1 /* $NetBSD: uipc_usrreq.c,v 1.126 2009/05/24 21:41:26 ad 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 1.126 2009/05/24 21:41:26 ad Exp $"); 100 101 #include <sys/param.h> 102 #include <sys/systm.h> 103 #include <sys/proc.h> 104 #include <sys/filedesc.h> 105 #include <sys/domain.h> 106 #include <sys/protosw.h> 107 #include <sys/socket.h> 108 #include <sys/socketvar.h> 109 #include <sys/unpcb.h> 110 #include <sys/un.h> 111 #include <sys/namei.h> 112 #include <sys/vnode.h> 113 #include <sys/file.h> 114 #include <sys/stat.h> 115 #include <sys/mbuf.h> 116 #include <sys/kauth.h> 117 #include <sys/kmem.h> 118 #include <sys/atomic.h> 119 #include <sys/uidinfo.h> 120 #include <sys/kernel.h> 121 #include <sys/kthread.h> 122 123 /* 124 * Unix communications domain. 125 * 126 * TODO: 127 * SEQPACKET, RDM 128 * rethink name space problems 129 * need a proper out-of-band 130 * 131 * Notes on locking: 132 * 133 * The generic rules noted in uipc_socket2.c apply. In addition: 134 * 135 * o We have a global lock, uipc_lock. 136 * 137 * o All datagram sockets are locked by uipc_lock. 138 * 139 * o For stream socketpairs, the two endpoints are created sharing the same 140 * independent lock. Sockets presented to PRU_CONNECT2 must already have 141 * matching locks. 142 * 143 * o Stream sockets created via socket() start life with their own 144 * independent lock. 145 * 146 * o Stream connections to a named endpoint are slightly more complicated. 147 * Sockets that have called listen() have their lock pointer mutated to 148 * the global uipc_lock. When establishing a connection, the connecting 149 * socket also has its lock mutated to uipc_lock, which matches the head 150 * (listening socket). We create a new socket for accept() to return, and 151 * that also shares the head's lock. Until the connection is completely 152 * done on both ends, all three sockets are locked by uipc_lock. Once the 153 * connection is complete, the association with the head's lock is broken. 154 * The connecting socket and the socket returned from accept() have their 155 * lock pointers mutated away from uipc_lock, and back to the connecting 156 * socket's original, independent lock. The head continues to be locked 157 * by uipc_lock. 158 * 159 * o If uipc_lock is determined to be a significant source of contention, 160 * it could easily be hashed out. It is difficult to simply make it an 161 * independent lock because of visibility / garbage collection issues: 162 * if a socket has been associated with a lock at any point, that lock 163 * must remain valid until the socket is no longer visible in the system. 164 * The lock must not be freed or otherwise destroyed until any sockets 165 * that had referenced it have also been destroyed. 166 */ 167 const struct sockaddr_un sun_noname = { 168 .sun_len = sizeof(sun_noname), 169 .sun_family = AF_LOCAL, 170 }; 171 ino_t unp_ino; /* prototype for fake inode numbers */ 172 173 struct mbuf *unp_addsockcred(struct lwp *, struct mbuf *); 174 static void unp_mark(file_t *); 175 static void unp_scan(struct mbuf *, void (*)(file_t *), int); 176 static void unp_discard_now(file_t *); 177 static void unp_discard_later(file_t *); 178 static void unp_thread(void *); 179 static void unp_thread_kick(void); 180 static kmutex_t *uipc_lock; 181 182 static kcondvar_t unp_thread_cv; 183 static lwp_t *unp_thread_lwp; 184 static SLIST_HEAD(,file) unp_thread_discard; 185 static int unp_defer; 186 187 /* 188 * Initialize Unix protocols. 189 */ 190 void 191 uipc_init(void) 192 { 193 int error; 194 195 uipc_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 196 cv_init(&unp_thread_cv, "unpgc"); 197 198 error = kthread_create(PRI_NONE, KTHREAD_MPSAFE, NULL, unp_thread, 199 NULL, &unp_thread_lwp, "unpgc"); 200 if (error != 0) 201 panic("uipc_init %d", error); 202 } 203 204 /* 205 * A connection succeeded: disassociate both endpoints from the head's 206 * lock, and make them share their own lock. There is a race here: for 207 * a very brief time one endpoint will be locked by a different lock 208 * than the other end. However, since the current thread holds the old 209 * lock (the listening socket's lock, the head) access can still only be 210 * made to one side of the connection. 211 */ 212 static void 213 unp_setpeerlocks(struct socket *so, struct socket *so2) 214 { 215 struct unpcb *unp; 216 kmutex_t *lock; 217 218 KASSERT(solocked2(so, so2)); 219 220 /* 221 * Bail out if either end of the socket is not yet fully 222 * connected or accepted. We only break the lock association 223 * with the head when the pair of sockets stand completely 224 * on their own. 225 */ 226 KASSERT(so->so_head == NULL); 227 if (so2->so_head != NULL) 228 return; 229 230 /* 231 * Drop references to old lock. A third reference (from the 232 * queue head) must be held as we still hold its lock. Bonus: 233 * we don't need to worry about garbage collecting the lock. 234 */ 235 lock = so->so_lock; 236 KASSERT(lock == uipc_lock); 237 mutex_obj_free(lock); 238 mutex_obj_free(lock); 239 240 /* 241 * Grab stream lock from the initiator and share between the two 242 * endpoints. Issue memory barrier to ensure all modifications 243 * become globally visible before the lock change. so2 is 244 * assumed not to have a stream lock, because it was created 245 * purely for the server side to accept this connection and 246 * started out life using the domain-wide lock. 