1 /* $NetBSD: uipc_usrreq.c,v 1.155 2014/07/01 05:49:18 rtr 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 1.155 2014/07/01 05:49:18 rtr 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 /* XXX: server side closed the socket */ 307 if (unp->unp_conn == NULL) 308 return ECONNREFUSED; 309 so2 = unp->unp_conn->unp_socket; 310 311 KASSERT(solocked(so2)); 312 313 if (unp->unp_addr) 314 sun = unp->unp_addr; 315 else 316 sun = &sun_noname; 317 if (unp->unp_conn->unp_flags & UNP_WANTCRED) 318 control = unp_addsockcred(l, control); 319 if (sbappendaddr(&so2->so_rcv, (const struct sockaddr *)sun, m, 320 control) == 0) { 321 so2->so_rcv.sb_overflowed++; 322 unp_dispose(control); 323 m_freem(control); 324 m_freem(m); 325 return (ENOBUFS); 326 } else { 327 sorwakeup(so2); 328 return (0); 329 } 330 } 331 332 void 333 unp_setaddr(struct socket *so, struct mbuf *nam, bool peeraddr) 334 { 335 const struct sockaddr_un *sun; 336 struct unpcb *unp; 337 bool ext; 338 339 KASSERT(solocked(so)); 340 unp = sotounpcb(so); 341 ext = false; 342 343 for (;;) { 344 sun = NULL; 345 if (peeraddr) { 346 if (unp->unp_conn && unp->unp_conn->unp_addr) 347 sun = unp->unp_conn->unp_addr; 348 } else { 349 if (unp->unp_addr) 350 sun = unp->unp_addr; 351 } 352 if (sun == NULL) 353 sun = &sun_noname; 354 nam->m_len = sun->sun_len; 355 if (nam->m_len > MLEN && !ext) { 356 sounlock(so); 357 MEXTMALLOC(nam, MAXPATHLEN * 2, M_WAITOK); 358 solock(so); 359 ext = true; 360 } else { 361 KASSERT(nam->m_len <= MAXPATHLEN * 2); 362 memcpy(mtod(nam, void *), sun, (size_t)nam->m_len); 363 break; 364 } 365 } 366 } 367 368 static int 369 unp_usrreq(struct socket *so, int req, struct mbuf *m, struct mbuf *nam, 370 struct mbuf *control, struct lwp *l) 371 { 372 struct unpcb *unp; 373 struct socket *so2; 374 u_int newhiwat; 375 int error = 0; 376 377 KASSERT(req != PRU_ATTACH); 378 KASSERT(req != PRU_DETACH); 379 KASSERT(req != PRU_CONTROL); 380 381 KASSERT(solocked(so)); 382 unp = sotounpcb(so); 383 384 KASSERT(!control || (req == PRU_SEND || req == PRU_SENDOOB)); 385 if (unp == NULL) { 386 error = EINVAL; 387 goto release; 388 } 389 390 switch (req) { 391 case PRU_BIND: 392 KASSERT(l != NULL); 393 error = unp_bind(so, nam, l); 394 break; 395 396 case PRU_LISTEN: 397 /* 398 * If the socket can accept a connection, it must be 399 * locked by uipc_lock. 400 */ 401 unp_resetlock(so); 402 if (unp->unp_vnode == NULL) 403 error = EINVAL; 404 break; 405 406 case PRU_CONNECT: 407 KASSERT(l != NULL); 408 error = unp_connect(so, nam, l); 409 break; 410 411 case PRU_CONNECT2: 412 error = unp_connect2(so, (struct socket *)nam, PRU_CONNECT2); 413 break; 414 415 case PRU_DISCONNECT: 416 unp_disconnect(unp); 417 break; 418 419 case PRU_ACCEPT: 420 KASSERT(so->so_lock == uipc_lock); 421 /* 422 * Mark the initiating STREAM socket as connected *ONLY* 423 * after it's been accepted. This prevents a client from 424 * overrunning a server and receiving ECONNREFUSED. 425 */ 426 if (unp->unp_conn == NULL) { 427 /* 428 * This will use the empty socket and will not 429 * allocate. 430 */ 431 unp_setaddr(so, nam, true); 432 break; 433 } 434 so2 = unp->unp_conn->unp_socket; 435 if (so2->so_state & SS_ISCONNECTING) { 436 KASSERT(solocked2(so, so->so_head)); 437 KASSERT(solocked2(so2, so->so_head)); 438 soisconnected(so2); 439 } 440 /* 441 * If the connection is fully established, break the 442 * association with uipc_lock and give the connected 443 * pair a separate lock to share. 444 * There is a race here: sotounpcb(so2)->unp_streamlock 445 * is not locked, so when changing so2->so_lock 446 * another thread can grab it while so->so_lock is still 447 * pointing to the (locked) uipc_lock. 448 * this should be harmless, except that this makes 449 * solocked2() and solocked() unreliable. 450 * Another problem is that unp_setaddr() expects the 451 * the socket locked. Grabing sotounpcb(so2)->unp_streamlock 452 * fixes both issues. 453 */ 454 mutex_enter(sotounpcb(so2)->unp_streamlock); 455 unp_setpeerlocks(so2, so); 456 /* 457 * Only now return peer's address, as we may need to 458 * block in order to allocate memory. 459 * 460 * XXX Minor race: connection can be broken while 461 * lock is dropped in unp_setaddr(). We will return 462 * error == 0 and sun_noname as the peer address. 463 */ 464 unp_setaddr(so, nam, true); 465 /* so_lock now points to unp_streamlock */ 466 mutex_exit(so2->so_lock); 467 break; 468 469 case PRU_SHUTDOWN: 470 socantsendmore(so); 471 unp_shutdown(unp); 472 break; 473 474 case PRU_RCVD: 475 switch (so->so_type) { 476 477 case SOCK_DGRAM: 478 panic("uipc 1"); 479 /*NOTREACHED*/ 480 481 case SOCK_SEQPACKET: /* FALLTHROUGH */ 482 case SOCK_STREAM: 483 #define rcv (&so->so_rcv) 484 #define snd (&so2->so_snd) 485 if (unp->unp_conn == 0) 486 break; 487 so2 = unp->unp_conn->unp_socket; 488 KASSERT(solocked2(so, so2)); 489 /* 490 * Adjust backpressure on sender 491 * and wakeup any waiting to write. 