uipc_usrreq.c revision 1.168
1 /* $NetBSD: uipc_usrreq.c,v 1.168 2014/08/08 03:05:45 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 1.168 2014/08/08 03:05:45 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 struct mbuf * unp_addsockcred(struct lwp *, struct mbuf *); 174 static void unp_discard_later(file_t *); 175 static void unp_discard_now(file_t *); 176 static void unp_disconnect1(struct unpcb *); 177 static bool unp_drop(struct unpcb *, int); 178 static int unp_internalize(struct mbuf **); 179 static void unp_mark(file_t *); 180 static void unp_scan(struct mbuf *, void (*)(file_t *), int); 181 static void unp_shutdown1(struct unpcb *); 182 static void unp_thread(void *); 183 static void unp_thread_kick(void); 184 185 static kmutex_t *uipc_lock; 186 187 static kcondvar_t unp_thread_cv; 188 static lwp_t *unp_thread_lwp; 189 static SLIST_HEAD(,file) unp_thread_discard; 190 static int unp_defer; 191 192 /* 193 * Initialize Unix protocols. 194 */ 195 void 196 uipc_init(void) 197 { 198 int error; 199 200 uipc_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 201 cv_init(&unp_thread_cv, "unpgc"); 202 203 error = kthread_create(PRI_NONE, KTHREAD_MPSAFE, NULL, unp_thread, 204 NULL, &unp_thread_lwp, "unpgc"); 205 if (error != 0) 206 panic("uipc_init %d", error); 207 } 208 209 /* 210 * A connection succeeded: disassociate both endpoints from the head's 211 * lock, and make them share their own lock. There is a race here: for 212 * a very brief time one endpoint will be locked by a different lock 213 * than the other end. However, since the current thread holds the old 214 * lock (the listening socket's lock, the head) access can still only be 215 * made to one side of the connection. 216 */ 217 static void 218 unp_setpeerlocks(struct socket *so, struct socket *so2) 219 { 220 struct unpcb *unp; 221 kmutex_t *lock; 222 223 KASSERT(solocked2(so, so2)); 224 225 /* 226 * Bail out if either end of the socket is not yet fully 227 * connected or accepted. We only break the lock association 228 * with the head when the pair of sockets stand completely 229 * on their own. 230 */ 231 KASSERT(so->so_head == NULL); 232 if (so2->so_head != NULL) 233 return; 234 235 /* 236 * Drop references to old lock. A third reference (from the 237 * queue head) must be held as we still hold its lock. Bonus: 238 * we don't need to worry about garbage collecting the lock. 239 */ 240 lock = so->so_lock; 241 KASSERT(lock == uipc_lock); 242 mutex_obj_free(lock); 243 mutex_obj_free(lock); 244 245 /* 246 * Grab stream lock from the initiator and share between the two 247 * endpoints. Issue memory barrier to ensure all modifications 248 * become globally visible before the lock change. so2 is 249 * assumed not to have a stream lock, because it was created 250 * purely for the server side to accept this connection and 251 * started out life using the domain-wide lock. 252 */ 253 unp = sotounpcb(so); 254 KASSERT(unp->unp_streamlock != NULL); 255 KASSERT(sotounpcb(so2)->unp_streamlock == NULL); 256 lock = unp->unp_streamlock; 257 unp->unp_streamlock = NULL; 258 mutex_obj_hold(lock); 259 membar_exit(); 260 /* 261 * possible race if lock is not held - see comment in 262 * uipc_usrreq(PRU_ACCEPT). 263 */ 264 KASSERT(mutex_owned(lock)); 265 solockreset(so, lock); 266 solockreset(so2, lock); 267 } 268 269 /* 270 * Reset a socket's lock back to the domain-wide lock. 271 */ 272 static void 273 unp_resetlock(struct socket *so) 274 { 275 kmutex_t *olock, *nlock; 276 struct unpcb *unp; 277 278 KASSERT(solocked(so)); 279 280 olock = so->so_lock; 281 nlock = uipc_lock; 282 if (olock == nlock) 283 return; 284 unp = sotounpcb(so); 285 KASSERT(unp->unp_streamlock == NULL); 286 unp->unp_streamlock = olock; 287 mutex_obj_hold(nlock); 288 mutex_enter(nlock); 289 solockreset(so, nlock); 290 mutex_exit(olock); 291 } 292 293 static void 294 unp_free(struct unpcb *unp) 295 { 296 if (unp->unp_addr) 297 free(unp->unp_addr, M_SONAME); 298 if (unp->unp_streamlock != NULL) 299 mutex_obj_free(unp->unp_streamlock); 300 kmem_free(unp, sizeof(*unp)); 301 } 302 303 static int 304 unp_output(struct mbuf *m, struct mbuf *control, struct unpcb *unp) 305 { 306 struct socket *so2; 307 const struct sockaddr_un *sun; 308 309 /* XXX: server side closed the socket */ 310 if (unp->unp_conn == NULL) 311 return ECONNREFUSED; 312 so2 = unp->unp_conn->unp_socket; 313 314 KASSERT(solocked(so2)); 315 316 if (unp->unp_addr) 317 sun = unp->unp_addr; 318 else 319 sun = &sun_noname; 320 if (unp->unp_conn->unp_flags & UNP_WANTCRED) 321 control = unp_addsockcred(curlwp, control); 322 if (sbappendaddr(&so2->so_rcv, (const struct sockaddr *)sun, m, 323 control) == 0) { 324 so2->so_rcv.sb_overflowed++; 325 unp_dispose(control); 326 m_freem(control); 327 m_freem(m); 328 return (ENOBUFS); 329 } else { 330 sorwakeup(so2); 331 return (0); 332 } 333 } 334 335 static void 336 unp_setaddr(struct socket *so, struct mbuf *nam, bool peeraddr) 337 { 338 const struct sockaddr_un *sun; 339 struct unpcb *unp; 340 bool ext; 341 342 KASSERT(solocked(so)); 343 unp = sotounpcb(so); 344 ext = false; 345 346 for (;;) { 347 sun = NULL; 348 if (peeraddr) { 349 if (unp->unp_conn && unp->unp_conn->unp_addr) 350 sun = unp->unp_conn->unp_addr; 351 } else { 352 if (unp->unp_addr) 353 sun = unp->unp_addr; 354 } 355 if (sun == NULL) 356 sun = &sun_noname; 357 nam->m_len = sun->sun_len; 358 if (nam->m_len > MLEN && !ext) { 359 sounlock(so); 360 MEXTMALLOC(nam, MAXPATHLEN * 2, M_WAITOK); 361 solock(so); 362 ext = true; 363 } else { 364 KASSERT(nam->m_len <= MAXPATHLEN * 2); 365 memcpy(mtod(nam, void *), sun, (size_t)nam->m_len); 366 break; 367 } 368 } 369 } 370 371 static int 372 unp_rcvd(struct socket *so, int flags, struct lwp *l) 373 { 374 struct unpcb *unp = sotounpcb(so); 375 struct socket *so2; 376 u_int newhiwat; 377 378 KASSERT(solocked(so)); 379 KASSERT(unp != NULL); 380 381 switch (so->so_type) { 382 383 case SOCK_DGRAM: 384 panic("uipc 1"); 385 /*NOTREACHED*/ 386 387 case SOCK_SEQPACKET: /* FALLTHROUGH */ 388 case SOCK_STREAM: 389 #define rcv (&so->so_rcv) 390 #define snd (&so2->so_snd) 391 if (unp->unp_conn == 0) 392 break; 393 so2 = unp->unp_conn->unp_socket; 394 KASSERT(solocked2(so, so2)); 395 /* 396 * Adjust backpressure on sender 397 * and wakeup any waiting to write. 398 */ 399 snd->sb_mbmax += unp->unp_mbcnt - rcv->sb_mbcnt; 400 unp->unp_mbcnt = rcv->sb_mbcnt; 401 newhiwat = snd->sb_hiwat + unp->unp_cc - rcv->sb_cc; 402 (void)chgsbsize(so2->so_uidinfo, 403 &snd->sb_hiwat, newhiwat, RLIM_INFINITY); 404 unp->unp_cc = rcv->sb_cc; 405 sowwakeup(so2); 406 #undef snd 407 #undef rcv 408 break; 409 410 default: 411 panic("uipc 2"); 412 } 413 414 return 0; 415 } 416 417 static int 418 unp_recvoob(struct socket *so, struct mbuf *m, int flags) 419 { 420 KASSERT(solocked(so)); 421 422 return EOPNOTSUPP; 423 } 424 425 static int 426 unp_send(struct socket *so, struct mbuf *m, struct mbuf *nam, 427 struct mbuf *control, struct lwp *l) 428 { 429 struct unpcb *unp = sotounpcb(so); 430 int error = 0; 431 u_int newhiwat; 432 struct socket *so2; 433 434 KASSERT(solocked(so)); 435 KASSERT(unp != NULL); 436 KASSERT(m != NULL); 437 438 /* 439 * Note: unp_internalize() rejects any control message 440 * other than SCM_RIGHTS, and only allows one. This 441 * has the side-effect of preventing a caller from 442 * forging SCM_CREDS. 443 */ 444 if (control) { 445 sounlock(so); 446 error = unp_internalize(&control); 447 solock(so); 448 if (error != 0) { 449 m_freem(control); 450 m_freem(m); 451 return error; 452 } 453 } 454 455 switch (so->so_type) { 456 457 case SOCK_DGRAM: { 458 KASSERT(so->so_lock == uipc_lock); 459 if (nam) { 460 if ((so->so_state & SS_ISCONNECTED) != 0) 461 error = EISCONN; 462 else { 463 /* 464 * Note: once connected, the 465 * socket's lock must not be 466 * dropped until we have sent 467 * the message and disconnected. 468 * This is necessary to prevent 469 * intervening control ops, like 470 * another connection. 471 */ 472 error = unp_connect(so, nam, l); 473 } 474 } else { 475 if ((so->so_state & SS_ISCONNECTED) == 0) 476 error = ENOTCONN; 477 } 478 if (error) { 479 unp_dispose(control); 480 m_freem(control); 481 m_freem(m); 482 return error; 483 } 484 error = unp_output(m, control, unp); 485 if (nam) 486 unp_disconnect1(unp); 487 break; 488 } 489 490 case SOCK_SEQPACKET: /* FALLTHROUGH */ 491 case SOCK_STREAM: 492 #define rcv (&so2->so_rcv) 493 #define snd (&so->so_snd) 494 if (unp->unp_conn == NULL) { 495 error = ENOTCONN; 496 break; 497 } 498 so2 = unp->unp_conn->unp_socket; 499 KASSERT(solocked2(so, so2)); 500 if (unp->unp_conn->unp_flags & UNP_WANTCRED) { 501 /* 502 * Credentials are passed only once on 503 * SOCK_STREAM and SOCK_SEQPACKET. 504 */ 505 unp->unp_conn->unp_flags &= ~UNP_WANTCRED; 506 control = unp_addsockcred(l, control); 507 } 508 /* 509 * Send to paired receive port, and then reduce 510 * send buffer hiwater marks to maintain backpressure. 511 * Wake up readers. 512 */ 513 if (control) { 514 if (sbappendcontrol(rcv, m, control) != 0) 515 control = NULL; 516 } else { 517 switch(so->so_type) { 518 case SOCK_SEQPACKET: 519 sbappendrecord(rcv, m); 520 break; 521 case SOCK_STREAM: 522 sbappend(rcv, m); 523 break; 524 default: 525 panic("uipc_usrreq"); 526 break; 527 } 528 } 529 snd->sb_mbmax -= 530 rcv->sb_mbcnt - unp->unp_conn->unp_mbcnt; 531 unp->unp_conn->unp_mbcnt = rcv->sb_mbcnt; 532 newhiwat = snd->sb_hiwat - 533 (rcv->sb_cc - unp->unp_conn->unp_cc); 534 (void)chgsbsize(so->so_uidinfo, 535 &snd->sb_hiwat, newhiwat, RLIM_INFINITY); 536 unp->unp_conn->unp_cc = rcv->sb_cc; 537 sorwakeup(so2); 538 #undef snd 539 #undef rcv 540 if (control != NULL) { 541 unp_dispose(control); 542 m_freem(control); 543 } 544 break; 545 546 default: 547 panic("uipc 4"); 548 } 549 550 return error; 551 } 552 553 static int 554 unp_sendoob(struct socket *so, struct mbuf *m, struct mbuf * control) 555 { 556 KASSERT(solocked(so)); 557 558 m_freem(m); 559 m_freem(control); 560 561 return EOPNOTSUPP; 562 } 563 564 static int 565 unp_usrreq(struct socket *so, int req, struct mbuf *m, struct mbuf *nam, 566 struct mbuf *control, struct lwp *l) 567 { 568 struct unpcb *unp; 569 int error = 0; 570 571 KASSERT(req != PRU_ATTACH); 572 KASSERT(req != PRU_DETACH); 573 KASSERT(req != PRU_ACCEPT); 574 KASSERT(req != PRU_BIND); 575 KASSERT(req != PRU_LISTEN); 576 KASSERT(req != PRU_CONNECT); 577 KASSERT(req != PRU_DISCONNECT); 578 KASSERT(req != PRU_SHUTDOWN); 579 KASSERT(req != PRU_ABORT); 580 KASSERT(req != PRU_CONTROL); 581 KASSERT(req != PRU_SENSE); 582 KASSERT(req != PRU_PEERADDR); 583 KASSERT(req != PRU_SOCKADDR); 584 KASSERT(req != PRU_RCVD); 585 KASSERT(req != PRU_RCVOOB); 586 KASSERT(req != PRU_SEND); 587 KASSERT(req != PRU_SENDOOB); 588 589 KASSERT(solocked(so)); 590 unp = sotounpcb(so); 591 592 KASSERT(!control); 593 if (unp == NULL) { 594 error = EINVAL; 595 goto release; 596 } 597 598 switch (req) { 599 case PRU_CONNECT2: 600 error = unp_connect2(so, (struct socket *)nam, PRU_CONNECT2); 601 break; 602 603 default: 604 panic("piusrreq"); 605 } 606 607 release: 608 return (error); 609 } 610 611 /* 612 * Unix domain socket option processing. 613 */ 614 int 615 uipc_ctloutput(int op, struct socket *so, struct sockopt *sopt) 616 { 617 struct unpcb *unp = sotounpcb(so); 618 int optval = 0, error = 0; 619 620 KASSERT(solocked(so)); 621 622 if (sopt->sopt_level != 0) { 623 error = ENOPROTOOPT; 624 } else switch (op) { 625 626 case PRCO_SETOPT: 627 switch (sopt->sopt_name) { 628 case LOCAL_CREDS: 629 case LOCAL_CONNWAIT: 630 error = sockopt_getint(sopt, &optval); 631 if (error) 632 break; 633 switch (sopt->sopt_name) { 634 #define OPTSET(bit) \ 635 if (optval) \ 636 unp->unp_flags |= (bit); \ 637 else \ 638 unp->unp_flags &= ~(bit); 639 640 case LOCAL_CREDS: 641 OPTSET(UNP_WANTCRED); 642 break; 643 case LOCAL_CONNWAIT: 644 OPTSET(UNP_CONNWAIT); 645 break; 646 } 647 break; 648 #undef OPTSET 649 650 default: 651 error = ENOPROTOOPT; 652 break; 653 } 654 break; 655 656 case PRCO_GETOPT: 657 sounlock(so); 658 switch (sopt->sopt_name) { 659 case LOCAL_PEEREID: 660 if (unp->unp_flags & UNP_EIDSVALID) { 661 error = sockopt_set(sopt, 662 &unp->unp_connid, sizeof(unp->unp_connid)); 663 } else { 664 error = EINVAL; 665 } 666 break; 667 case LOCAL_CREDS: 668 #define OPTBIT(bit) (unp->unp_flags & (bit) ? 1 : 0) 669 670 optval = OPTBIT(UNP_WANTCRED); 671 error = sockopt_setint(sopt, optval); 672 break; 673 #undef OPTBIT 674 675 default: 676 error = ENOPROTOOPT; 677 break; 678 } 679 solock(so); 680 break; 681 } 682 return (error); 683 } 684 685 /* 686 * Both send and receive buffers are allocated PIPSIZ bytes of buffering 687 * for stream sockets, although the total for sender and receiver is 688 * actually only PIPSIZ. 689 * Datagram sockets really use the sendspace as the maximum datagram size, 690 * and don't really want to reserve the sendspace. Their recvspace should 691 * be large enough for at least one max-size datagram plus address. 692 */ 693 #define PIPSIZ 4096 694 u_long unpst_sendspace = PIPSIZ; 695 u_long unpst_recvspace = PIPSIZ; 696 u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 697 u_long unpdg_recvspace = 4*1024; 698 699 u_int unp_rights; /* files in flight */ 700 u_int unp_rights_ratio = 2; /* limit, fraction of maxfiles */ 701 702 static int 703 unp_attach(struct socket *so, int proto) 704 { 705 struct unpcb *unp = sotounpcb(so); 706 u_long sndspc, rcvspc; 707 int error; 708 709 KASSERT(unp == NULL); 710 711 switch (so->so_type) { 712 case SOCK_SEQPACKET: 713 /* FALLTHROUGH */ 714 case SOCK_STREAM: 715 if (so->so_lock == NULL) { 716 so->so_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 717 solock(so); 718 } 719 sndspc = unpst_sendspace; 720 rcvspc = unpst_recvspace; 721 break; 722 723 case SOCK_DGRAM: 724 if (so->so_lock == NULL) { 725 mutex_obj_hold(uipc_lock); 726 so->so_lock = uipc_lock; 727 solock(so); 728 } 729 sndspc = unpdg_sendspace; 730 rcvspc = unpdg_recvspace; 731 break; 732 733 default: 734 panic("unp_attach"); 735 } 736 737 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 738 error = soreserve(so, sndspc, rcvspc); 739 if (error) { 740 return error; 741 } 742 } 743 744 unp = kmem_zalloc(sizeof(*unp), KM_SLEEP); 745 nanotime(&unp->unp_ctime); 746 unp->unp_socket = so; 747 so->so_pcb = unp; 748 749 KASSERT(solocked(so)); 750 return 0; 751 } 752 753 static void 754 unp_detach(struct socket *so) 755 { 756 struct unpcb *unp; 757 vnode_t *vp; 758 759 unp = sotounpcb(so); 760 KASSERT(unp != NULL); 761 KASSERT(solocked(so)); 762 retry: 763 if ((vp = unp->unp_vnode) != NULL) { 764 sounlock(so); 765 /* Acquire v_interlock to protect against unp_connect(). */ 766 /* XXXAD racy */ 767 mutex_enter(vp->v_interlock); 768 vp->v_socket = NULL; 769 mutex_exit(vp->v_interlock); 770 vrele(vp); 771 solock(so); 772 unp->unp_vnode = NULL; 773 } 774 if (unp->unp_conn) 775 unp_disconnect1(unp); 776 while (unp->unp_refs) { 777 KASSERT(solocked2(so, unp->unp_refs->unp_socket)); 778 if (unp_drop(unp->unp_refs, ECONNRESET)) { 779 solock(so); 780 goto retry; 781 } 782 } 783 soisdisconnected(so); 784 so->so_pcb = NULL; 785 if (unp_rights) { 786 /* 787 * Normally the receive buffer is flushed later, in sofree, 788 * but if our receive buffer holds references to files that 789 * are now garbage, we will enqueue those file references to 790 * the garbage collector and kick it into action. 