247 */ 248 unp = sotounpcb(so); 249 KASSERT(unp->unp_streamlock != NULL); 250 KASSERT(sotounpcb(so2)->unp_streamlock == NULL); 251 lock = unp->unp_streamlock; 252 unp->unp_streamlock = NULL; 253 mutex_obj_hold(lock); 254 membar_exit(); 255 solockreset(so, lock); 256 solockreset(so2, lock); 257 } 258 259 /* 260 * Reset a socket's lock back to the domain-wide lock. 261 */ 262 static void 263 unp_resetlock(struct socket *so) 264 { 265 kmutex_t *olock, *nlock; 266 struct unpcb *unp; 267 268 KASSERT(solocked(so)); 269 270 olock = so->so_lock; 271 nlock = uipc_lock; 272 if (olock == nlock) 273 return; 274 unp = sotounpcb(so); 275 KASSERT(unp->unp_streamlock == NULL); 276 unp->unp_streamlock = olock; 277 mutex_obj_hold(nlock); 278 mutex_enter(nlock); 279 solockreset(so, nlock); 280 mutex_exit(olock); 281 } 282 283 static void 284 unp_free(struct unpcb *unp) 285 { 286 287 if (unp->unp_addr) 288 free(unp->unp_addr, M_SONAME); 289 if (unp->unp_streamlock != NULL) 290 mutex_obj_free(unp->unp_streamlock); 291 free(unp, M_PCB); 292 } 293 294 int 295 unp_output(struct mbuf *m, struct mbuf *control, struct unpcb *unp, 296 struct lwp *l) 297 { 298 struct socket *so2; 299 const struct sockaddr_un *sun; 300 301 so2 = unp->unp_conn->unp_socket; 302 303 KASSERT(solocked(so2)); 304 305 if (unp->unp_addr) 306 sun = unp->unp_addr; 307 else 308 sun = &sun_noname; 309 if (unp->unp_conn->unp_flags & UNP_WANTCRED) 310 control = unp_addsockcred(l, control); 311 if (sbappendaddr(&so2->so_rcv, (const struct sockaddr *)sun, m, 312 control) == 0) { 313 so2->so_rcv.sb_overflowed++; 314 unp_dispose(control); 315 m_freem(control); 316 m_freem(m); 317 return (ENOBUFS); 318 } else { 319 sorwakeup(so2); 320 return (0); 321 } 322 } 323 324 void 325 unp_setaddr(struct socket *so, struct mbuf *nam, bool peeraddr) 326 { 327 const struct sockaddr_un *sun; 328 struct unpcb *unp; 329 bool ext; 330 331 unp = sotounpcb(so); 332 ext = false; 333 334 for (;;) { 335 sun = NULL; 336 if (peeraddr) { 337 if (unp->unp_conn && unp->unp_conn->unp_addr) 338 sun = unp->unp_conn->unp_addr; 339 } else { 340 if (unp->unp_addr) 341 sun = unp->unp_addr; 342 } 343 if (sun == NULL) 344 sun = &sun_noname; 345 nam->m_len = sun->sun_len; 346 if (nam->m_len > MLEN && !ext) { 347 sounlock(so); 348 MEXTMALLOC(nam, MAXPATHLEN * 2, M_WAITOK); 349 solock(so); 350 ext = true; 351 } else { 352 KASSERT(nam->m_len <= MAXPATHLEN * 2); 353 memcpy(mtod(nam, void *), sun, (size_t)nam->m_len); 354 break; 355 } 356 } 357 } 358 359 /*ARGSUSED*/ 360 int 361 uipc_usrreq(struct socket *so, int req, struct mbuf *m, struct mbuf *nam, 362 struct mbuf *control, struct lwp *l) 363 { 364 struct unpcb *unp = sotounpcb(so); 365 struct socket *so2; 366 struct proc *p; 367 u_int newhiwat; 368 int error = 0; 369 370 if (req == PRU_CONTROL) 371 return (EOPNOTSUPP); 372 373 #ifdef DIAGNOSTIC 374 if (req != PRU_SEND && req != PRU_SENDOOB && control) 375 panic("uipc_usrreq: unexpected control mbuf"); 376 #endif 377 p = l ? l->l_proc : NULL; 378 if (req != PRU_ATTACH) { 379 if (unp == NULL) { 380 error = EINVAL; 381 goto release; 382 } 383 KASSERT(solocked(so)); 384 } 385 386 switch (req) { 387 388 case PRU_ATTACH: 389 if (unp != NULL) { 390 error = EISCONN; 391 break; 392 } 393 error = unp_attach(so); 394 break; 395 396 case PRU_DETACH: 397 unp_detach(unp); 398 break; 399 400 case PRU_BIND: 401 KASSERT(l != NULL); 402 error = unp_bind(so, nam, l); 403 break; 404 405 case PRU_LISTEN: 406 /* 407 * If the socket can accept a connection, it must be 408 * locked by uipc_lock. 409 */ 410 unp_resetlock(so); 411 if (unp->unp_vnode == NULL) 412 error = EINVAL; 413 break; 414 415 case PRU_CONNECT: 416 KASSERT(l != NULL); 417 error = unp_connect(so, nam, l); 418 break; 419 420 case PRU_CONNECT2: 421 error = unp_connect2(so, (struct socket *)nam, PRU_CONNECT2); 422 break; 423 424 case PRU_DISCONNECT: 425 unp_disconnect(unp); 426 break; 427 428 case PRU_ACCEPT: 429 KASSERT(so->so_lock == uipc_lock); 430 /* 431 * Mark the initiating STREAM socket as connected *ONLY* 432 * after it's been accepted. This prevents a client from 433 * overrunning a server and receiving ECONNREFUSED. 434 */ 435 if (unp->unp_conn == NULL) 436 break; 437 so2 = unp->unp_conn->unp_socket; 438 if (so2->so_state & SS_ISCONNECTING) { 439 KASSERT(solocked2(so, so->so_head)); 440 KASSERT(solocked2(so2, so->so_head)); 441 soisconnected(so2); 442 } 443 /* 444 * If the connection is fully established, break the 445 * association with uipc_lock and give the connected 446 * pair a seperate lock to share. 447 */ 448 unp_setpeerlocks(so2, so); 449 /* 450 * Only now return peer's address, as we may need to 451 * block in order to allocate memory. 452 * 453 * XXX Minor race: connection can be broken while 454 * lock is dropped in unp_setaddr(). We will return 455 * error == 0 and sun_noname as the peer address. 456 */ 457 unp_setaddr(so, nam, true); 458 break; 459 460 case PRU_SHUTDOWN: 461 socantsendmore(so); 462 unp_shutdown(unp); 463 break; 464 465 case PRU_RCVD: 466 switch (so->so_type) { 467 468 case SOCK_DGRAM: 469 panic("uipc 1"); 470 /*NOTREACHED*/ 471 472 case SOCK_STREAM: 473 #define rcv (&so->so_rcv) 474 #define snd (&so2->so_snd) 475 if (unp->unp_conn == 0) 476 break; 477 so2 = unp->unp_conn->unp_socket; 478 KASSERT(solocked2(so, so2)); 479 /* 480 * Adjust backpressure on sender 481 * and wakeup any waiting to write. 