492 */ 493 snd->sb_mbmax += unp->unp_mbcnt - rcv->sb_mbcnt; 494 unp->unp_mbcnt = rcv->sb_mbcnt; 495 newhiwat = snd->sb_hiwat + unp->unp_cc - rcv->sb_cc; 496 (void)chgsbsize(so2->so_uidinfo, 497 &snd->sb_hiwat, newhiwat, RLIM_INFINITY); 498 unp->unp_cc = rcv->sb_cc; 499 sowwakeup(so2); 500 #undef snd 501 #undef rcv 502 break; 503 504 default: 505 panic("uipc 2"); 506 } 507 break; 508 509 case PRU_SEND: 510 /* 511 * Note: unp_internalize() rejects any control message 512 * other than SCM_RIGHTS, and only allows one. This 513 * has the side-effect of preventing a caller from 514 * forging SCM_CREDS. 515 */ 516 if (control) { 517 sounlock(so); 518 error = unp_internalize(&control); 519 solock(so); 520 if (error != 0) { 521 m_freem(control); 522 m_freem(m); 523 break; 524 } 525 } 526 switch (so->so_type) { 527 528 case SOCK_DGRAM: { 529 KASSERT(so->so_lock == uipc_lock); 530 if (nam) { 531 if ((so->so_state & SS_ISCONNECTED) != 0) 532 error = EISCONN; 533 else { 534 /* 535 * Note: once connected, the 536 * socket's lock must not be 537 * dropped until we have sent 538 * the message and disconnected. 539 * This is necessary to prevent 540 * intervening control ops, like 541 * another connection. 542 */ 543 error = unp_connect(so, nam, l); 544 } 545 } else { 546 if ((so->so_state & SS_ISCONNECTED) == 0) 547 error = ENOTCONN; 548 } 549 if (error) { 550 unp_dispose(control); 551 m_freem(control); 552 m_freem(m); 553 break; 554 } 555 KASSERT(l != NULL); 556 error = unp_output(m, control, unp, l); 557 if (nam) 558 unp_disconnect(unp); 559 break; 560 } 561 562 case SOCK_SEQPACKET: /* FALLTHROUGH */ 563 case SOCK_STREAM: 564 #define rcv (&so2->so_rcv) 565 #define snd (&so->so_snd) 566 if (unp->unp_conn == NULL) { 567 error = ENOTCONN; 568 break; 569 } 570 so2 = unp->unp_conn->unp_socket; 571 KASSERT(solocked2(so, so2)); 572 if (unp->unp_conn->unp_flags & UNP_WANTCRED) { 573 /* 574 * Credentials are passed only once on 575 * SOCK_STREAM and SOCK_SEQPACKET. 576 */ 577 unp->unp_conn->unp_flags &= ~UNP_WANTCRED; 578 control = unp_addsockcred(l, control); 579 } 580 /* 581 * Send to paired receive port, and then reduce 582 * send buffer hiwater marks to maintain backpressure. 583 * Wake up readers. 584 */ 585 if (control) { 586 if (sbappendcontrol(rcv, m, control) != 0) 587 control = NULL; 588 } else { 589 switch(so->so_type) { 590 case SOCK_SEQPACKET: 591 sbappendrecord(rcv, m); 592 break; 593 case SOCK_STREAM: 594 sbappend(rcv, m); 595 break; 596 default: 597 panic("uipc_usrreq"); 598 break; 599 } 600 } 601 snd->sb_mbmax -= 602 rcv->sb_mbcnt - unp->unp_conn->unp_mbcnt; 603 unp->unp_conn->unp_mbcnt = rcv->sb_mbcnt; 604 newhiwat = snd->sb_hiwat - 605 (rcv->sb_cc - unp->unp_conn->unp_cc); 606 (void)chgsbsize(so->so_uidinfo, 607 &snd->sb_hiwat, newhiwat, RLIM_INFINITY); 608 unp->unp_conn->unp_cc = rcv->sb_cc; 609 sorwakeup(so2); 610 #undef snd 611 #undef rcv 612 if (control != NULL) { 613 unp_dispose(control); 614 m_freem(control); 615 } 616 break; 617 618 default: 619 panic("uipc 4"); 620 } 621 break; 622 623 case PRU_ABORT: 624 (void)unp_drop(unp, ECONNABORTED); 625 KASSERT(so->so_head == NULL); 626 KASSERT(so->so_pcb != NULL); 627 unp_detach(so); 628 break; 629 630 case PRU_SENSE: 631 ((struct stat *) m)->st_blksize = so->so_snd.sb_hiwat; 632 switch (so->so_type) { 633 case SOCK_SEQPACKET: /* FALLTHROUGH */ 634 case SOCK_STREAM: 635 if (unp->unp_conn == 0) 636 break; 637 638 so2 = unp->unp_conn->unp_socket; 639 KASSERT(solocked2(so, so2)); 640 ((struct stat *) m)->st_blksize += so2->so_rcv.sb_cc; 641 break; 642 default: 643 break; 644 } 645 ((struct stat *) m)->st_dev = NODEV; 646 if (unp->unp_ino == 0) 647 unp->unp_ino = unp_ino++; 648 ((struct stat *) m)->st_atimespec = 649 ((struct stat *) m)->st_mtimespec = 650 ((struct stat *) m)->st_ctimespec = unp->unp_ctime; 651 ((struct stat *) m)->st_ino = unp->unp_ino; 652 return (0); 653 654 case PRU_RCVOOB: 655 error = EOPNOTSUPP; 656 break; 657 658 case PRU_SENDOOB: 659 m_freem(control); 660 m_freem(m); 661 error = EOPNOTSUPP; 662 break; 663 664 case PRU_SOCKADDR: 665 unp_setaddr(so, nam, false); 666 break; 667 668 case PRU_PEERADDR: 669 unp_setaddr(so, nam, true); 670 break; 671 672 default: 673 panic("piusrreq"); 674 } 675 676 release: 677 return (error); 678 } 679 680 /* 681 * Unix domain socket option processing. 