uipc_usrreq.c revision 1.156
1 /* $NetBSD: uipc_usrreq.c,v 1.156 2014/07/06 03:33:33 rtr Exp $ */ 2 3 /*- 4 * Copyright (c) 1998, 2000, 2004, 2008, 2009 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility, 9 * NASA Ames Research Center, and by Andrew Doran. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 30 * POSSIBILITY OF SUCH DAMAGE. 31 */ 32 33 /* 34 * Copyright (c) 1982, 1986, 1989, 1991, 1993 35 * The Regents of the University of California. All rights reserved. 36 * 37 * Redistribution and use in source and binary forms, with or without 38 * modification, are permitted provided that the following conditions 39 * are met: 40 * 1. Redistributions of source code must retain the above copyright 41 * notice, this list of conditions and the following disclaimer. 42 * 2. Redistributions in binary form must reproduce the above copyright 43 * notice, this list of conditions and the following disclaimer in the 44 * documentation and/or other materials provided with the distribution. 45 * 3. Neither the name of the University nor the names of its contributors 46 * may be used to endorse or promote products derived from this software 47 * without specific prior written permission. 48 * 49 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 50 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 51 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 52 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 53 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 54 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 55 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 56 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 57 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 58 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 59 * SUCH DAMAGE. 60 * 61 * @(#)uipc_usrreq.c 8.9 (Berkeley) 5/14/95 62 */ 63 64 /* 65 * Copyright (c) 1997 Christopher G. Demetriou. All rights reserved. 66 * 67 * Redistribution and use in source and binary forms, with or without 68 * modification, are permitted provided that the following conditions 69 * are met: 70 * 1. Redistributions of source code must retain the above copyright 71 * notice, this list of conditions and the following disclaimer. 72 * 2. Redistributions in binary form must reproduce the above copyright 73 * notice, this list of conditions and the following disclaimer in the 74 * documentation and/or other materials provided with the distribution. 75 * 3. All advertising materials mentioning features or use of this software 76 * must display the following acknowledgement: 77 * This product includes software developed by the University of 78 * California, Berkeley and its contributors. 79 * 4. Neither the name of the University nor the names of its contributors 80 * may be used to endorse or promote products derived from this software 81 * without specific prior written permission. 82 * 83 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 84 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 85 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 86 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 87 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 88 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 89 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 90 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 91 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 92 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 93 * SUCH DAMAGE. 94 * 95 * @(#)uipc_usrreq.c 8.9 (Berkeley) 5/14/95 96 */ 97 98 #include <sys/cdefs.h> 99 __KERNEL_RCSID(0, "$NetBSD: uipc_usrreq.c,v 1.156 2014/07/06 03:33:33 rtr Exp $"); 100 101 #include <sys/param.h> 102 #include <sys/systm.h> 103 #include <sys/proc.h> 104 #include <sys/filedesc.h> 105 #include <sys/domain.h> 106 #include <sys/protosw.h> 107 #include <sys/socket.h> 108 #include <sys/socketvar.h> 109 #include <sys/unpcb.h> 110 #include <sys/un.h> 111 #include <sys/namei.h> 112 #include <sys/vnode.h> 113 #include <sys/file.h> 114 #include <sys/stat.h> 115 #include <sys/mbuf.h> 116 #include <sys/kauth.h> 117 #include <sys/kmem.h> 118 #include <sys/atomic.h> 119 #include <sys/uidinfo.h> 120 #include <sys/kernel.h> 121 #include <sys/kthread.h> 122 123 /* 124 * Unix communications domain. 125 * 126 * TODO: 127 * RDM 128 * rethink name space problems 129 * need a proper out-of-band 130 * 131 * Notes on locking: 132 * 133 * The generic rules noted in uipc_socket2.c apply. In addition: 134 * 135 * o We have a global lock, uipc_lock. 136 * 137 * o All datagram sockets are locked by uipc_lock. 138 * 139 * o For stream socketpairs, the two endpoints are created sharing the same 140 * independent lock. Sockets presented to PRU_CONNECT2 must already have 141 * matching locks. 142 * 143 * o Stream sockets created via socket() start life with their own 144 * independent lock. 145 * 146 * o Stream connections to a named endpoint are slightly more complicated. 147 * Sockets that have called listen() have their lock pointer mutated to 148 * the global uipc_lock. When establishing a connection, the connecting 149 * socket also has its lock mutated to uipc_lock, which matches the head 150 * (listening socket). We create a new socket for accept() to return, and 151 * that also shares the head's lock. Until the connection is completely 152 * done on both ends, all three sockets are locked by uipc_lock. Once the 153 * connection is complete, the association with the head's lock is broken. 154 * The connecting socket and the socket returned from accept() have their 155 * lock pointers mutated away from uipc_lock, and back to the connecting 156 * socket's original, independent lock. The head continues to be locked 157 * by uipc_lock. 158 * 159 * o If uipc_lock is determined to be a significant source of contention, 160 * it could easily be hashed out. It is difficult to simply make it an 161 * independent lock because of visibility / garbage collection issues: 162 * if a socket has been associated with a lock at any point, that lock 163 * must remain valid until the socket is no longer visible in the system. 