1 /* $NetBSD: uipc_usrreq.c,v 1.133 2010/11/19 06:44:43 dholland 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 1.133 2010/11/19 06:44:43 dholland 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 * SEQPACKET, 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 = sizeof(sun_noname), 169 .sun_family = AF_LOCAL, 170 }; 171 ino_t unp_ino; /* prototype for fake inode numbers */ 172 173 struct mbuf *unp_addsockcred(struct lwp *, struct mbuf *); 174 static void unp_mark(file_t *); 175 static void unp_scan(struct mbuf *, void (*)(file_t *), int); 176 static void unp_discard_now(file_t *); 177 static void unp_discard_later(file_t *); 178 static void unp_thread(void *); 179 static void unp_thread_kick(void); 180 static kmutex_t *uipc_lock; 181 182 static kcondvar_t unp_thread_cv; 183 static lwp_t *unp_thread_lwp; 184 static SLIST_HEAD(,file) unp_thread_discard; 185 static int unp_defer; 186 187 /* 188 * Initialize Unix protocols. 189 */ 190 void 191 uipc_init(void) 192 { 193 int error; 194 195 uipc_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 196 cv_init(&unp_thread_cv, "unpgc"); 197 198 error = kthread_create(PRI_NONE, KTHREAD_MPSAFE, NULL, unp_thread, 199 NULL, &unp_thread_lwp, "unpgc"); 200 if (error != 0) 201 panic("uipc_init %d", error); 202 } 203 204 /* 205 * A connection succeeded: disassociate both endpoints from the head's 206 * lock, and make them share their own lock. There is a race here: for 207 * a very brief time one endpoint will be locked by a different lock 208 * than the other end. However, since the current thread holds the old 209 * lock (the listening socket's lock, the head) access can still only be 210 * made to one side of the connection. 211 */ 212 static void 213 unp_setpeerlocks(struct socket *so, struct socket *so2) 214 { 215 struct unpcb *unp; 216 kmutex_t *lock; 217 218 KASSERT(solocked2(so, so2)); 219 220 /* 221 * Bail out if either end of the socket is not yet fully 222 * connected or accepted. We only break the lock association 223 * with the head when the pair of sockets stand completely 224 * on their own. 225 */ 226 KASSERT(so->so_head == NULL); 227 if (so2->so_head != NULL) 228 return; 229 230 /* 231 * Drop references to old lock. A third reference (from the 232 * queue head) must be held as we still hold its lock. Bonus: 233 * we don't need to worry about garbage collecting the lock. 234 */ 235 lock = so->so_lock; 236 KASSERT(lock == uipc_lock); 237 mutex_obj_free(lock); 238 mutex_obj_free(lock); 239 240 /* 241 * Grab stream lock from the initiator and share between the two 242 * endpoints. Issue memory barrier to ensure all modifications 243 * become globally visible before the lock change. so2 is 244 * assumed not to have a stream lock, because it was created 245 * purely for the server side to accept this connection and 246 * started out life using the domain-wide lock. 247 */ 248 unp = sotounpcb(so); 249 KASSERT(unp->unp_streamlock != NULL); 250 KASSERT(sotounpcb(so2)->unp_streamlock == NULL); 251 lock = unp->unp_streamlock; 252 unp->unp_streamlock = NULL; 253 mutex_obj_hold(lock); 254 membar_exit(); 255 /* 256 * possible race if lock is not held - see comment in 257 * uipc_usrreq(PRU_ACCEPT). 258 */ 259 KASSERT(mutex_owned(lock)); 260 solockreset(so, lock); 261 solockreset(so2, lock); 262 } 263 264 /* 265 * Reset a socket's lock back to the domain-wide lock. 266 */ 267 static void 268 unp_resetlock(struct socket *so) 269 { 270 kmutex_t *olock, *nlock; 271 struct unpcb *unp; 272 273 KASSERT(solocked(so)); 274 275 olock = so->so_lock; 276 nlock = uipc_lock; 277 if (olock == nlock) 278 return; 279 unp = sotounpcb(so); 280 KASSERT(unp->unp_streamlock == NULL); 281 unp->unp_streamlock = olock; 282 mutex_obj_hold(nlock); 283 mutex_enter(nlock); 284 solockreset(so, nlock); 285 mutex_exit(olock); 286 } 287 288 static void 289 unp_free(struct unpcb *unp) 290 { 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 free(unp, M_PCB); 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 /*ARGSUSED*/ 366 int 367 uipc_usrreq(struct socket *so, int req, struct mbuf *m, struct mbuf *nam, 368 struct mbuf *control, struct lwp *l) 369 { 370 struct unpcb *unp = sotounpcb(so); 371 struct socket *so2; 372 struct proc *p; 373 u_int newhiwat; 374 int error = 0; 375 376 if (req == PRU_CONTROL) 377 return (EOPNOTSUPP); 378 379 #ifdef DIAGNOSTIC 380 if (req != PRU_SEND && req != PRU_SENDOOB && control) 381 panic("uipc_usrreq: unexpected control mbuf"); 382 #endif 383 p = l ? l->l_proc : NULL; 384 if (req != PRU_ATTACH) { 385 if (unp == NULL) { 386 error = EINVAL; 387 goto release; 388 } 389 KASSERT(solocked(so)); 390 } 391 392 switch (req) { 393 394 case PRU_ATTACH: 395 if (unp != NULL) { 396 error = EISCONN; 397 break; 398 } 399 error = unp_attach(so); 400 break; 401 402 case PRU_DETACH: 403 unp_detach(unp); 404 break; 405 406 case PRU_BIND: 407 KASSERT(l != NULL); 408 error = unp_bind(so, nam, l); 409 break; 410 411 case PRU_LISTEN: 412 /* 413 * If the socket can accept a connection, it must be 414 * locked by uipc_lock. 415 */ 416 unp_resetlock(so); 417 if (unp->unp_vnode == NULL) 418 error = EINVAL; 419 break; 420 421 case PRU_CONNECT: 422 KASSERT(l != NULL); 423 error = unp_connect(so, nam, l); 424 break; 425 426 case PRU_CONNECT2: 427 error = unp_connect2(so, (struct socket *)nam, PRU_CONNECT2); 428 break; 429 430 case PRU_DISCONNECT: 431 unp_disconnect(unp); 432 break; 433 434 case PRU_ACCEPT: 435 KASSERT(so->so_lock == uipc_lock); 436 /* 437 * Mark the initiating STREAM socket as connected *ONLY* 438 * after it's been accepted. This prevents a client from 439 * overrunning a server and receiving ECONNREFUSED. 