682 */ 683 int 684 uipc_ctloutput(int op, struct socket *so, struct sockopt *sopt) 685 { 686 struct unpcb *unp = sotounpcb(so); 687 int optval = 0, error = 0; 688 689 KASSERT(solocked(so)); 690 691 if (sopt->sopt_level != 0) { 692 error = ENOPROTOOPT; 693 } else switch (op) { 694 695 case PRCO_SETOPT: 696 switch (sopt->sopt_name) { 697 case LOCAL_CREDS: 698 case LOCAL_CONNWAIT: 699 error = sockopt_getint(sopt, &optval); 700 if (error) 701 break; 702 switch (sopt->sopt_name) { 703 #define OPTSET(bit) \ 704 if (optval) \ 705 unp->unp_flags |= (bit); \ 706 else \ 707 unp->unp_flags &= ~(bit); 708 709 case LOCAL_CREDS: 710 OPTSET(UNP_WANTCRED); 711 break; 712 case LOCAL_CONNWAIT: 713 OPTSET(UNP_CONNWAIT); 714 break; 715 } 716 break; 717 #undef OPTSET 718 719 default: 720 error = ENOPROTOOPT; 721 break; 722 } 723 break; 724 725 case PRCO_GETOPT: 726 sounlock(so); 727 switch (sopt->sopt_name) { 728 case LOCAL_PEEREID: 729 if (unp->unp_flags & UNP_EIDSVALID) { 730 error = sockopt_set(sopt, 731 &unp->unp_connid, sizeof(unp->unp_connid)); 732 } else { 733 error = EINVAL; 734 } 735 break; 736 case LOCAL_CREDS: 737 #define OPTBIT(bit) (unp->unp_flags & (bit) ? 1 : 0) 738 739 optval = OPTBIT(UNP_WANTCRED); 740 error = sockopt_setint(sopt, optval); 741 break; 742 #undef OPTBIT 743 744 default: 745 error = ENOPROTOOPT; 746 break; 747 } 748 solock(so); 749 break; 750 } 751 return (error); 752 } 753 754 /* 755 * Both send and receive buffers are allocated PIPSIZ bytes of buffering 756 * for stream sockets, although the total for sender and receiver is 757 * actually only PIPSIZ. 758 * Datagram sockets really use the sendspace as the maximum datagram size, 759 * and don't really want to reserve the sendspace. Their recvspace should 760 * be large enough for at least one max-size datagram plus address. 761 */ 762 #define PIPSIZ 4096 763 u_long unpst_sendspace = PIPSIZ; 764 u_long unpst_recvspace = PIPSIZ; 765 u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 766 u_long unpdg_recvspace = 4*1024; 767 768 u_int unp_rights; /* files in flight */ 769 u_int unp_rights_ratio = 2; /* limit, fraction of maxfiles */ 770 771 static int 772 unp_attach(struct socket *so, int proto) 773 { 774 struct unpcb *unp = sotounpcb(so); 775 u_long sndspc, rcvspc; 776 int error; 777 778 KASSERT(unp == NULL); 779 780 switch (so->so_type) { 781 case SOCK_SEQPACKET: 782 /* FALLTHROUGH */ 783 case SOCK_STREAM: 784 if (so->so_lock == NULL) { 785 so->so_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 786 solock(so); 787 } 788 sndspc = unpst_sendspace; 789 rcvspc = unpst_recvspace; 790 break; 791 792 case SOCK_DGRAM: 793 if (so->so_lock == NULL) { 794 mutex_obj_hold(uipc_lock); 795 so->so_lock = uipc_lock; 796 solock(so); 797 } 798 sndspc = unpdg_sendspace; 799 rcvspc = unpdg_recvspace; 800 break; 801 802 default: 803 panic("unp_attach"); 804 } 805 806 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 807 error = soreserve(so, sndspc, rcvspc); 808 if (error) { 809 return error; 810 } 811 } 812 813 unp = kmem_zalloc(sizeof(*unp), KM_SLEEP); 814 nanotime(&unp->unp_ctime); 815 unp->unp_socket = so; 816 so->so_pcb = unp; 817 818 KASSERT(solocked(so)); 819 return 0; 820 } 821 822 static void 823 unp_detach(struct socket *so) 824 { 825 struct unpcb *unp; 826 vnode_t *vp; 827 828 unp = sotounpcb(so); 829 KASSERT(unp != NULL); 830 KASSERT(solocked(so)); 831 retry: 832 if ((vp = unp->unp_vnode) != NULL) { 833 sounlock(so); 834 /* Acquire v_interlock to protect against unp_connect(). */ 835 /* XXXAD racy */ 836 mutex_enter(vp->v_interlock); 837 vp->v_socket = NULL; 838 mutex_exit(vp->v_interlock); 839 vrele(vp); 840 solock(so); 841 unp->unp_vnode = NULL; 842 } 843 if (unp->unp_conn) 844 unp_disconnect(unp); 845 while (unp->unp_refs) { 846 KASSERT(solocked2(so, unp->unp_refs->unp_socket)); 847 if (unp_drop(unp->unp_refs, ECONNRESET)) { 848 solock(so); 849 goto retry; 850 } 851 } 852 soisdisconnected(so); 853 so->so_pcb = NULL; 854 if (unp_rights) { 855 /* 856 * Normally the receive buffer is flushed later, in sofree, 857 * but if our receive buffer holds references to files that 858 * are now garbage, we will enqueue those file references to 859 * the garbage collector and kick it into action. 860 */ 861 sorflush(so); 862 unp_free(unp); 863 unp_thread_kick(); 864 } else 865 unp_free(unp); 866 } 867 868 static int 869 unp_ioctl(struct socket *so, u_long cmd, void *nam, struct ifnet *ifp) 870 { 871 return EOPNOTSUPP; 872 } 873 874 /* 875 * Allocate the new sockaddr. We have to allocate one 876 * extra byte so that we can ensure that the pathname 877 * is nul-terminated. Note that unlike linux, we don't 878 * include in the address length the NUL in the path 879 * component, because doing so, would exceed sizeof(sockaddr_un) 880 * for fully occupied pathnames. Linux is also inconsistent, 881 * because it does not include the NUL in the length of 882 * what it calls "abstract" unix sockets. 