164 * The lock must not be freed or otherwise destroyed until any sockets 165 * that had referenced it have also been destroyed. 166 */ 167 const struct sockaddr_un sun_noname = { 168 .sun_len = offsetof(struct sockaddr_un, sun_path), 169 .sun_family = AF_LOCAL, 170 }; 171 ino_t unp_ino; /* prototype for fake inode numbers */ 172 173 static void unp_detach(struct socket *); 174 struct mbuf *unp_addsockcred(struct lwp *, struct mbuf *); 175 static void unp_mark(file_t *); 176 static void unp_scan(struct mbuf *, void (*)(file_t *), int); 177 static void unp_discard_now(file_t *); 178 static void unp_discard_later(file_t *); 179 static void unp_thread(void *); 180 static void unp_thread_kick(void); 181 static kmutex_t *uipc_lock; 182 183 static kcondvar_t unp_thread_cv; 184 static lwp_t *unp_thread_lwp; 185 static SLIST_HEAD(,file) unp_thread_discard; 186 static int unp_defer; 187 188 /* 189 * Initialize Unix protocols. 190 */ 191 void 192 uipc_init(void) 193 { 194 int error; 195 196 uipc_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 197 cv_init(&unp_thread_cv, "unpgc"); 198 199 error = kthread_create(PRI_NONE, KTHREAD_MPSAFE, NULL, unp_thread, 200 NULL, &unp_thread_lwp, "unpgc"); 201 if (error != 0) 202 panic("uipc_init %d", error); 203 } 204 205 /* 206 * A connection succeeded: disassociate both endpoints from the head's 207 * lock, and make them share their own lock. There is a race here: for 208 * a very brief time one endpoint will be locked by a different lock 209 * than the other end. However, since the current thread holds the old 210 * lock (the listening socket's lock, the head) access can still only be 211 * made to one side of the connection. 212 */ 213 static void 214 unp_setpeerlocks(struct socket *so, struct socket *so2) 215 { 216 struct unpcb *unp; 217 kmutex_t *lock; 218 219 KASSERT(solocked2(so, so2)); 220 221 /* 222 * Bail out if either end of the socket is not yet fully 223 * connected or accepted. We only break the lock association 224 * with the head when the pair of sockets stand completely 225 * on their own. 226 */ 227 KASSERT(so->so_head == NULL); 228 if (so2->so_head != NULL) 229 return; 230 231 /* 232 * Drop references to old lock. A third reference (from the 233 * queue head) must be held as we still hold its lock. Bonus: 234 * we don't need to worry about garbage collecting the lock. 235 */ 236 lock = so->so_lock; 237 KASSERT(lock == uipc_lock); 238 mutex_obj_free(lock); 239 mutex_obj_free(lock); 240 241 /* 242 * Grab stream lock from the initiator and share between the two 243 * endpoints. Issue memory barrier to ensure all modifications 244 * become globally visible before the lock change. so2 is 245 * assumed not to have a stream lock, because it was created 246 * purely for the server side to accept this connection and 247 * started out life using the domain-wide lock. 248 */ 249 unp = sotounpcb(so); 250 KASSERT(unp->unp_streamlock != NULL); 251 KASSERT(sotounpcb(so2)->unp_streamlock == NULL); 252 lock = unp->unp_streamlock; 253 unp->unp_streamlock = NULL; 254 mutex_obj_hold(lock); 255 membar_exit(); 256 /* 257 * possible race if lock is not held - see comment in 258 * uipc_usrreq(PRU_ACCEPT). 259 */ 260 KASSERT(mutex_owned(lock)); 261 solockreset(so, lock); 262 solockreset(so2, lock); 263 } 264 265 /* 266 * Reset a socket's lock back to the domain-wide lock. 267 */ 268 static void 269 unp_resetlock(struct socket *so) 270 { 271 kmutex_t *olock, *nlock; 272 struct unpcb *unp; 273 274 KASSERT(solocked(so)); 275 276 olock = so->so_lock; 277 nlock = uipc_lock; 278 if (olock == nlock) 279 return; 280 unp = sotounpcb(so); 281 KASSERT(unp->unp_streamlock == NULL); 282 unp->unp_streamlock = olock; 283 mutex_obj_hold(nlock); 284 mutex_enter(nlock); 285 solockreset(so, nlock); 286 mutex_exit(olock); 287 } 288 289 static void 290 unp_free(struct unpcb *unp) 291 { 292 if (unp->unp_addr) 293 free(unp->unp_addr, M_SONAME); 294 if (unp->unp_streamlock != NULL) 295 mutex_obj_free(unp->unp_streamlock); 296 kmem_free(unp, sizeof(*unp)); 297 } 298 299 int 300 unp_output(struct mbuf *m, struct mbuf *control, struct unpcb *unp, 301 struct lwp *l) 302 { 303 struct socket *so2; 304 const struct sockaddr_un *sun; 305 306 /* XXX: server side closed the socket */ 307 if (unp->unp_conn == NULL) 308 return ECONNREFUSED; 309 so2 = unp->unp_conn->unp_socket; 310 311 KASSERT(solocked(so2)); 312 313 if (unp->unp_addr) 314 sun = unp->unp_addr; 315 else 316 sun = &sun_noname; 317 if (unp->unp_conn->unp_flags & UNP_WANTCRED) 318 control = unp_addsockcred(l, control); 319 if (sbappendaddr(&so2->so_rcv, (const struct sockaddr *)sun, m, 320 control) == 0) { 321 so2->so_rcv.sb_overflowed++; 322 unp_dispose(control); 323 m_freem(control); 324 m_freem(m); 325 return (ENOBUFS); 326 } else { 327 sorwakeup(so2); 328 return (0); 329 } 330 } 331 332 void 333 unp_setaddr(struct socket *so, struct mbuf *nam, bool peeraddr) 334 { 335 const struct sockaddr_un *sun; 336 struct unpcb *unp; 337 bool ext; 338 339 KASSERT(solocked(so)); 340 unp = sotounpcb(so); 341 ext = false; 342 343 for (;;) { 344 sun = NULL; 345 if (peeraddr) { 346 if (unp->unp_conn && unp->unp_conn->unp_addr) 347 sun = unp->unp_conn->unp_addr; 348 } else { 349 if (unp->unp_addr) 350 sun = unp->unp_addr; 351 } 352 if (sun == NULL) 353 sun = &sun_noname; 354 nam->m_len = sun->sun_len; 355 if (nam->m_len > MLEN && !ext) { 356 sounlock(so); 357 MEXTMALLOC(nam, MAXPATHLEN * 2, M_WAITOK); 358 solock(so); 359 ext = true; 360 } else { 361 KASSERT(nam->m_len <= MAXPATHLEN * 2); 362 memcpy(mtod(nam, void *), sun, (size_t)nam->m_len); 363 break; 364 } 365 } 366 } 367 368 static int 369 unp_usrreq(struct socket *so, int req, struct mbuf *m, struct mbuf *nam, 370 struct mbuf *control, struct lwp *l) 371 { 372 struct unpcb *unp; 373 struct socket *so2; 374 u_int newhiwat; 375 int error = 0; 376 377 KASSERT(req != PRU_ATTACH); 378 KASSERT(req != PRU_DETACH); 379 KASSERT(req != PRU_CONTROL); 380 KASSERT(req != PRU_SENSE); 381 382 KASSERT(solocked(so)); 383 unp = sotounpcb(so); 384 385 KASSERT(!control || (req == PRU_SEND || req == PRU_SENDOOB)); 386 if (unp == NULL) { 387 error = EINVAL; 388 goto release; 389 } 390 391 switch (req) { 392 case PRU_BIND: 393 KASSERT(l != NULL); 394 error = unp_bind(so, nam, l); 395 break; 396 397 case PRU_LISTEN: 398 /* 399 * If the socket can accept a connection, it must be 400 * locked by uipc_lock. 401 */ 402 unp_resetlock(so); 403 if (unp->unp_vnode == NULL) 404 error = EINVAL; 405 break; 406 407 case PRU_CONNECT: 408 KASSERT(l != NULL); 409 error = unp_connect(so, nam, l); 410 break; 411 412 case PRU_CONNECT2: 413 error = unp_connect2(so, (struct socket *)nam, PRU_CONNECT2); 414 break; 415 416 case PRU_DISCONNECT: 417 unp_disconnect(unp); 418 break; 419 420 case PRU_ACCEPT: 421 KASSERT(so->so_lock == uipc_lock); 422 /* 423 * Mark the initiating STREAM socket as connected *ONLY* 424 * after it's been accepted. This prevents a client from 425 * overrunning a server and receiving ECONNREFUSED. 