1 /* $NetBSD: uipc_usrreq.c,v 1.150 2014/01/23 10:13:56 hannken 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 1.150 2014/01/23 10:13:56 hannken 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 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 u_int newhiwat; 373 int error = 0; 374 375 if (req == PRU_CONTROL) 376 return (EOPNOTSUPP); 377 378 #ifdef DIAGNOSTIC 379 if (req != PRU_SEND && req != PRU_SENDOOB && control) 380 panic("uipc_usrreq: unexpected control mbuf"); 381 #endif 382 if (req != PRU_ATTACH) { 383 if (unp == NULL) { 384 error = EINVAL; 385 goto release; 386 } 387 KASSERT(solocked(so)); 388 } 389 390 switch (req) { 391 392 case PRU_ATTACH: 393 if (unp != NULL) { 394 error = EISCONN; 395 break; 396 } 397 error = unp_attach(so); 398 break; 399 400 case PRU_DETACH: 401 unp_detach(unp); 402 break; 403 404 case PRU_BIND: 405 KASSERT(l != NULL); 406 error = unp_bind(so, nam, l); 407 break; 408 409 case PRU_LISTEN: 410 /* 411 * If the socket can accept a connection, it must be 412 * locked by uipc_lock. 413 */ 414 unp_resetlock(so); 415 if (unp->unp_vnode == NULL) 416 error = EINVAL; 417 break; 418 419 case PRU_CONNECT: 420 KASSERT(l != NULL); 421 error = unp_connect(so, nam, l); 422 break; 423 424 case PRU_CONNECT2: 425 error = unp_connect2(so, (struct socket *)nam, PRU_CONNECT2); 426 break; 427 428 case PRU_DISCONNECT: 429 unp_disconnect(unp); 430 break; 431 432 case PRU_ACCEPT: 433 KASSERT(so->so_lock == uipc_lock); 434 /* 435 * Mark the initiating STREAM socket as connected *ONLY* 436 * after it's been accepted. This prevents a client from 437 * overrunning a server and receiving ECONNREFUSED. 438 */ 439 if (unp->unp_conn == NULL) { 440 /* 441 * This will use the empty socket and will not 442 * allocate. 443 */ 444 unp_setaddr(so, nam, true); 445 break; 446 } 447 so2 = unp->unp_conn->unp_socket; 448 if (so2->so_state & SS_ISCONNECTING) { 449 KASSERT(solocked2(so, so->so_head)); 450 KASSERT(solocked2(so2, so->so_head)); 451 soisconnected(so2); 452 } 453 /* 454 * If the connection is fully established, break the 455 * association with uipc_lock and give the connected 456 * pair a seperate lock to share. 457 * There is a race here: sotounpcb(so2)->unp_streamlock 458 * is not locked, so when changing so2->so_lock 459 * another thread can grab it while so->so_lock is still 460 * pointing to the (locked) uipc_lock. 461 * this should be harmless, except that this makes 462 * solocked2() and solocked() unreliable. 463 * Another problem is that unp_setaddr() expects the 464 * the socket locked. Grabing sotounpcb(so2)->unp_streamlock 465 * fixes both issues. 466 */ 467 mutex_enter(sotounpcb(so2)->unp_streamlock); 468 unp_setpeerlocks(so2, so); 469 /* 470 * Only now return peer's address, as we may need to 471 * block in order to allocate memory. 472 * 473 * XXX Minor race: connection can be broken while 474 * lock is dropped in unp_setaddr(). We will return 475 * error == 0 and sun_noname as the peer address. 476 */ 477 unp_setaddr(so, nam, true); 478 /* so_lock now points to unp_streamlock */ 479 mutex_exit(so2->so_lock); 480 break; 481 482 case PRU_SHUTDOWN: 483 socantsendmore(so); 484 unp_shutdown(unp); 485 break; 486 487 case PRU_RCVD: 488 switch (so->so_type) { 489 490 case SOCK_DGRAM: 491 panic("uipc 1"); 492 /*NOTREACHED*/ 493 494 case SOCK_SEQPACKET: /* FALLTHROUGH */ 495 case SOCK_STREAM: 496 #define rcv (&so->so_rcv) 497 #define snd (&so2->so_snd) 498 if (unp->unp_conn == 0) 499 break; 500 so2 = unp->unp_conn->unp_socket; 501 KASSERT(solocked2(so, so2)); 502 /* 503 * Adjust backpressure on sender 504 * and wakeup any waiting to write. 505 */ 506 snd->sb_mbmax += unp->unp_mbcnt - rcv->sb_mbcnt; 507 unp->unp_mbcnt = rcv->sb_mbcnt; 508 newhiwat = snd->sb_hiwat + unp->unp_cc - rcv->sb_cc; 509 (void)chgsbsize(so2->so_uidinfo, 510 &snd->sb_hiwat, newhiwat, RLIM_INFINITY); 511 unp->unp_cc = rcv->sb_cc; 512 sowwakeup(so2); 513 #undef snd 514 #undef rcv 515 break; 516 517 default: 518 panic("uipc 2"); 519 } 520 break; 521 522 case PRU_SEND: 523 /* 524 * Note: unp_internalize() rejects any control message 525 * other than SCM_RIGHTS, and only allows one. This 526 * has the side-effect of preventing a caller from 527 * forging SCM_CREDS. 528 */ 529 if (control) { 530 sounlock(so); 531 error = unp_internalize(&control); 532 solock(so); 533 if (error != 0) { 534 m_freem(control); 535 m_freem(m); 536 break; 537 } 538 } 539 switch (so->so_type) { 540 541 case SOCK_DGRAM: { 542 KASSERT(so->so_lock == uipc_lock); 543 if (nam) { 544 if ((so->so_state & SS_ISCONNECTED) != 0) 545 error = EISCONN; 546 else { 547 /* 548 * Note: once connected, the 549 * socket's lock must not be 550 * dropped until we have sent 551 * the message and disconnected. 