883 */ 884 static struct sockaddr_un * 885 makeun(struct mbuf *nam, size_t *addrlen) { 886 struct sockaddr_un *sun; 887 888 *addrlen = nam->m_len + 1; 889 sun = malloc(*addrlen, M_SONAME, M_WAITOK); 890 m_copydata(nam, 0, nam->m_len, (void *)sun); 891 *(((char *)sun) + nam->m_len) = '\0'; 892 sun->sun_len = strlen(sun->sun_path) + 893 offsetof(struct sockaddr_un, sun_path); 894 return sun; 895 } 896 897 int 898 unp_bind(struct socket *so, struct mbuf *nam, struct lwp *l) 899 { 900 struct sockaddr_un *sun; 901 struct unpcb *unp; 902 vnode_t *vp; 903 struct vattr vattr; 904 size_t addrlen; 905 int error; 906 struct pathbuf *pb; 907 struct nameidata nd; 908 proc_t *p; 909 910 unp = sotounpcb(so); 911 if (unp->unp_vnode != NULL) 912 return (EINVAL); 913 if ((unp->unp_flags & UNP_BUSY) != 0) { 914 /* 915 * EALREADY may not be strictly accurate, but since this 916 * is a major application error it's hardly a big deal. 917 */ 918 return (EALREADY); 919 } 920 unp->unp_flags |= UNP_BUSY; 921 sounlock(so); 922 923 p = l->l_proc; 924 sun = makeun(nam, &addrlen); 925 926 pb = pathbuf_create(sun->sun_path); 927 if (pb == NULL) { 928 error = ENOMEM; 929 goto bad; 930 } 931 NDINIT(&nd, CREATE, FOLLOW | LOCKPARENT | TRYEMULROOT, pb); 932 933 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 934 if ((error = namei(&nd)) != 0) { 935 pathbuf_destroy(pb); 936 goto bad; 937 } 938 vp = nd.ni_vp; 939 if (vp != NULL) { 940 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 941 if (nd.ni_dvp == vp) 942 vrele(nd.ni_dvp); 943 else 944 vput(nd.ni_dvp); 945 vrele(vp); 946 pathbuf_destroy(pb); 947 error = EADDRINUSE; 948 goto bad; 949 } 950 vattr_null(&vattr); 951 vattr.va_type = VSOCK; 952 vattr.va_mode = ACCESSPERMS & ~(p->p_cwdi->cwdi_cmask); 953 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 954 if (error) { 955 vput(nd.ni_dvp); 956 pathbuf_destroy(pb); 957 goto bad; 958 } 959 vp = nd.ni_vp; 960 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 961 solock(so); 962 vp->v_socket = unp->unp_socket; 963 unp->unp_vnode = vp; 964 unp->unp_addrlen = addrlen; 965 unp->unp_addr = sun; 966 unp->unp_connid.unp_pid = p->p_pid; 967 unp->unp_connid.unp_euid = kauth_cred_geteuid(l->l_cred); 968 unp->unp_connid.unp_egid = kauth_cred_getegid(l->l_cred); 969 unp->unp_flags |= UNP_EIDSBIND; 970 VOP_UNLOCK(vp); 971 vput(nd.ni_dvp); 972 unp->unp_flags &= ~UNP_BUSY; 973 pathbuf_destroy(pb); 974 return (0); 975 976 bad: 977 free(sun, M_SONAME); 978 solock(so); 979 unp->unp_flags &= ~UNP_BUSY; 980 return (error); 981 } 982 983 int 984 unp_connect(struct socket *so, struct mbuf *nam, struct lwp *l) 985 { 986 struct sockaddr_un *sun; 987 vnode_t *vp; 988 struct socket *so2, *so3; 989 struct unpcb *unp, *unp2, *unp3; 990 size_t addrlen; 991 int error; 992 struct pathbuf *pb; 993 struct nameidata nd; 994 995 unp = sotounpcb(so); 996 if ((unp->unp_flags & UNP_BUSY) != 0) { 997 /* 998 * EALREADY may not be strictly accurate, but since this 999 * is a major application error it's hardly a big deal. 1000 */ 1001 return (EALREADY); 1002 } 1003 unp->unp_flags |= UNP_BUSY; 1004 sounlock(so); 1005 1006 sun = makeun(nam, &addrlen); 1007 pb = pathbuf_create(sun->sun_path); 1008 if (pb == NULL) { 1009 error = ENOMEM; 1010 goto bad2; 1011 } 1012 1013 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb); 1014 1015 if ((error = namei(&nd)) != 0) { 1016 pathbuf_destroy(pb); 1017 goto bad2; 1018 } 1019 vp = nd.ni_vp; 1020 if (vp->v_type != VSOCK) { 1021 error = ENOTSOCK; 1022 goto bad; 1023 } 1024 pathbuf_destroy(pb); 1025 if ((error = VOP_ACCESS(vp, VWRITE, l->l_cred)) != 0) 1026 goto bad; 1027 /* Acquire v_interlock to protect against unp_detach(). */ 1028 mutex_enter(vp->v_interlock); 1029 so2 = vp->v_socket; 1030 if (so2 == NULL) { 1031 mutex_exit(vp->v_interlock); 1032 error = ECONNREFUSED; 1033 goto bad; 1034 } 1035 if (so->so_type != so2->so_type) { 1036 mutex_exit(vp->v_interlock); 1037 error = EPROTOTYPE; 1038 goto bad; 1039 } 1040 solock(so); 1041 unp_resetlock(so); 1042 mutex_exit(vp->v_interlock); 1043 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) { 1044 /* 1045 * This may seem somewhat fragile but is OK: if we can 1046 * see SO_ACCEPTCONN set on the endpoint, then it must 1047 * be locked by the domain-wide uipc_lock. 1048 */ 1049 KASSERT((so2->so_options & SO_ACCEPTCONN) == 0 || 1050 so2->so_lock == uipc_lock); 1051 if ((so2->so_options & SO_ACCEPTCONN) == 0 || 1052 (so3 = sonewconn(so2, false)) == NULL) { 1053 error = ECONNREFUSED; 1054 sounlock(so); 1055 goto bad; 1056 } 1057 unp2 = sotounpcb(so2); 1058 unp3 = sotounpcb(so3); 1059 if (unp2->unp_addr) { 1060 unp3->unp_addr = malloc(unp2->unp_addrlen, 1061 M_SONAME, M_WAITOK); 1062 memcpy(unp3->unp_addr, unp2->unp_addr, 1063 unp2->unp_addrlen); 1064 unp3->unp_addrlen = unp2->unp_addrlen; 1065 } 1066 unp3->unp_flags = unp2->unp_flags; 1067 unp3->unp_connid.unp_pid = l->l_proc->p_pid; 1068 unp3->unp_connid.unp_euid = kauth_cred_geteuid(l->l_cred); 1069 unp3->unp_connid.unp_egid = kauth_cred_getegid(l->l_cred); 1070 unp3->unp_flags |= UNP_EIDSVALID; 1071 if (unp2->unp_flags & UNP_EIDSBIND) { 1072 unp->unp_connid = unp2->unp_connid; 1073 unp->unp_flags |= UNP_EIDSVALID; 1074 } 1075 so2 = so3; 1076 } 1077 error = unp_connect2(so, so2, PRU_CONNECT); 1078 