426 */ 427 if (unp->unp_conn == NULL) { 428 /* 429 * This will use the empty socket and will not 430 * allocate. 431 */ 432 unp_setaddr(so, nam, true); 433 break; 434 } 435 so2 = unp->unp_conn->unp_socket; 436 if (so2->so_state & SS_ISCONNECTING) { 437 KASSERT(solocked2(so, so->so_head)); 438 KASSERT(solocked2(so2, so->so_head)); 439 soisconnected(so2); 440 } 441 /* 442 * If the connection is fully established, break the 443 * association with uipc_lock and give the connected 444 * pair a separate lock to share. 445 * There is a race here: sotounpcb(so2)->unp_streamlock 446 * is not locked, so when changing so2->so_lock 447 * another thread can grab it while so->so_lock is still 448 * pointing to the (locked) uipc_lock. 449 * this should be harmless, except that this makes 450 * solocked2() and solocked() unreliable. 451 * Another problem is that unp_setaddr() expects the 452 * the socket locked. Grabing sotounpcb(so2)->unp_streamlock 453 * fixes both issues. 454 */ 455 mutex_enter(sotounpcb(so2)->unp_streamlock); 456 unp_setpeerlocks(so2, so); 457 /* 458 * Only now return peer's address, as we may need to 459 * block in order to allocate memory. 460 * 461 * XXX Minor race: connection can be broken while 462 * lock is dropped in unp_setaddr(). We will return 463 * error == 0 and sun_noname as the peer address. 464 */ 465 unp_setaddr(so, nam, true); 466 /* so_lock now points to unp_streamlock */ 467 mutex_exit(so2->so_lock); 468 break; 469 470 case PRU_SHUTDOWN: 471 socantsendmore(so); 472 unp_shutdown(unp); 473 break; 474 475 case PRU_RCVD: 476 switch (so->so_type) { 477 478 case SOCK_DGRAM: 479 panic("uipc 1"); 480 /*NOTREACHED*/ 481 482 case SOCK_SEQPACKET: /* FALLTHROUGH */ 483 case SOCK_STREAM: 484 #define rcv (&so->so_rcv) 485 #define snd (&so2->so_snd) 486 if (unp->unp_conn == 0) 487 break; 488 so2 = unp->unp_conn->unp_socket; 489 KASSERT(solocked2(so, so2)); 490 /* 491 * Adjust backpressure on sender 492 * and wakeup any waiting to write. 493 */ 494 snd->sb_mbmax += unp->unp_mbcnt - rcv->sb_mbcnt; 495 unp->unp_mbcnt = rcv->sb_mbcnt; 496 newhiwat = snd->sb_hiwat + unp->unp_cc - rcv->sb_cc; 497 (void)chgsbsize(so2->so_uidinfo, 498 &snd->sb_hiwat, newhiwat, RLIM_INFINITY); 499 unp->unp_cc = rcv->sb_cc; 500 sowwakeup(so2); 501 #undef snd 502 #undef rcv 503 break; 504 505 default: 506 panic("uipc 2"); 507 } 508 break; 509 510 case PRU_SEND: 511 /* 512 * Note: unp_internalize() rejects any control message 513 * other than SCM_RIGHTS, and only allows one. This 514 * has the side-effect of preventing a caller from 515 * forging SCM_CREDS. 516 */ 517 if (control) { 518 sounlock(so); 519 error = unp_internalize(&control); 520 solock(so); 521 if (error != 0) { 522 m_freem(control); 523 m_freem(m); 524 break; 525 } 526 } 527 switch (so->so_type) { 528 529 case SOCK_DGRAM: { 530 KASSERT(so->so_lock == uipc_lock); 531 if (nam) { 532 if ((so->so_state & SS_ISCONNECTED) != 0) 533 error = EISCONN; 534 else { 535 /* 536 * Note: once connected, the 537 * socket's lock must not be 538 * dropped until we have sent 539 * the message and disconnected. 540 * This is necessary to prevent 541 * intervening control ops, like 542 * another connection. 543 */ 544 error = unp_connect(so, nam, l); 545 } 546 } else { 547 if ((so->so_state & SS_ISCONNECTED) == 0) 548 error = ENOTCONN; 549 } 550 if (error) { 551 unp_dispose(control); 552 m_freem(control); 553 m_freem(m); 554 break; 555 } 556 KASSERT(l != NULL); 557 error = unp_output(m, control, unp, l); 558 if (nam) 559 unp_disconnect(unp); 560 break; 561 } 562 563 case SOCK_SEQPACKET: /* FALLTHROUGH */ 564 case SOCK_STREAM: 565 #define rcv (&so2->so_rcv) 566 #define snd (&so->so_snd) 567 if (unp->unp_conn == NULL) { 568 error = ENOTCONN; 569 break; 570 } 571 so2 = unp->unp_conn->unp_socket; 572 KASSERT(solocked2(so, so2)); 573 if (unp->unp_conn->unp_flags & UNP_WANTCRED) { 574 /* 575 * Credentials are passed only once on 576 * SOCK_STREAM and SOCK_SEQPACKET. 577 */ 578 unp->unp_conn->unp_flags &= ~UNP_WANTCRED; 579 control = unp_addsockcred(l, control); 580 } 581 /* 582 * Send to paired receive port, and then reduce 583 * send buffer hiwater marks to maintain backpressure. 584 * Wake up readers. 585 */ 586 if (control) { 587 if (sbappendcontrol(rcv, m, control) != 0) 588 control = NULL; 589 } else { 590 switch(so->so_type) { 591 case SOCK_SEQPACKET: 592 sbappendrecord(rcv, m); 593 break; 594 case SOCK_STREAM: 595 sbappend(rcv, m); 596 break; 597 default: 598 panic("uipc_usrreq"); 599 break; 600 } 601 } 602 snd->sb_mbmax -= 603 rcv->sb_mbcnt - unp->unp_conn->unp_mbcnt; 604 unp->unp_conn->unp_mbcnt = rcv->sb_mbcnt; 605 newhiwat = snd->sb_hiwat - 606 (rcv->sb_cc - unp->unp_conn->unp_cc); 607 (void)chgsbsize(so->so_uidinfo, 608 &snd->sb_hiwat, newhiwat, RLIM_INFINITY); 609 unp->unp_conn->unp_cc = rcv->sb_cc; 610 sorwakeup(so2); 611 #undef snd 612 #undef rcv 613 if (control != NULL) { 614 unp_dispose(control); 615 m_freem(control); 616 } 617 break; 618 619 default: 620 panic("uipc 4"); 621 } 622 break; 623 624 case PRU_ABORT: 625 (void)unp_drop(unp, ECONNABORTED); 626 KASSERT(so->so_head == NULL); 627 KASSERT(so->so_pcb != NULL); 628 unp_detach(so); 629 break; 630 631 case PRU_RCVOOB: 632 error = EOPNOTSUPP; 633 break; 634 635 case PRU_SENDOOB: 636 m_freem(control); 637 m_freem(m); 638 error = EOPNOTSUPP; 639 break; 640 641 case PRU_SOCKADDR: 642 unp_setaddr(so, nam, false); 643 break; 644 645 case PRU_PEERADDR: 646 unp_setaddr(so, nam, true); 647 break; 648 649 default: 650 panic("piusrreq"); 651 } 652 653 release: 654 return (error); 655 } 656 657 /* 658 * Unix domain socket option processing. 