552 * This is necessary to prevent 553 * intervening control ops, like 554 * another connection. 555 */ 556 error = unp_connect(so, nam, l); 557 } 558 } else { 559 if ((so->so_state & SS_ISCONNECTED) == 0) 560 error = ENOTCONN; 561 } 562 if (error) { 563 unp_dispose(control); 564 m_freem(control); 565 m_freem(m); 566 break; 567 } 568 KASSERT(l != NULL); 569 error = unp_output(m, control, unp, l); 570 if (nam) 571 unp_disconnect(unp); 572 break; 573 } 574 575 case SOCK_SEQPACKET: /* FALLTHROUGH */ 576 case SOCK_STREAM: 577 #define rcv (&so2->so_rcv) 578 #define snd (&so->so_snd) 579 if (unp->unp_conn == NULL) { 580 error = ENOTCONN; 581 break; 582 } 583 so2 = unp->unp_conn->unp_socket; 584 KASSERT(solocked2(so, so2)); 585 if (unp->unp_conn->unp_flags & UNP_WANTCRED) { 586 /* 587 * Credentials are passed only once on 588 * SOCK_STREAM and SOCK_SEQPACKET. 589 */ 590 unp->unp_conn->unp_flags &= ~UNP_WANTCRED; 591 control = unp_addsockcred(l, control); 592 } 593 /* 594 * Send to paired receive port, and then reduce 595 * send buffer hiwater marks to maintain backpressure. 596 * Wake up readers. 597 */ 598 if (control) { 599 if (sbappendcontrol(rcv, m, control) != 0) 600 control = NULL; 601 } else { 602 switch(so->so_type) { 603 case SOCK_SEQPACKET: 604 sbappendrecord(rcv, m); 605 break; 606 case SOCK_STREAM: 607 sbappend(rcv, m); 608 break; 609 default: 610 panic("uipc_usrreq"); 611 break; 612 } 613 } 614 snd->sb_mbmax -= 615 rcv->sb_mbcnt - unp->unp_conn->unp_mbcnt; 616 unp->unp_conn->unp_mbcnt = rcv->sb_mbcnt; 617 newhiwat = snd->sb_hiwat - 618 (rcv->sb_cc - unp->unp_conn->unp_cc); 619 (void)chgsbsize(so->so_uidinfo, 620 &snd->sb_hiwat, newhiwat, RLIM_INFINITY); 621 unp->unp_conn->unp_cc = rcv->sb_cc; 622 sorwakeup(so2); 623 #undef snd 624 #undef rcv 625 if (control != NULL) { 626 unp_dispose(control); 627 m_freem(control); 628 } 629 break; 630 631 default: 632 panic("uipc 4"); 633 } 634 break; 635 636 case PRU_ABORT: 637 (void)unp_drop(unp, ECONNABORTED); 638 639 KASSERT(so->so_head == NULL); 640 #ifdef DIAGNOSTIC 641 if (so->so_pcb == NULL) 642 panic("uipc 5: drop killed pcb"); 643 #endif 644 unp_detach(unp); 645 break; 646 647 case PRU_SENSE: 648 ((struct stat *) m)->st_blksize = so->so_snd.sb_hiwat; 649 switch (so->so_type) { 650 case SOCK_SEQPACKET: /* FALLTHROUGH */ 651 case SOCK_STREAM: 652 if (unp->unp_conn == 0) 653 break; 654 655 so2 = unp->unp_conn->unp_socket; 656 KASSERT(solocked2(so, so2)); 657 ((struct stat *) m)->st_blksize += so2->so_rcv.sb_cc; 658 break; 659 default: 660 break; 661 } 662 ((struct stat *) m)->st_dev = NODEV; 663 if (unp->unp_ino == 0) 664 unp->unp_ino = unp_ino++; 665 ((struct stat *) m)->st_atimespec = 666 ((struct stat *) m)->st_mtimespec = 667 ((struct stat *) m)->st_ctimespec = unp->unp_ctime; 668 ((struct stat *) m)->st_ino = unp->unp_ino; 669 return (0); 670 671 case PRU_RCVOOB: 672 error = EOPNOTSUPP; 673 break; 674 675 case PRU_SENDOOB: 676 m_freem(control); 677 m_freem(m); 678 error = EOPNOTSUPP; 679 break; 680 681 case PRU_SOCKADDR: 682 unp_setaddr(so, nam, false); 683 break; 684 685 case PRU_PEERADDR: 686 unp_setaddr(so, nam, true); 687 break; 688 689 default: 690 panic("piusrreq"); 691 } 692 693 release: 694 return (error); 695 } 696 697 /* 698 * Unix domain socket option processing. 699 */ 700 int 701 uipc_ctloutput(int op, struct socket *so, struct sockopt *sopt) 702 { 703 struct unpcb *unp = sotounpcb(so); 704 int optval = 0, error = 0; 705 706 KASSERT(solocked(so)); 707 708 if (sopt->sopt_level != 0) { 709 error = ENOPROTOOPT; 710 } else switch (op) { 711 712 case PRCO_SETOPT: 713 switch (sopt->sopt_name) { 714 case LOCAL_CREDS: 715 case LOCAL_CONNWAIT: 716 error = sockopt_getint(sopt, &optval); 717 if (error) 718 break; 719 switch (sopt->sopt_name) { 720 #define OPTSET(bit) \ 721 if (optval) \ 722 unp->unp_flags |= (bit); \ 723 else \ 724 unp->unp_flags &= ~(bit); 725 726 case LOCAL_CREDS: 727 OPTSET(UNP_WANTCRED); 728 break; 729 case LOCAL_CONNWAIT: 730 OPTSET(UNP_CONNWAIT); 731 break; 732 } 733 break; 734 #undef OPTSET 735 736 default: 737 error = ENOPROTOOPT; 738 break; 739 } 740 break; 741 742 case PRCO_GETOPT: 743 sounlock(so); 744 switch (sopt->sopt_name) { 745 case LOCAL_PEEREID: 746 if (unp->unp_flags & UNP_EIDSVALID) { 747 error = sockopt_set(sopt, 748 &unp->unp_connid, sizeof(unp->unp_connid)); 749 } else { 750 error = EINVAL; 751 } 752 break; 753 case LOCAL_CREDS: 754 #define OPTBIT(bit) (unp->unp_flags & (bit) ? 