440 */ 441 if (unp->unp_conn == NULL) 442 break; 443 so2 = unp->unp_conn->unp_socket; 444 if (so2->so_state & SS_ISCONNECTING) { 445 KASSERT(solocked2(so, so->so_head)); 446 KASSERT(solocked2(so2, so->so_head)); 447 soisconnected(so2); 448 } 449 /* 450 * If the connection is fully established, break the 451 * association with uipc_lock and give the connected 452 * pair a seperate lock to share. 453 * There is a race here: sotounpcb(so2)->unp_streamlock 454 * is not locked, so when changing so2->so_lock 455 * another thread can grab it while so->so_lock is still 456 * pointing to the (locked) uipc_lock. 457 * this should be harmless, except that this makes 458 * solocked2() and solocked() unreliable. 459 * Another problem is that unp_setaddr() expects the 460 * the socket locked. Grabing sotounpcb(so2)->unp_streamlock 461 * fixes both issues. 462 */ 463 mutex_enter(sotounpcb(so2)->unp_streamlock); 464 unp_setpeerlocks(so2, so); 465 /* 466 * Only now return peer's address, as we may need to 467 * block in order to allocate memory. 468 * 469 * XXX Minor race: connection can be broken while 470 * lock is dropped in unp_setaddr(). We will return 471 * error == 0 and sun_noname as the peer address. 472 */ 473 unp_setaddr(so, nam, true); 474 /* so_lock now points to unp_streamlock */ 475 mutex_exit(so2->so_lock); 476 break; 477 478 case PRU_SHUTDOWN: 479 socantsendmore(so); 480 unp_shutdown(unp); 481 break; 482 483 case PRU_RCVD: 484 switch (so->so_type) { 485 486 case SOCK_DGRAM: 487 panic("uipc 1"); 488 /*NOTREACHED*/ 489 490 case SOCK_STREAM: 491 #define rcv (&so->so_rcv) 492 #define snd (&so2->so_snd) 493 if (unp->unp_conn == 0) 494 break; 495 so2 = unp->unp_conn->unp_socket; 496 KASSERT(solocked2(so, so2)); 497 /* 498 * Adjust backpressure on sender 499 * and wakeup any waiting to write. 500 */ 501 snd->sb_mbmax += unp->unp_mbcnt - rcv->sb_mbcnt; 502 unp->unp_mbcnt = rcv->sb_mbcnt; 503 newhiwat = snd->sb_hiwat + unp->unp_cc - rcv->sb_cc; 504 (void)chgsbsize(so2->so_uidinfo, 505 &snd->sb_hiwat, newhiwat, RLIM_INFINITY); 506 unp->unp_cc = rcv->sb_cc; 507 sowwakeup(so2); 508 #undef snd 509 #undef rcv 510 break; 511 512 default: 513 panic("uipc 2"); 514 } 515 break; 516 517 case PRU_SEND: 518 /* 519 * Note: unp_internalize() rejects any control message 520 * other than SCM_RIGHTS, and only allows one. This 521 * has the side-effect of preventing a caller from 522 * forging SCM_CREDS. 523 */ 524 if (control) { 525 sounlock(so); 526 error = unp_internalize(&control); 527 solock(so); 528 if (error != 0) { 529 m_freem(control); 530 m_freem(m); 531 break; 532 } 533 } 534 switch (so->so_type) { 535 536 case SOCK_DGRAM: { 537 KASSERT(so->so_lock == uipc_lock); 538 if (nam) { 539 if ((so->so_state & SS_ISCONNECTED) != 0) 540 error = EISCONN; 541 else { 542 /* 543 * Note: once connected, the 544 * socket's lock must not be 545 * dropped until we have sent 546 * the message and disconnected. 547 * This is necessary to prevent 548 * intervening control ops, like 549 * another connection. 550 */ 551 error = unp_connect(so, nam, l); 552 } 553 } else { 554 if ((so->so_state & SS_ISCONNECTED) == 0) 555 error = ENOTCONN; 556 } 557 if (error) { 558 unp_dispose(control); 559 m_freem(control); 560 m_freem(m); 561 break; 562 } 563 KASSERT(p != NULL); 564 error = unp_output(m, control, unp, l); 565 if (nam) 566 unp_disconnect(unp); 567 break; 568 } 569 570 case SOCK_STREAM: 571 #define rcv (&so2->so_rcv) 572 #define snd (&so->so_snd) 573 if (unp->unp_conn == NULL) { 574 error = ENOTCONN; 575 break; 576 } 577 so2 = unp->unp_conn->unp_socket; 578 KASSERT(solocked2(so, so2)); 579 if (unp->unp_conn->unp_flags & UNP_WANTCRED) { 580 /* 581 * Credentials are passed only once on 582 * SOCK_STREAM. 583 */ 584 unp->unp_conn->unp_flags &= ~UNP_WANTCRED; 585 control = unp_addsockcred(l, control); 586 } 587 /* 588 * Send to paired receive port, and then reduce 589 * send buffer hiwater marks to maintain backpressure. 590 * Wake up readers. 