659 */ 660 int 661 uipc_ctloutput(int op, struct socket *so, struct sockopt *sopt) 662 { 663 struct unpcb *unp = sotounpcb(so); 664 int optval = 0, error = 0; 665 666 KASSERT(solocked(so)); 667 668 if (sopt->sopt_level != 0) { 669 error = ENOPROTOOPT; 670 } else switch (op) { 671 672 case PRCO_SETOPT: 673 switch (sopt->sopt_name) { 674 case LOCAL_CREDS: 675 case LOCAL_CONNWAIT: 676 error = sockopt_getint(sopt, &optval); 677 if (error) 678 break; 679 switch (sopt->sopt_name) { 680 #define OPTSET(bit) \ 681 if (optval) \ 682 unp->unp_flags |= (bit); \ 683 else \ 684 unp->unp_flags &= ~(bit); 685 686 case LOCAL_CREDS: 687 OPTSET(UNP_WANTCRED); 688 break; 689 case LOCAL_CONNWAIT: 690 OPTSET(UNP_CONNWAIT); 691 break; 692 } 693 break; 694 #undef OPTSET 695 696 default: 697 error = ENOPROTOOPT; 698 break; 699 } 700 break; 701 702 case PRCO_GETOPT: 703 sounlock(so); 704 switch (sopt->sopt_name) { 705 case LOCAL_PEEREID: 706 if (unp->unp_flags & UNP_EIDSVALID) { 707 error = sockopt_set(sopt, 708 &unp->unp_connid, sizeof(unp->unp_connid)); 709 } else { 710 error = EINVAL; 711 } 712 break; 713 case LOCAL_CREDS: 714 #define OPTBIT(bit) (unp->unp_flags & (bit) ? 1 : 0) 715 716 optval = OPTBIT(UNP_WANTCRED); 717 error = sockopt_setint(sopt, optval); 718 break; 719 #undef OPTBIT 720 721 default: 722 error = ENOPROTOOPT; 723 break; 724 } 725 solock(so); 726 break; 727 } 728 return (error); 729 } 730 731 /* 732 * Both send and receive buffers are allocated PIPSIZ bytes of buffering 733 * for stream sockets, although the total for sender and receiver is 734 * actually only PIPSIZ. 735 * Datagram sockets really use the sendspace as the maximum datagram size, 736 * and don't really want to reserve the sendspace. Their recvspace should 737 * be large enough for at least one max-size datagram plus address. 738 */ 739 #define PIPSIZ 4096 740 u_long unpst_sendspace = PIPSIZ; 741 u_long unpst_recvspace = PIPSIZ; 742 u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 743 u_long unpdg_recvspace = 4*1024; 744 745 u_int unp_rights; /* files in flight */ 746 u_int unp_rights_ratio = 2; /* limit, fraction of maxfiles */ 747 748 static int 749 unp_attach(struct socket *so, int proto) 750 { 751 struct unpcb *unp = sotounpcb(so); 752 u_long sndspc, rcvspc; 753 int error; 754 755 KASSERT(unp == NULL); 756 757 switch (so->so_type) { 758 case SOCK_SEQPACKET: 759 /* FALLTHROUGH */ 760 case SOCK_STREAM: 761 if (so->so_lock == NULL) { 762 so->so_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 763 solock(so); 764 } 765 sndspc = unpst_sendspace; 766 rcvspc = unpst_recvspace; 767 break; 768 769 case SOCK_DGRAM: 770 if (so->so_lock == NULL) { 771 mutex_obj_hold(uipc_lock); 772 so->so_lock = uipc_lock; 773 solock(so); 774 } 775 sndspc = unpdg_sendspace; 776 rcvspc = unpdg_recvspace; 777 break; 778 779 default: 780 panic("unp_attach"); 781 } 782 783 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 784 error = soreserve(so, sndspc, rcvspc); 785 if (error) { 786 return error; 787 } 788 } 789 790 unp = kmem_zalloc(sizeof(*unp), KM_SLEEP); 791 nanotime(&unp->unp_ctime); 792 unp->unp_socket = so; 793 so->so_pcb = unp; 794 795 KASSERT(solocked(so)); 796 return 0; 797 } 798 799 static void 800 unp_detach(struct socket *so) 801 { 802 struct unpcb *unp; 803 vnode_t *vp; 804 805 unp = sotounpcb(so); 806 KASSERT(unp != NULL); 807 KASSERT(solocked(so)); 808 retry: 809 if ((vp = unp->unp_vnode) != NULL) { 810 sounlock(so); 811 /* Acquire v_interlock to protect against unp_connect(). */ 812 /* XXXAD racy */ 813 mutex_enter(vp->v_interlock); 814 vp->v_socket = NULL; 815 mutex_exit(vp->v_interlock); 816 vrele(vp); 817 solock(so); 818 unp->unp_vnode = NULL; 819 } 820 if (unp->unp_conn) 821 unp_disconnect(unp); 822 while (unp->unp_refs) { 823 KASSERT(solocked2(so, unp->unp_refs->unp_socket)); 824 if (unp_drop(unp->unp_refs, ECONNRESET)) { 825 solock(so); 826 goto retry; 827 } 828 } 829 soisdisconnected(so); 830 so->so_pcb = NULL; 831 if (unp_rights) { 832 /* 833 * Normally the receive buffer is flushed later, in sofree, 834 * but if our receive buffer holds references to files that 835 * are now garbage, we will enqueue those file references to 836 * the garbage collector and kick it into action. 837 */ 838 sorflush(so); 839 unp_free(unp); 840 unp_thread_kick(); 841 } else 842 unp_free(unp); 843 } 844 845 static int 846 unp_ioctl(struct socket *so, u_long cmd, void *nam, struct ifnet *ifp) 847 { 848 return EOPNOTSUPP; 849 } 850 851 static int 852 unp_stat(struct socket *so, struct stat *ub) 853 { 854 struct unpcb *unp; 855 struct socket *so2; 856 857 unp = sotounpcb(so); 858 if (unp == NULL) 859 return EINVAL; 860 861 ub->st_blksize = so->so_snd.sb_hiwat; 862 switch (so->so_type) { 863 case SOCK_SEQPACKET: /* FALLTHROUGH */ 864 case SOCK_STREAM: 865 if (unp->unp_conn == 0) 866 break; 867 868 so2 = unp->unp_conn->unp_socket; 869 KASSERT(solocked2(so, so2)); 870 ub->st_blksize += so2->so_rcv.sb_cc; 871 break; 872 default: 873 break; 874 } 875 ub->st_dev = NODEV; 876 if (unp->unp_ino == 0) 877 unp->unp_ino = unp_ino++; 878 ub->st_atimespec = ub->st_mtimespec = ub->st_ctimespec = unp->unp_ctime; 879 ub->st_ino = unp->unp_ino; 880 return (0); 881 } 882 883 /* 884 * Allocate the new sockaddr. We have to allocate one 885 * extra byte so that we can ensure that the pathname 886 * is nul-terminated. Note that unlike linux, we don't 887 * include in the address length the NUL in the path 888 * component, because doing so, would exceed sizeof(sockaddr_un) 889 * for fully occupied pathnames. Linux is also inconsistent, 890 * because it does not include the NUL in the length of 891 * what it calls "abstract" unix sockets. 