sounlock(so); 1079 bad: 1080 vput(vp); 1081 bad2: 1082 free(sun, M_SONAME); 1083 solock(so); 1084 unp->unp_flags &= ~UNP_BUSY; 1085 return (error); 1086 } 1087 1088 int 1089 unp_connect2(struct socket *so, struct socket *so2, int req) 1090 { 1091 struct unpcb *unp = sotounpcb(so); 1092 struct unpcb *unp2; 1093 1094 if (so2->so_type != so->so_type) 1095 return (EPROTOTYPE); 1096 1097 /* 1098 * All three sockets involved must be locked by same lock: 1099 * 1100 * local endpoint (so) 1101 * remote endpoint (so2) 1102 * queue head (so2->so_head, only if PR_CONNREQUIRED) 1103 */ 1104 KASSERT(solocked2(so, so2)); 1105 KASSERT(so->so_head == NULL); 1106 if (so2->so_head != NULL) { 1107 KASSERT(so2->so_lock == uipc_lock); 1108 KASSERT(solocked2(so2, so2->so_head)); 1109 } 1110 1111 unp2 = sotounpcb(so2); 1112 unp->unp_conn = unp2; 1113 switch (so->so_type) { 1114 1115 case SOCK_DGRAM: 1116 unp->unp_nextref = unp2->unp_refs; 1117 unp2->unp_refs = unp; 1118 soisconnected(so); 1119 break; 1120 1121 case SOCK_SEQPACKET: /* FALLTHROUGH */ 1122 case SOCK_STREAM: 1123 unp2->unp_conn = unp; 1124 if (req == PRU_CONNECT && 1125 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 1126 soisconnecting(so); 1127 else 1128 soisconnected(so); 1129 soisconnected(so2); 1130 /* 1131 * If the connection is fully established, break the 1132 * association with uipc_lock and give the connected 1133 * pair a seperate lock to share. For CONNECT2, we 1134 * require that the locks already match (the sockets 1135 * are created that way). 1136 */ 1137 if (req == PRU_CONNECT) { 1138 KASSERT(so2->so_head != NULL); 1139 unp_setpeerlocks(so, so2); 1140 } 1141 break; 1142 1143 default: 1144 panic("unp_connect2"); 1145 } 1146 return (0); 1147 } 1148 1149 void 1150 unp_disconnect(struct unpcb *unp) 1151 { 1152 struct unpcb *unp2 = unp->unp_conn; 1153 struct socket *so; 1154 1155 if (unp2 == 0) 1156 return; 1157 unp->unp_conn = 0; 1158 so = unp->unp_socket; 1159 switch (so->so_type) { 1160 case SOCK_DGRAM: 1161 if (unp2->unp_refs == unp) 1162 unp2->unp_refs = unp->unp_nextref; 1163 else { 1164 unp2 = unp2->unp_refs; 1165 for (;;) { 1166 KASSERT(solocked2(so, unp2->unp_socket)); 1167 if (unp2 == 0) 1168 panic("unp_disconnect"); 1169 if (unp2->unp_nextref == unp) 1170 break; 1171 unp2 = unp2->unp_nextref; 1172 } 1173 unp2->unp_nextref = unp->unp_nextref; 1174 } 1175 unp->unp_nextref = 0; 1176 so->so_state &= ~SS_ISCONNECTED; 1177 break; 1178 1179 case SOCK_SEQPACKET: /* FALLTHROUGH */ 1180 case SOCK_STREAM: 1181 KASSERT(solocked2(so, unp2->unp_socket)); 1182 soisdisconnected(so); 1183 unp2->unp_conn = 0; 1184 soisdisconnected(unp2->unp_socket); 1185 break; 1186 } 1187 } 1188 1189 void 1190 unp_shutdown(struct unpcb *unp) 1191 { 1192 struct socket *so; 1193 1194 switch(unp->unp_socket->so_type) { 1195 case SOCK_SEQPACKET: /* FALLTHROUGH */ 1196 case SOCK_STREAM: 1197 if (unp->unp_conn && (so = unp->unp_conn->unp_socket)) 1198 socantrcvmore(so); 1199 break; 1200 default: 1201 break; 1202 } 1203 } 1204 1205 bool 1206 unp_drop(struct unpcb *unp, int errno) 1207 { 1208 struct socket *so = unp->unp_socket; 1209 1210 KASSERT(solocked(so)); 1211 1212 so->so_error = errno; 1213 unp_disconnect(unp); 1214 if (so->so_head) { 1215 so->so_pcb = NULL; 1216 /* sofree() drops the socket lock */ 1217 sofree(so); 1218 unp_free(unp); 1219 return true; 1220 } 1221 return false; 1222 } 1223 1224 #ifdef notdef 1225 unp_drain(void) 1226 { 1227 1228 } 1229 #endif 1230 1231 int 1232 unp_externalize(struct mbuf *rights, struct lwp *l, int flags) 1233 { 1234 struct cmsghdr * const cm = mtod(rights, struct cmsghdr *); 1235 struct proc * const p = l->l_proc; 1236 file_t **rp; 1237 int error = 0; 1238 1239 const size_t nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) / 1240 sizeof(file_t *); 1241 if (nfds == 0) 1242 goto noop; 1243 1244 int * const fdp = kmem_alloc(nfds * sizeof(int), KM_SLEEP); 1245 rw_enter(&p->p_cwdi->cwdi_lock, RW_READER); 1246 1247 /* Make sure the recipient should be able to see the files.. */ 1248 rp = (file_t **)CMSG_DATA(cm); 1249 for (size_t i = 0; i < nfds; i++) { 1250 file_t * const fp = *rp++; 1251 if (fp == NULL) { 1252 error = EINVAL; 1253 goto out; 1254 } 1255 /* 1256 * If we are in a chroot'ed directory, and 1257 * someone wants to pass us a directory, make 1258 * sure it's inside the subtree we're allowed 1259 * to access. 1260 */ 1261 if (p->p_cwdi->cwdi_rdir != NULL && fp->f_type == DTYPE_VNODE) { 1262 vnode_t *vp = (vnode_t *)fp->f_data; 1263 if ((vp->v_type == VDIR) && 1264 !vn_isunder(vp, p->p_cwdi->cwdi_rdir, l)) { 1265 error = EPERM; 1266 goto out; 1267 } 1268 } 1269 } 1270 1271 restart: 1272 /* 1273 * First loop -- allocate file descriptor table slots for the 1274 * new files. 