482 */ 483 snd->sb_mbmax += unp->unp_mbcnt - rcv->sb_mbcnt; 484 unp->unp_mbcnt = rcv->sb_mbcnt; 485 newhiwat = snd->sb_hiwat + unp->unp_cc - rcv->sb_cc; 486 (void)chgsbsize(so2->so_uidinfo, 487 &snd->sb_hiwat, newhiwat, RLIM_INFINITY); 488 unp->unp_cc = rcv->sb_cc; 489 sowwakeup(so2); 490 #undef snd 491 #undef rcv 492 break; 493 494 default: 495 panic("uipc 2"); 496 } 497 break; 498 499 case PRU_SEND: 500 /* 501 * Note: unp_internalize() rejects any control message 502 * other than SCM_RIGHTS, and only allows one. This 503 * has the side-effect of preventing a caller from 504 * forging SCM_CREDS. 505 */ 506 if (control) { 507 sounlock(so); 508 error = unp_internalize(&control); 509 solock(so); 510 if (error != 0) { 511 m_freem(control); 512 m_freem(m); 513 break; 514 } 515 } 516 switch (so->so_type) { 517 518 case SOCK_DGRAM: { 519 KASSERT(so->so_lock == uipc_lock); 520 if (nam) { 521 if ((so->so_state & SS_ISCONNECTED) != 0) 522 error = EISCONN; 523 else { 524 /* 525 * Note: once connected, the 526 * socket's lock must not be 527 * dropped until we have sent 528 * the message and disconnected. 529 * This is necessary to prevent 530 * intervening control ops, like 531 * another connection. 532 */ 533 error = unp_connect(so, nam, l); 534 } 535 } else { 536 if ((so->so_state & SS_ISCONNECTED) == 0) 537 error = ENOTCONN; 538 } 539 if (error) { 540 unp_dispose(control); 541 m_freem(control); 542 m_freem(m); 543 break; 544 } 545 KASSERT(p != NULL); 546 error = unp_output(m, control, unp, l); 547 if (nam) 548 unp_disconnect(unp); 549 break; 550 } 551 552 case SOCK_STREAM: 553 #define rcv (&so2->so_rcv) 554 #define snd (&so->so_snd) 555 if (unp->unp_conn == NULL) { 556 error = ENOTCONN; 557 break; 558 } 559 so2 = unp->unp_conn->unp_socket; 560 KASSERT(solocked2(so, so2)); 561 if (unp->unp_conn->unp_flags & UNP_WANTCRED) { 562 /* 563 * Credentials are passed only once on 564 * SOCK_STREAM. 565 */ 566 unp->unp_conn->unp_flags &= ~UNP_WANTCRED; 567 control = unp_addsockcred(l, control); 568 } 569 /* 570 * Send to paired receive port, and then reduce 571 * send buffer hiwater marks to maintain backpressure. 572 * Wake up readers. 573 */ 574 if (control) { 575 if (sbappendcontrol(rcv, m, control) != 0) 576 control = NULL; 577 } else 578 sbappend(rcv, m); 579 snd->sb_mbmax -= 580 rcv->sb_mbcnt - unp->unp_conn->unp_mbcnt; 581 unp->unp_conn->unp_mbcnt = rcv->sb_mbcnt; 582 newhiwat = snd->sb_hiwat - 583 (rcv->sb_cc - unp->unp_conn->unp_cc); 584 (void)chgsbsize(so->so_uidinfo, 585 &snd->sb_hiwat, newhiwat, RLIM_INFINITY); 586 unp->unp_conn->unp_cc = rcv->sb_cc; 587 sorwakeup(so2); 588 #undef snd 589 #undef rcv 590 if (control != NULL) { 591 unp_dispose(control); 592 m_freem(control); 593 } 594 break; 595 596 default: 597 panic("uipc 4"); 598 } 599 break; 600 601 case PRU_ABORT: 602 (void)unp_drop(unp, ECONNABORTED); 603 604 KASSERT(so->so_head == NULL); 605 #ifdef DIAGNOSTIC 606 if (so->so_pcb == NULL) 607 panic("uipc 5: drop killed pcb"); 608 #endif 609 unp_detach(unp); 610 break; 611 612 case PRU_SENSE: 613 ((struct stat *) m)->st_blksize = so->so_snd.sb_hiwat; 614 if (so->so_type == SOCK_STREAM && unp->unp_conn != 0) { 615 so2 = unp->unp_conn->unp_socket; 616 KASSERT(solocked2(so, so2)); 617 ((struct stat *) m)->st_blksize += so2->so_rcv.sb_cc; 618 } 619 ((struct stat *) m)->st_dev = NODEV; 620 if (unp->unp_ino == 0) 621 unp->unp_ino = unp_ino++; 622 ((struct stat *) m)->st_atimespec = 623 ((struct stat *) m)->st_mtimespec = 624 ((struct stat *) m)->st_ctimespec = unp->unp_ctime; 625 ((struct stat *) m)->st_ino = unp->unp_ino; 626 return (0); 627 628 case PRU_RCVOOB: 629 error = EOPNOTSUPP; 630 break; 631 632 case PRU_SENDOOB: 633 m_freem(control); 634 m_freem(m); 635 error = EOPNOTSUPP; 636 break; 637 638 case PRU_SOCKADDR: 639 unp_setaddr(so, nam, false); 640 break; 641 642 case PRU_PEERADDR: 643 unp_setaddr(so, nam, true); 644 break; 645 646 default: 647 panic("piusrreq"); 648 } 649 650 release: 651 return (error); 652 } 653 654 /* 655 * Unix domain socket option processing. 656 */ 657 int 658 uipc_ctloutput(int op, struct socket *so, struct sockopt *sopt) 659 { 660 struct unpcb *unp = sotounpcb(so); 661 int optval = 0, error = 0; 662 663 KASSERT(solocked(so)); 664 665 if (sopt->sopt_level != 0) { 666 error = ENOPROTOOPT; 667 } else switch (op) { 668 669 case PRCO_SETOPT: 670 switch (sopt->sopt_name) { 671 case LOCAL_CREDS: 672 case LOCAL_CONNWAIT: 673 error = sockopt_getint(sopt, &optval); 674 if (error) 675 break; 676 switch (sopt->sopt_name) { 677 #define OPTSET(bit) \ 678 if (optval) \ 679 unp->unp_flags |= (bit); \ 680 else \ 681 unp->unp_flags &= ~(bit); 682 683 case LOCAL_CREDS: 684 OPTSET(UNP_WANTCRED); 685 break; 686 case LOCAL_CONNWAIT: 687 OPTSET(UNP_CONNWAIT); 688 break; 689 } 690 break; 691 #undef OPTSET 692 693 default: 694 error = ENOPROTOOPT; 695 break; 696 } 697 break; 698 699 case PRCO_GETOPT: 700 sounlock(so); 701 switch (sopt->sopt_name) { 702 case LOCAL_PEEREID: 703 if (unp->unp_flags & UNP_EIDSVALID) { 704 error = sockopt_set(sopt, 705 &unp->unp_connid, sizeof(unp->unp_connid)); 706 } else { 707 error = EINVAL; 708 } 709 break; 710 case LOCAL_CREDS: 711 #define OPTBIT(bit) (unp->unp_flags & (bit) ? 1 : 0) 712 713 optval = OPTBIT(UNP_WANTCRED); 714 error = sockopt_setint(sopt, optval); 715 break; 716 #undef OPTBIT 717 718 default: 719 error = ENOPROTOOPT; 720 break; 721 } 722 solock(so); 723 break; 724 } 725 return (error); 726 } 727 728 /* 729 * Both send and receive buffers are allocated PIPSIZ bytes of buffering 730 * for stream sockets, although the total for sender and receiver is 731 * actually only PIPSIZ. 