791 */ 792 sorflush(so); 793 unp_free(unp); 794 unp_thread_kick(); 795 } else 796 unp_free(unp); 797 } 798 799 static int 800 unp_accept(struct socket *so, struct mbuf *nam) 801 { 802 struct unpcb *unp = sotounpcb(so); 803 struct socket *so2; 804 805 KASSERT(solocked(so)); 806 KASSERT(nam != NULL); 807 808 /* XXX code review required to determine if unp can ever be NULL */ 809 if (unp == NULL) 810 return EINVAL; 811 812 KASSERT(so->so_lock == uipc_lock); 813 /* 814 * Mark the initiating STREAM socket as connected *ONLY* 815 * after it's been accepted. This prevents a client from 816 * overrunning a server and receiving ECONNREFUSED. 817 */ 818 if (unp->unp_conn == NULL) { 819 /* 820 * This will use the empty socket and will not 821 * allocate. 822 */ 823 unp_setaddr(so, nam, true); 824 return 0; 825 } 826 so2 = unp->unp_conn->unp_socket; 827 if (so2->so_state & SS_ISCONNECTING) { 828 KASSERT(solocked2(so, so->so_head)); 829 KASSERT(solocked2(so2, so->so_head)); 830 soisconnected(so2); 831 } 832 /* 833 * If the connection is fully established, break the 834 * association with uipc_lock and give the connected 835 * pair a separate lock to share. 836 * There is a race here: sotounpcb(so2)->unp_streamlock 837 * is not locked, so when changing so2->so_lock 838 * another thread can grab it while so->so_lock is still 839 * pointing to the (locked) uipc_lock. 840 * this should be harmless, except that this makes 841 * solocked2() and solocked() unreliable. 842 * Another problem is that unp_setaddr() expects the 843 * the socket locked. Grabing sotounpcb(so2)->unp_streamlock 844 * fixes both issues. 845 */ 846 mutex_enter(sotounpcb(so2)->unp_streamlock); 847 unp_setpeerlocks(so2, so); 848 /* 849 * Only now return peer's address, as we may need to 850 * block in order to allocate memory. 851 * 852 * XXX Minor race: connection can be broken while 853 * lock is dropped in unp_setaddr(). We will return 854 * error == 0 and sun_noname as the peer address. 855 */ 856 unp_setaddr(so, nam, true); 857 /* so_lock now points to unp_streamlock */ 858 mutex_exit(so2->so_lock); 859 return 0; 860 } 861 862 static int 863 unp_ioctl(struct socket *so, u_long cmd, void *nam, struct ifnet *ifp) 864 { 865 return EOPNOTSUPP; 866 } 867 868 static int 869 unp_stat(struct socket *so, struct stat *ub) 870 { 871 struct unpcb *unp; 872 struct socket *so2; 873 874 KASSERT(solocked(so)); 875 876 unp = sotounpcb(so); 877 if (unp == NULL) 878 return EINVAL; 879 880 ub->st_blksize = so->so_snd.sb_hiwat; 881 switch (so->so_type) { 882 case SOCK_SEQPACKET: /* FALLTHROUGH */ 883 case SOCK_STREAM: 884 if (unp->unp_conn == 0) 885 break; 886 887 so2 = unp->unp_conn->unp_socket; 888 KASSERT(solocked2(so, so2)); 889 ub->st_blksize += so2->so_rcv.sb_cc; 890 break; 891 default: 892 break; 893 } 894 ub->st_dev = NODEV; 895 if (unp->unp_ino == 0) 896 unp->unp_ino = unp_ino++; 897 ub->st_atimespec = ub->st_mtimespec = ub->st_ctimespec = unp->unp_ctime; 898 ub->st_ino = unp->unp_ino; 899 return (0); 900 } 901 902 static int 903 unp_peeraddr(struct socket *so, struct mbuf *nam) 904 { 905 KASSERT(solocked(so)); 906 KASSERT(sotounpcb(so) != NULL); 907 KASSERT(nam != NULL); 908 909 unp_setaddr(so, nam, true); 910 return 0; 911 } 912 913 static int 914 unp_sockaddr(struct socket *so, struct mbuf *nam) 915 { 916 KASSERT(solocked(so)); 917 KASSERT(sotounpcb(so) != NULL); 918 KASSERT(nam != NULL); 919 920 unp_setaddr(so, nam, false); 921 return 0; 922 } 923 924 /* 925 * Allocate the new sockaddr. We have to allocate one 926 * extra byte so that we can ensure that the pathname 927 * is nul-terminated. Note that unlike linux, we don't 928 * include in the address length the NUL in the path 929 * component, because doing so, would exceed sizeof(sockaddr_un) 930 * for fully occupied pathnames. Linux is also inconsistent, 931 * because it does not include the NUL in the length of 932 * what it calls "abstract" unix sockets. 933 */ 934 static struct sockaddr_un * 935 makeun(struct mbuf *nam, size_t *addrlen) { 936 struct sockaddr_un *sun; 937 938 *addrlen = nam->m_len + 1; 939 sun = malloc(*addrlen, M_SONAME, M_WAITOK); 940 m_copydata(nam, 0, nam->m_len, (void *)sun); 941 *(((char *)sun) + nam->m_len) = '\0'; 942 sun->sun_len = strlen(sun->sun_path) + 943 offsetof(struct sockaddr_un, sun_path); 944 return sun; 945 } 946 947 static int 948 unp_bind(struct socket *so, struct mbuf *nam, struct lwp *l) 949 { 950 struct sockaddr_un *sun; 951 struct unpcb *unp; 952 vnode_t *vp; 953 struct vattr vattr; 954 size_t addrlen; 955 int error; 956 struct pathbuf *pb; 957 struct nameidata nd; 958 proc_t *p; 959 960 unp = sotounpcb(so); 961 962 KASSERT(solocked(so)); 963 KASSERT(unp != NULL); 964 KASSERT(nam != NULL); 965 966 if (unp->unp_vnode != NULL) 967 return (EINVAL); 968 if ((unp->unp_flags & UNP_BUSY) != 0) { 969 /* 970 * EALREADY may not be strictly accurate, but since