892 */ 893 static struct sockaddr_un * 894 makeun(struct mbuf *nam, size_t *addrlen) { 895 struct sockaddr_un *sun; 896 897 *addrlen = nam->m_len + 1; 898 sun = malloc(*addrlen, M_SONAME, M_WAITOK); 899 m_copydata(nam, 0, nam->m_len, (void *)sun); 900 *(((char *)sun) + nam->m_len) = '\0'; 901 sun->sun_len = strlen(sun->sun_path) + 902 offsetof(struct sockaddr_un, sun_path); 903 return sun; 904 } 905 906 int 907 unp_bind(struct socket *so, struct mbuf *nam, struct lwp *l) 908 { 909 struct sockaddr_un *sun; 910 struct unpcb *unp; 911 vnode_t *vp; 912 struct vattr vattr; 913 size_t addrlen; 914 int error; 915 struct pathbuf *pb; 916 struct nameidata nd; 917 proc_t *p; 918 919 unp = sotounpcb(so); 920 if (unp->unp_vnode != NULL) 921 return (EINVAL); 922 if ((unp->unp_flags & UNP_BUSY) != 0) { 923 /* 924 * EALREADY may not be strictly accurate, but since this 925 * is a major application error it's hardly a big deal. 926 */ 927 return (EALREADY); 928 } 929 unp->unp_flags |= UNP_BUSY; 930 sounlock(so); 931 932 p = l->l_proc; 933 sun = makeun(nam, &addrlen); 934 935 pb = pathbuf_create(sun->sun_path); 936 if (pb == NULL) { 937 error = ENOMEM; 938 goto bad; 939 } 940 NDINIT(&nd, CREATE, FOLLOW | LOCKPARENT | TRYEMULROOT, pb); 941 942 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 943 if ((error = namei(&nd)) != 0) { 944 pathbuf_destroy(pb); 945 goto bad; 946 } 947 vp = nd.ni_vp; 948 if (vp != NULL) { 949 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 950 if (nd.ni_dvp == vp) 951 vrele(nd.ni_dvp); 952 else 953 vput(nd.ni_dvp); 954 vrele(vp); 955 pathbuf_destroy(pb); 956 error = EADDRINUSE; 957 goto bad; 958 } 959 vattr_null(&vattr); 960 vattr.va_type = VSOCK; 961 vattr.va_mode = ACCESSPERMS & ~(p->p_cwdi->cwdi_cmask); 962 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 963 if (error) { 964 vput(nd.ni_dvp); 965 pathbuf_destroy(pb); 966 goto bad; 967 } 968 vp = nd.ni_vp; 969 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 970 solock(so); 971 vp->v_socket = unp->unp_socket; 972 unp->unp_vnode = vp; 973 unp->unp_addrlen = addrlen; 974 unp->unp_addr = sun; 975 unp->unp_connid.unp_pid = p->p_pid; 976 unp->unp_connid.unp_euid = kauth_cred_geteuid(l->l_cred); 977 unp->unp_connid.unp_egid = kauth_cred_getegid(l->l_cred); 978 unp->unp_flags |= UNP_EIDSBIND; 979 VOP_UNLOCK(vp); 980 vput(nd.ni_dvp); 981 unp->unp_flags &= ~UNP_BUSY; 982 pathbuf_destroy(pb); 983 return (0); 984 985 bad: 986 free(sun, M_SONAME); 987 solock(so); 988 unp->unp_flags &= ~UNP_BUSY; 989 return (error); 990 } 991 992 int 993 unp_connect(struct socket *so, struct mbuf *nam, struct lwp *l) 994 { 995 struct sockaddr_un *sun; 996 vnode_t *vp; 997 struct socket *so2, *so3; 998 struct unpcb *unp, *unp2, *unp3; 999 size_t addrlen; 1000 int error; 1001 struct pathbuf *pb; 1002 struct nameidata nd; 1003 1004 unp = sotounpcb(so); 1005 if ((unp->unp_flags & UNP_BUSY) != 0) { 1006 /* 1007 * EALREADY may not be strictly accurate, but since this 1008 * is a major application error it's hardly a big deal. 1009 */ 1010 return (EALREADY); 1011 } 1012 unp->unp_flags |= UNP_BUSY; 1013 sounlock(so); 1014 1015 sun = makeun(nam, &addrlen); 1016 pb = pathbuf_create(sun->sun_path); 1017 if (pb == NULL) { 1018 error = ENOMEM; 1019 goto bad2; 1020 } 1021 1022 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb); 1023 1024 if ((error = namei(&nd)) != 0) { 1025 pathbuf_destroy(pb); 1026 goto bad2; 1027 } 1028 vp = nd.ni_vp; 1029 if (vp->v_type != VSOCK) { 1030 error = ENOTSOCK; 1031 goto bad; 1032 } 1033 pathbuf_destroy(pb); 1034 if ((error = VOP_ACCESS(vp, VWRITE, l->l_cred)) != 0) 1035 goto bad; 1036 /* Acquire v_interlock to protect against unp_detach(). */ 1037 mutex_enter(vp->v_interlock); 1038 so2 = vp->v_socket; 1039 if (so2 == NULL) { 1040 mutex_exit(vp->v_interlock); 1041 error = ECONNREFUSED; 1042 goto bad; 1043 } 1044 if (so->so_type != so2->so_type) { 1045 mutex_exit(vp->v_interlock); 1046 error = EPROTOTYPE; 1047 goto bad; 1048 } 1049 solock(so); 1050 unp_resetlock(so); 1051 mutex_exit(vp->v_interlock); 1052 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) { 1053 /* 1054 * This may seem somewhat fragile but is OK: if we can 1055 * see SO_ACCEPTCONN set on the endpoint, then it must 1056 * be locked by the domain-wide uipc_lock. 1057 */ 1058 KASSERT((so2->so_options & SO_ACCEPTCONN) == 0 || 1059 so2->so_lock == uipc_lock); 1060 if ((so2->so_options & SO_ACCEPTCONN) == 0 || 1061 (so3 = sonewconn(so2, false)) == NULL) { 1062 error = ECONNREFUSED; 1063 sounlock(so); 1064 goto bad; 1065 } 1066 unp2 = sotounpcb(so2); 1067 unp3 = sotounpcb(so3); 1068 if (unp2->unp_addr) { 1069 unp3->unp_addr = malloc(unp2->unp_addrlen, 1070 M_SONAME, M_WAITOK); 1071 memcpy(unp3->unp_addr, unp2->unp_addr, 1072 unp2->unp_addrlen); 1073 unp3->unp_addrlen = unp2->unp_addrlen; 1074 } 1075 unp3->unp_flags = unp2->unp_flags; 1076 unp3->unp_connid.unp_pid = l->l_proc->p_pid; 1077 unp3->unp_connid.unp_euid = kauth_cred_geteuid(l->l_cred); 1078 unp3->unp_connid.unp_egid = kauth_cred_getegid(l->l_cred); 1079 unp3->unp_flags |= UNP_EIDSVALID; 1080 if (unp2->unp_flags & UNP_EIDSBIND) { 1081 unp->unp_connid = unp2->unp_connid; 1082 unp->unp_flags |= UNP_EIDSVALID; 1083 } 1084 so2 = so3; 1085 } 1086 error = unp_connect2(so, so2, PRU_CONNECT); 1087 sounlock(so); 1088 bad: 1089 vput(vp); 1090 bad2: 1091 free(sun, M_SONAME); 1092 solock(so); 1093 unp->unp_flags &= ~UNP_BUSY; 1094 return (error); 1095 } 1096 1097 int 1098 unp_connect2(struct socket *so, struct socket *so2, int req) 1099 { 1100 struct unpcb *unp = sotounpcb(so); 1101 struct unpcb *unp2; 1102 1103 if (so2->so_type != so->so_type) 1104 return (EPROTOTYPE); 1105 1106 /* 1107 * All three sockets involved must be locked by same lock: 1108 * 1109 * local endpoint (so) 1110 * remote endpoint (so2) 1111 * queue head (so2->so_head, only if PR_CONNREQUIRED) 1112 */ 1113 KASSERT(solocked2(so, so2)); 1114 KASSERT(so->so_head == NULL); 1115 if (so2->so_head != NULL) { 1116 KASSERT(so2->so_lock == uipc_lock); 1117 KASSERT(solocked2(so2, so2->so_head)); 1118 } 1119 1120 unp2 = sotounpcb(so2); 1121 unp->unp_conn = unp2; 1122 switch (so->so_type) { 1123 1124 case SOCK_DGRAM: 1125 unp->unp_nextref = unp2->unp_refs; 1126 unp2->unp_refs = unp; 1127 soisconnected(so); 1128 break; 1129 1130 case SOCK_SEQPACKET: /* FALLTHROUGH */ 1131 case SOCK_STREAM: 1132 unp2->unp_conn = unp; 1133 if (req == PRU_CONNECT && 1134 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 1135 soisconnecting(so); 1136 else 1137 soisconnected(so); 1138 soisconnected(so2); 1139 /* 1140 * If the connection is fully established, break the 1141 * association with uipc_lock and give the connected 1142 * pair a seperate lock to share. For CONNECT2, we 1143 * require that the locks already match (the sockets 1144 * are created that way). 