732 * Datagram sockets really use the sendspace as the maximum datagram size, 733 * and don't really want to reserve the sendspace. Their recvspace should 734 * be large enough for at least one max-size datagram plus address. 735 */ 736 #define PIPSIZ 4096 737 u_long unpst_sendspace = PIPSIZ; 738 u_long unpst_recvspace = PIPSIZ; 739 u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 740 u_long unpdg_recvspace = 4*1024; 741 742 u_int unp_rights; /* files in flight */ 743 u_int unp_rights_ratio = 2; /* limit, fraction of maxfiles */ 744 745 int 746 unp_attach(struct socket *so) 747 { 748 struct unpcb *unp; 749 int error; 750 751 switch (so->so_type) { 752 case SOCK_STREAM: 753 if (so->so_lock == NULL) { 754 /* 755 * XXX Assuming that no socket locks are held, 756 * as this call may sleep. 757 */ 758 so->so_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 759 solock(so); 760 } 761 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 762 error = soreserve(so, unpst_sendspace, unpst_recvspace); 763 if (error != 0) 764 return (error); 765 } 766 break; 767 768 case SOCK_DGRAM: 769 if (so->so_lock == NULL) { 770 mutex_obj_hold(uipc_lock); 771 so->so_lock = uipc_lock; 772 solock(so); 773 } 774 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 775 error = soreserve(so, unpdg_sendspace, unpdg_recvspace); 776 if (error != 0) 777 return (error); 778 } 779 break; 780 781 default: 782 panic("unp_attach"); 783 } 784 KASSERT(solocked(so)); 785 unp = malloc(sizeof(*unp), M_PCB, M_NOWAIT); 786 if (unp == NULL) 787 return (ENOBUFS); 788 memset(unp, 0, sizeof(*unp)); 789 unp->unp_socket = so; 790 so->so_pcb = unp; 791 nanotime(&unp->unp_ctime); 792 return (0); 793 } 794 795 void 796 unp_detach(struct unpcb *unp) 797 { 798 struct socket *so; 799 vnode_t *vp; 800 801 so = unp->unp_socket; 802 803 retry: 804 if ((vp = unp->unp_vnode) != NULL) { 805 sounlock(so); 806 /* Acquire v_interlock to protect against unp_connect(). */ 807 /* XXXAD racy */ 808 mutex_enter(&vp->v_interlock); 809 vp->v_socket = NULL; 810 vrelel(vp, 0); 811 solock(so); 812 unp->unp_vnode = NULL; 813 } 814 if (unp->unp_conn) 815 unp_disconnect(unp); 816 while (unp->unp_refs) { 817 KASSERT(solocked2(so, unp->unp_refs->unp_socket)); 818 if (unp_drop(unp->unp_refs, ECONNRESET)) { 819 solock(so); 820 goto retry; 821 } 822 } 823 soisdisconnected(so); 824 so->so_pcb = NULL; 825 if (unp_rights) { 826 /* 827 * Normally the receive buffer is flushed later, in sofree, 828 * but if our receive buffer holds references to files that 829 * are now garbage, we will enqueue those file references to 830 * the garbage collector and kick it into action. 831 */ 832 sorflush(so); 833 unp_free(unp); 834 unp_thread_kick(); 835 } else 836 unp_free(unp); 837 } 838 839 int 840 unp_bind(struct socket *so, struct mbuf *nam, struct lwp *l) 841 { 842 struct sockaddr_un *sun; 843 struct unpcb *unp; 844 vnode_t *vp; 845 struct vattr vattr; 846 size_t addrlen; 847 int error; 848 struct nameidata nd; 849 proc_t *p; 850 851 unp = sotounpcb(so); 852 if (unp->unp_vnode != NULL) 853 return (EINVAL); 854 if ((unp->unp_flags & UNP_BUSY) != 0) { 855 /* 856 * EALREADY may not be strictly accurate, but since this 857 * is a major application error it's hardly a big deal. 858 */ 859 return (EALREADY); 860 } 861 unp->unp_flags |= UNP_BUSY; 862 sounlock(so); 863 864 /* 865 * Allocate the new sockaddr. We have to allocate one 866 * extra byte so that we can ensure that the pathname 867 * is nul-terminated. 868 */ 869 p = l->l_proc; 870 addrlen = nam->m_len + 1; 871 sun = malloc(addrlen, M_SONAME, M_WAITOK); 872 m_copydata(nam, 0, nam->m_len, (void *)sun); 873 *(((char *)sun) + nam->m_len) = '\0'; 874 875 NDINIT(&nd, CREATE, FOLLOW | LOCKPARENT | TRYEMULROOT, UIO_SYSSPACE, 876 sun->sun_path); 877 878 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 879 if ((error = namei(&nd)) != 0) 880 goto bad; 881 vp = nd.ni_vp; 882 if (vp != NULL) { 883 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 884 if (nd.ni_dvp == vp) 885 vrele(nd.ni_dvp); 886 else 887 vput(nd.ni_dvp); 888 vrele(vp); 889 error = EADDRINUSE; 890 goto bad; 891 } 892 VATTR_NULL(&vattr); 893 vattr.va_type = VSOCK; 894 vattr.va_mode = ACCESSPERMS & ~(p->p_cwdi->cwdi_cmask); 895 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 896 if (error) 897 goto bad; 898 vp = nd.ni_vp; 899 solock(so); 900 vp->v_socket = unp->unp_socket; 901 unp->unp_vnode = vp; 902 unp->unp_addrlen = addrlen; 903 unp->unp_addr = sun; 904 unp->unp_connid.unp_pid = p->p_pid; 905 unp->unp_connid.unp_euid = kauth_cred_geteuid(l->l_cred); 906 unp->unp_connid.unp_egid = kauth_cred_getegid(l->l_cred); 907 unp->unp_flags |= UNP_EIDSBIND; 908 VOP_UNLOCK(vp, 0); 909 unp->unp_flags &= ~UNP_BUSY; 910 return (0); 911 912 bad: 913 free(sun, M_SONAME); 914 solock(so); 915 unp->unp_flags &= ~UNP_BUSY; 916 return (error); 917 } 918 919 int 920 unp_connect(struct socket *so, struct mbuf *nam, struct lwp *l) 921 { 922 struct sockaddr_un *sun; 923 vnode_t *vp; 924 struct socket *so2, *so3; 925 struct unpcb *unp, *unp2, *unp3; 926 size_t addrlen; 927 int error; 928 struct nameidata nd; 929 930 unp = sotounpcb(so); 931 if ((unp->unp_flags & UNP_BUSY) != 0) { 932 /* 933 * EALREADY may not be strictly accurate, but since this 934 * is a major application error it's hardly a big deal. 935 */ 936 return (EALREADY); 937 } 938 unp->unp_flags |= UNP_BUSY; 939 sounlock(so); 940 941 /* 942 * Allocate a temporary sockaddr. We have to allocate one extra 943 * byte so that we can ensure that the pathname is nul-terminated. 