this 971 * is a major application error it's hardly a big deal. 972 */ 973 return (EALREADY); 974 } 975 unp->unp_flags |= UNP_BUSY; 976 sounlock(so); 977 978 p = l->l_proc; 979 sun = makeun(nam, &addrlen); 980 981 pb = pathbuf_create(sun->sun_path); 982 if (pb == NULL) { 983 error = ENOMEM; 984 goto bad; 985 } 986 NDINIT(&nd, CREATE, FOLLOW | LOCKPARENT | TRYEMULROOT, pb); 987 988 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 989 if ((error = namei(&nd)) != 0) { 990 pathbuf_destroy(pb); 991 goto bad; 992 } 993 vp = nd.ni_vp; 994 if (vp != NULL) { 995 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 996 if (nd.ni_dvp == vp) 997 vrele(nd.ni_dvp); 998 else 999 vput(nd.ni_dvp); 1000 vrele(vp); 1001 pathbuf_destroy(pb); 1002 error = EADDRINUSE; 1003 goto bad; 1004 } 1005 vattr_null(&vattr); 1006 vattr.va_type = VSOCK; 1007 vattr.va_mode = ACCESSPERMS & ~(p->p_cwdi->cwdi_cmask); 1008 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 1009 if (error) { 1010 vput(nd.ni_dvp); 1011 pathbuf_destroy(pb); 1012 goto bad; 1013 } 1014 vp = nd.ni_vp; 1015 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1016 solock(so); 1017 vp->v_socket = unp->unp_socket; 1018 unp->unp_vnode = vp; 1019 unp->unp_addrlen = addrlen; 1020 unp->unp_addr = sun; 1021 unp->unp_connid.unp_pid = p->p_pid; 1022 unp->unp_connid.unp_euid = kauth_cred_geteuid(l->l_cred); 1023 unp->unp_connid.unp_egid = kauth_cred_getegid(l->l_cred); 1024 unp->unp_flags |= UNP_EIDSBIND; 1025 VOP_UNLOCK(vp); 1026 vput(nd.ni_dvp); 1027 unp->unp_flags &= ~UNP_BUSY; 1028 pathbuf_destroy(pb); 1029 return (0); 1030 1031 bad: 1032 free(sun, M_SONAME); 1033 solock(so); 1034 unp->unp_flags &= ~UNP_BUSY; 1035 return (error); 1036 } 1037 1038 static int 1039 unp_listen(struct socket *so, struct lwp *l) 1040 { 1041 struct unpcb *unp = sotounpcb(so); 1042 1043 KASSERT(solocked(so)); 1044 KASSERT(unp != NULL); 1045 1046 /* 1047 * If the socket can accept a connection, it must be 1048 * locked by uipc_lock. 1049 */ 1050 unp_resetlock(so); 1051 if (unp->unp_vnode == NULL) 1052 return EINVAL; 1053 1054 return 0; 1055 } 1056 1057 static int 1058 unp_disconnect(struct socket *so) 1059 { 1060 KASSERT(solocked(so)); 1061 KASSERT(sotounpcb(so) != NULL); 1062 1063 unp_disconnect1(sotounpcb(so)); 1064 return 0; 1065 } 1066 1067 static int 1068 unp_shutdown(struct socket *so) 1069 { 1070 KASSERT(solocked(so)); 1071 KASSERT(sotounpcb(so) != NULL); 1072 1073 socantsendmore(so); 1074 unp_shutdown1(sotounpcb(so)); 1075 return 0; 1076 } 1077 1078 static int 1079 unp_abort(struct socket *so) 1080 { 1081 KASSERT(solocked(so)); 1082 KASSERT(sotounpcb(so) != NULL); 1083 1084 (void)unp_drop(sotounpcb(so), ECONNABORTED); 1085 KASSERT(so->so_head == NULL); 1086 KASSERT(so->so_pcb != NULL); 1087 unp_detach(so); 1088 return 0; 1089 } 1090 1091 int 1092 unp_connect(struct socket *so, struct mbuf *nam, struct lwp *l) 1093 { 1094 struct sockaddr_un *sun; 1095 vnode_t *vp; 1096 struct socket *so2, *so3; 1097 struct unpcb *unp, *unp2, *unp3; 1098 size_t addrlen; 1099 int error; 1100 struct pathbuf *pb; 1101 struct nameidata nd; 1102 1103 unp = sotounpcb(so); 1104 if ((unp->unp_flags & UNP_BUSY) != 0) { 1105 /* 1106 * EALREADY may not be strictly accurate, but since this 1107 * is a major application error it's hardly a big deal. 1108 */ 1109 return (EALREADY); 1110 } 1111 unp->unp_flags |= UNP_BUSY; 1112 sounlock(so); 1113 1114 sun = makeun(nam, &addrlen); 1115 pb = pathbuf_create(sun->sun_path); 1116 if (pb == NULL) { 1117 error = ENOMEM; 1118 goto bad2; 1119 } 1120 1121 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb); 1122 1123 if ((error = namei(&nd)) != 0) { 1124 pathbuf_destroy(pb); 1125 goto bad2; 1126 } 1127 vp = nd.ni_vp; 1128 if (vp->v_type != VSOCK) { 1129 error = ENOTSOCK; 1130 goto bad; 1131 } 1132 pathbuf_destroy(pb); 1133 if ((error = VOP_ACCESS(vp, VWRITE, l->l_cred)) != 0) 1134 goto bad; 1135 /* Acquire v_interlock to protect against unp_detach(). */ 1136 mutex_enter(vp->v_interlock); 1137 so2 = vp->v_socket; 1138 if (so2 == NULL) { 1139 mutex_exit(vp->v_interlock); 1140 error = ECONNREFUSED; 1141 goto bad; 1142 } 1143 if (so->so_type != so2->so_type) { 1144 mutex_exit(vp->v_interlock); 1145 error = EPROTOTYPE; 1146 goto bad; 1147 } 1148 solock(so); 1149 unp_resetlock(so); 1150 mutex_exit(vp->v_interlock); 1151 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) { 1152 /* 1153 * This may seem somewhat fragile but is OK: if we can 1154 * see SO_ACCEPTCONN set on the endpoint, then it must 1155 * be locked by the domain-wide uipc_lock. 1156 */ 1157 KASSERT((so2->so_options & SO_ACCEPTCONN) == 0 || 1158 so2->so_lock == uipc_lock); 1159 if ((so2->so_options & SO_ACCEPTCONN) == 0 || 1160 (so3 = sonewconn(so2, false)) == NULL) { 1161 error = ECONNREFUSED; 1162 sounlock(so); 1163 goto bad; 1164 } 1165 unp2 = sotounpcb(so2); 1166 unp3 = sotounpcb(so3); 1167 if (unp2->unp_addr) { 1168 unp3->unp_addr = malloc(unp2->unp_addrlen, 1169 M_SONAME, M_WAITOK); 1170 memcpy(unp3->unp_addr, unp2->unp_addr, 1171 unp2->unp_addrlen); 1172 unp3->unp_addrlen = unp2->unp_addrlen; 1173 } 1174 unp3->unp_flags = unp2->unp_flags; 1175 unp3->unp_connid.unp_pid = l->l_proc->p_pid; 1176 unp3->unp_connid.unp_euid = kauth_cred_geteuid(l->l_cred); 1177 unp3->unp_connid.unp_egid = kauth_cred_getegid(l->l_cred); 1178 unp3->unp_flags |= UNP_EIDSVALID; 1179 if (unp2->unp_flags & UNP_EIDSBIND) { 1180 unp->unp_connid = unp2->unp_connid; 1181 unp->unp_flags |= UNP_EIDSVALID; 1182 } 1183 so2 = so3; 1184 } 1185 error = unp_connect2(so, so2, PRU_CONNECT); 1186 sounlock(so); 1187 bad: 1188 vput(vp); 1189 bad2: 1190 free(sun, M_SONAME); 1191 solock(so); 1192 unp->unp_flags &= ~UNP_BUSY; 1193 return (error); 1194 } 1195 1196 int 1197 unp_connect2(struct socket *so, struct socket *so2, int req) 1198 { 1199 struct unpcb *unp = sotounpcb(so); 1200 struct unpcb *unp2; 1201 1202 if (so2->so_type != so->so_type) 1203 return (EPROTOTYPE); 1204 1205 /* 1206 * All three sockets involved must be locked by same lock: 1207 * 1208 * local endpoint (so) 1209 * remote endpoint (so2) 1210 * queue head (so2->so_head, only if PR_CONNREQUIRED) 1211 */ 1212 KASSERT(solocked2(so, so2)); 1213 KASSERT(so->so_head == NULL); 1214 if (so2->so_head != NULL) { 1215 KASSERT(so2->so_lock == uipc_lock); 1216 KASSERT(solocked2(so2, so2->so_head)); 1217 } 1218 1219 unp2 = sotounpcb(so2); 1220 unp->unp_conn = unp2; 1221 switch (so->so_type) { 1222 1223 case SOCK_DGRAM: 1224 unp->unp_nextref = unp2->unp_refs; 1225 unp2->unp_refs = unp; 1226 soisconnected(so); 1227 break; 1228 1229 case SOCK_SEQPACKET: /* FALLTHROUGH */ 1230 case SOCK_STREAM: 1231 unp2->unp_conn = unp; 1232 if (req == PRU_CONNECT && 1233 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 1234 soisconnecting(so); 1235 else 1236 soisconnected(so); 1237 soisconnected(so2); 1238 /* 1239 * If the connection is fully established, break the 1240 * association with uipc_lock and give the connected 1241 * pair a seperate lock to share. For CONNECT2, we 1242 * require that the locks already match (the sockets 1243 * are created that way). 1244 */ 1245 if (req == PRU_CONNECT) { 1246 KASSERT(so2->so_head != NULL); 1247 unp_setpeerlocks(so, so2); 1248 } 1249 break; 1250 1251 default: 1252 panic("unp_connect2"); 1253 } 1254 return (0); 1255 } 1256 1257 static void 1258 unp_disconnect1(struct unpcb *unp) 1259 { 1260 struct unpcb *unp2 = unp->unp_conn; 1261 struct socket *so; 1262 1263 if (unp2 == 0) 1264 return; 1265 unp->unp_conn = 0; 1266 so = unp->unp_socket; 1267 switch (so->so_type) { 1268 case SOCK_DGRAM: 1269 if (unp2->unp_refs == unp) 1270 unp2->unp_refs = unp->unp_nextref; 1271 else { 1272 unp2 = unp2->unp_refs; 1273 for (;;) { 1274 KASSERT(solocked2(so, unp2->unp_socket)); 1275 if (unp2 == 0) 1276 panic("unp_disconnect1"); 1277 if (unp2->unp_nextref == unp) 1278 break; 1279 unp2 = unp2->unp_nextref; 1280 } 1281 unp2->unp_nextref = unp->unp_nextref; 1282 } 1283 unp->unp_nextref = 0; 1284 so->so_state &= ~SS_ISCONNECTED; 1285 break; 1286 1287 case SOCK_SEQPACKET: /* FALLTHROUGH */ 1288 case SOCK_STREAM: 1289 KASSERT(solocked2(so, unp2->unp_socket)); 1290 soisdisconnected(so); 1291 unp2->unp_conn = 0; 1292 soisdisconnected(unp2->unp_socket); 1293 break; 1294 } 1295 } 1296 1297 static void 1298 unp_shutdown1(struct unpcb *unp) 1299 { 1300 struct socket *so; 1301 1302 switch(unp->unp_socket->so_type) { 1303 case SOCK_SEQPACKET: /* FALLTHROUGH */ 1304 case SOCK_STREAM: 1305 if (unp->unp_conn && (so = unp->unp_conn->unp_socket)) 1306 socantrcvmore(so); 1307 break; 1308 default: 1309 break; 1310 } 1311 } 1312 1313 static bool 1314 unp_drop(struct unpcb *unp, int errno) 1315 { 1316 struct socket *so = unp->unp_socket; 1317 1318 KASSERT(solocked(so)); 1319 1320 so->so_error = errno; 1321 unp_disconnect1(unp); 1322 if (so->so_head) { 1323 so->so_pcb = NULL; 1324 /* sofree() drops the socket lock */ 1325 sofree(so); 1326 unp_free(unp); 1327 return true; 1328 } 1329 return false; 1330 } 1331 1332 #ifdef notdef 1333 unp_drain(void) 1334 { 1335 1336 } 1337 #endif 1338 1339 int 1340 unp_externalize(struct mbuf *rights, struct lwp *l, int flags) 1341 { 1342 struct cmsghdr * const cm = mtod(rights, struct cmsghdr *); 1343 struct proc * const p = l->l_proc; 1344 file_t **rp; 1345 int error = 0; 1346 1347 const size_t nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) / 1348 sizeof(file_t *); 1349 if (nfds == 0) 1350 goto noop; 1351 1352 int * const fdp = kmem_alloc(nfds * sizeof(int), KM_SLEEP); 1353 rw_enter(&p->p_cwdi->cwdi_lock, RW_READER); 1354 1355 /* Make sure the recipient should be able to see the files.. */ 1356 rp = (file_t **)CMSG_DATA(cm); 1357 for (size_t i = 0; i < nfds; i++) { 1358 file_t * const fp = *rp++; 1359 if (fp == NULL) { 1360 error = EINVAL; 1361 goto out; 1362 } 1363 /* 1364 * If we are in a chroot'ed directory, and 1365 * someone wants to pass us a directory, make 1366 * sure it's inside the subtree we're allowed 1367 * to access. 