1275 */ 1276 for (size_t i = 0; i < nfds; i++) { 1277 if ((error = fd_alloc(p, 0, &fdp[i])) != 0) { 1278 /* 1279 * Back out what we've done so far. 1280 */ 1281 while (i-- > 0) { 1282 fd_abort(p, NULL, fdp[i]); 1283 } 1284 if (error == ENOSPC) { 1285 fd_tryexpand(p); 1286 error = 0; 1287 goto restart; 1288 } 1289 /* 1290 * This is the error that has historically 1291 * been returned, and some callers may 1292 * expect it. 1293 */ 1294 error = EMSGSIZE; 1295 goto out; 1296 } 1297 } 1298 1299 /* 1300 * Now that adding them has succeeded, update all of the 1301 * file passing state and affix the descriptors. 1302 */ 1303 rp = (file_t **)CMSG_DATA(cm); 1304 int *ofdp = (int *)CMSG_DATA(cm); 1305 for (size_t i = 0; i < nfds; i++) { 1306 file_t * const fp = *rp++; 1307 const int fd = fdp[i]; 1308 atomic_dec_uint(&unp_rights); 1309 fd_set_exclose(l, fd, (flags & O_CLOEXEC) != 0); 1310 fd_affix(p, fp, fd); 1311 /* 1312 * Done with this file pointer, replace it with a fd; 1313 */ 1314 *ofdp++ = fd; 1315 mutex_enter(&fp->f_lock); 1316 fp->f_msgcount--; 1317 mutex_exit(&fp->f_lock); 1318 /* 1319 * Note that fd_affix() adds a reference to the file. 1320 * The file may already have been closed by another 1321 * LWP in the process, so we must drop the reference 1322 * added by unp_internalize() with closef(). 1323 */ 1324 closef(fp); 1325 } 1326 1327 /* 1328 * Adjust length, in case of transition from large file_t 1329 * pointers to ints. 1330 */ 1331 if (sizeof(file_t *) != sizeof(int)) { 1332 cm->cmsg_len = CMSG_LEN(nfds * sizeof(int)); 1333 rights->m_len = CMSG_SPACE(nfds * sizeof(int)); 1334 } 1335 out: 1336 if (__predict_false(error != 0)) { 1337 file_t **const fpp = (file_t **)CMSG_DATA(cm); 1338 for (size_t i = 0; i < nfds; i++) 1339 unp_discard_now(fpp[i]); 1340 /* 1341 * Truncate the array so that nobody will try to interpret 1342 * what is now garbage in it. 1343 */ 1344 cm->cmsg_len = CMSG_LEN(0); 1345 rights->m_len = CMSG_SPACE(0); 1346 } 1347 rw_exit(&p->p_cwdi->cwdi_lock); 1348 kmem_free(fdp, nfds * sizeof(int)); 1349 1350 noop: 1351 /* 1352 * Don't disclose kernel memory in the alignment space. 1353 */ 1354 KASSERT(cm->cmsg_len <= rights->m_len); 1355 memset(&mtod(rights, char *)[cm->cmsg_len], 0, rights->m_len - 1356 cm->cmsg_len); 1357 return error; 1358 } 1359 1360 int 1361 unp_internalize(struct mbuf **controlp) 1362 { 1363 filedesc_t *fdescp = curlwp->l_fd; 1364 struct mbuf *control = *controlp; 1365 struct cmsghdr *newcm, *cm = mtod(control, struct cmsghdr *); 1366 file_t **rp, **files; 1367 file_t *fp; 1368 int i, fd, *fdp; 1369 int nfds, error; 1370 u_int maxmsg; 1371 1372 error = 0; 1373 newcm = NULL; 1374 1375 /* Sanity check the control message header. */ 1376 if (cm->cmsg_type != SCM_RIGHTS || cm->cmsg_level != SOL_SOCKET || 1377 cm->cmsg_len > control->m_len || 1378 cm->cmsg_len < CMSG_ALIGN(sizeof(*cm))) 1379 return (EINVAL); 1380 1381 /* 1382 * Verify that the file descriptors are valid, and acquire 1383 * a reference to each. 1384 */ 1385 nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) / sizeof(int); 1386 fdp = (int *)CMSG_DATA(cm); 1387 maxmsg = maxfiles / unp_rights_ratio; 1388 for (i = 0; i < nfds; i++) { 1389 fd = *fdp++; 1390 if (atomic_inc_uint_nv(&unp_rights) > maxmsg) { 1391 atomic_dec_uint(&unp_rights); 1392 nfds = i; 1393 error = EAGAIN; 1394 goto out; 1395 } 1396 if ((fp = fd_getfile(fd)) == NULL 1397 || fp->f_type == DTYPE_KQUEUE) { 1398 if (fp) 1399 fd_putfile(fd); 1400 atomic_dec_uint(&unp_rights); 1401 nfds = i; 1402 error = EBADF; 1403 goto out; 1404 } 1405 } 1406 1407 /* Allocate new space and copy header into it. */ 1408 newcm = malloc(CMSG_SPACE(nfds * sizeof(file_t *)), M_MBUF, M_WAITOK); 1409 if (newcm == NULL) { 1410 error = E2BIG; 1411 goto out; 1412 } 1413 memcpy(newcm, cm, sizeof(struct cmsghdr)); 1414 files = (file_t **)CMSG_DATA(newcm); 1415 1416 /* 1417 * Transform the file descriptors into file_t pointers, in 1418 * reverse order so that if pointers are bigger than ints, the 1419 * int won't get until we're done. No need to lock, as we have 1420 * already validated the descriptors with fd_getfile(). 1421 */ 1422 fdp = (int *)CMSG_DATA(cm) + nfds; 1423 rp = files + nfds; 1424 for (i = 0; i < nfds; i++) { 1425 fp = fdescp->fd_dt->dt_ff[*--fdp]->ff_file; 1426 KASSERT(fp != NULL); 1427 mutex_enter(&fp->f_lock); 1428 *--rp = fp; 1429 fp->f_count++; 1430 fp->f_msgcount++; 1431 mutex_exit(&fp->f_lock); 1432 } 1433 1434 out: 1435 /* Release descriptor references. */ 1436 fdp = (int *)CMSG_DATA(cm); 1437 for (i = 0; i < nfds; i++) { 1438 fd_putfile(*fdp++); 1439 if (error != 0) { 1440 atomic_dec_uint(&unp_rights); 1441 } 1442 } 1443 1444 if (error == 0) { 1445 if (control->m_flags & M_EXT) { 1446 m_freem(control); 1447 *controlp = control = m_get(M_WAIT, MT_CONTROL); 1448 } 1449 MEXTADD(control, newcm, CMSG_SPACE(nfds * sizeof(file_t *)), 1450 M_MBUF, NULL, NULL); 1451 cm = newcm; 1452 /* 1453 * Adjust message & mbuf to note amount of space 1454 * actually used. 