591 */ 592 if (control) { 593 if (sbappendcontrol(rcv, m, control) != 0) 594 control = NULL; 595 } else 596 sbappend(rcv, m); 597 snd->sb_mbmax -= 598 rcv->sb_mbcnt - unp->unp_conn->unp_mbcnt; 599 unp->unp_conn->unp_mbcnt = rcv->sb_mbcnt; 600 newhiwat = snd->sb_hiwat - 601 (rcv->sb_cc - unp->unp_conn->unp_cc); 602 (void)chgsbsize(so->so_uidinfo, 603 &snd->sb_hiwat, newhiwat, RLIM_INFINITY); 604 unp->unp_conn->unp_cc = rcv->sb_cc; 605 sorwakeup(so2); 606 #undef snd 607 #undef rcv 608 if (control != NULL) { 609 unp_dispose(control); 610 m_freem(control); 611 } 612 break; 613 614 default: 615 panic("uipc 4"); 616 } 617 break; 618 619 case PRU_ABORT: 620 (void)unp_drop(unp, ECONNABORTED); 621 622 KASSERT(so->so_head == NULL); 623 #ifdef DIAGNOSTIC 624 if (so->so_pcb == NULL) 625 panic("uipc 5: drop killed pcb"); 626 #endif 627 unp_detach(unp); 628 break; 629 630 case PRU_SENSE: 631 ((struct stat *) m)->st_blksize = so->so_snd.sb_hiwat; 632 if (so->so_type == SOCK_STREAM && unp->unp_conn != 0) { 633 so2 = unp->unp_conn->unp_socket; 634 KASSERT(solocked2(so, so2)); 635 ((struct stat *) m)->st_blksize += so2->so_rcv.sb_cc; 636 } 637 ((struct stat *) m)->st_dev = NODEV; 638 if (unp->unp_ino == 0) 639 unp->unp_ino = unp_ino++; 640 ((struct stat *) m)->st_atimespec = 641 ((struct stat *) m)->st_mtimespec = 642 ((struct stat *) m)->st_ctimespec = unp->unp_ctime; 643 ((struct stat *) m)->st_ino = unp->unp_ino; 644 return (0); 645 646 case PRU_RCVOOB: 647 error = EOPNOTSUPP; 648 break; 649 650 case PRU_SENDOOB: 651 m_freem(control); 652 m_freem(m); 653 error = EOPNOTSUPP; 654 break; 655 656 case PRU_SOCKADDR: 657 unp_setaddr(so, nam, false); 658 break; 659 660 case PRU_PEERADDR: 661 unp_setaddr(so, nam, true); 662 break; 663 664 default: 665 panic("piusrreq"); 666 } 667 668 release: 669 return (error); 670 } 671 672 /* 673 * Unix domain socket option processing. 674 */ 675 int 676 uipc_ctloutput(int op, struct socket *so, struct sockopt *sopt) 677 { 678 struct unpcb *unp = sotounpcb(so); 679 int optval = 0, error = 0; 680 681 KASSERT(solocked(so)); 682 683 if (sopt->sopt_level != 0) { 684 error = ENOPROTOOPT; 685 } else switch (op) { 686 687 case PRCO_SETOPT: 688 switch (sopt->sopt_name) { 689 case LOCAL_CREDS: 690 case LOCAL_CONNWAIT: 691 error = sockopt_getint(sopt, &optval); 692 if (error) 693 break; 694 switch (sopt->sopt_name) { 695 #define OPTSET(bit) \ 696 if (optval) \ 697 unp->unp_flags |= (bit); \ 698 else \ 699 unp->unp_flags &= ~(bit); 700 701 case LOCAL_CREDS: 702 OPTSET(UNP_WANTCRED); 703 break; 704 case LOCAL_CONNWAIT: 705 OPTSET(UNP_CONNWAIT); 706 break; 707 } 708 break; 709 #undef OPTSET 710 711 default: 712 error = ENOPROTOOPT; 713 break; 714 } 715 break; 716 717 case PRCO_GETOPT: 718 sounlock(so); 719 switch (sopt->sopt_name) { 720 case LOCAL_PEEREID: 721 if (unp->unp_flags & UNP_EIDSVALID) { 722 error = sockopt_set(sopt, 723 &unp->unp_connid, sizeof(unp->unp_connid)); 724 } else { 725 error = EINVAL; 726 } 727 break; 728 case LOCAL_CREDS: 729 #define OPTBIT(bit) (unp->unp_flags & (bit) ? 1 : 0) 730 731 optval = OPTBIT(UNP_WANTCRED); 732 error = sockopt_setint(sopt, optval); 733 break; 734 #undef OPTBIT 735 736 default: 737 error = ENOPROTOOPT; 738 break; 739 } 740 solock(so); 741 break; 742 } 743 return (error); 744 } 745 746 /* 747 * Both send and receive buffers are allocated PIPSIZ bytes of buffering 748 * for stream sockets, although the total for sender and receiver is 749 * actually only PIPSIZ. 750 * Datagram sockets really use the sendspace as the maximum datagram size, 751 * and don't really want to reserve the sendspace. Their recvspace should 752 * be large enough for at least one max-size datagram plus address. 753 */ 754 #define PIPSIZ 4096 755 u_long unpst_sendspace = PIPSIZ; 756 u_long unpst_recvspace = PIPSIZ; 757 u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 758 u_long unpdg_recvspace = 4*1024; 759 760 u_int unp_rights; /* files in flight */ 761 u_int unp_rights_ratio = 2; /* limit, fraction of maxfiles */ 762 763 int 764 unp_attach(struct socket *so) 765 { 766 struct unpcb *unp; 767 int error; 768 769 switch (so->so_type) { 770 case SOCK_STREAM: 771 if (so->so_lock == NULL) { 772 /* 773 * XXX Assuming that no socket locks are held, 774 * as this call may sleep. 775 */ 776 so->so_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 777 solock(so); 778 } 779 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 780 error = soreserve(so, unpst_sendspace, unpst_recvspace); 781 if (error != 0) 782 return (error); 783 } 784 break; 785 786 case SOCK_DGRAM: 787 if (so->so_lock == NULL) { 788 mutex_obj_hold(uipc_lock); 789 so->so_lock = uipc_lock; 790 solock(so); 791 } 792 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 793 error = soreserve(so, unpdg_sendspace, unpdg_recvspace); 794 if (error != 0) 795 return (error); 796 } 797 break; 798 799 default: 800 panic("unp_attach"); 801 } 802 KASSERT(solocked(so)); 803 unp = malloc(sizeof(*unp), M_PCB, M_NOWAIT); 804 if (unp == NULL) 805 return (ENOBUFS); 806 memset(unp, 0, sizeof(*unp)); 807 unp->unp_socket = so; 808 so->so_pcb = unp; 809 nanotime(&unp->unp_ctime); 810 return (0); 811 } 812 813 void 814 unp_detach(struct unpcb *unp) 815 { 816 struct socket *so; 817 vnode_t *vp; 818 819 so = unp->unp_socket; 820 821 retry: 822 if ((vp = unp->unp_vnode) != NULL) { 823 sounlock(so); 824 /* Acquire v_interlock to protect against unp_connect(). */ 825 /* XXXAD racy */ 826 mutex_enter(&vp->v_interlock); 827 vp->v_socket = NULL; 828 vrelel(vp, 0); 829 solock(so); 830 unp->unp_vnode = NULL; 831 } 832 if (unp->unp_conn) 833 unp_disconnect(unp); 834 while (unp->unp_refs) { 835 KASSERT(solocked2(so, unp->unp_refs->unp_socket)); 836 if (unp_drop(unp->unp_refs, ECONNRESET)) { 837 solock(so); 838 goto retry; 839 } 840 } 841 soisdisconnected(so); 842 so->so_pcb = NULL; 843 if (unp_rights) { 844 /* 845 * Normally the receive buffer is flushed later, in sofree, 846 * but if our receive buffer holds references to files that 847 * are now garbage, we will enqueue those file references to 848 * the garbage collector and kick it into action. 