1368 */ 1369 if (p->p_cwdi->cwdi_rdir != NULL && fp->f_type == DTYPE_VNODE) { 1370 vnode_t *vp = (vnode_t *)fp->f_data; 1371 if ((vp->v_type == VDIR) && 1372 !vn_isunder(vp, p->p_cwdi->cwdi_rdir, l)) { 1373 error = EPERM; 1374 goto out; 1375 } 1376 } 1377 } 1378 1379 restart: 1380 /* 1381 * First loop -- allocate file descriptor table slots for the 1382 * new files. 1383 */ 1384 for (size_t i = 0; i < nfds; i++) { 1385 if ((error = fd_alloc(p, 0, &fdp[i])) != 0) { 1386 /* 1387 * Back out what we've done so far. 1388 */ 1389 while (i-- > 0) { 1390 fd_abort(p, NULL, fdp[i]); 1391 } 1392 if (error == ENOSPC) { 1393 fd_tryexpand(p); 1394 error = 0; 1395 goto restart; 1396 } 1397 /* 1398 * This is the error that has historically 1399 * been returned, and some callers may 1400 * expect it. 1401 */ 1402 error = EMSGSIZE; 1403 goto out; 1404 } 1405 } 1406 1407 /* 1408 * Now that adding them has succeeded, update all of the 1409 * file passing state and affix the descriptors. 1410 */ 1411 rp = (file_t **)CMSG_DATA(cm); 1412 int *ofdp = (int *)CMSG_DATA(cm); 1413 for (size_t i = 0; i < nfds; i++) { 1414 file_t * const fp = *rp++; 1415 const int fd = fdp[i]; 1416 atomic_dec_uint(&unp_rights); 1417 fd_set_exclose(l, fd, (flags & O_CLOEXEC) != 0); 1418 fd_affix(p, fp, fd); 1419 /* 1420 * Done with this file pointer, replace it with a fd; 1421 */ 1422 *ofdp++ = fd; 1423 mutex_enter(&fp->f_lock); 1424 fp->f_msgcount--; 1425 mutex_exit(&fp->f_lock); 1426 /* 1427 * Note that fd_affix() adds a reference to the file. 1428 * The file may already have been closed by another 1429 * LWP in the process, so we must drop the reference 1430 * added by unp_internalize() with closef(). 1431 */ 1432 closef(fp); 1433 } 1434 1435 /* 1436 * Adjust length, in case of transition from large file_t 1437 * pointers to ints. 1438 */ 1439 if (sizeof(file_t *) != sizeof(int)) { 1440 cm->cmsg_len = CMSG_LEN(nfds * sizeof(int)); 1441 rights->m_len = CMSG_SPACE(nfds * sizeof(int)); 1442 } 1443 out: 1444 if (__predict_false(error != 0)) { 1445 file_t **const fpp = (file_t **)CMSG_DATA(cm); 1446 for (size_t i = 0; i < nfds; i++) 1447 unp_discard_now(fpp[i]); 1448 /* 1449 * Truncate the array so that nobody will try to interpret 1450 * what is now garbage in it. 1451 */ 1452 cm->cmsg_len = CMSG_LEN(0); 1453 rights->m_len = CMSG_SPACE(0); 1454 } 1455 rw_exit(&p->p_cwdi->cwdi_lock); 1456 kmem_free(fdp, nfds * sizeof(int)); 1457 1458 noop: 1459 /* 1460 * Don't disclose kernel memory in the alignment space. 1461 */ 1462 KASSERT(cm->cmsg_len <= rights->m_len); 1463 memset(&mtod(rights, char *)[cm->cmsg_len], 0, rights->m_len - 1464 cm->cmsg_len); 1465 return error; 1466 } 1467 1468 static int 1469 unp_internalize(struct mbuf **controlp) 1470 { 1471 filedesc_t *fdescp = curlwp->l_fd; 1472 struct mbuf *control = *controlp; 1473 struct cmsghdr *newcm, *cm = mtod(control, struct cmsghdr *); 1474 file_t **rp, **files; 1475 file_t *fp; 1476 int i, fd, *fdp; 1477 int nfds, error; 1478 u_int maxmsg; 1479 1480 error = 0; 1481 newcm = NULL; 1482 1483 /* Sanity check the control message header. */ 1484 if (cm->cmsg_type != SCM_RIGHTS || cm->cmsg_level != SOL_SOCKET || 1485 cm->cmsg_len > control->m_len || 1486 cm->cmsg_len < CMSG_ALIGN(sizeof(*cm))) 1487 return (EINVAL); 1488 1489 /* 1490 * Verify that the file descriptors are valid, and acquire 1491 * a reference to each. 1492 */ 1493 nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) / sizeof(int); 1494 fdp = (int *)CMSG_DATA(cm); 1495 maxmsg = maxfiles / unp_rights_ratio; 1496 for (i = 0; i < nfds; i++) { 1497 fd = *fdp++; 1498 if (atomic_inc_uint_nv(&unp_rights) > maxmsg) { 1499 atomic_dec_uint(&unp_rights); 1500 nfds = i; 1501 error = EAGAIN; 1502 goto out; 1503 } 1504 if ((fp = fd_getfile(fd)) == NULL 1505 || fp->f_type == DTYPE_KQUEUE) { 1506 if (fp) 1507 fd_putfile(fd); 1508 atomic_dec_uint(&unp_rights); 1509 nfds = i; 1510 error = EBADF; 1511 goto out; 1512 } 1513 } 1514 1515 /* Allocate new space and copy header into it. */ 1516 newcm = malloc(CMSG_SPACE(nfds * sizeof(file_t *)), M_MBUF, M_WAITOK); 1517 if (newcm == NULL) { 1518 error = E2BIG; 1519 goto out; 1520 } 1521 memcpy(newcm, cm, sizeof(struct cmsghdr)); 1522 files = (file_t **)CMSG_DATA(newcm); 1523 1524 /* 1525 * Transform the file descriptors into file_t pointers, in 1526 * reverse order so that if pointers are bigger than ints, the 1527 * int won't get until we're done. No need to lock, as we have 1528 * already validated the descriptors with fd_getfile(). 