1455 */ 1456 cm->cmsg_len = CMSG_LEN(nfds * sizeof(file_t *)); 1457 control->m_len = CMSG_SPACE(nfds * sizeof(file_t *)); 1458 } 1459 1460 return error; 1461 } 1462 1463 struct mbuf * 1464 unp_addsockcred(struct lwp *l, struct mbuf *control) 1465 { 1466 struct sockcred *sc; 1467 struct mbuf *m; 1468 void *p; 1469 1470 m = sbcreatecontrol1(&p, SOCKCREDSIZE(kauth_cred_ngroups(l->l_cred)), 1471 SCM_CREDS, SOL_SOCKET, M_WAITOK); 1472 if (m == NULL) 1473 return control; 1474 1475 sc = p; 1476 sc->sc_uid = kauth_cred_getuid(l->l_cred); 1477 sc->sc_euid = kauth_cred_geteuid(l->l_cred); 1478 sc->sc_gid = kauth_cred_getgid(l->l_cred); 1479 sc->sc_egid = kauth_cred_getegid(l->l_cred); 1480 sc->sc_ngroups = kauth_cred_ngroups(l->l_cred); 1481 1482 for (int i = 0; i < sc->sc_ngroups; i++) 1483 sc->sc_groups[i] = kauth_cred_group(l->l_cred, i); 1484 1485 return m_add(control, m); 1486 } 1487 1488 /* 1489 * Do a mark-sweep GC of files in the system, to free up any which are 1490 * caught in flight to an about-to-be-closed socket. Additionally, 1491 * process deferred file closures. 1492 */ 1493 static void 1494 unp_gc(file_t *dp) 1495 { 1496 extern struct domain unixdomain; 1497 file_t *fp, *np; 1498 struct socket *so, *so1; 1499 u_int i, old, new; 1500 bool didwork; 1501 1502 KASSERT(curlwp == unp_thread_lwp); 1503 KASSERT(mutex_owned(&filelist_lock)); 1504 1505 /* 1506 * First, process deferred file closures. 1507 */ 1508 while (!SLIST_EMPTY(&unp_thread_discard)) { 1509 fp = SLIST_FIRST(&unp_thread_discard); 1510 KASSERT(fp->f_unpcount > 0); 1511 KASSERT(fp->f_count > 0); 1512 KASSERT(fp->f_msgcount > 0); 1513 KASSERT(fp->f_count >= fp->f_unpcount); 1514 KASSERT(fp->f_count >= fp->f_msgcount); 1515 KASSERT(fp->f_msgcount >= fp->f_unpcount); 1516 SLIST_REMOVE_HEAD(&unp_thread_discard, f_unplist); 1517 i = fp->f_unpcount; 1518 fp->f_unpcount = 0; 1519 mutex_exit(&filelist_lock); 1520 for (; i != 0; i--) { 1521 unp_discard_now(fp); 1522 } 1523 mutex_enter(&filelist_lock); 1524 } 1525 1526 /* 1527 * Clear mark bits. Ensure that we don't consider new files 1528 * entering the file table during this loop (they will not have 1529 * FSCAN set). 1530 */ 1531 unp_defer = 0; 1532 LIST_FOREACH(fp, &filehead, f_list) { 1533 for (old = fp->f_flag;; old = new) { 1534 new = atomic_cas_uint(&fp->f_flag, old, 1535 (old | FSCAN) & ~(FMARK|FDEFER)); 1536 if (__predict_true(old == new)) { 1537 break; 1538 } 1539 } 1540 } 1541 1542 /* 1543 * Iterate over the set of sockets, marking ones believed (based on 1544 * refcount) to be referenced from a process, and marking for rescan 1545 * sockets which are queued on a socket. Recan continues descending 1546 * and searching for sockets referenced by sockets (FDEFER), until 1547 * there are no more socket->socket references to be discovered. 1548 */ 1549 do { 1550 didwork = false; 1551 for (fp = LIST_FIRST(&filehead); fp != NULL; fp = np) { 1552 KASSERT(mutex_owned(&filelist_lock)); 1553 np = LIST_NEXT(fp, f_list); 1554 mutex_enter(&fp->f_lock); 1555 if ((fp->f_flag & FDEFER) != 0) { 1556 atomic_and_uint(&fp->f_flag, ~FDEFER); 1557 unp_defer--; 1558 KASSERT(fp->f_count != 0); 1559 } else { 1560 if (fp->f_count == 0 || 1561 (fp->f_flag & FMARK) != 0 || 1562 fp->f_count == fp->f_msgcount || 1563 fp->f_unpcount != 0) { 1564 mutex_exit(&fp->f_lock); 1565 continue; 1566 } 1567 } 1568 atomic_or_uint(&fp->f_flag, FMARK); 1569 1570 if (fp->f_type != DTYPE_SOCKET || 1571 (so = fp->f_data) == NULL || 1572 so->so_proto->pr_domain != &unixdomain || 1573 (so->so_proto->pr_flags & PR_RIGHTS) == 0) { 1574 mutex_exit(&fp->f_lock); 1575 continue; 1576 } 1577 1578 /* Gain file ref, mark our position, and unlock. */ 1579 didwork = true; 1580 LIST_INSERT_AFTER(fp, dp, f_list); 1581 fp->f_count++; 1582 mutex_exit(&fp->f_lock); 1583 mutex_exit(&filelist_lock); 1584 1585 /* 1586 * Mark files referenced from sockets queued on the 1587 * accept queue as well. 1588 */ 1589 solock(so); 1590 unp_scan(so->so_rcv.sb_mb, unp_mark, 0); 1591 if ((so->so_options & SO_ACCEPTCONN) != 0) { 1592 TAILQ_FOREACH(so1, &so->so_q0, so_qe) { 1593 unp_scan(so1->so_rcv.sb_mb, unp_mark, 0); 1594 } 1595 TAILQ_FOREACH(so1, &so->so_q, so_qe) { 1596 unp_scan(so1->so_rcv.sb_mb, unp_mark, 0); 1597 } 1598 } 1599 sounlock(so); 1600 1601 /* Re-lock and restart from where we left off. */ 1602 closef(fp); 1603 mutex_enter(&filelist_lock); 1604 np = LIST_NEXT(dp, f_list); 1605 LIST_REMOVE(dp, f_list); 1606 } 1607 /* 1608 * Bail early if we did nothing in the loop above. Could 1609 * happen because of concurrent activity causing unp_defer 1610 * to get out of sync. 1611 */ 1612 } while (unp_defer != 0 && didwork); 1613 1614 /* 1615 * Sweep pass. 1616 * 1617 * We grab an extra reference to each of the files that are 1618 * not otherwise accessible and then free the rights that are 1619 * stored in messages on them. 1620 */ 1621 for (fp = LIST_FIRST(&filehead); fp != NULL; fp = np) { 1622 KASSERT(mutex_owned(&filelist_lock)); 1623 np = LIST_NEXT(fp, f_list); 1624 mutex_enter(&fp->f_lock); 1625 1626 /* 1627 * Ignore non-sockets. 