1 : 0) 755 756 optval = OPTBIT(UNP_WANTCRED); 757 error = sockopt_setint(sopt, optval); 758 break; 759 #undef OPTBIT 760 761 default: 762 error = ENOPROTOOPT; 763 break; 764 } 765 solock(so); 766 break; 767 } 768 return (error); 769 } 770 771 /* 772 * Both send and receive buffers are allocated PIPSIZ bytes of buffering 773 * for stream sockets, although the total for sender and receiver is 774 * actually only PIPSIZ. 775 * Datagram sockets really use the sendspace as the maximum datagram size, 776 * and don't really want to reserve the sendspace. Their recvspace should 777 * be large enough for at least one max-size datagram plus address. 778 */ 779 #define PIPSIZ 4096 780 u_long unpst_sendspace = PIPSIZ; 781 u_long unpst_recvspace = PIPSIZ; 782 u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 783 u_long unpdg_recvspace = 4*1024; 784 785 u_int unp_rights; /* files in flight */ 786 u_int unp_rights_ratio = 2; /* limit, fraction of maxfiles */ 787 788 int 789 unp_attach(struct socket *so) 790 { 791 struct unpcb *unp; 792 int error; 793 794 switch (so->so_type) { 795 case SOCK_SEQPACKET: /* FALLTHROUGH */ 796 case SOCK_STREAM: 797 if (so->so_lock == NULL) { 798 /* 799 * XXX Assuming that no socket locks are held, 800 * as this call may sleep. 801 */ 802 so->so_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 803 solock(so); 804 } 805 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 806 error = soreserve(so, unpst_sendspace, unpst_recvspace); 807 if (error != 0) 808 return (error); 809 } 810 break; 811 812 case SOCK_DGRAM: 813 if (so->so_lock == NULL) { 814 mutex_obj_hold(uipc_lock); 815 so->so_lock = uipc_lock; 816 solock(so); 817 } 818 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 819 error = soreserve(so, unpdg_sendspace, unpdg_recvspace); 820 if (error != 0) 821 return (error); 822 } 823 break; 824 825 default: 826 panic("unp_attach"); 827 } 828 KASSERT(solocked(so)); 829 unp = malloc(sizeof(*unp), M_PCB, M_NOWAIT); 830 if (unp == NULL) 831 return (ENOBUFS); 832 memset(unp, 0, sizeof(*unp)); 833 unp->unp_socket = so; 834 so->so_pcb = unp; 835 nanotime(&unp->unp_ctime); 836 return (0); 837 } 838 839 void 840 unp_detach(struct unpcb *unp) 841 { 842 struct socket *so; 843 vnode_t *vp; 844 845 so = unp->unp_socket; 846 847 retry: 848 if ((vp = unp->unp_vnode) != NULL) { 849 sounlock(so); 850 /* Acquire v_interlock to protect against unp_connect(). */ 851 /* XXXAD racy */ 852 mutex_enter(vp->v_interlock); 853 vp->v_socket = NULL; 854 mutex_exit(vp->v_interlock); 855 vrele(vp); 856 solock(so); 857 unp->unp_vnode = NULL; 858 } 859 if (unp->unp_conn) 860 unp_disconnect(unp); 861 while (unp->unp_refs) { 862 KASSERT(solocked2(so, unp->unp_refs->unp_socket)); 863 if (unp_drop(unp->unp_refs, ECONNRESET)) { 864 solock(so); 865 goto retry; 866 } 867 } 868 soisdisconnected(so); 869 so->so_pcb = NULL; 870 if (unp_rights) { 871 /* 872 * Normally the receive buffer is flushed later, in sofree, 873 * but if our receive buffer holds references to files that 874 * are now garbage, we will enqueue those file references to 875 * the garbage collector and kick it into action. 876 */ 877 sorflush(so); 878 unp_free(unp); 879 unp_thread_kick(); 880 } else 881 unp_free(unp); 882 } 883 884 /* 885 * Allocate the new sockaddr. We have to allocate one 886 * extra byte so that we can ensure that the pathname 887 * is nul-terminated. Note that unlike linux, we don't 888 * include in the address length the NUL in the path 889 * component, because doing so, would exceed sizeof(sockaddr_un) 890 * for fully occupied pathnames. Linux is also inconsistent, 891 * because it does not include the NUL in the length of 892 * what it calls "abstract" unix sockets. 893 */ 894 static struct sockaddr_un * 895 makeun(struct mbuf *nam, size_t *addrlen) { 896 struct sockaddr_un *sun; 897 898 *addrlen = nam->m_len + 1; 899 sun = malloc(*addrlen, M_SONAME, M_WAITOK); 900 m_copydata(nam, 0, nam->m_len, (void *)sun); 901 *(((char *)sun) + nam->m_len) = '\0'; 902 sun->sun_len = strlen(sun->sun_path) + 903 offsetof(struct sockaddr_un, sun_path); 904 return sun; 905 } 906 907 int 908 unp_bind(struct socket *so, struct mbuf *nam, struct lwp *l) 909 { 910 struct sockaddr_un *sun; 911 struct unpcb *unp; 912 vnode_t *vp; 913 struct vattr vattr; 914 size_t addrlen; 915 int error; 916 struct pathbuf *pb; 917 struct nameidata nd; 918 proc_t *p; 919 920 unp = sotounpcb(so); 921 if (unp->unp_vnode != NULL) 922 return (EINVAL); 923 if ((unp->unp_flags & UNP_BUSY) != 0) { 924 /* 925 * EALREADY may not be strictly accurate, but since this 926 * is a major application error it's hardly a big deal. 927 */ 928 return (EALREADY); 929 } 930 unp->unp_flags |= UNP_BUSY; 931 sounlock(so); 932 933 p = l->l_proc; 934 sun = makeun(nam, &addrlen); 935 936 pb = pathbuf_create(sun->sun_path); 937 if (pb == NULL) { 938 error = ENOMEM; 939 