849 */ 850 sorflush(so); 851 unp_free(unp); 852 unp_thread_kick(); 853 } else 854 unp_free(unp); 855 } 856 857 int 858 unp_bind(struct socket *so, struct mbuf *nam, struct lwp *l) 859 { 860 struct sockaddr_un *sun; 861 struct unpcb *unp; 862 vnode_t *vp; 863 struct vattr vattr; 864 size_t addrlen; 865 int error; 866 struct pathbuf *pb; 867 struct nameidata nd; 868 proc_t *p; 869 870 unp = sotounpcb(so); 871 if (unp->unp_vnode != NULL) 872 return (EINVAL); 873 if ((unp->unp_flags & UNP_BUSY) != 0) { 874 /* 875 * EALREADY may not be strictly accurate, but since this 876 * is a major application error it's hardly a big deal. 877 */ 878 return (EALREADY); 879 } 880 unp->unp_flags |= UNP_BUSY; 881 sounlock(so); 882 883 /* 884 * Allocate the new sockaddr. We have to allocate one 885 * extra byte so that we can ensure that the pathname 886 * is nul-terminated. 887 */ 888 p = l->l_proc; 889 addrlen = nam->m_len + 1; 890 sun = malloc(addrlen, M_SONAME, M_WAITOK); 891 m_copydata(nam, 0, nam->m_len, (void *)sun); 892 *(((char *)sun) + nam->m_len) = '\0'; 893 894 pb = pathbuf_create(sun->sun_path); 895 if (pb == NULL) { 896 error = ENOMEM; 897 goto bad; 898 } 899 NDINIT(&nd, CREATE, FOLLOW | LOCKPARENT | TRYEMULROOT, pb); 900 901 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 902 if ((error = namei(&nd)) != 0) { 903 pathbuf_destroy(pb); 904 goto bad; 905 } 906 vp = nd.ni_vp; 907 if (vp != NULL) { 908 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 909 if (nd.ni_dvp == vp) 910 vrele(nd.ni_dvp); 911 else 912 vput(nd.ni_dvp); 913 vrele(vp); 914 pathbuf_destroy(pb); 915 error = EADDRINUSE; 916 goto bad; 917 } 918 vattr_null(&vattr); 919 vattr.va_type = VSOCK; 920 vattr.va_mode = ACCESSPERMS & ~(p->p_cwdi->cwdi_cmask); 921 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 922 if (error) { 923 pathbuf_destroy(pb); 924 goto bad; 925 } 926 vp = nd.ni_vp; 927 solock(so); 928 vp->v_socket = unp->unp_socket; 929 unp->unp_vnode = vp; 930 unp->unp_addrlen = addrlen; 931 unp->unp_addr = sun; 932 unp->unp_connid.unp_pid = p->p_pid; 933 unp->unp_connid.unp_euid = kauth_cred_geteuid(l->l_cred); 934 unp->unp_connid.unp_egid = kauth_cred_getegid(l->l_cred); 935 unp->unp_flags |= UNP_EIDSBIND; 936 VOP_UNLOCK(vp); 937 unp->unp_flags &= ~UNP_BUSY; 938 pathbuf_destroy(pb); 939 return (0); 940 941 bad: 942 free(sun, M_SONAME); 943 solock(so); 944 unp->unp_flags &= ~UNP_BUSY; 945 return (error); 946 } 947 948 int 949 unp_connect(struct socket *so, struct mbuf *nam, struct lwp *l) 950 { 951 struct sockaddr_un *sun; 952 vnode_t *vp; 953 struct socket *so2, *so3; 954 struct unpcb *unp, *unp2, *unp3; 955 size_t addrlen; 956 int error; 957 struct pathbuf *pb; 958 struct nameidata nd; 959 960 unp = sotounpcb(so); 961 if ((unp->unp_flags & UNP_BUSY) != 0) { 962 /* 963 * EALREADY may not be strictly accurate, but since this 964 * is a major application error it's hardly a big deal. 965 */ 966 return (EALREADY); 967 } 968 unp->unp_flags |= UNP_BUSY; 969 sounlock(so); 970 971 /* 972 * Allocate a temporary sockaddr. We have to allocate one extra 973 * byte so that we can ensure that the pathname is nul-terminated. 974 * When we establish the connection, we copy the other PCB's 975 * sockaddr to our own. 976 */ 977 addrlen = nam->m_len + 1; 978 sun = malloc(addrlen, M_SONAME, M_WAITOK); 979 m_copydata(nam, 0, nam->m_len, (void *)sun); 980 *(((char *)sun) + nam->m_len) = '\0'; 981 982 pb = pathbuf_create(sun->sun_path); 983 if (pb == NULL) { 984 error = ENOMEM; 985 goto bad2; 986 } 987 988 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb); 989 990 if ((error = namei(&nd)) != 0) { 991 pathbuf_destroy(pb); 992 goto bad2; 993 } 994 vp = nd.ni_vp; 995 if (vp->v_type != VSOCK) { 996 error = ENOTSOCK; 997 goto bad; 998 } 999 pathbuf_destroy(pb); 1000 if ((error = VOP_ACCESS(vp, VWRITE, l->l_cred)) != 0) 1001 goto bad; 1002 /* Acquire v_interlock to protect against unp_detach(). */ 1003 mutex_enter(&vp->v_interlock); 1004 so2 = vp->v_socket; 1005 if (so2 == NULL) { 1006 mutex_exit(&vp->v_interlock); 1007 error = ECONNREFUSED; 1008 goto bad; 1009 } 1010 if (so->so_type != so2->so_type) { 1011 mutex_exit(&vp->v_interlock); 1012 error = EPROTOTYPE; 1013 goto bad; 1014 } 1015 solock(so); 1016 unp_resetlock(so); 1017 mutex_exit(&vp->v_interlock); 1018 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) { 1019 /* 1020 * This may seem somewhat fragile but is OK: if we can 1021 * see SO_ACCEPTCONN set on the endpoint, then it must 