1145 */ 1146 if (req == PRU_CONNECT) { 1147 KASSERT(so2->so_head != NULL); 1148 unp_setpeerlocks(so, so2); 1149 } 1150 break; 1151 1152 default: 1153 panic("unp_connect2"); 1154 } 1155 return (0); 1156 } 1157 1158 void 1159 unp_disconnect(struct unpcb *unp) 1160 { 1161 struct unpcb *unp2 = unp->unp_conn; 1162 struct socket *so; 1163 1164 if (unp2 == 0) 1165 return; 1166 unp->unp_conn = 0; 1167 so = unp->unp_socket; 1168 switch (so->so_type) { 1169 case SOCK_DGRAM: 1170 if (unp2->unp_refs == unp) 1171 unp2->unp_refs = unp->unp_nextref; 1172 else { 1173 unp2 = unp2->unp_refs; 1174 for (;;) { 1175 KASSERT(solocked2(so, unp2->unp_socket)); 1176 if (unp2 == 0) 1177 panic("unp_disconnect"); 1178 if (unp2->unp_nextref == unp) 1179 break; 1180 unp2 = unp2->unp_nextref; 1181 } 1182 unp2->unp_nextref = unp->unp_nextref; 1183 } 1184 unp->unp_nextref = 0; 1185 so->so_state &= ~SS_ISCONNECTED; 1186 break; 1187 1188 case SOCK_SEQPACKET: /* FALLTHROUGH */ 1189 case SOCK_STREAM: 1190 KASSERT(solocked2(so, unp2->unp_socket)); 1191 soisdisconnected(so); 1192 unp2->unp_conn = 0; 1193 soisdisconnected(unp2->unp_socket); 1194 break; 1195 } 1196 } 1197 1198 void 1199 unp_shutdown(struct unpcb *unp) 1200 { 1201 struct socket *so; 1202 1203 switch(unp->unp_socket->so_type) { 1204 case SOCK_SEQPACKET: /* FALLTHROUGH */ 1205 case SOCK_STREAM: 1206 if (unp->unp_conn && (so = unp->unp_conn->unp_socket)) 1207 socantrcvmore(so); 1208 break; 1209 default: 1210 break; 1211 } 1212 } 1213 1214 bool 1215 unp_drop(struct unpcb *unp, int errno) 1216 { 1217 struct socket *so = unp->unp_socket; 1218 1219 KASSERT(solocked(so)); 1220 1221 so->so_error = errno; 1222 unp_disconnect(unp); 1223 if (so->so_head) { 1224 so->so_pcb = NULL; 1225 /* sofree() drops the socket lock */ 1226 sofree(so); 1227 unp_free(unp); 1228 return true; 1229 } 1230 return false; 1231 } 1232 1233 #ifdef notdef 1234 unp_drain(void) 1235 { 1236 1237 } 1238 #endif 1239 1240 int 1241 unp_externalize(struct mbuf *rights, struct lwp *l, int flags) 1242 { 1243 struct cmsghdr * const cm = mtod(rights, struct cmsghdr *); 1244 struct proc * const p = l->l_proc; 1245 file_t **rp; 1246 int error = 0; 1247 1248 const size_t nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) / 1249 sizeof(file_t *); 1250 if (nfds == 0) 1251 goto noop; 1252 1253 int * const fdp = kmem_alloc(nfds * sizeof(int), KM_SLEEP); 1254 rw_enter(&p->p_cwdi->cwdi_lock, RW_READER); 1255 1256 /* Make sure the recipient should be able to see the files.. */ 1257 rp = (file_t **)CMSG_DATA(cm); 1258 for (size_t i = 0; i < nfds; i++) { 1259 file_t * const fp = *rp++; 1260 if (fp == NULL) { 1261 error = EINVAL; 1262 goto out; 1263 } 1264 /* 1265 * If we are in a chroot'ed directory, and 1266 * someone wants to pass us a directory, make 1267 * sure it's inside the subtree we're allowed 1268 * to access. 1269 */ 1270 if (p->p_cwdi->cwdi_rdir != NULL && fp->f_type == DTYPE_VNODE) { 1271 vnode_t *vp = (vnode_t *)fp->f_data; 1272 if ((vp->v_type == VDIR) && 1273 !vn_isunder(vp, p->p_cwdi->cwdi_rdir, l)) { 1274 error = EPERM; 1275 goto out; 1276 } 1277 } 1278 } 1279 1280 restart: 1281 /* 1282 * First loop -- allocate file descriptor table slots for the 1283 * new files. 1284 */ 1285 for (size_t i = 0; i < nfds; i++) { 1286 if ((error = fd_alloc(p, 0, &fdp[i])) != 0) { 1287 /* 1288 * Back out what we've done so far. 1289 */ 1290 while (i-- > 0) { 1291 fd_abort(p, NULL, fdp[i]); 1292 } 1293 if (error == ENOSPC) { 1294 fd_tryexpand(p); 1295 error = 0; 1296 goto restart; 1297 } 1298 /* 1299 * This is the error that has historically 1300 * been returned, and some callers may 1301 * expect it. 1302 */ 1303 error = EMSGSIZE; 1304 goto out; 1305 } 1306 } 1307 1308 /* 1309 * Now that adding them has succeeded, update all of the 1310 * file passing state and affix the descriptors. 1311 */ 1312 rp = (file_t **)CMSG_DATA(cm); 1313 int *ofdp = (int *)CMSG_DATA(cm); 1314 for (size_t i = 0; i < nfds; i++) { 1315 file_t * const fp = *rp++; 1316 const int fd = fdp[i]; 1317 atomic_dec_uint(&unp_rights); 1318 fd_set_exclose(l, fd, (flags & O_CLOEXEC) != 0); 1319 fd_affix(p, fp, fd); 1320 /* 1321 * Done with this file pointer, replace it with a fd; 1322 */ 1323 *ofdp++ = fd; 1324 mutex_enter(&fp->f_lock); 1325 fp->f_msgcount--; 1326 mutex_exit(&fp->f_lock); 1327 /* 1328 * Note that fd_affix() adds a reference to the file. 1329 * The file may already have been closed by another 1330 * LWP in the process, so we must drop the reference 1331 * added by unp_internalize() with closef(). 1332 */ 1333 closef(fp); 1334 } 1335 1336 /* 1337 * Adjust length, in case of transition from large file_t 1338 * pointers to ints. 1339 */ 1340 if (sizeof(file_t *) != sizeof(int)) { 1341 cm->cmsg_len = CMSG_LEN(nfds * sizeof(int)); 1342 rights->m_len = CMSG_SPACE(nfds * sizeof(int)); 1343 } 1344 out: 1345 if (__predict_false(error != 0)) { 1346 file_t **const fpp = (file_t **)CMSG_DATA(cm); 1347 for (size_t i = 0; i < nfds; i++) 1348 unp_discard_now(fpp[i]); 1349 /* 1350 * Truncate the array so that nobody will try to interpret 1351 * what is now garbage in it. 1352 */ 1353 cm->cmsg_len = CMSG_LEN(0); 1354 rights->m_len = CMSG_SPACE(0); 1355 } 1356 rw_exit(&p->p_cwdi->cwdi_lock); 1357 kmem_free(fdp, nfds * sizeof(int)); 1358 1359 noop: 1360 /* 1361 * Don't disclose kernel memory in the alignment space. 