas_uint(&fp->f_flag, old, 1528 (old | FSCAN) & ~(FMARK|FDEFER)); 1529 if (__predict_true(old == new)) { 1530 break; 1531 } 1532 } 1533 } 1534 1535 /* 1536 * Iterate over the set of sockets, marking ones believed (based on 1537 * refcount) to be referenced from a process, and marking for rescan 1538 * sockets which are queued on a socket. Recan continues descending 1539 * and searching for sockets referenced by sockets (FDEFER), until 1540 * there are no more socket->socket references to be discovered. 1541 */ 1542 do { 1543 didwork = false; 1544 for (fp = LIST_FIRST(&filehead); fp != NULL; fp = np) { 1545 KASSERT(mutex_owned(&filelist_lock)); 1546 np = LIST_NEXT(fp, f_list); 1547 mutex_enter(&fp->f_lock); 1548 if ((fp->f_flag & FDEFER) != 0) { 1549 atomic_and_uint(&fp->f_flag, ~FDEFER); 1550 unp_defer--; 1551 KASSERT(fp->f_count != 0); 1552 } else { 1553 if (fp->f_count == 0 || 1554 (fp->f_flag & FMARK) != 0 || 1555 fp->f_count == fp->f_msgcount || 1556 fp->f_unpcount != 0) { 1557 mutex_exit(&fp->f_lock); 1558 continue; 1559 } 1560 } 1561 atomic_or_uint(&fp->f_flag, FMARK); 1562 1563 if (fp->f_type != DTYPE_SOCKET || 1564 (so = fp->f_data) == NULL || 1565 so->so_proto->pr_domain != &unixdomain || 1566 (so->so_proto->pr_flags & PR_RIGHTS) == 0) { 1567 mutex_exit(&fp->f_lock); 1568 continue; 1569 } 1570 1571 /* Gain file ref, mark our position, and unlock. */ 1572 didwork = true; 1573 LIST_INSERT_AFTER(fp, dp, f_list); 1574 fp->f_count++; 1575 mutex_exit(&fp->f_lock); 1576 mutex_exit(&filelist_lock); 1577 1578 /* 1579 * Mark files referenced from sockets queued on the 1580 * accept queue as well. 1581 */ 1582 solock(so); 1583 unp_scan(so->so_rcv.sb_mb, unp_mark, 0); 1584 if ((so->so_options & SO_ACCEPTCONN) != 0) { 1585 TAILQ_FOREACH(so1, &so->so_q0, so_qe) { 1586 unp_scan(so1->so_rcv.sb_mb, unp_mark, 0); 1587 } 1588 TAILQ_FOREACH(so1, &so->so_q, so_qe) { 1589 unp_scan(so1->so_rcv.sb_mb, unp_mark, 0); 1590 } 1591 } 1592 sounlock(so); 1593 1594 /* Re-lock and restart from where we left off. */ 1595 closef(fp); 1596 mutex_enter(&filelist_lock); 1597 np = LIST_NEXT(dp, f_list); 1598 LIST_REMOVE(dp, f_list); 1599 } 1600 /* 1601 * Bail early if we did nothing in the loop above. Could 1602 * happen because of concurrent activity causing unp_defer 1603 * to get out of sync. 1604 */ 1605 } while (unp_defer != 0 && didwork); 1606 1607 /* 1608 * Sweep pass. 1609 * 1610 * We grab an extra reference to each of the files that are 1611 * not otherwise accessible and then free the rights that are 1612 * stored in messages on them. 1613 */ 1614 for (fp = LIST_FIRST(&filehead); fp != NULL; fp = np) { 1615 KASSERT(mutex_owned(&filelist_lock)); 1616 np = LIST_NEXT(fp, f_list); 1617 mutex_enter(&fp->f_lock); 1618 1619 /* 1620 * Ignore non-sockets. 1621 * Ignore dead sockets, or sockets with pending close. 1622 * Ignore sockets obviously referenced elsewhere. 1623 * Ignore sockets marked as referenced by our scan. 1624 * Ignore new sockets that did not exist during the scan. 1625 */ 1626 if (fp->f_type != DTYPE_SOCKET || 1627 fp->f_count == 0 || fp->f_unpcount != 0 || 1628 fp->f_count != fp->f_msgcount || 1629 (fp->f_flag & (FMARK | FSCAN)) != FSCAN) { 1630 mutex_exit(&fp->f_lock); 1631 continue; 1632 } 1633 1634 /* Gain file ref, mark our position, and unlock. */ 1635 LIST_INSERT_AFTER(fp, dp, f_list); 1636 fp->f_count++; 1637 mutex_exit(&fp->f_lock); 1638 mutex_exit(&filelist_lock); 1639 1640 /* 1641 * Flush all data from the socket's receive buffer. 1642 * This will cause files referenced only by the 1643 * socket to be queued for close. 