goto bad; 940 } 941 NDINIT(&nd, CREATE, FOLLOW | LOCKPARENT | TRYEMULROOT, pb); 942 943 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 944 if ((error = namei(&nd)) != 0) { 945 pathbuf_destroy(pb); 946 goto bad; 947 } 948 vp = nd.ni_vp; 949 if (vp != NULL) { 950 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 951 if (nd.ni_dvp == vp) 952 vrele(nd.ni_dvp); 953 else 954 vput(nd.ni_dvp); 955 vrele(vp); 956 pathbuf_destroy(pb); 957 error = EADDRINUSE; 958 goto bad; 959 } 960 vattr_null(&vattr); 961 vattr.va_type = VSOCK; 962 vattr.va_mode = ACCESSPERMS & ~(p->p_cwdi->cwdi_cmask); 963 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 964 if (error) { 965 vput(nd.ni_dvp); 966 pathbuf_destroy(pb); 967 goto bad; 968 } 969 vp = nd.ni_vp; 970 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 971 solock(so); 972 vp->v_socket = unp->unp_socket; 973 unp->unp_vnode = vp; 974 unp->unp_addrlen = addrlen; 975 unp->unp_addr = sun; 976 unp->unp_connid.unp_pid = p->p_pid; 977 unp->unp_connid.unp_euid = kauth_cred_geteuid(l->l_cred); 978 unp->unp_connid.unp_egid = kauth_cred_getegid(l->l_cred); 979 unp->unp_flags |= UNP_EIDSBIND; 980 VOP_UNLOCK(vp); 981 vput(nd.ni_dvp); 982 unp->unp_flags &= ~UNP_BUSY; 983 pathbuf_destroy(pb); 984 return (0); 985 986 bad: 987 free(sun, M_SONAME); 988 solock(so); 989 unp->unp_flags &= ~UNP_BUSY; 990 return (error); 991 } 992 993 int 994 unp_connect(struct socket *so, struct mbuf *nam, struct lwp *l) 995 { 996 struct sockaddr_un *sun; 997 vnode_t *vp; 998 struct socket *so2, *so3; 999 struct unpcb *unp, *unp2, *unp3; 1000 size_t addrlen; 1001 int error; 1002 struct pathbuf *pb; 1003 struct nameidata nd; 1004 1005 unp = sotounpcb(so); 1006 if ((unp->unp_flags & UNP_BUSY) != 0) { 1007 /* 1008 * EALREADY may not be strictly accurate, but since this 1009 * is a major application error it's hardly a big deal. 1010 */ 1011 return (EALREADY); 1012 } 1013 unp->unp_flags |= UNP_BUSY; 1014 sounlock(so); 1015 1016 sun = makeun(nam, &addrlen); 1017 pb = pathbuf_create(sun->sun_path); 1018 if (pb == NULL) { 1019 error = ENOMEM; 1020 goto bad2; 1021 } 1022 1023 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb); 1024 1025 if ((error = namei(&nd)) != 0) { 1026 pathbuf_destroy(pb); 1027 goto bad2; 1028 } 1029 vp = nd.ni_vp; 1030 if (vp->v_type != VSOCK) { 1031 error = ENOTSOCK; 1032 goto bad; 1033 } 1034 pathbuf_destroy(pb); 1035 if ((error = VOP_ACCESS(vp, VWRITE, l->l_cred)) != 0) 1036 goto bad; 1037 /* Acquire v_interlock to protect against unp_detach(). */ 1038 mutex_enter(vp->v_interlock); 1039 so2 = vp->v_socket; 1040 if (so2 == NULL) { 1041 mutex_exit(vp->v_interlock); 1042 error = ECONNREFUSED; 1043 goto bad; 1044 } 1045 if (so->so_type != so2->so_type) { 1046 mutex_exit(vp->v_interlock); 1047 error = EPROTOTYPE; 1048 goto bad; 1049 } 1050 solock(so); 1051 unp_resetlock(so); 1052 mutex_exit(vp->v_interlock); 1053 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) { 1054 /* 1055 * This may seem somewhat fragile but is OK: if we can 1056 * see SO_ACCEPTCONN set on the endpoint, then it must 1057 * be locked by the domain-wide uipc_lock. 1058 */ 1059 KASSERT((so2->so_options & SO_ACCEPTCONN) == 0 || 1060 so2->so_lock == uipc_lock); 1061 if ((so2->so_options & SO_ACCEPTCONN) == 0 || 1062 (so3 = sonewconn(so2, false)) == NULL) { 1063 error = ECONNREFUSED; 1064 sounlock(so); 1065 goto bad; 1066 } 1067 unp2 = sotounpcb(so2); 1068 unp3 = sotounpcb(so3); 1069 if (unp2->unp_addr) { 1070 unp3->unp_addr = malloc(unp2->unp_addrlen, 1071 M_SONAME, M_WAITOK); 1072 memcpy(unp3->unp_addr, unp2->unp_addr, 1073 unp2->unp_addrlen); 1074 unp3->unp_addrlen = unp2->unp_addrlen; 1075 } 1076 unp3->unp_flags = unp2->unp_flags; 1077 unp3->unp_connid.unp_pid = l->l_proc->p_pid; 1078 unp3->unp_connid.unp_euid = kauth_cred_geteuid(l->l_cred); 1079 unp3->unp_connid.unp_egid = kauth_cred_getegid(l->l_cred); 1080 unp3->unp_flags |= UNP_EIDSVALID; 1081 if (unp2->unp_flags & UNP_EIDSBIND) { 1082 unp->unp_connid = unp2->unp_connid; 1083 unp->unp_flags |= UNP_EIDSVALID; 1084 } 1085 so2 = so3; 1086 } 1087 error = unp_connect2(so, so2, PRU_CONNECT); 1088 sounlock(so); 1089 bad: 1090 vput(vp); 1091 bad2: 1092 free(sun, M_SONAME); 1093 solock(so); 1094 unp->unp_flags &= ~UNP_BUSY; 1095 return (error); 1096 } 1097 1098 int 1099 unp_connect2(struct socket *so, struct socket *so2, int req) 1100 { 1101 struct unpcb *unp = sotounpcb(so); 1102 struct unpcb *unp2; 1103 1104 if (so2->so_type != so->so_type) 1105 return (EPROTOTYPE); 1106 1107 /* 1108 * All three sockets involved must be locked by same lock: 1109 * 1110 * local endpoint (so) 1111 * remote endpoint (so2) 1112 * queue head (so2->so_head, only if