1362 */ 1363 KASSERT(cm->cmsg_len <= rights->m_len); 1364 memset(&mtod(rights, char *)[cm->cmsg_len], 0, rights->m_len - 1365 cm->cmsg_len); 1366 return error; 1367 } 1368 1369 int 1370 unp_internalize(struct mbuf **controlp) 1371 { 1372 filedesc_t *fdescp = curlwp->l_fd; 1373 struct mbuf *control = *controlp; 1374 struct cmsghdr *newcm, *cm = mtod(control, struct cmsghdr *); 1375 file_t **rp, **files; 1376 file_t *fp; 1377 int i, fd, *fdp; 1378 int nfds, error; 1379 u_int maxmsg; 1380 1381 error = 0; 1382 newcm = NULL; 1383 1384 /* Sanity check the control message header. */ 1385 if (cm->cmsg_type != SCM_RIGHTS || cm->cmsg_level != SOL_SOCKET || 1386 cm->cmsg_len > control->m_len || 1387 cm->cmsg_len < CMSG_ALIGN(sizeof(*cm))) 1388 return (EINVAL); 1389 1390 /* 1391 * Verify that the file descriptors are valid, and acquire 1392 * a reference to each. 1393 */ 1394 nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) / sizeof(int); 1395 fdp = (int *)CMSG_DATA(cm); 1396 maxmsg = maxfiles / unp_rights_ratio; 1397 for (i = 0; i < nfds; i++) { 1398 fd = *fdp++; 1399 if (atomic_inc_uint_nv(&unp_rights) > maxmsg) { 1400 atomic_dec_uint(&unp_rights); 1401 nfds = i; 1402 error = EAGAIN; 1403 goto out; 1404 } 1405 if ((fp = fd_getfile(fd)) == NULL 1406 || fp->f_type == DTYPE_KQUEUE) { 1407 if (fp) 1408 fd_putfile(fd); 1409 atomic_dec_uint(&unp_rights); 1410 nfds = i; 1411 error = EBADF; 1412 goto out; 1413 } 1414 } 1415 1416 /* Allocate new space and copy header into it. */ 1417 newcm = malloc(CMSG_SPACE(nfds * sizeof(file_t *)), M_MBUF, M_WAITOK); 1418 if (newcm == NULL) { 1419 error = E2BIG; 1420 goto out; 1421 } 1422 memcpy(newcm, cm, sizeof(struct cmsghdr)); 1423 files = (file_t **)CMSG_DATA(newcm); 1424 1425 /* 1426 * Transform the file descriptors into file_t pointers, in 1427 * reverse order so that if pointers are bigger than ints, the 1428 * int won't get until we're done. No need to lock, as we have 1429 * already validated the descriptors with fd_getfile(). 1430 */ 1431 fdp = (int *)CMSG_DATA(cm) + nfds; 1432 rp = files + nfds; 1433 for (i = 0; i < nfds; i++) { 1434 fp = fdescp->fd_dt->dt_ff[*--fdp]->ff_file; 1435 KASSERT(fp != NULL); 1436 mutex_enter(&fp->f_lock); 1437 *--rp = fp; 1438 fp->f_count++; 1439 fp->f_msgcount++; 1440 mutex_exit(&fp->f_lock); 1441 } 1442 1443 out: 1444 /* Release descriptor references. */ 1445 fdp = (int *)CMSG_DATA(cm); 1446 for (i = 0; i < nfds; i++) { 1447 fd_putfile(*fdp++); 1448 if (error != 0) { 1449 atomic_dec_uint(&unp_rights); 1450 } 1451 } 1452 1453 if (error == 0) { 1454 if (control->m_flags & M_EXT) { 1455 m_freem(control); 1456 *controlp = control = m_get(M_WAIT, MT_CONTROL); 1457 } 1458 MEXTADD(control, newcm, CMSG_SPACE(nfds * sizeof(file_t *)), 1459 M_MBUF, NULL, NULL); 1460 cm = newcm; 1461 /* 1462 * Adjust message & mbuf to note amount of space 1463 * actually used. 1464 */ 1465 cm->cmsg_len = CMSG_LEN(nfds * sizeof(file_t *)); 1466 control->m_len = CMSG_SPACE(nfds * sizeof(file_t *)); 1467 } 1468 1469 return error; 1470 } 1471 1472 struct mbuf * 1473 unp_addsockcred(struct lwp *l, struct mbuf *control) 1474 { 1475 struct sockcred *sc; 1476 struct mbuf *m; 1477 void *p; 1478 1479 m = sbcreatecontrol1(&p, SOCKCREDSIZE(kauth_cred_ngroups(l->l_cred)), 1480 SCM_CREDS, SOL_SOCKET, M_WAITOK); 1481 if (m == NULL) 1482 return control; 1483 1484 sc = p; 1485 sc->sc_uid = kauth_cred_getuid(l->l_cred); 1486 sc->sc_euid = kauth_cred_geteuid(l->l_cred); 1487 sc->sc_gid = kauth_cred_getgid(l->l_cred); 1488 sc->sc_egid = kauth_cred_getegid(l->l_cred); 1489 sc->sc_ngroups = kauth_cred_ngroups(l->l_cred); 1490 1491 for (int i = 0; i < sc->sc_ngroups; i++) 1492 sc->sc_groups[i] = kauth_cred_group(l->l_cred, i); 1493 1494 return m_add(control, m); 1495 } 1496 1497 /* 1498 * Do a mark-sweep GC of files in the system, to free up any which are 1499 * caught in flight to an about-to-be-closed socket. Additionally, 1500 * process deferred file closures. 1501 */ 1502 static void 1503 unp_gc(file_t *dp) 1504 { 1505 extern struct domain unixdomain; 1506 file_t *fp, *np; 1507 struct socket *so, *so1; 1508 u_int i, old, new; 1509 bool didwork; 1510 1511 KASSERT(curlwp == unp_thread_lwp); 1512 KASSERT(mutex_owned(&filelist_lock)); 1513 1514 /* 1515 * First, process deferred file closures. 1516 */ 1517 while (!SLIST_EMPTY(&unp_thread_discard)) { 1518 fp = SLIST_FIRST(&unp_thread_discard); 1519 KASSERT(fp->f_unpcount > 0); 1520 KASSERT(fp->f_count > 0); 1521 KASSERT(fp->f_msgcount > 0); 1522 KASSERT(fp->f_count >= fp->f_unpcount); 1523 KASSERT(fp->f_count >= fp->f_msgcount); 1524 KASSERT(fp->f_msgcount >= fp->f_unpcount); 1525 SLIST_REMOVE_HEAD(&unp_thread_discard, f_unplist); 1526 i = fp->f_unpcount; 1527 fp->f_unpcount = 0; 1528 mutex_exit(&filelist_lock); 1529 for (; i != 0; i--) { 1530 unp_discard_now(fp); 1531 } 1532 mutex_enter(&filelist_lock); 1533 } 1534 1535 /* 1536 * Clear mark bits. Ensure that we don't consider new files 1537 * entering the file table during this loop (they will not have 1538 * FSCAN set). 1539 */ 1540 unp_defer = 0; 1541 LIST_FOREACH(fp, &filehead, f_list) { 1542 for (old = fp->f_flag;; old = new) { 1543 new = atomic_cas_uint(&fp->f_flag, old, 1544 (old | FSCAN) & ~(FMARK|FDEFER)); 1545 if (__predict_true(old == new)) { 1546 break; 1547 } 1548 } 1549 } 1550 1551 /* 1552 * Iterate over the set of sockets, marking ones believed (based on 1553 * refcount) to be referenced from a process, and marking for rescan 1554 * sockets which are queued on a socket. Recan continues descending 1555 * and searching for sockets referenced by sockets (FDEFER), until 1556 * there are no more socket->socket references to be discovered. 