944 * When we establish the connection, we copy the other PCB's 945 * sockaddr to our own. 946 */ 947 addrlen = nam->m_len + 1; 948 sun = malloc(addrlen, M_SONAME, M_WAITOK); 949 m_copydata(nam, 0, nam->m_len, (void *)sun); 950 *(((char *)sun) + nam->m_len) = '\0'; 951 952 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, UIO_SYSSPACE, 953 sun->sun_path); 954 955 if ((error = namei(&nd)) != 0) 956 goto bad2; 957 vp = nd.ni_vp; 958 if (vp->v_type != VSOCK) { 959 error = ENOTSOCK; 960 goto bad; 961 } 962 if ((error = VOP_ACCESS(vp, VWRITE, l->l_cred)) != 0) 963 goto bad; 964 /* Acquire v_interlock to protect against unp_detach(). */ 965 mutex_enter(&vp->v_interlock); 966 so2 = vp->v_socket; 967 if (so2 == NULL) { 968 mutex_exit(&vp->v_interlock); 969 error = ECONNREFUSED; 970 goto bad; 971 } 972 if (so->so_type != so2->so_type) { 973 mutex_exit(&vp->v_interlock); 974 error = EPROTOTYPE; 975 goto bad; 976 } 977 solock(so); 978 unp_resetlock(so); 979 mutex_exit(&vp->v_interlock); 980 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) { 981 /* 982 * This may seem somewhat fragile but is OK: if we can 983 * see SO_ACCEPTCONN set on the endpoint, then it must 984 * be locked by the domain-wide uipc_lock. 985 */ 986 KASSERT((so->so_options & SO_ACCEPTCONN) == 0 || 987 so2->so_lock == uipc_lock); 988 if ((so2->so_options & SO_ACCEPTCONN) == 0 || 989 (so3 = sonewconn(so2, 0)) == NULL) { 990 error = ECONNREFUSED; 991 sounlock(so); 992 goto bad; 993 } 994 unp2 = sotounpcb(so2); 995 unp3 = sotounpcb(so3); 996 if (unp2->unp_addr) { 997 unp3->unp_addr = malloc(unp2->unp_addrlen, 998 M_SONAME, M_WAITOK); 999 memcpy(unp3->unp_addr, unp2->unp_addr, 1000 unp2->unp_addrlen); 1001 unp3->unp_addrlen = unp2->unp_addrlen; 1002 } 1003 unp3->unp_flags = unp2->unp_flags; 1004 unp3->unp_connid.unp_pid = l->l_proc->p_pid; 1005 unp3->unp_connid.unp_euid = kauth_cred_geteuid(l->l_cred); 1006 unp3->unp_connid.unp_egid = kauth_cred_getegid(l->l_cred); 1007 unp3->unp_flags |= UNP_EIDSVALID; 1008 if (unp2->unp_flags & UNP_EIDSBIND) { 1009 unp->unp_connid = unp2->unp_connid; 1010 unp->unp_flags |= UNP_EIDSVALID; 1011 } 1012 so2 = so3; 1013 } 1014 error = unp_connect2(so, so2, PRU_CONNECT); 1015 sounlock(so); 1016 bad: 1017 vput(vp); 1018 bad2: 1019 free(sun, M_SONAME); 1020 solock(so); 1021 unp->unp_flags &= ~UNP_BUSY; 1022 return (error); 1023 } 1024 1025 int 1026 unp_connect2(struct socket *so, struct socket *so2, int req) 1027 { 1028 struct unpcb *unp = sotounpcb(so); 1029 struct unpcb *unp2; 1030 1031 if (so2->so_type != so->so_type) 1032 return (EPROTOTYPE); 1033 1034 /* 1035 * All three sockets involved must be locked by same lock: 1036 * 1037 * local endpoint (so) 1038 * remote endpoint (so2) 1039 * queue head (so->so_head, only if PR_CONNREQUIRED) 1040 */ 1041 KASSERT(solocked2(so, so2)); 1042 KASSERT(so->so_head == NULL); 1043 if (so2->so_head != NULL) { 1044 KASSERT(so2->so_lock == uipc_lock); 1045 KASSERT(solocked2(so2, so2->so_head)); 1046 } 1047 1048 unp2 = sotounpcb(so2); 1049 unp->unp_conn = unp2; 1050 switch (so->so_type) { 1051 1052 case SOCK_DGRAM: 1053 unp->unp_nextref = unp2->unp_refs; 1054 unp2->unp_refs = unp; 1055 soisconnected(so); 1056 break; 1057 1058 case SOCK_STREAM: 1059 unp2->unp_conn = unp; 1060 if (req == PRU_CONNECT && 1061 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 1062 soisconnecting(so); 1063 else 1064 soisconnected(so); 1065 soisconnected(so2); 1066 /* 1067 * If the connection is fully established, break the 1068 * association with uipc_lock and give the connected 1069 * pair a seperate lock to share. For CONNECT2, we 1070 * require that the locks already match (the sockets 1071 * are created that way). 1072 */ 1073 if (req == PRU_CONNECT) { 1074 KASSERT(so2->so_head != NULL); 1075 unp_setpeerlocks(so, so2); 1076 } 1077 break; 1078 1079 default: 1080 panic("unp_connect2"); 1081 } 1082 return (0); 1083 } 1084 1085 void 1086 unp_disconnect(struct unpcb *unp) 1087 { 1088 struct unpcb *unp2 = unp->unp_conn; 1089 struct socket *so; 1090 1091 if (unp2 == 0) 1092 return; 1093 unp->unp_conn = 0; 1094 so = unp->unp_socket; 1095 switch (so->so_type) { 1096 case SOCK_DGRAM: 1097 if (unp2->unp_refs == unp) 1098 unp2->unp_refs = unp->unp_nextref; 1099 else { 1100 unp2 = unp2->unp_refs; 1101 for (;;) { 1102 KASSERT(solocked2(so, unp2->unp_socket)); 1103 if (unp2 == 0) 1104 panic("unp_disconnect"); 1105 if (unp2->unp_nextref == unp) 1106 break; 1107 unp2 = unp2->unp_nextref; 1108 } 1109 unp2->unp_nextref = unp->unp_nextref; 1110 } 1111 unp->unp_nextref = 0; 1112 so->so_state &= ~SS_ISCONNECTED; 1113 break; 1114 1115 case SOCK_STREAM: 1116 KASSERT(solocked2(so, unp2->unp_socket)); 1117 soisdisconnected(so); 1118 unp2->unp_conn = 0; 1119 soisdisconnected(unp2->unp_socket); 1120 break; 1121 } 1122 } 1123 1124 #ifdef notdef 1125 unp_abort(struct unpcb *unp) 1126 { 1127 unp_detach(unp); 1128 } 1129 #endif 1130 1131 void 1132 unp_shutdown(struct unpcb *unp) 1133 { 1134 struct socket *so; 1135 1136 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn && 1137 (so = unp->unp_conn->unp_socket)) 1138 socantrcvmore(so); 1139 } 1140 