1022 * be locked by the domain-wide uipc_lock. 1023 */ 1024 KASSERT((so2->so_options & SO_ACCEPTCONN) == 0 || 1025 so2->so_lock == uipc_lock); 1026 if ((so2->so_options & SO_ACCEPTCONN) == 0 || 1027 (so3 = sonewconn(so2, 0)) == NULL) { 1028 error = ECONNREFUSED; 1029 sounlock(so); 1030 goto bad; 1031 } 1032 unp2 = sotounpcb(so2); 1033 unp3 = sotounpcb(so3); 1034 if (unp2->unp_addr) { 1035 unp3->unp_addr = malloc(unp2->unp_addrlen, 1036 M_SONAME, M_WAITOK); 1037 memcpy(unp3->unp_addr, unp2->unp_addr, 1038 unp2->unp_addrlen); 1039 unp3->unp_addrlen = unp2->unp_addrlen; 1040 } 1041 unp3->unp_flags = unp2->unp_flags; 1042 unp3->unp_connid.unp_pid = l->l_proc->p_pid; 1043 unp3->unp_connid.unp_euid = kauth_cred_geteuid(l->l_cred); 1044 unp3->unp_connid.unp_egid = kauth_cred_getegid(l->l_cred); 1045 unp3->unp_flags |= UNP_EIDSVALID; 1046 if (unp2->unp_flags & UNP_EIDSBIND) { 1047 unp->unp_connid = unp2->unp_connid; 1048 unp->unp_flags |= UNP_EIDSVALID; 1049 } 1050 so2 = so3; 1051 } 1052 error = unp_connect2(so, so2, PRU_CONNECT); 1053 sounlock(so); 1054 bad: 1055 vput(vp); 1056 bad2: 1057 free(sun, M_SONAME); 1058 solock(so); 1059 unp->unp_flags &= ~UNP_BUSY; 1060 return (error); 1061 } 1062 1063 int 1064 unp_connect2(struct socket *so, struct socket *so2, int req) 1065 { 1066 struct unpcb *unp = sotounpcb(so); 1067 struct unpcb *unp2; 1068 1069 if (so2->so_type != so->so_type) 1070 return (EPROTOTYPE); 1071 1072 /* 1073 * All three sockets involved must be locked by same lock: 1074 * 1075 * local endpoint (so) 1076 * remote endpoint (so2) 1077 * queue head (so2->so_head, only if PR_CONNREQUIRED) 1078 */ 1079 KASSERT(solocked2(so, so2)); 1080 KASSERT(so->so_head == NULL); 1081 if (so2->so_head != NULL) { 1082 KASSERT(so2->so_lock == uipc_lock); 1083 KASSERT(solocked2(so2, so2->so_head)); 1084 } 1085 1086 unp2 = sotounpcb(so2); 1087 unp->unp_conn = unp2; 1088 switch (so->so_type) { 1089 1090 case SOCK_DGRAM: 1091 unp->unp_nextref = unp2->unp_refs; 1092 unp2->unp_refs = unp; 1093 soisconnected(so); 1094 break; 1095 1096 case SOCK_STREAM: 1097 unp2->unp_conn = unp; 1098 if (req == PRU_CONNECT && 1099 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 1100 soisconnecting(so); 1101 else 1102 soisconnected(so); 1103 soisconnected(so2); 1104 /* 1105 * If the connection is fully established, break the 1106 * association with uipc_lock and give the connected 1107 * pair a seperate lock to share. For CONNECT2, we 1108 * require that the locks already match (the sockets 1109 * are created that way). 1110 */ 1111 if (req == PRU_CONNECT) { 1112 KASSERT(so2->so_head != NULL); 1113 unp_setpeerlocks(so, so2); 1114 } 1115 break; 1116 1117 default: 1118 panic("unp_connect2"); 1119 } 1120 return (0); 1121 } 1122 1123 void 1124 unp_disconnect(struct unpcb *unp) 1125 { 1126 struct unpcb *unp2 = unp->unp_conn; 1127 struct socket *so; 1128 1129 if (unp2 == 0) 1130 return; 1131 unp->unp_conn = 0; 1132 so = unp->unp_socket; 1133 switch (so->so_type) { 1134 case SOCK_DGRAM: 1135 if (unp2->unp_refs == unp) 1136 unp2->unp_refs = unp->unp_nextref; 1137 else { 1138 unp2 = unp2->unp_refs; 1139 for (;;) { 1140 KASSERT(solocked2(so, unp2->unp_socket)); 1141 if (unp2 == 0) 1142 panic("unp_disconnect"); 1143 if (unp2->unp_nextref == unp) 1144 break; 1145 unp2 = unp2->unp_nextref; 1146 } 1147 unp2->unp_nextref = unp->unp_nextref; 1148 } 1149 unp->unp_nextref = 0; 1150 so->so_state &= ~SS_ISCONNECTED; 1151 break; 1152 1153 case SOCK_STREAM: 1154 KASSERT(solocked2(so, unp2->unp_socket)); 1155 soisdisconnected(so); 1156 unp2->unp_conn = 0; 1157 soisdisconnected(unp2->unp_socket); 1158 break; 1159 } 1160 } 1161 1162 #ifdef notdef 1163 unp_abort(struct unpcb *unp) 1164 { 1165 unp_detach(unp); 1166 } 1167 #endif 1168 1169 void 1170 unp_shutdown(struct unpcb *unp) 1171 { 1172 struct socket *so; 1173 1174 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn && 1175 (so = unp->unp_conn->unp_socket)) 1176 socantrcvmore(so); 1177 } 1178 1179 bool 1180 unp_drop(struct unpcb *unp, int errno) 1181 { 1182 struct socket *so = unp->unp_socket; 1183 1184 KASSERT(solocked(so)); 1185 1186 so->so_error = errno; 1187 unp_disconnect(unp); 1188 if (so->so_head) { 1189 so->so_pcb = NULL; 1190 /* sofree() drops the socket lock */ 1191 sofree(so); 1192 unp_free(unp); 1193 return true; 1194 } 1195 return false; 1196 } 1197 1198 #ifdef notdef 1199 unp_drain(void) 1200 { 1201 1202 } 1203 #endif 1204 1205 int 1206 unp_externalize(struct mbuf *rights, struct lwp *l) 1207 { 1208 struct cmsghdr *cm = mtod(rights, struct cmsghdr *); 1209 struct proc *p = l->l_proc; 1210 int i, *fdp; 1211 file_t **rp; 1212 file_t *fp; 1213 int nfds, error = 0; 1214 1215 nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) / 1216 sizeof(file_t *); 1217 rp = (file_t **)CMSG_DATA(cm); 1218 1219 fdp = malloc(nfds * sizeof(int), M_TEMP, M_WAITOK); 1220 rw_enter(&p->p_cwdi->cwdi_lock, RW_READER); 1221 1222 /* Make sure the recipient should be able to see the files.. */ 1223 if (p->p_cwdi->cwdi_rdir != NULL) { 1224 rp = (file_t **)CMSG_DATA(cm); 1225 for (i = 0; i < nfds; i++) { 1226 fp = *rp++; 1227 /* 1228 * If we are in a chroot'ed directory, and 1229 * someone wants to pass us a directory, make 1230 * sure it's inside the subtree we're allowed 1231 * to access. 