PR_CONNREQUIRED) 1113 */ 1114 KASSERT(solocked2(so, so2)); 1115 KASSERT(so->so_head == NULL); 1116 if (so2->so_head != NULL) { 1117 KASSERT(so2->so_lock == uipc_lock); 1118 KASSERT(solocked2(so2, so2->so_head)); 1119 } 1120 1121 unp2 = sotounpcb(so2); 1122 unp->unp_conn = unp2; 1123 switch (so->so_type) { 1124 1125 case SOCK_DGRAM: 1126 unp->unp_nextref = unp2->unp_refs; 1127 unp2->unp_refs = unp; 1128 soisconnected(so); 1129 break; 1130 1131 case SOCK_SEQPACKET: /* FALLTHROUGH */ 1132 case SOCK_STREAM: 1133 unp2->unp_conn = unp; 1134 if (req == PRU_CONNECT && 1135 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 1136 soisconnecting(so); 1137 else 1138 soisconnected(so); 1139 soisconnected(so2); 1140 /* 1141 * If the connection is fully established, break the 1142 * association with uipc_lock and give the connected 1143 * pair a seperate lock to share. For CONNECT2, we 1144 * require that the locks already match (the sockets 1145 * are created that way). 1146 */ 1147 if (req == PRU_CONNECT) { 1148 KASSERT(so2->so_head != NULL); 1149 unp_setpeerlocks(so, so2); 1150 } 1151 break; 1152 1153 default: 1154 panic("unp_connect2"); 1155 } 1156 return (0); 1157 } 1158 1159 void 1160 unp_disconnect(struct unpcb *unp) 1161 { 1162 struct unpcb *unp2 = unp->unp_conn; 1163 struct socket *so; 1164 1165 if (unp2 == 0) 1166 return; 1167 unp->unp_conn = 0; 1168 so = unp->unp_socket; 1169 switch (so->so_type) { 1170 case SOCK_DGRAM: 1171 if (unp2->unp_refs == unp) 1172 unp2->unp_refs = unp->unp_nextref; 1173 else { 1174 unp2 = unp2->unp_refs; 1175 for (;;) { 1176 KASSERT(solocked2(so, unp2->unp_socket)); 1177 if (unp2 == 0) 1178 panic("unp_disconnect"); 1179 if (unp2->unp_nextref == unp) 1180 break; 1181 unp2 = unp2->unp_nextref; 1182 } 1183 unp2->unp_nextref = unp->unp_nextref; 1184 } 1185 unp->unp_nextref = 0; 1186 so->so_state &= ~SS_ISCONNECTED; 1187 break; 1188 1189 case SOCK_SEQPACKET: /* FALLTHROUGH */ 1190 case SOCK_STREAM: 1191 KASSERT(solocked2(so, unp2->unp_socket)); 1192 soisdisconnected(so); 1193 unp2->unp_conn = 0; 1194 soisdisconnected(unp2->unp_socket); 1195 break; 1196 } 1197 } 1198 1199 #ifdef notdef 1200 unp_abort(struct unpcb *unp) 1201 { 1202 unp_detach(unp); 1203 } 1204 #endif 1205 1206 void 1207 unp_shutdown(struct unpcb *unp) 1208 { 1209 struct socket *so; 1210 1211 switch(unp->unp_socket->so_type) { 1212 case SOCK_SEQPACKET: /* FALLTHROUGH */ 1213 case SOCK_STREAM: 1214 if (unp->unp_conn && (so = unp->unp_conn->unp_socket)) 1215 socantrcvmore(so); 1216 break; 1217 default: 1218 break; 1219 } 1220 } 1221 1222 bool 1223 unp_drop(struct unpcb *unp, int errno) 1224 { 1225 struct socket *so = unp->unp_socket; 1226 1227 KASSERT(solocked(so)); 1228 1229 so->so_error = errno; 1230 unp_disconnect(unp); 1231 if (so->so_head) { 1232 so->so_pcb = NULL; 1233 /* sofree() drops the socket lock */ 1234 sofree(so); 1235 unp_free(unp); 1236 return true; 1237 } 1238 return false; 1239 } 1240 1241 #ifdef notdef 1242 unp_drain(void) 1243 { 1244 1245 } 1246 #endif 1247 1248 int 1249 unp_externalize(struct mbuf *rights, struct lwp *l, int flags) 1250 { 1251 struct cmsghdr * const cm = mtod(rights, struct cmsghdr *); 1252 struct proc * const p = l->l_proc; 1253 file_t **rp; 1254 int error = 0; 1255 1256 const size_t nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) / 1257 sizeof(file_t *); 1258 if (nfds == 0) 1259 goto noop; 1260 1261 int * const fdp = kmem_alloc(nfds * sizeof(int), KM_SLEEP); 1262 rw_enter(&p->p_cwdi->cwdi_lock, RW_READER); 1263 1264 /* Make sure the recipient should be able to see the files.. */ 1265 rp = (file_t **)CMSG_DATA(cm); 1266 for (size_t i = 0; i < nfds; i++) { 1267 file_t * const fp = *rp++; 1268 if (fp == NULL) { 1269 error = EINVAL; 1270 goto out; 1271 } 1272 /* 1273 * If we are in a chroot'ed directory, and 1274 * someone wants to pass us a directory, make 1275 * sure it's inside the subtree we're allowed 1276 * to access. 1277 */ 1278 if (p->p_cwdi->cwdi_rdir != NULL && fp->f_type == DTYPE_VNODE) { 1279 vnode_t *vp = (vnode_t *)fp->f_data; 1280 if ((vp->v_type == VDIR) && 1281 !vn_isunder(vp, p->p_cwdi->cwdi_rdir, l)) { 1282 error = EPERM; 1283 goto out; 1284 } 1285 } 1286 } 1287 1288 restart: 1289 /* 1290 * First loop -- allocate file descriptor table slots for the 1291 * new files. 1292 */ 1293 for (size_t i = 0; i < nfds; i++) { 1294 if ((error = fd_alloc(p, 0, &fdp[i])) != 0) { 1295 /* 1296 * Back out what we've done so far. 1297 */ 1298 while (i-- > 0) { 1299 fd_abort(p, NULL, fdp[i]); 1300 } 1301 if (error == ENOSPC) { 1302 fd_tryexpand(p); 1303 error = 0; 1304 goto restart; 1305 } 1306 /* 1307 * This is the error that has historically 1308 * been returned, and some callers may 1309 * expect it. 1310 */ 1311 error = EMSGSIZE; 1312 goto out; 1313 } 1314 } 1315 1316 /* 1317 * Now that adding them has succeeded, update all of the 1318 * file passing state and affix the descriptors. 