1141 bool 1142 unp_drop(struct unpcb *unp, int errno) 1143 { 1144 struct socket *so = unp->unp_socket; 1145 1146 KASSERT(solocked(so)); 1147 1148 so->so_error = errno; 1149 unp_disconnect(unp); 1150 if (so->so_head) { 1151 so->so_pcb = NULL; 1152 /* sofree() drops the socket lock */ 1153 sofree(so); 1154 unp_free(unp); 1155 return true; 1156 } 1157 return false; 1158 } 1159 1160 #ifdef notdef 1161 unp_drain(void) 1162 { 1163 1164 } 1165 #endif 1166 1167 int 1168 unp_externalize(struct mbuf *rights, struct lwp *l) 1169 { 1170 struct cmsghdr *cm = mtod(rights, struct cmsghdr *); 1171 struct proc *p = l->l_proc; 1172 int i, *fdp; 1173 file_t **rp; 1174 file_t *fp; 1175 int nfds, error = 0; 1176 1177 nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) / 1178 sizeof(file_t *); 1179 rp = (file_t **)CMSG_DATA(cm); 1180 1181 fdp = malloc(nfds * sizeof(int), M_TEMP, M_WAITOK); 1182 rw_enter(&p->p_cwdi->cwdi_lock, RW_READER); 1183 1184 /* Make sure the recipient should be able to see the files.. */ 1185 if (p->p_cwdi->cwdi_rdir != NULL) { 1186 rp = (file_t **)CMSG_DATA(cm); 1187 for (i = 0; i < nfds; i++) { 1188 fp = *rp++; 1189 /* 1190 * If we are in a chroot'ed directory, and 1191 * someone wants to pass us a directory, make 1192 * sure it's inside the subtree we're allowed 1193 * to access. 1194 */ 1195 if (fp->f_type == DTYPE_VNODE) { 1196 vnode_t *vp = (vnode_t *)fp->f_data; 1197 if ((vp->v_type == VDIR) && 1198 !vn_isunder(vp, p->p_cwdi->cwdi_rdir, l)) { 1199 error = EPERM; 1200 break; 1201 } 1202 } 1203 } 1204 } 1205 1206 restart: 1207 rp = (file_t **)CMSG_DATA(cm); 1208 if (error != 0) { 1209 for (i = 0; i < nfds; i++) { 1210 fp = *rp; 1211 *rp++ = 0; 1212 unp_discard_now(fp); 1213 } 1214 goto out; 1215 } 1216 1217 /* 1218 * First loop -- allocate file descriptor table slots for the 1219 * new files. 1220 */ 1221 for (i = 0; i < nfds; i++) { 1222 fp = *rp++; 1223 if ((error = fd_alloc(p, 0, &fdp[i])) != 0) { 1224 /* 1225 * Back out what we've done so far. 1226 */ 1227 for (--i; i >= 0; i--) { 1228 fd_abort(p, NULL, fdp[i]); 1229 } 1230 if (error == ENOSPC) { 1231 fd_tryexpand(p); 1232 error = 0; 1233 } else { 1234 /* 1235 * This is the error that has historically 1236 * been returned, and some callers may 1237 * expect it. 1238 */ 1239 error = EMSGSIZE; 1240 } 1241 goto restart; 1242 } 1243 } 1244 1245 /* 1246 * Now that adding them has succeeded, update all of the 1247 * file passing state and affix the descriptors. 1248 */ 1249 rp = (file_t **)CMSG_DATA(cm); 1250 for (i = 0; i < nfds; i++) { 1251 fp = *rp++; 1252 atomic_dec_uint(&unp_rights); 1253 fd_affix(p, fp, fdp[i]); 1254 mutex_enter(&fp->f_lock); 1255 fp->f_msgcount--; 1256 mutex_exit(&fp->f_lock); 1257 /* 1258 * Note that fd_affix() adds a reference to the file. 1259 * The file may already have been closed by another 1260 * LWP in the process, so we must drop the reference 1261 * added by unp_internalize() with closef(). 1262 */ 1263 closef(fp); 1264 } 1265 1266 /* 1267 * Copy temporary array to message and adjust length, in case of 1268 * transition from large file_t pointers to ints. 1269 */ 1270 memcpy(CMSG_DATA(cm), fdp, nfds * sizeof(int)); 1271 cm->cmsg_len = CMSG_LEN(nfds * sizeof(int)); 1272 rights->m_len = CMSG_SPACE(nfds * sizeof(int)); 1273 out: 1274 rw_exit(&p->p_cwdi->cwdi_lock); 1275 free(fdp, M_TEMP); 1276 return (error); 1277 } 1278 1279 int 1280 unp_internalize(struct mbuf **controlp) 1281 { 1282 filedesc_t *fdescp = curlwp->l_fd; 1283 struct mbuf *control = *controlp; 1284 struct cmsghdr *newcm, *cm = mtod(control, struct cmsghdr *); 1285 file_t **rp, **files; 1286 file_t *fp; 1287 int i, fd, *fdp; 1288 int nfds, error; 1289 u_int maxmsg; 1290 1291 error = 0; 1292 newcm = NULL; 1293 1294 /* Sanity check the control message header. */ 1295 if (cm->cmsg_type != SCM_RIGHTS || cm->cmsg_level != SOL_SOCKET || 1296 cm->cmsg_len > control->m_len || 1297 cm->cmsg_len < CMSG_ALIGN(sizeof(*cm))) 1298 return (EINVAL); 1299 1300 /* 1301 * Verify that the file descriptors are valid, and acquire 1302 * a reference to each. 1303 */ 1304 nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) / sizeof(int); 1305 fdp = (int *)CMSG_DATA(cm); 1306 maxmsg = maxfiles / unp_rights_ratio; 1307 for (i = 0; i < nfds; i++) { 1308 fd = *fdp++; 1309 if (atomic_inc_uint_nv(&unp_rights) > maxmsg) { 1310 atomic_dec_uint(&unp_rights); 1311 nfds = i; 1312 error = EAGAIN; 1313 goto out; 1314 } 1315 if ((fp = fd_getfile(fd)) == NULL) { 1316 atomic_dec_uint(&unp_rights); 1317 nfds = i; 1318 error = EBADF; 1319 goto out; 1320 } 1321 } 1322 1323 /* Allocate new space and copy header into it. */ 1324 newcm = malloc(CMSG_SPACE(nfds * sizeof(file_t *)), M_MBUF, M_WAITOK); 1325 if (newcm == NULL) { 1326 error = E2BIG; 1327 goto out; 1328 } 1329 memcpy(newcm, cm, sizeof(struct cmsghdr)); 1330 files = (file_t **)CMSG_DATA(newcm); 1331 1332 /* 1333 * Transform the file descriptors into file_t pointers, in 1334 * reverse order so that if pointers are bigger than ints, the 1335 * int won't get until we're done. No need to lock, as we have 1336 * already validated the descriptors with fd_getfile(). 