1557 */ 1558 do { 1559 didwork = false; 1560 for (fp = LIST_FIRST(&filehead); fp != NULL; fp = np) { 1561 KASSERT(mutex_owned(&filelist_lock)); 1562 np = LIST_NEXT(fp, f_list); 1563 mutex_enter(&fp->f_lock); 1564 if ((fp->f_flag & FDEFER) != 0) { 1565 atomic_and_uint(&fp->f_flag, ~FDEFER); 1566 unp_defer--; 1567 KASSERT(fp->f_count != 0); 1568 } else { 1569 if (fp->f_count == 0 || 1570 (fp->f_flag & FMARK) != 0 || 1571 fp->f_count == fp->f_msgcount || 1572 fp->f_unpcount != 0) { 1573 mutex_exit(&fp->f_lock); 1574 continue; 1575 } 1576 } 1577 atomic_or_uint(&fp->f_flag, FMARK); 1578 1579 if (fp->f_type != DTYPE_SOCKET || 1580 (so = fp->f_data) == NULL || 1581 so->so_proto->pr_domain != &unixdomain || 1582 (so->so_proto->pr_flags & PR_RIGHTS) == 0) { 1583 mutex_exit(&fp->f_lock); 1584 continue; 1585 } 1586 1587 /* Gain file ref, mark our position, and unlock. */ 1588 didwork = true; 1589 LIST_INSERT_AFTER(fp, dp, f_list); 1590 fp->f_count++; 1591 mutex_exit(&fp->f_lock); 1592 mutex_exit(&filelist_lock); 1593 1594 /* 1595 * Mark files referenced from sockets queued on the 1596 * accept queue as well. 1597 */ 1598 solock(so); 1599 unp_scan(so->so_rcv.sb_mb, unp_mark, 0); 1600 if ((so->so_options & SO_ACCEPTCONN) != 0) { 1601 TAILQ_FOREACH(so1, &so->so_q0, so_qe) { 1602 unp_scan(so1->so_rcv.sb_mb, unp_mark, 0); 1603 } 1604 TAILQ_FOREACH(so1, &so->so_q, so_qe) { 1605 unp_scan(so1->so_rcv.sb_mb, unp_mark, 0); 1606 } 1607 } 1608 sounlock(so); 1609 1610 /* Re-lock and restart from where we left off. */ 1611 closef(fp); 1612 mutex_enter(&filelist_lock); 1613 np = LIST_NEXT(dp, f_list); 1614 LIST_REMOVE(dp, f_list); 1615 } 1616 /* 1617 * Bail early if we did nothing in the loop above. Could 1618 * happen because of concurrent activity causing unp_defer 1619 * to get out of sync. 1620 */ 1621 } while (unp_defer != 0 && didwork); 1622 1623 /* 1624 * Sweep pass. 1625 * 1626 * We grab an extra reference to each of the files that are 1627 * not otherwise accessible and then free the rights that are 1628 * stored in messages on them. 1629 */ 1630 for (fp = LIST_FIRST(&filehead); fp != NULL; fp = np) { 1631 KASSERT(mutex_owned(&filelist_lock)); 1632 np = LIST_NEXT(fp, f_list); 1633 mutex_enter(&fp->f_lock); 1634 1635 /* 1636 * Ignore non-sockets. 1637 * Ignore dead sockets, or sockets with pending close. 1638 * Ignore sockets obviously referenced elsewhere. 1639 * Ignore sockets marked as referenced by our scan. 1640 * Ignore new sockets that did not exist during the scan. 1641 */ 1642 if (fp->f_type != DTYPE_SOCKET || 1643 fp->f_count == 0 || fp->f_unpcount != 0 || 1644 fp->f_count != fp->f_msgcount || 1645 (fp->f_flag & (FMARK | FSCAN)) != FSCAN) { 1646 mutex_exit(&fp->f_lock); 1647 continue; 1648 } 1649 1650 /* Gain file ref, mark our position, and unlock. */ 1651 LIST_INSERT_AFTER(fp, dp, f_list); 1652 fp->f_count++; 1653 mutex_exit(&fp->f_lock); 1654 mutex_exit(&filelist_lock); 1655 1656 /* 1657 * Flush all data from the socket's receive buffer. 1658 * This will cause files referenced only by the 1659 * socket to be queued for close. 1660 */ 1661 so = fp->f_data; 1662 solock(so); 1663 sorflush(so); 1664 sounlock(so); 1665 1666 /* Re-lock and restart from where we left off. */ 1667 closef(fp); 1668 mutex_enter(&filelist_lock); 1669 np = LIST_NEXT(dp, f_list); 1670 LIST_REMOVE(dp, f_list); 1671 } 1672 } 1673 1674 /* 1675 * Garbage collector thread. While SCM_RIGHTS messages are in transit, 1676 * wake once per second to garbage collect. Run continually while we 1677 * have deferred closes to process. 1678 */ 1679 static void 1680 unp_thread(void *cookie) 1681 { 1682 file_t *dp; 1683 1684 /* Allocate a dummy file for our scans. */ 1685 if ((dp = fgetdummy()) == NULL) { 1686 panic("unp_thread"); 1687 } 1688 1689 mutex_enter(&filelist_lock); 1690 for (;;) { 1691 KASSERT(mutex_owned(&filelist_lock)); 1692 if (SLIST_EMPTY(&unp_thread_discard)) { 1693 if (unp_rights != 0) { 1694 (void)cv_timedwait(&unp_thread_cv, 1695 &filelist_lock, hz); 1696 } else { 1697 cv_wait(&unp_thread_cv, &filelist_lock); 1698 } 1699 } 1700 unp_gc(dp); 1701 } 1702 /* NOTREACHED */ 1703 } 1704 1705 /* 1706 * Kick the garbage collector into action if there is something for 1707 * it to process. 1708 */ 1709 static void 1710 unp_thread_kick(void) 1711 { 1712 1713 if (!SLIST_EMPTY(&unp_thread_discard) || unp_rights != 0) { 1714 mutex_enter(&filelist_lock); 1715 cv_signal(&unp_thread_cv); 1716 mutex_exit(&filelist_lock); 1717 } 1718 } 1719 1720 void 1721 unp_dispose(struct mbuf *m) 1722 { 1723 1724 if (m) 1725 unp_scan(m, unp_discard_later, 1); 1726 } 1727 1728 void 1729 unp_scan(struct mbuf *m0, void (*op)(file_t *), int discard) 1730 { 1731 struct mbuf *m; 1732 file_t **rp, *fp; 1733 struct cmsghdr *cm; 1734 int i, qfds; 1735 1736 while (m0) { 1737 for (m = m0; m; m = m->m_next) { 1738 if (m->m_type != MT_CONTROL || 1739 m->m_len < sizeof(*cm)) { 1740 continue; 1741 } 1742 cm = mtod(m, struct cmsghdr *); 1743 if (cm->cmsg_level != SOL_SOCKET || 1744 cm->cmsg_type != SCM_RIGHTS) 1745 continue; 1746 qfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) 1747 / sizeof(file_t *); 1748 rp = (file_t **)CMSG_DATA(cm); 1749 for (i = 0; i < qfds; i++) { 1750 fp = *rp; 1751 if (discard) { 1752 *rp = 0; 1753 } 1754 (*op)(fp); 1755 rp++; 1756 } 1757 } 1758 m0 = m0->m_nextpkt; 1759 } 1760 } 1761 1762 void 1763 unp_mark(file_t *fp) 1764 { 1765 1766 if (fp == NULL) 1767 return; 1768 1769 /* If we're already deferred, don't screw up the defer count */ 1770 mutex_enter(&fp->f_lock); 1771 if (fp->f_flag & (FMARK | FDEFER)) { 1772 mutex_exit(&fp->f_lock); 1773 return; 1774 } 1775 1776 /* 1777 * Minimize the number of deferrals... Sockets are the only type of 1778 * file which can hold references to another file, so just mark 1779 * other files, and defer unmarked sockets for the next pass. 1780 */ 1781 if (fp->f_type == DTYPE_SOCKET) { 1782 unp_defer++; 1783 KASSERT(fp->f_count != 0); 1784 atomic_or_uint(&fp->f_flag, FDEFER); 1785 } else { 1786 atomic_or_uint(&fp->f_flag, FMARK); 1787 } 1788 mutex_exit(&fp->f_lock); 1789 } 1790 1791 static void 1792 unp_discard_now(file_t *fp) 1793 { 1794 1795 if (fp == NULL) 1796 return; 1797 1798 KASSERT(fp->f_count > 0); 1799 KASSERT(fp->f_msgcount > 0); 1800 1801 mutex_enter(&fp->f_lock); 1802 fp->f_msgcount--; 1803 mutex_exit(&fp->f_lock); 1804 atomic_dec_uint(&unp_rights); 1805 (void)closef(fp); 1806 } 1807 1808 static void 1809 unp_discard_later(file_t *fp) 1810 { 1811 1812 if (fp == NULL) 1813 return; 1814 1815 KASSERT(fp->f_count > 0); 1816 KASSERT(fp->f_msgcount > 0); 1817 1818 mutex_enter(&filelist_lock); 1819 if (fp->f_unpcount++ == 0) { 1820 SLIST_INSERT_HEAD(&unp_thread_discard, fp, f_unplist); 1821 } 1822 mutex_exit(&filelist_lock); 1823 } 1824 1825 const struct pr_usrreqs unp_usrreqs = { 1826 .pr_attach = unp_attach, 1827 .pr_detach = unp_detach, 1828 .pr_ioctl = unp_ioctl, 1829 .pr_stat = unp_stat, 1830 .pr_generic = unp_usrreq, 1831 }; 1832
Indexes created Sun Sep 21 20:09:37 GMT 2025