1232 */ 1233 if (fp->f_type == DTYPE_VNODE) { 1234 vnode_t *vp = (vnode_t *)fp->f_data; 1235 if ((vp->v_type == VDIR) && 1236 !vn_isunder(vp, p->p_cwdi->cwdi_rdir, l)) { 1237 error = EPERM; 1238 break; 1239 } 1240 } 1241 } 1242 } 1243 1244 restart: 1245 rp = (file_t **)CMSG_DATA(cm); 1246 if (error != 0) { 1247 for (i = 0; i < nfds; i++) { 1248 fp = *rp; 1249 *rp++ = 0; 1250 unp_discard_now(fp); 1251 } 1252 goto out; 1253 } 1254 1255 /* 1256 * First loop -- allocate file descriptor table slots for the 1257 * new files. 1258 */ 1259 for (i = 0; i < nfds; i++) { 1260 fp = *rp++; 1261 if ((error = fd_alloc(p, 0, &fdp[i])) != 0) { 1262 /* 1263 * Back out what we've done so far. 1264 */ 1265 for (--i; i >= 0; i--) { 1266 fd_abort(p, NULL, fdp[i]); 1267 } 1268 if (error == ENOSPC) { 1269 fd_tryexpand(p); 1270 error = 0; 1271 } else { 1272 /* 1273 * This is the error that has historically 1274 * been returned, and some callers may 1275 * expect it. 1276 */ 1277 error = EMSGSIZE; 1278 } 1279 goto restart; 1280 } 1281 } 1282 1283 /* 1284 * Now that adding them has succeeded, update all of the 1285 * file passing state and affix the descriptors. 1286 */ 1287 rp = (file_t **)CMSG_DATA(cm); 1288 for (i = 0; i < nfds; i++) { 1289 fp = *rp++; 1290 atomic_dec_uint(&unp_rights); 1291 fd_affix(p, fp, fdp[i]); 1292 mutex_enter(&fp->f_lock); 1293 fp->f_msgcount--; 1294 mutex_exit(&fp->f_lock); 1295 /* 1296 * Note that fd_affix() adds a reference to the file. 1297 * The file may already have been closed by another 1298 * LWP in the process, so we must drop the reference 1299 * added by unp_internalize() with closef(). 1300 */ 1301 closef(fp); 1302 } 1303 1304 /* 1305 * Copy temporary array to message and adjust length, in case of 1306 * transition from large file_t pointers to ints. 1307 */ 1308 memcpy(CMSG_DATA(cm), fdp, nfds * sizeof(int)); 1309 cm->cmsg_len = CMSG_LEN(nfds * sizeof(int)); 1310 rights->m_len = CMSG_SPACE(nfds * sizeof(int)); 1311 out: 1312 rw_exit(&p->p_cwdi->cwdi_lock); 1313 free(fdp, M_TEMP); 1314 return (error); 1315 } 1316 1317 int 1318 unp_internalize(struct mbuf **controlp) 1319 { 1320 filedesc_t *fdescp = curlwp->l_fd; 1321 struct mbuf *control = *controlp; 1322 struct cmsghdr *newcm, *cm = mtod(control, struct cmsghdr *); 1323 file_t **rp, **files; 1324 file_t *fp; 1325 int i, fd, *fdp; 1326 int nfds, error; 1327 u_int maxmsg; 1328 1329 error = 0; 1330 newcm = NULL; 1331 1332 /* Sanity check the control message header. */ 1333 if (cm->cmsg_type != SCM_RIGHTS || cm->cmsg_level != SOL_SOCKET || 1334 cm->cmsg_len > control->m_len || 1335 cm->cmsg_len < CMSG_ALIGN(sizeof(*cm))) 1336 return (EINVAL); 1337 1338 /* 1339 * Verify that the file descriptors are valid, and acquire 1340 * a reference to each. 1341 */ 1342 nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) / sizeof(int); 1343 fdp = (int *)CMSG_DATA(cm); 1344 maxmsg = maxfiles / unp_rights_ratio; 1345 for (i = 0; i < nfds; i++) { 1346 fd = *fdp++; 1347 if (atomic_inc_uint_nv(&unp_rights) > maxmsg) { 1348 atomic_dec_uint(&unp_rights); 1349 nfds = i; 1350 error = EAGAIN; 1351 goto out; 1352 } 1353 if ((fp = fd_getfile(fd)) == NULL) { 1354 atomic_dec_uint(&unp_rights); 1355 nfds = i; 1356 error = EBADF; 1357 goto out; 1358 } 1359 } 1360 1361 /* Allocate new space and copy header into it. */ 1362 newcm = malloc(CMSG_SPACE(nfds * sizeof(file_t *)), M_MBUF, M_WAITOK); 1363 if (newcm == NULL) { 1364 error = E2BIG; 1365 goto out; 1366 } 1367 memcpy(newcm, cm, sizeof(struct cmsghdr)); 1368 files = (file_t **)CMSG_DATA(newcm); 1369 1370 /* 1371 * Transform the file descriptors into file_t pointers, in 1372 * reverse order so that if pointers are bigger than ints, the 1373 * int won't get until we're done. No need to lock, as we have 1374 * already validated the descriptors with fd_getfile(). 1375 */ 1376 fdp = (int *)CMSG_DATA(cm) + nfds; 1377 rp = files + nfds; 1378 for (i = 0; i < nfds; i++) { 1379 fp = fdescp->fd_dt->dt_ff[*--fdp]->ff_file; 1380 KASSERT(fp != NULL); 1381 mutex_enter(&fp->f_lock); 1382 *--rp = fp; 1383 fp->f_count++; 1384 fp->f_msgcount++; 1385 mutex_exit(&fp->f_lock); 1386 } 1387 1388 out: 1389 /* Release descriptor references. */ 1390 fdp = (int *)CMSG_DATA(cm); 1391 for (i = 0; i < nfds; i++) { 1392 fd_putfile(*fdp++); 1393 if (error != 0) { 1394 atomic_dec_uint(&unp_rights); 1395 } 1396 } 1397 1398 if (error == 0) { 1399 if (control->m_flags & M_EXT) { 1400 m_freem(control); 1401 *controlp = control = m_get(M_WAIT, MT_CONTROL); 1402 } 1403 MEXTADD(control, newcm, CMSG_SPACE(nfds * sizeof(file_t *)), 1404 M_MBUF, NULL, NULL); 1405 cm = newcm; 1406 /* 1407 * Adjust message & mbuf to note amount of space 1408 * actually used. 1409 */ 1410 cm->cmsg_len = CMSG_LEN(nfds * sizeof(file_t *)); 1411 control->m_len = CMSG_SPACE(nfds * sizeof(file_t *)); 1412 } 1413 1414 return