1337 */ 1338 fdp = (int *)CMSG_DATA(cm) + nfds; 1339 rp = files + nfds; 1340 for (i = 0; i < nfds; i++) { 1341 fp = fdescp->fd_dt->dt_ff[*--fdp]->ff_file; 1342 KASSERT(fp != NULL); 1343 mutex_enter(&fp->f_lock); 1344 *--rp = fp; 1345 fp->f_count++; 1346 fp->f_msgcount++; 1347 mutex_exit(&fp->f_lock); 1348 } 1349 1350 out: 1351 /* Release descriptor references. */ 1352 fdp = (int *)CMSG_DATA(cm); 1353 for (i = 0; i < nfds; i++) { 1354 fd_putfile(*fdp++); 1355 if (error != 0) { 1356 atomic_dec_uint(&unp_rights); 1357 } 1358 } 1359 1360 if (error == 0) { 1361 if (control->m_flags & M_EXT) { 1362 m_freem(control); 1363 *controlp = control = m_get(M_WAIT, MT_CONTROL); 1364 } 1365 MEXTADD(control, newcm, CMSG_SPACE(nfds * sizeof(file_t *)), 1366 M_MBUF, NULL, NULL); 1367 cm = newcm; 1368 /* 1369 * Adjust message & mbuf to note amount of space 1370 * actually used. 1371 */ 1372 cm->cmsg_len = CMSG_LEN(nfds * sizeof(file_t *)); 1373 control->m_len = CMSG_SPACE(nfds * sizeof(file_t *)); 1374 } 1375 1376 return error; 1377 } 1378 1379 struct mbuf * 1380 unp_addsockcred(struct lwp *l, struct mbuf *control) 1381 { 1382 struct cmsghdr *cmp; 1383 struct sockcred *sc; 1384 struct mbuf *m, *n; 1385 int len, space, i; 1386 1387 len = CMSG_LEN(SOCKCREDSIZE(kauth_cred_ngroups(l->l_cred))); 1388 space = CMSG_SPACE(SOCKCREDSIZE(kauth_cred_ngroups(l->l_cred))); 1389 1390 m = m_get(M_WAIT, MT_CONTROL); 1391 if (space > MLEN) { 1392 if (space > MCLBYTES) 1393 MEXTMALLOC(m, space, M_WAITOK); 1394 else 1395 m_clget(m, M_WAIT); 1396 if ((m->m_flags & M_EXT) == 0) { 1397 m_free(m); 1398 return (control); 1399 } 1400 } 1401 1402 m->m_len = space; 1403 m->m_next = NULL; 1404 cmp = mtod(m, struct cmsghdr *); 1405 sc = (struct sockcred *)CMSG_DATA(cmp); 1406 cmp->cmsg_len = len; 1407 cmp->cmsg_level = SOL_SOCKET; 1408 cmp->cmsg_type = SCM_CREDS; 1409 sc->sc_uid = kauth_cred_getuid(l->l_cred); 1410 sc->sc_euid = kauth_cred_geteuid(l->l_cred); 1411 sc->sc_gid = kauth_cred_getgid(l->l_cred); 1412 sc->sc_egid = kauth_cred_getegid(l->l_cred); 1413 sc->sc_ngroups = kauth_cred_ngroups(l->l_cred); 1414 for (i = 0; i < sc->sc_ngroups; i++) 1415 sc->sc_groups[i] = kauth_cred_group(l->l_cred, i); 1416 1417 /* 1418 * If a control message already exists, append us to the end. 1419 */ 1420 if (control != NULL) { 1421 for (n = control; n->m_next != NULL; n = n->m_next) 1422 ; 1423 n->m_next = m; 1424 } else 1425 control = m; 1426 1427 return (control); 1428 } 1429 1430 /* 1431 * Do a mark-sweep GC of files in the system, to free up any which are 1432 * caught in flight to an about-to-be-closed socket. Additionally, 1433 * process deferred file closures. 1434 */ 1435 static void 1436 unp_gc(file_t *dp) 1437 { 1438 extern struct domain unixdomain; 1439 file_t *fp, *np; 1440 struct socket *so, *so1; 1441 u_int i, old, new; 1442 bool didwork; 1443 1444 KASSERT(curlwp == unp_thread_lwp); 1445 KASSERT(mutex_owned(&filelist_lock)); 1446 1447 /* 1448 * First, process deferred file closures. 1449 */ 1450 while (!SLIST_EMPTY(&unp_thread_discard)) { 1451 fp = SLIST_FIRST(&unp_thread_discard); 1452 KASSERT(fp->f_unpcount > 0); 1453 KASSERT(fp->f_count > 0); 1454 KASSERT(fp->f_msgcount > 0); 1455 KASSERT(fp->f_count >= fp->f_unpcount); 1456 KASSERT(fp->f_count >= fp->f_msgcount); 1457 KASSERT(fp->f_msgcount >= fp->f_unpcount); 1458 SLIST_REMOVE_HEAD(&unp_thread_discard, f_unplist); 1459 i = fp->f_unpcount; 1460 fp->f_unpcount = 0; 1461 mutex_exit(&filelist_lock); 1462 for (; i != 0; i--) { 1463 unp_discard_now(fp); 1464 } 1465 mutex_enter(&filelist_lock); 1466 } 1467 1468 /* 1469 * Clear mark bits. Ensure that we don't consider new files 1470 * entering the file table during this loop (they will not have 1471 * FSCAN set). 1472 */ 1473 unp_defer = 0; 1474 LIST_FOREACH(fp, &filehead, f_list) { 1475 for (old = fp->f_flag;; old = new) { 1476 new = atomic_cas_uint(&fp->f_flag, old, 1477 (old | FSCAN) & ~(FMARK|FDEFER)); 1478 if (__predict_true(old == new)) { 1479 break; 1480 } 1481 } 1482 } 1483 1484 /* 1485 * Iterate over the set of sockets, marking ones believed (based on 1486 * refcount) to be referenced from a process, and marking for rescan 1487 * sockets which are queued on a socket. Recan continues descending 1488 * and searching for sockets referenced by sockets (FDEFER), until 1489 * there are no more socket->socket references to be discovered. 1490 */ 1491 do { 1492 didwork = false; 1493 for (fp = LIST_FIRST(&filehead); fp != NULL; fp = np) { 1494 KASSERT(mutex_owned(&filelist_lock)); 1495 np = LIST_NEXT(fp, f_list); 1496 mutex_enter(&fp->f_lock); 1497 if ((fp->f_flag & FDEFER) != 0) { 1498 atomic_and_uint(&fp->f_flag, ~FDEFER); 1499 unp_defer--; 1500 KASSERT(fp->f_count != 0); 1501 } else { 1502 if (fp->f_count == 0 || 1503 (fp->f_flag & FMARK) != 0 || 1504 fp->f_count == fp->f_msgcount || 1505 fp->f_unpcount != 0) { 1506 mutex_exit(&fp->f_lock); 1507 continue; 1508 } 1509 } 1510 atomic_or_uint(&fp->f_flag, FMARK); 1511 1512 if (fp->f_type != DTYPE_SOCKET || 1513 (so = fp->f_data) == NULL || 1514 so->so_proto->pr_domain != &unixdomain || 1515 (so->so_proto->pr_flags & PR_RIGHTS) == 0) { 1516 mutex_exit(&fp->f_lock); 1517 continue; 1518 } 1519 1520 /* Gain file ref, mark our position, and unlock. */ 1521 didwork = true; 1522 LIST_INSERT_AFTER(fp, dp, f_list); 1523 fp->f_count++; 1524 mutex_exit(&fp->f_lock); 1525 mutex_exit(&filelist_lock); 1526 1527 /* 1528 * Mark files referenced from sockets queued on the 1529 * accept queue as well. 1530 */ 1531 solock(so); 1532 unp_scan(so->so_rcv.sb_mb, unp_mark, 0); 1533 if ((so->so_options & SO_ACCEPTCONN) != 0) { 1534 TAILQ_FOREACH(so1, &so->so_q0, so_qe) { 1535 unp_scan(so1->so_rcv.sb_mb, unp_mark, 0); 1536 } 