1628 * Ignore dead sockets, or sockets with pending close. 1629 * Ignore sockets obviously referenced elsewhere. 1630 * Ignore sockets marked as referenced by our scan. 1631 * Ignore new sockets that did not exist during the scan. 1632 */ 1633 if (fp->f_type != DTYPE_SOCKET || 1634 fp->f_count == 0 || fp->f_unpcount != 0 || 1635 fp->f_count != fp->f_msgcount || 1636 (fp->f_flag & (FMARK | FSCAN)) != FSCAN) { 1637 mutex_exit(&fp->f_lock); 1638 continue; 1639 } 1640 1641 /* Gain file ref, mark our position, and unlock. */ 1642 LIST_INSERT_AFTER(fp, dp, f_list); 1643 fp->f_count++; 1644 mutex_exit(&fp->f_lock); 1645 mutex_exit(&filelist_lock); 1646 1647 /* 1648 * Flush all data from the socket's receive buffer. 1649 * This will cause files referenced only by the 1650 * socket to be queued for close. 1651 */ 1652 so = fp->f_data; 1653 solock(so); 1654 sorflush(so); 1655 sounlock(so); 1656 1657 /* Re-lock and restart from where we left off. */ 1658 closef(fp); 1659 mutex_enter(&filelist_lock); 1660 np = LIST_NEXT(dp, f_list); 1661 LIST_REMOVE(dp, f_list); 1662 } 1663 } 1664 1665 /* 1666 * Garbage collector thread. While SCM_RIGHTS messages are in transit, 1667 * wake once per second to garbage collect. Run continually while we 1668 * have deferred closes to process. 1669 */ 1670 static void 1671 unp_thread(void *cookie) 1672 { 1673 file_t *dp; 1674 1675 /* Allocate a dummy file for our scans. */ 1676 if ((dp = fgetdummy()) == NULL) { 1677 panic("unp_thread"); 1678 } 1679 1680 mutex_enter(&filelist_lock); 1681 for (;;) { 1682 KASSERT(mutex_owned(&filelist_lock)); 1683 if (SLIST_EMPTY(&unp_thread_discard)) { 1684 if (unp_rights != 0) { 1685 (void)cv_timedwait(&unp_thread_cv, 1686 &filelist_lock, hz); 1687 } else { 1688 cv_wait(&unp_thread_cv, &filelist_lock); 1689 } 1690 } 1691 unp_gc(dp); 1692 } 1693 /* NOTREACHED */ 1694 } 1695 1696 /* 1697 * Kick the garbage collector into action if there is something for 1698 * it to process. 1699 */ 1700 static void 1701 unp_thread_kick(void) 1702 { 1703 1704 if (!SLIST_EMPTY(&unp_thread_discard) || unp_rights != 0) { 1705 mutex_enter(&filelist_lock); 1706 cv_signal(&unp_thread_cv); 1707 mutex_exit(&filelist_lock); 1708 } 1709 } 1710 1711 void 1712 unp_dispose(struct mbuf *m) 1713 { 1714 1715 if (m) 1716 unp_scan(m, unp_discard_later, 1); 1717 } 1718 1719 void 1720 unp_scan(struct mbuf *m0, void (*op)(file_t *), int discard) 1721 { 1722 struct mbuf *m; 1723 file_t **rp, *fp; 1724 struct cmsghdr *cm; 1725 int i, qfds; 1726 1727 while (m0) { 1728 for (m = m0; m; m = m->m_next) { 1729 if (m->m_type != MT_CONTROL || 1730 m->m_len < sizeof(*cm)) { 1731 continue; 1732 } 1733 cm = mtod(m, struct cmsghdr *); 1734 if (cm->cmsg_level != SOL_SOCKET || 1735 cm->cmsg_type != SCM_RIGHTS) 1736 continue; 1737 qfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) 1738 / sizeof(file_t *); 1739 rp = (file_t **)CMSG_DATA(cm); 1740 for (i = 0; i < qfds; i++) { 1741 fp = *rp; 1742 if (discard) { 1743 *rp = 0; 1744 } 1745 (*op)(fp); 1746 rp++; 1747 } 1748 } 1749 m0 = m0->m_nextpkt; 1750 } 1751 } 1752 1753 void 1754 unp_mark(file_t *fp) 1755 { 1756 1757 if (fp == NULL) 1758 return; 1759 1760 /* If we're already deferred, don't screw up the defer count */ 1761 mutex_enter(&fp->f_lock); 1762 if (fp->f_flag & (FMARK | FDEFER)) { 1763 mutex_exit(&fp->f_lock); 1764 return; 1765 } 1766 1767 /* 1768 * Minimize the number of deferrals... Sockets are the only type of 1769 * file which can hold references to another file, so just mark 1770 * other files, and defer unmarked sockets for the next pass. 1771 */ 1772 if (fp->f_type == DTYPE_SOCKET) { 1773 unp_defer++; 1774 KASSERT(fp->f_count != 0); 1775 atomic_or_uint(&fp->f_flag, FDEFER); 1776 } else { 1777 atomic_or_uint(&fp->f_flag, FMARK); 1778 } 1779 mutex_exit(&fp->f_lock); 1780 } 1781 1782 static void 1783 unp_discard_now(file_t *fp) 1784 { 1785 1786 if (fp == NULL) 1787 return; 1788 1789 KASSERT(fp->f_count > 0); 1790 KASSERT(fp->f_msgcount > 0); 1791 1792 mutex_enter(&fp->f_lock); 1793 fp->f_msgcount--; 1794 mutex_exit(&fp->f_lock); 1795 atomic_dec_uint(&unp_rights); 1796 (void)closef(fp); 1797 } 1798 1799 static void 1800 unp_discard_later(file_t *fp) 1801 { 1802 1803 if (fp == NULL) 1804 return; 1805 1806 KASSERT(fp->f_count > 0); 1807 KASSERT(fp->f_msgcount > 0); 1808 1809 mutex_enter(&filelist_lock); 1810 if (fp->f_unpcount++ == 0) { 1811 SLIST_INSERT_HEAD(&unp_thread_discard, fp, f_unplist); 1812 } 1813 mutex_exit(&filelist_lock); 1814 } 1815 1816 const struct pr_usrreqs unp_usrreqs = { 1817 .pr_attach = unp_attach, 1818 .pr_detach = unp_detach, 1819 .pr_ioctl = unp_ioctl, 1820 .pr_generic = unp_usrreq, 1821 }; 1822