1644 */ 1645 so = fp->f_data; 1646 solock(so); 1647 sorflush(so); 1648 sounlock(so); 1649 1650 /* Re-lock and restart from where we left off. */ 1651 closef(fp); 1652 mutex_enter(&filelist_lock); 1653 np = LIST_NEXT(dp, f_list); 1654 LIST_REMOVE(dp, f_list); 1655 } 1656 } 1657 1658 /* 1659 * Garbage collector thread. While SCM_RIGHTS messages are in transit, 1660 * wake once per second to garbage collect. Run continually while we 1661 * have deferred closes to process. 1662 */ 1663 static void 1664 unp_thread(void *cookie) 1665 { 1666 file_t *dp; 1667 1668 /* Allocate a dummy file for our scans. */ 1669 if ((dp = fgetdummy()) == NULL) { 1670 panic("unp_thread"); 1671 } 1672 1673 mutex_enter(&filelist_lock); 1674 for (;;) { 1675 KASSERT(mutex_owned(&filelist_lock)); 1676 if (SLIST_EMPTY(&unp_thread_discard)) { 1677 if (unp_rights != 0) { 1678 (void)cv_timedwait(&unp_thread_cv, 1679 &filelist_lock, hz); 1680 } else { 1681 cv_wait(&unp_thread_cv, &filelist_lock); 1682 } 1683 } 1684 unp_gc(dp); 1685 } 1686 /* NOTREACHED */ 1687 } 1688 1689 /* 1690 * Kick the garbage collector into action if there is something for 1691 * it to process. 1692 */ 1693 static void 1694 unp_thread_kick(void) 1695 { 1696 1697 if (!SLIST_EMPTY(&unp_thread_discard) || unp_rights != 0) { 1698 mutex_enter(&filelist_lock); 1699 cv_signal(&unp_thread_cv); 1700 mutex_exit(&filelist_lock); 1701 } 1702 } 1703 1704 void 1705 unp_dispose(struct mbuf *m) 1706 { 1707 1708 if (m) 1709 unp_scan(m, unp_discard_later, 1); 1710 } 1711 1712 void 1713 unp_scan(struct mbuf *m0, void (*op)(file_t *), int discard) 1714 { 1715 struct mbuf *m; 1716 file_t **rp, *fp; 1717 struct cmsghdr *cm; 1718 int i, qfds; 1719 1720 while (m0) { 1721 for (m = m0; m; m = m->m_next) { 1722 if (m->m_type != MT_CONTROL || 1723 m->m_len < sizeof(*cm)) { 1724 continue; 1725 } 1726 cm = mtod(m, struct cmsghdr *); 1727 if (cm->cmsg_level != SOL_SOCKET || 1728 cm->cmsg_type != SCM_RIGHTS) 1729 continue; 1730 qfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) 1731 / sizeof(file_t *); 1732 rp = (file_t **)CMSG_DATA(cm); 1733 for (i = 0; i < qfds; i++) { 1734 fp = *rp; 1735 if (discard) { 1736 *rp = 0; 1737 } 1738 (*op)(fp); 1739 rp++; 1740 } 1741 } 1742 m0 = m0->m_nextpkt; 1743 } 1744 } 1745 1746 void 1747 unp_mark(file_t *fp) 1748 { 1749 1750 if (fp == NULL) 1751 return; 1752 1753 /* If we're already deferred, don't screw up the defer count */ 1754 mutex_enter(&fp->f_lock); 1755 if (fp->f_flag & (FMARK | FDEFER)) { 1756 mutex_exit(&fp->f_lock); 1757 return; 1758 } 1759 1760 /* 1761 * Minimize the number of deferrals... Sockets are the only type of 1762 * file which can hold references to another file, so just mark 1763 * other files, and defer unmarked sockets for the next pass. 1764 */ 1765 if (fp->f_type == DTYPE_SOCKET) { 1766 unp_defer++; 1767 KASSERT(fp->f_count != 0); 1768 atomic_or_uint(&fp->f_flag, FDEFER); 1769 } else { 1770 atomic_or_uint(&fp->f_flag, FMARK); 1771 } 1772 mutex_exit(&fp->f_lock); 1773 } 1774 1775 static void 1776 unp_discard_now(file_t *fp) 1777 { 1778 1779 if (fp == NULL) 1780 return; 1781 1782 KASSERT(fp->f_count > 0); 1783 KASSERT(fp->f_msgcount > 0); 1784 1785 mutex_enter(&fp->f_lock); 1786 fp->f_msgcount--; 1787 mutex_exit(&fp->f_lock); 1788 atomic_dec_uint(&unp_rights); 1789 (void)closef(fp); 1790 } 1791 1792 static void 1793 unp_discard_later(file_t *fp) 1794 { 1795 1796 if (fp == NULL) 1797 return; 1798 1799 KASSERT(fp->f_count > 0); 1800 KASSERT(fp->f_msgcount > 0); 1801 1802 mutex_enter(&filelist_lock); 1803 if (fp->f_unpcount++ == 0) { 1804 SLIST_INSERT_HEAD(&unp_thread_discard, fp, f_unplist); 1805 } 1806 mutex_exit(&filelist_lock); 1807 } 1808 1809 const struct pr_usrreqs unp_usrreqs = { 1810 .pr_attach = unp_attach, 1811 .pr_detach = unp_detach, 1812 .pr_generic = unp_usrreq, 1813 }; 1814