1319 */ 1320 rp = (file_t **)CMSG_DATA(cm); 1321 int *ofdp = (int *)CMSG_DATA(cm); 1322 for (size_t i = 0; i < nfds; i++) { 1323 file_t * const fp = *rp++; 1324 const int fd = fdp[i]; 1325 atomic_dec_uint(&unp_rights); 1326 fd_set_exclose(l, fd, (flags & O_CLOEXEC) != 0); 1327 fd_affix(p, fp, fd); 1328 /* 1329 * Done with this file pointer, replace it with a fd; 1330 */ 1331 *ofdp++ = fd; 1332 mutex_enter(&fp->f_lock); 1333 fp->f_msgcount--; 1334 mutex_exit(&fp->f_lock); 1335 /* 1336 * Note that fd_affix() adds a reference to the file. 1337 * The file may already have been closed by another 1338 * LWP in the process, so we must drop the reference 1339 * added by unp_internalize() with closef(). 1340 */ 1341 closef(fp); 1342 } 1343 1344 /* 1345 * Adjust length, in case of transition from large file_t 1346 * pointers to ints. 1347 */ 1348 if (sizeof(file_t *) != sizeof(int)) { 1349 cm->cmsg_len = CMSG_LEN(nfds * sizeof(int)); 1350 rights->m_len = CMSG_SPACE(nfds * sizeof(int)); 1351 } 1352 out: 1353 if (__predict_false(error != 0)) { 1354 file_t **const fpp = (file_t **)CMSG_DATA(cm); 1355 for (size_t i = 0; i < nfds; i++) 1356 unp_discard_now(fpp[i]); 1357 /* 1358 * Truncate the array so that nobody will try to interpret 1359 * what is now garbage in it. 1360 */ 1361 cm->cmsg_len = CMSG_LEN(0); 1362 rights->m_len = CMSG_SPACE(0); 1363 } 1364 rw_exit(&p->p_cwdi->cwdi_lock); 1365 kmem_free(fdp, nfds * sizeof(int)); 1366 1367 noop: 1368 /* 1369 * Don't disclose kernel memory in the alignment space. 1370 */ 1371 KASSERT(cm->cmsg_len <= rights->m_len); 1372 memset(&mtod(rights, char *)[cm->cmsg_len], 0, rights->m_len - 1373 cm->cmsg_len); 1374 return error; 1375 } 1376 1377 int 1378 unp_internalize(struct mbuf **controlp) 1379 { 1380 filedesc_t *fdescp = curlwp->l_fd; 1381 struct mbuf *control = *controlp; 1382 struct cmsghdr *newcm, *cm = mtod(control, struct cmsghdr *); 1383 file_t **rp, **files; 1384 file_t *fp; 1385 int i, fd, *fdp; 1386 int nfds, error; 1387 u_int maxmsg; 1388 1389 error = 0; 1390 newcm = NULL; 1391 1392 /* Sanity check the control message header. */ 1393 if (cm->cmsg_type != SCM_RIGHTS || cm->cmsg_level != SOL_SOCKET || 1394 cm->cmsg_len > control->m_len || 1395 cm->cmsg_len < CMSG_ALIGN(sizeof(*cm))) 1396 return (EINVAL); 1397 1398 /* 1399 * Verify that the file descriptors are valid, and acquire 1400 * a reference to each. 1401 */ 1402 nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) / sizeof(int); 1403 fdp = (int *)CMSG_DATA(cm); 1404 maxmsg = maxfiles / unp_rights_ratio; 1405 for (i = 0; i < nfds; i++) { 1406 fd = *fdp++; 1407 if (atomic_inc_uint_nv(&unp_rights) > maxmsg) { 1408 atomic_dec_uint(&unp_rights); 1409 nfds = i; 1410 error = EAGAIN; 1411 goto out; 1412 } 1413 if ((fp = fd_getfile(fd)) == NULL 1414 || fp->f_type == DTYPE_KQUEUE) { 1415 if (fp) 1416 fd_putfile(fd); 1417 atomic_dec_uint(&unp_rights); 1418 nfds = i; 1419 error = EBADF; 1420 goto out; 1421 } 1422 } 1423 1424 /* Allocate new space and copy header into it. */ 1425 newcm = malloc(CMSG_SPACE(nfds * sizeof(file_t *)), M_MBUF, M_WAITOK); 1426 if (newcm == NULL) { 1427 error = E2BIG; 1428 goto out; 1429 } 1430 memcpy(newcm, cm, sizeof(struct cmsghdr)); 1431 files = (file_t **)CMSG_DATA(newcm); 1432 1433 /* 1434 * Transform the file descriptors into file_t pointers, in 1435 * reverse order so that if pointers are bigger than ints, the 1436 * int won't get until we're done. No need to lock, as we have 1437 * already validated the descriptors with fd_getfile(). 1438 */ 1439 fdp = (int *)CMSG_DATA(cm) + nfds; 1440 rp = files + nfds; 1441 for (i = 0; i < nfds; i++) { 1442 fp = fdescp->fd_dt->dt_ff[*--fdp]->ff_file; 1443 KASSERT(fp != NULL); 1444 mutex_enter(&fp->f_lock); 1445 *--rp = fp; 1446 fp->f_count++; 1447 fp->f_msgcount++; 1448 mutex_exit(&fp->f_lock); 1449 } 1450 1451 out: 1452 /* Release descriptor references. */ 1453 fdp = (int *)CMSG_DATA(cm); 1454 for (i = 0; i < nfds; i++) { 1455 fd_putfile(*fdp++); 1456 if (error != 0) { 1457 atomic_dec_uint(&unp_rights); 1458 } 1459 } 1460 1461 if (error == 0) { 1462 if (control->m_flags & M_EXT) { 1463 m_freem(control); 1464 *controlp = control = m_get(M_WAIT, MT_CONTROL); 1465 } 1466 MEXTADD(control, newcm, CMSG_SPACE(nfds * sizeof(file_t *)), 1467 M_MBUF, NULL, NULL); 1468 cm = newcm; 1469 /* 1470 * Adjust message & mbuf to note amount of space 1471 * actually used. 1472 */ 1473 cm->cmsg_len = CMSG_LEN(nfds * sizeof(file_t *)); 1474 control->m_len = CMSG_SPACE(nfds * sizeof(file_t *)); 1475 } 1476 1477 return error; 1478 } 1479 1480 struct mbuf * 1481 unp_addsockcred(struct lwp *l, struct mbuf *control) 1482 { 1483 struct sockcred *sc; 1484 struct mbuf *m; 1485 void *p; 1486 1487 m = sbcreatecontrol1(&p, SOCKCREDSIZE(kauth_cred_ngroups(l->l_cred)), 1488 SCM_CREDS, SOL_SOCKET, M_WAITOK); 1489 if (m == NULL) 1490 return control; 1491 1492 sc = p; 1493 sc->sc_uid = kauth_cred_getuid(l->l_cred); 1494 sc->sc_euid = kauth_cred_geteuid(l->l_cred); 1495 sc->sc_gid = kauth_cred_getgid(l->l_cred); 1496 sc->sc_egid = kauth_cred_getegid(l->l_cred); 1497 sc->sc_ngroups = kauth_cred_ngroups(l->l_cred); 1498 1499 for (int i = 0; i < sc->sc_ngroups; i++) 1500 sc->sc_groups[i] = kauth_cred_group(l->l_cred, i); 1501 1502 return m_add(control, m); 1503 } 1504 1505 /* 1506 * Do a mark-sweep GC of files in the system, to free up any which are 1507 * caught in flight to an about-to-be-closed socket. Additionally, 1508 * process deferred file closures. 1509 */ 1510 static void 1511 unp_gc(file_t *dp) 1512 { 1513 extern struct domain unixdomain; 1514 file_t *fp, *np; 1515 struct socket *so, *so1; 1516 u_int i, old, new; 1517 bool didwork; 1518 1519 KASSERT(curlwp == unp_thread_lwp); 1520 KASSERT(mutex_owned(&filelist_lock)); 1521 1522 /* 1523 * First, process deferred file closures. 1524 */ 1525 while (!SLIST_EMPTY(&unp_thread_discard)) { 1526 fp = SLIST_FIRST(&unp_thread_discard); 1527 KASSERT(fp->f_unpcount > 0); 1528 KASSERT(fp->f_count > 0); 1529 KASSERT(fp->f_msgcount > 0); 1530 KASSERT(fp->f_count >= fp->f_unpcount); 1531 KASSERT(fp->f_count >= fp->f_msgcount); 1532 KASSERT(fp->f_msgcount >= fp->f_unpcount); 1533 SLIST_REMOVE_HEAD(&unp_thread_discard, f_unplist); 1534 i = fp->f_unpcount; 1535 fp->f_unpcount = 0; 1536 mutex_exit(&filelist_lock); 1537 for (; i != 0; i--) { 1538 unp_discard_now(fp); 1539 } 1540 mutex_enter(&filelist_lock); 1541 } 1542 1543 /* 1544 * Clear mark bits. Ensure that we don't consider new files 1545 * entering the file table during this loop (they will not have 1546 * FSCAN set). 1547 */ 1548 unp_defer = 0; 1549 LIST_FOREACH(fp, &filehead, f_list) { 1550 for (old = fp->f_flag;; old = new) { 1551 new = atomic_cas_uint(&fp->f_flag, old, 1552 (old | FSCAN) & ~(FMARK|FDEFER)); 1553 if (__predict_true(old == new)) { 1554 break; 1555 } 1556 } 1557 } 1558 1559 /* 1560 * Iterate over the set of sockets, marking ones believed (based on 1561 * refcount) to be referenced from a process, and marking for rescan 1562 * sockets which are queued on a socket. Recan continues descending 1563 * and searching for sockets referenced by sockets (FDEFER), until 1564 * there are no more socket->socket references to be discovered. 1565 */ 1566 do { 1567 didwork = false; 1568 for (fp = LIST_FIRST(&filehead); fp != NULL; fp = np) { 1569 KASSERT(mutex_owned(&filelist_lock)); 1570 np = LIST_NEXT(fp, f_list); 1571 mutex_enter(&fp->f_lock); 1572 if ((fp->f_flag & FDEFER) != 0) { 1573 atomic_and_uint(&fp->f_flag, ~FDEFER); 1574 unp_defer--; 1575 KASSERT(fp->f_count != 0); 1576 } else { 1577 if (fp->f_count == 0 || 1578 (fp->f_flag & FMARK) != 0 || 1579 fp->f_count == fp->f_msgcount || 1580 fp->f_unpcount != 0) { 1581 mutex_exit(&fp->f_lock); 1582 continue; 1583 } 1584 } 1585 atomic_or_uint(&fp->f_flag, FMARK); 1586 1587 if (fp->f_type != DTYPE_SOCKET || 1588 (so = fp->f_data) == NULL || 1589 so->so_proto->pr_domain != &unixdomain || 1590 (so->so_proto->pr_flags & PR_RIGHTS) == 0) { 1591 mutex_exit(&fp->f_lock); 1592 continue; 1593 } 1594 1595 /* Gain file ref, mark our position, and unlock. */ 1596 didwork = true; 1597 LIST_INSERT_AFTER(fp, dp, f_list); 1598 fp->f_count++; 1599 mutex_exit(&fp->f_lock); 1600 mutex_exit(&filelist_lock); 1601 1602 /* 1603 * Mark files referenced from sockets queued on the 1604 * accept queue as well. 1605 */ 1606 solock(so); 1607 unp_scan(so->so_rcv.sb_mb, unp_mark, 0); 1608 if ((so->so_options & SO_ACCEPTCONN) != 0) { 1609 TAILQ_FOREACH(so1, &so->so_q0, so_qe) { 1610 unp_scan(so1->so_rcv.sb_mb, unp_mark, 0); 1611 } 1612 TAILQ_FOREACH(so1, &so->so_q, so_qe) { 1613 unp_scan(so1->so_rcv.sb_mb, unp_mark, 0); 1614 } 1615 } 1616 sounlock(so); 1617 1618 /* Re-lock and restart from where we left off. */ 1619 closef(fp); 1620 mutex_enter(&filelist_lock); 1621 np = LIST_NEXT(dp, f_list); 1622 LIST_REMOVE(dp, f_list); 1623 } 1624 /* 1625 * Bail early if we did nothing in the loop above. Could 1626 * happen because of concurrent activity causing unp_defer 1627 * to get out of sync. 1628 */ 1629 } while (unp_defer != 0 && didwork); 1630 1631 /* 1632 * Sweep pass. 1633 * 1634 * We grab an extra reference to each of the files that are 1635 * not otherwise accessible and then free the rights that are 1636 * stored in messages on them. 1637 */ 1638 for (fp = LIST_FIRST(&filehead); fp != NULL; fp = np) { 1639 KASSERT(mutex_owned(&filelist_lock)); 1640 np = LIST_NEXT(fp, f_list); 1641 mutex_enter(&fp->f_lock); 1642 1643 /* 1644 * Ignore non-sockets. 