error; 1415 } 1416 1417 struct mbuf * 1418 unp_addsockcred(struct lwp *l, struct mbuf *control) 1419 { 1420 struct cmsghdr *cmp; 1421 struct sockcred *sc; 1422 struct mbuf *m, *n; 1423 int len, space, i; 1424 1425 len = CMSG_LEN(SOCKCREDSIZE(kauth_cred_ngroups(l->l_cred))); 1426 space = CMSG_SPACE(SOCKCREDSIZE(kauth_cred_ngroups(l->l_cred))); 1427 1428 m = m_get(M_WAIT, MT_CONTROL); 1429 if (space > MLEN) { 1430 if (space > MCLBYTES) 1431 MEXTMALLOC(m, space, M_WAITOK); 1432 else 1433 m_clget(m, M_WAIT); 1434 if ((m->m_flags & M_EXT) == 0) { 1435 m_free(m); 1436 return (control); 1437 } 1438 } 1439 1440 m->m_len = space; 1441 m->m_next = NULL; 1442 cmp = mtod(m, struct cmsghdr *); 1443 sc = (struct sockcred *)CMSG_DATA(cmp); 1444 cmp->cmsg_len = len; 1445 cmp->cmsg_level = SOL_SOCKET; 1446 cmp->cmsg_type = SCM_CREDS; 1447 sc->sc_uid = kauth_cred_getuid(l->l_cred); 1448 sc->sc_euid = kauth_cred_geteuid(l->l_cred); 1449 sc->sc_gid = kauth_cred_getgid(l->l_cred); 1450 sc->sc_egid = kauth_cred_getegid(l->l_cred); 1451 sc->sc_ngroups = kauth_cred_ngroups(l->l_cred); 1452 for (i = 0; i < sc->sc_ngroups; i++) 1453 sc->sc_groups[i] = kauth_cred_group(l->l_cred, i); 1454 1455 /* 1456 * If a control message already exists, append us to the end. 1457 */ 1458 if (control != NULL) { 1459 for (n = control; n->m_next != NULL; n = n->m_next) 1460 ; 1461 n->m_next = m; 1462 } else 1463 control = m; 1464 1465 return (control); 1466 } 1467 1468 /* 1469 * Do a mark-sweep GC of files in the system, to free up any which are 1470 * caught in flight to an about-to-be-closed socket. Additionally, 1471 * process deferred file closures. 1472 */ 1473 static void 1474 unp_gc(file_t *dp) 1475 { 1476 extern struct domain unixdomain; 1477 file_t *fp, *np; 1478 struct socket *so, *so1; 1479 u_int i, old, new; 1480 bool didwork; 1481 1482 KASSERT(curlwp == unp_thread_lwp); 1483 KASSERT(mutex_owned(&filelist_lock)); 1484 1485 /* 1486 * First, process deferred file closures. 1487 */ 1488 while (!SLIST_EMPTY(&unp_thread_discard)) { 1489 fp = SLIST_FIRST(&unp_thread_discard); 1490 KASSERT(fp->f_unpcount > 0); 1491 KASSERT(fp->f_count > 0); 1492 KASSERT(fp->f_msgcount > 0); 1493 KASSERT(fp->f_count >= fp->f_unpcount); 1494 KASSERT(fp->f_count >= fp->f_msgcount); 1495 KASSERT(fp->f_msgcount >= fp->f_unpcount); 1496 SLIST_REMOVE_HEAD(&unp_thread_discard, f_unplist); 1497 i = fp->f_unpcount; 1498 fp->f_unpcount = 0; 1499 mutex_exit(&filelist_lock); 1500 for (; i != 0; i--) { 1501 unp_discard_now(fp); 1502 } 1503 mutex_enter(&filelist_lock); 1504 } 1505 1506 /* 1507 * Clear mark bits. Ensure that we don't consider new files 1508 * entering the file table during this loop (they will not have 1509 * FSCAN set). 1510 */ 1511 unp_defer = 0; 1512 LIST_FOREACH(fp, &filehead, f_list) { 1513 for (old = fp->f_flag;; old = new) { 1514 new = atomic_cas_uint(&fp->f_flag, old, 1515 (old | FSCAN) & ~(FMARK|FDEFER)); 1516 if (__predict_true(old == new)) { 1517 break; 1518 } 1519 } 1520 } 1521 1522 /* 1523 * Iterate over the set of sockets, marking ones believed (based on 1524 * refcount) to be referenced from a process, and marking for rescan 1525 * sockets which are queued on a socket. Recan continues descending 1526 * and searching for sockets referenced by sockets (FDEFER), until 1527 * there are no more socket->socket references to be discovered. 1528 */ 1529 do { 1530 didwork = false; 1531 for (fp = LIST_FIRST(&filehead); fp != NULL; fp = np) { 1532 KASSERT(mutex_owned(&filelist_lock)); 1533 np = LIST_NEXT(fp, f_list); 1534 mutex_enter(&fp->f_lock); 1535 if ((fp->f_flag & FDEFER) != 0) { 1536 atomic_and_uint(&fp->f_flag, ~FDEFER); 1537 unp_defer--; 1538 KASSERT(fp->f_count != 0); 1539 } else { 1540 if (fp->f_count == 0 || 1541 (fp->f_flag & FMARK) != 0 || 1542 fp->f_count == fp->f_msgcount || 1543 fp->f_unpcount != 0) { 1544 mutex_exit(&fp->f_lock); 1545 continue; 1546 } 1547 } 1548 atomic_or_uint(&fp->f_flag, FMARK); 1549 1550 if (fp->f_type != DTYPE_SOCKET || 1551 (so = fp->f_data) == NULL || 1552 so->so_proto->pr_domain != &unixdomain || 1553 (so->so_proto->pr_flags & PR_RIGHTS) == 0) { 1554 mutex_exit(&fp->f_lock); 1555 continue; 1556 } 1557 1558 /* Gain file ref, mark our position, and unlock. */ 1559 didwork = true; 1560 LIST_INSERT_AFTER(fp, dp, f_list); 1561 fp->f_count++; 1562 mutex_exit(&fp->f_lock); 1563 mutex_exit(&filelist_lock); 1564 1565 /* 1566 * Mark files referenced from sockets queued on the 1567 * accept queue as well. 1568 */ 1569 solock(so); 1570 unp_scan(so->so_rcv.sb_mb, unp_mark, 0); 1571 if ((so->so_options & SO_ACCEPTCONN) != 0) { 1572 TAILQ_FOREACH(so1, &so->so_q0, so_qe) { 1573 unp_scan(so1->so_rcv.sb_mb, unp_mark, 0); 1574 } 1575 TAILQ_FOREACH(so1, &so->so_q, so_qe) { 1576 unp_scan(so1->so_rcv.sb_mb, unp_mark, 0); 1577 } 1578 } 1579 sounlock(so); 1580 1581 /* Re-lock and restart from where we left off. */ 1582 closef(fp); 1583 mutex_enter(&filelist_lock); 1584 np = LIST_NEXT(dp, f_list); 1585 LIST_REMOVE(dp, f_list); 1586 } 1587 /* 1588 * Bail early if we did nothing in the loop above. Could 1589 * happen because of concurrent activity causing unp_defer 1590 * to get out of sync. 1591 */ 1592 } while (unp_defer != 0 && didwork); 1593 1594 /* 1595 * Sweep pass. 1596 * 1597 * We grab an extra reference to each of the files that are 1598 * not otherwise accessible and then free the rights that are 1599 * stored in messages on them. 1600 */ 1601 for (fp = LIST_FIRST(&filehead); fp != NULL; fp = np) { 1602 KASSERT(mutex_owned(&filelist_lock)); 1603 np = LIST_NEXT(fp, f_list); 1604 mutex_enter(&fp->f_lock); 1605 1606 /* 1607 * Ignore non-sockets. 