1537 TAILQ_FOREACH(so1, &so->so_q, so_qe) { 1538 unp_scan(so1->so_rcv.sb_mb, unp_mark, 0); 1539 } 1540 } 1541 sounlock(so); 1542 1543 /* Re-lock and restart from where we left off. */ 1544 closef(fp); 1545 mutex_enter(&filelist_lock); 1546 np = LIST_NEXT(dp, f_list); 1547 LIST_REMOVE(dp, f_list); 1548 } 1549 /* 1550 * Bail early if we did nothing in the loop above. Could 1551 * happen because of concurrent activity causing unp_defer 1552 * to get out of sync. 1553 */ 1554 } while (unp_defer != 0 && didwork); 1555 1556 /* 1557 * Sweep pass. 1558 * 1559 * We grab an extra reference to each of the files that are 1560 * not otherwise accessible and then free the rights that are 1561 * stored in messages on them. 1562 */ 1563 for (fp = LIST_FIRST(&filehead); fp != NULL; fp = np) { 1564 KASSERT(mutex_owned(&filelist_lock)); 1565 np = LIST_NEXT(fp, f_list); 1566 mutex_enter(&fp->f_lock); 1567 1568 /* 1569 * Ignore non-sockets. 1570 * Ignore dead sockets, or sockets with pending close. 1571 * Ignore sockets obviously referenced elsewhere. 1572 * Ignore sockets marked as referenced by our scan. 1573 * Ignore new sockets that did not exist during the scan. 1574 */ 1575 if (fp->f_type != DTYPE_SOCKET || 1576 fp->f_count == 0 || fp->f_unpcount != 0 || 1577 fp->f_count != fp->f_msgcount || 1578 (fp->f_flag & (FMARK | FSCAN)) != FSCAN) { 1579 mutex_exit(&fp->f_lock); 1580 continue; 1581 } 1582 1583 /* Gain file ref, mark our position, and unlock. */ 1584 LIST_INSERT_AFTER(fp, dp, f_list); 1585 fp->f_count++; 1586 mutex_exit(&fp->f_lock); 1587 mutex_exit(&filelist_lock); 1588 1589 /* 1590 * Flush all data from the socket's receive buffer. 1591 * This will cause files referenced only by the 1592 * socket to be queued for close. 1593 */ 1594 so = fp->f_data; 1595 solock(so); 1596 sorflush(so); 1597 sounlock(so); 1598 1599 /* Re-lock and restart from where we left off. */ 1600 closef(fp); 1601 mutex_enter(&filelist_lock); 1602 np = LIST_NEXT(dp, f_list); 1603 LIST_REMOVE(dp, f_list); 1604 } 1605 } 1606 1607 /* 1608 * Garbage collector thread. While SCM_RIGHTS messages are in transit, 1609 * wake once per second to garbage collect. Run continually while we 1610 * have deferred closes to process. 1611 */ 1612 static void 1613 unp_thread(void *cookie) 1614 { 1615 file_t *dp; 1616 1617 /* Allocate a dummy file for our scans. */ 1618 if ((dp = fgetdummy()) == NULL) { 1619 panic("unp_thread"); 1620 } 1621 1622 mutex_enter(&filelist_lock); 1623 for (;;) { 1624 KASSERT(mutex_owned(&filelist_lock)); 1625 if (SLIST_EMPTY(&unp_thread_discard)) { 1626 if (unp_rights != 0) { 1627 (void)cv_timedwait(&unp_thread_cv, 1628 &filelist_lock, hz); 1629 } else { 1630 cv_wait(&unp_thread_cv, &filelist_lock); 1631 } 1632 } 1633 unp_gc(dp); 1634 } 1635 /* NOTREACHED */ 1636 } 1637 1638 /* 1639 * Kick the garbage collector into action if there is something for 1640 * it to process. 1641 */ 1642 static void 1643 unp_thread_kick(void) 1644 { 1645 1646 if (!SLIST_EMPTY(&unp_thread_discard) || unp_rights != 0) { 1647 mutex_enter(&filelist_lock); 1648 cv_signal(&unp_thread_cv); 1649 mutex_exit(&filelist_lock); 1650 } 1651 } 1652 1653 void 1654 unp_dispose(struct mbuf *m) 1655 { 1656 1657 if (m) 1658 unp_scan(m, unp_discard_later, 1); 1659 } 1660 1661 void 1662 unp_scan(struct mbuf *m0, void (*op)(file_t *), int discard) 1663 { 1664 struct mbuf *m; 1665 file_t **rp, *fp; 1666 struct cmsghdr *cm; 1667 int i, qfds; 1668 1669 while (m0) { 1670 for (m = m0; m; m = m->m_next) { 1671 if (m->m_type != MT_CONTROL || 1672 m->m_len < sizeof(*cm)) { 1673 continue; 1674 } 1675 cm = mtod(m, struct cmsghdr *); 1676 if (cm->cmsg_level != SOL_SOCKET || 1677 cm->cmsg_type != SCM_RIGHTS) 1678 continue; 1679 qfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) 1680 / sizeof(file_t *); 1681 rp = (file_t **)CMSG_DATA(cm); 1682 for (i = 0; i < qfds; i++) { 1683 fp = *rp; 1684 if (discard) { 1685 *rp = 0; 1686 } 1687 (*op)(fp); 1688 rp++; 1689 } 1690 } 1691 m0 = m0->m_nextpkt; 1692 } 1693 } 1694 1695 void 1696 unp_mark(file_t *fp) 1697 { 1698 1699 if (fp == NULL) 1700 return; 1701 1702 /* If we're already deferred, don't screw up the defer count */ 1703 mutex_enter(&fp->f_lock); 1704 if (fp->f_flag & (FMARK | FDEFER)) { 1705 mutex_exit(&fp->f_lock); 1706 return; 1707 } 1708 1709 /* 1710 * Minimize the number of deferrals... Sockets are the only type of 1711 * file which can hold references to another file, so just mark 1712 * other files, and defer unmarked sockets for the next pass. 1713 */ 1714 if (fp->f_type == DTYPE_SOCKET) { 1715 unp_defer++; 1716 KASSERT(fp->f_count != 0); 1717 atomic_or_uint(&fp->f_flag, FDEFER); 1718 } else { 1719 atomic_or_uint(&fp->f_flag, FMARK); 1720 } 1721 mutex_exit(&fp->f_lock); 1722 } 1723 1724 static void 1725 unp_discard_now(file_t *fp) 1726 { 1727 1728 if (fp == NULL) 1729 return; 1730 1731 KASSERT(fp->f_count > 0); 1732 KASSERT(fp->f_msgcount > 0); 1733 1734 mutex_enter(&fp->f_lock); 1735 fp->f_msgcount--; 1736 mutex_exit(&fp->f_lock); 1737 atomic_dec_uint(&unp_rights); 1738 (void)closef(fp); 1739 } 1740 1741 static void 1742 unp_discard_later(file_t *fp) 1743 { 1744 1745 if (fp == NULL) 1746 return; 1747 1748 KASSERT(fp->f_count > 0); 1749 KASSERT(fp->f_msgcount > 0); 1750 1751 mutex_enter(&filelist_lock); 1752 if (fp->f_unpcount++ == 0) { 1753 SLIST_INSERT_HEAD(&unp_thread_discard, fp, f_unplist); 1754 } 1755 mutex_exit(&filelist_lock); 1756 } 1757