1645 * Ignore dead sockets, or sockets with pending close. 1646 * Ignore sockets obviously referenced elsewhere. 1647 * Ignore sockets marked as referenced by our scan. 1648 * Ignore new sockets that did not exist during the scan. 1649 */ 1650 if (fp->f_type != DTYPE_SOCKET || 1651 fp->f_count == 0 || fp->f_unpcount != 0 || 1652 fp->f_count != fp->f_msgcount || 1653 (fp->f_flag & (FMARK | FSCAN)) != FSCAN) { 1654 mutex_exit(&fp->f_lock); 1655 continue; 1656 } 1657 1658 /* Gain file ref, mark our position, and unlock. */ 1659 LIST_INSERT_AFTER(fp, dp, f_list); 1660 fp->f_count++; 1661 mutex_exit(&fp->f_lock); 1662 mutex_exit(&filelist_lock); 1663 1664 /* 1665 * Flush all data from the socket's receive buffer. 1666 * This will cause files referenced only by the 1667 * socket to be queued for close. 1668 */ 1669 so = fp->f_data; 1670 solock(so); 1671 sorflush(so); 1672 sounlock(so); 1673 1674 /* Re-lock and restart from where we left off. */ 1675 closef(fp); 1676 mutex_enter(&filelist_lock); 1677 np = LIST_NEXT(dp, f_list); 1678 LIST_REMOVE(dp, f_list); 1679 } 1680 } 1681 1682 /* 1683 * Garbage collector thread. While SCM_RIGHTS messages are in transit, 1684 * wake once per second to garbage collect. Run continually while we 1685 * have deferred closes to process. 1686 */ 1687 static void 1688 unp_thread(void *cookie) 1689 { 1690 file_t *dp; 1691 1692 /* Allocate a dummy file for our scans. */ 1693 if ((dp = fgetdummy()) == NULL) { 1694 panic("unp_thread"); 1695 } 1696 1697 mutex_enter(&filelist_lock); 1698 for (;;) { 1699 KASSERT(mutex_owned(&filelist_lock)); 1700 if (SLIST_EMPTY(&unp_thread_discard)) { 1701 if (unp_rights != 0) { 1702 (void)cv_timedwait(&unp_thread_cv, 1703 &filelist_lock, hz); 1704 } else { 1705 cv_wait(&unp_thread_cv, &filelist_lock); 1706 } 1707 } 1708 unp_gc(dp); 1709 } 1710 /* NOTREACHED */ 1711 } 1712 1713 /* 1714 * Kick the garbage collector into action if there is something for 1715 * it to process. 1716 */ 1717 static void 1718 unp_thread_kick(void) 1719 { 1720 1721 if (!SLIST_EMPTY(&unp_thread_discard) || unp_rights != 0) { 1722 mutex_enter(&filelist_lock); 1723 cv_signal(&unp_thread_cv); 1724 mutex_exit(&filelist_lock); 1725 } 1726 } 1727 1728 void 1729 unp_dispose(struct mbuf *m) 1730 { 1731 1732 if (m) 1733 unp_scan(m, unp_discard_later, 1); 1734 } 1735 1736 void 1737 unp_scan(struct mbuf *m0, void (*op)(file_t *), int discard) 1738 { 1739 struct mbuf *m; 1740 file_t **rp, *fp; 1741 struct cmsghdr *cm; 1742 int i, qfds; 1743 1744 while (m0) { 1745 for (m = m0; m; m = m->m_next) { 1746 if (m->m_type != MT_CONTROL || 1747 m->m_len < sizeof(*cm)) { 1748 continue; 1749 } 1750 cm = mtod(m, struct cmsghdr *); 1751 if (cm->cmsg_level != SOL_SOCKET || 1752 cm->cmsg_type != SCM_RIGHTS) 1753 continue; 1754 qfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) 1755 / sizeof(file_t *); 1756 rp = (file_t **)CMSG_DATA(cm); 1757 for (i = 0; i < qfds; i++) { 1758 fp = *rp; 1759 if (discard) { 1760 *rp = 0; 1761 } 1762 (*op)(fp); 1763 rp++; 1764 } 1765 } 1766 m0 = m0->m_nextpkt; 1767 } 1768 } 1769 1770 void 1771 unp_mark(file_t *fp) 1772 { 1773 1774 if (fp == NULL) 1775 return; 1776 1777 /* If we're already deferred, don't screw up the defer count */ 1778 mutex_enter(&fp->f_lock); 1779 if (fp->f_flag & (FMARK | FDEFER)) { 1780 mutex_exit(&fp->f_lock); 1781 return; 1782 } 1783 1784 /* 1785 * Minimize the number of deferrals... Sockets are the only type of 1786 * file which can hold references to another file, so just mark 1787 * other files, and defer unmarked sockets for the next pass. 1788 */ 1789 if (fp->f_type == DTYPE_SOCKET) { 1790 unp_defer++; 1791 KASSERT(fp->f_count != 0); 1792 atomic_or_uint(&fp->f_flag, FDEFER); 1793 } else { 1794 atomic_or_uint(&fp->f_flag, FMARK); 1795 } 1796 mutex_exit(&fp->f_lock); 1797 } 1798 1799 static void 1800 unp_discard_now(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(&fp->f_lock); 1810 fp->f_msgcount--; 1811 mutex_exit(&fp->f_lock); 1812 atomic_dec_uint(&unp_rights); 1813 (void)closef(fp); 1814 } 1815 1816 static void 1817 unp_discard_later(file_t *fp) 1818 { 1819 1820 if (fp == NULL) 1821 return; 1822 1823 KASSERT(fp->f_count > 0); 1824 KASSERT(fp->f_msgcount > 0); 1825 1826 mutex_enter(&filelist_lock); 1827 if (fp->f_unpcount++ == 0) { 1828 SLIST_INSERT_HEAD(&unp_thread_discard, fp, f_unplist); 1829 } 1830 mutex_exit(&filelist_lock); 1831 } 1832