1608 * Ignore dead sockets, or sockets with pending close. 1609 * Ignore sockets obviously referenced elsewhere. 1610 * Ignore sockets marked as referenced by our scan. 1611 * Ignore new sockets that did not exist during the scan. 1612 */ 1613 if (fp->f_type != DTYPE_SOCKET || 1614 fp->f_count == 0 || fp->f_unpcount != 0 || 1615 fp->f_count != fp->f_msgcount || 1616 (fp->f_flag & (FMARK | FSCAN)) != FSCAN) { 1617 mutex_exit(&fp->f_lock); 1618 continue; 1619 } 1620 1621 /* Gain file ref, mark our position, and unlock. */ 1622 LIST_INSERT_AFTER(fp, dp, f_list); 1623 fp->f_count++; 1624 mutex_exit(&fp->f_lock); 1625 mutex_exit(&filelist_lock); 1626 1627 /* 1628 * Flush all data from the socket's receive buffer. 1629 * This will cause files referenced only by the 1630 * socket to be queued for close. 1631 */ 1632 so = fp->f_data; 1633 solock(so); 1634 sorflush(so); 1635 sounlock(so); 1636 1637 /* Re-lock and restart from where we left off. */ 1638 closef(fp); 1639 mutex_enter(&filelist_lock); 1640 np = LIST_NEXT(dp, f_list); 1641 LIST_REMOVE(dp, f_list); 1642 } 1643 } 1644 1645 /* 1646 * Garbage collector thread. While SCM_RIGHTS messages are in transit, 1647 * wake once per second to garbage collect. Run continually while we 1648 * have deferred closes to process. 1649 */ 1650 static void 1651 unp_thread(void *cookie) 1652 { 1653 file_t *dp; 1654 1655 /* Allocate a dummy file for our scans. */ 1656 if ((dp = fgetdummy()) == NULL) { 1657 panic("unp_thread"); 1658 } 1659 1660 mutex_enter(&filelist_lock); 1661 for (;;) { 1662 KASSERT(mutex_owned(&filelist_lock)); 1663 if (SLIST_EMPTY(&unp_thread_discard)) { 1664 if (unp_rights != 0) { 1665 (void)cv_timedwait(&unp_thread_cv, 1666 &filelist_lock, hz); 1667 } else { 1668 cv_wait(&unp_thread_cv, &filelist_lock); 1669 } 1670 } 1671 unp_gc(dp); 1672 } 1673 /* NOTREACHED */ 1674 } 1675 1676 /* 1677 * Kick the garbage collector into action if there is something for 1678 * it to process. 1679 */ 1680 static void 1681 unp_thread_kick(void) 1682 { 1683 1684 if (!SLIST_EMPTY(&unp_thread_discard) || unp_rights != 0) { 1685 mutex_enter(&filelist_lock); 1686 cv_signal(&unp_thread_cv); 1687 mutex_exit(&filelist_lock); 1688 } 1689 } 1690 1691 void 1692 unp_dispose(struct mbuf *m) 1693 { 1694 1695 if (m) 1696 unp_scan(m, unp_discard_later, 1); 1697 } 1698 1699 void 1700 unp_scan(struct mbuf *m0, void (*op)(file_t *), int discard) 1701 { 1702 struct mbuf *m; 1703 file_t **rp, *fp; 1704 struct cmsghdr *cm; 1705 int i, qfds; 1706 1707 while (m0) { 1708 for (m = m0; m; m = m->m_next) { 1709 if (m->m_type != MT_CONTROL || 1710 m->m_len < sizeof(*cm)) { 1711 continue; 1712 } 1713 cm = mtod(m, struct cmsghdr *); 1714 if (cm->cmsg_level != SOL_SOCKET || 1715 cm->cmsg_type != SCM_RIGHTS) 1716 continue; 1717 qfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) 1718 / sizeof(file_t *); 1719 rp = (file_t **)CMSG_DATA(cm); 1720 for (i = 0; i < qfds; i++) { 1721 fp = *rp; 1722 if (discard) { 1723 *rp = 0; 1724 } 1725 (*op)(fp); 1726 rp++; 1727 } 1728 } 1729 m0 = m0->m_nextpkt; 1730 } 1731 } 1732 1733 void 1734 unp_mark(file_t *fp) 1735 { 1736 1737 if (fp == NULL) 1738 return; 1739 1740 /* If we're already deferred, don't screw up the defer count */ 1741 mutex_enter(&fp->f_lock); 1742 if (fp->f_flag & (FMARK | FDEFER)) { 1743 mutex_exit(&fp->f_lock); 1744 return; 1745 } 1746 1747 /* 1748 * Minimize the number of deferrals... Sockets are the only type of 1749 * file which can hold references to another file, so just mark 1750 * other files, and defer unmarked sockets for the next pass. 1751 */ 1752 if (fp->f_type == DTYPE_SOCKET) { 1753 unp_defer++; 1754 KASSERT(fp->f_count != 0); 1755 atomic_or_uint(&fp->f_flag, FDEFER); 1756 } else { 1757 atomic_or_uint(&fp->f_flag, FMARK); 1758 } 1759 mutex_exit(&fp->f_lock); 1760 } 1761 1762 static void 1763 unp_discard_now(file_t *fp) 1764 { 1765 1766 if (fp == NULL) 1767 return; 1768 1769 KASSERT(fp->f_count > 0); 1770 KASSERT(fp->f_msgcount > 0); 1771 1772 mutex_enter(&fp->f_lock); 1773 fp->f_msgcount--; 1774 mutex_exit(&fp->f_lock); 1775 atomic_dec_uint(&unp_rights); 1776 (void)closef(fp); 1777 } 1778 1779 static void 1780 unp_discard_later(file_t *fp) 1781 { 1782 1783 if (fp == NULL) 1784 return; 1785 1786 KASSERT(fp->f_count > 0); 1787 KASSERT(fp->f_msgcount > 0); 1788 1789 mutex_enter(&filelist_lock); 1790 if (fp->f_unpcount++ == 0) { 1791 SLIST_INSERT_HEAD(&unp_thread_discard, fp, f_unplist); 1792 } 1793 mutex_exit(&filelist_lock); 1794 } 1795