1529 */ 1530 fdp = (int *)CMSG_DATA(cm) + nfds; 1531 rp = files + nfds; 1532 for (i = 0; i < nfds; i++) { 1533 fp = fdescp->fd_dt->dt_ff[*--fdp]->ff_file; 1534 KASSERT(fp != NULL); 1535 mutex_enter(&fp->f_lock); 1536 *--rp = fp; 1537 fp->f_count++; 1538 fp->f_msgcount++; 1539 mutex_exit(&fp->f_lock); 1540 } 1541 1542 out: 1543 /* Release descriptor references. */ 1544 fdp = (int *)CMSG_DATA(cm); 1545 for (i = 0; i < nfds; i++) { 1546 fd_putfile(*fdp++); 1547 if (error != 0) { 1548 atomic_dec_uint(&unp_rights); 1549 } 1550 } 1551 1552 if (error == 0) { 1553 if (control->m_flags & M_EXT) { 1554 m_freem(control); 1555 *controlp = control = m_get(M_WAIT, MT_CONTROL); 1556 } 1557 MEXTADD(control, newcm, CMSG_SPACE(nfds * sizeof(file_t *)), 1558 M_MBUF, NULL, NULL); 1559 cm = newcm; 1560 /* 1561 * Adjust message & mbuf to note amount of space 1562 * actually used. 1563 */ 1564 cm->cmsg_len = CMSG_LEN(nfds * sizeof(file_t *)); 1565 control->m_len = CMSG_SPACE(nfds * sizeof(file_t *)); 1566 } 1567 1568 return error; 1569 } 1570 1571 struct mbuf * 1572 unp_addsockcred(struct lwp *l, struct mbuf *control) 1573 { 1574 struct sockcred *sc; 1575 struct mbuf *m; 1576 void *p; 1577 1578 m = sbcreatecontrol1(&p, SOCKCREDSIZE(kauth_cred_ngroups(l->l_cred)), 1579 SCM_CREDS, SOL_SOCKET, M_WAITOK); 1580 if (m == NULL) 1581 return control; 1582 1583 sc = p; 1584 sc->sc_uid = kauth_cred_getuid(l->l_cred); 1585 sc->sc_euid = kauth_cred_geteuid(l->l_cred); 1586 sc->sc_gid = kauth_cred_getgid(l->l_cred); 1587 sc->sc_egid = kauth_cred_getegid(l->l_cred); 1588 sc->sc_ngroups = kauth_cred_ngroups(l->l_cred); 1589 1590 for (int i = 0; i < sc->sc_ngroups; i++) 1591 sc->sc_groups[i] = kauth_cred_group(l->l_cred, i); 1592 1593 return m_add(control, m); 1594 } 1595 1596 /* 1597 * Do a mark-sweep GC of files in the system, to free up any which are 1598 * caught in flight to an about-to-be-closed socket. Additionally, 1599 * process deferred file closures. 1600 */ 1601 static void 1602 unp_gc(file_t *dp) 1603 { 1604 extern struct domain unixdomain; 1605 file_t *fp, *np; 1606 struct socket *so, *so1; 1607 u_int i, old, new; 1608 bool didwork; 1609 1610 KASSERT(curlwp == unp_thread_lwp); 1611 KASSERT(mutex_owned(&filelist_lock)); 1612 1613 /* 1614 * First, process deferred file closures. 1615 */ 1616 while (!SLIST_EMPTY(&unp_thread_discard)) { 1617 fp = SLIST_FIRST(&unp_thread_discard); 1618 KASSERT(fp->f_unpcount > 0); 1619 KASSERT(fp->f_count > 0); 1620 KASSERT(fp->f_msgcount > 0); 1621 KASSERT(fp->f_count >= fp->f_unpcount); 1622 KASSERT(fp->f_count >= fp->f_msgcount); 1623 KASSERT(fp->f_msgcount >= fp->f_unpcount); 1624 SLIST_REMOVE_HEAD(&unp_thread_discard, f_unplist); 1625 i = fp->f_unpcount; 1626 fp->f_unpcount = 0; 1627 mutex_exit(&filelist_lock); 1628 for (; i != 0; i--) { 1629 unp_discard_now(fp); 1630 } 1631 mutex_enter(&filelist_lock); 1632 } 1633 1634 /* 1635 * Clear mark bits. Ensure that we don't consider new files 1636 * entering the file table during this loop (they will not have 1637 * FSCAN set). 1638 */ 1639 unp_defer = 0; 1640 LIST_FOREACH(fp, &filehead, f_list) { 1641 for (old = fp->f_flag;; old = new) { 1642 new = atomic_cas_uint(&fp->f_flag, old, 1643 (old | FSCAN) & ~(FMARK|FDEFER)); 1644 if (__predict_true(old == new)) { 1645 break; 1646 } 1647 } 1648 } 1649 1650 /* 1651 * Iterate over the set of sockets, marking ones believed (based on 1652 * refcount) to be referenced from a process, and marking for rescan 1653 * sockets which are queued on a socket. Recan continues descending 1654 * and searching for sockets referenced by sockets (FDEFER), until 1655 * there are no more socket->socket references to be discovered. 1656 */ 1657 do { 1658 didwork = false; 1659 for (fp = LIST_FIRST(&filehead); fp != NULL; fp = np) { 1660 KASSERT(mutex_owned(&filelist_lock)); 1661 np = LIST_NEXT(fp, f_list); 1662 mutex_enter(&fp->f_lock); 1663 if ((fp->f_flag & FDEFER) != 0) { 1664 atomic_and_uint(&fp->f_flag, ~FDEFER); 1665 unp_defer--; 1666 KASSERT(fp->f_count != 0); 1667 } else { 1668 if (fp->f_count == 0 || 1669 (fp->f_flag & FMARK) != 0 || 1670 fp->f_count == fp->f_msgcount || 1671 fp->f_unpcount != 0) { 1672 mutex_exit(&fp->f_lock); 1673 continue; 1674 } 1675 } 1676 atomic_or_uint(&fp->f_flag, FMARK); 1677 1678 if (fp->f_type != DTYPE_SOCKET || 1679 (so = fp->f_data) == NULL || 1680 so->so_proto->pr_domain != &unixdomain || 1681 (so->so_proto->pr_flags & PR_RIGHTS) == 0) { 1682 mutex_exit(&fp->f_lock); 1683 continue; 1684 } 1685 1686 /* Gain file ref, mark our position, and unlock. */ 1687 didwork = true; 1688 LIST_INSERT_AFTER(fp, dp, f_list); 1689 fp->f_count++; 1690 mutex_exit(&fp->f_lock); 1691 mutex_exit(&filelist_lock); 1692 1693 /* 1694 * Mark files referenced from sockets queued on the 1695 * accept queue as well. 1696 */ 1697 solock(so); 1698 unp_scan(so->so_rcv.sb_mb, unp_mark, 0); 1699 if ((so->so_options & SO_ACCEPTCONN) != 0) { 1700 TAILQ_FOREACH(so1, &so->so_q0, so_qe) { 1701 unp_scan(so1->so_rcv.sb_mb, unp_mark, 0); 1702 } 1703 TAILQ_FOREACH(so1, &so->so_q, so_qe) { 1704 unp_scan(so1->so_rcv.sb_mb, unp_mark, 0); 1705 } 1706 } 1707 sounlock(so); 1708 1709 /* Re-lock and restart from where we left off. */ 1710 closef(fp); 1711 mutex_enter(&filelist_lock); 1712 np = LIST_NEXT(dp, f_list); 1713 LIST_REMOVE(dp, f_list); 1714 } 1715 /* 1716 * Bail early if we did nothing in the loop above. Could 1717 * happen because of concurrent activity causing unp_defer 1718 * to get out of sync. 1719 */ 1720 } while (unp_defer != 0 && didwork); 1721 1722 /* 1723 * Sweep pass. 1724 * 1725 * We grab an extra reference to each of the files that are 1726 * not otherwise accessible and then free the rights that are 1727 * stored in messages on them. 1728 */ 1729 for (fp = LIST_FIRST(&filehead); fp != NULL; fp = np) { 1730 KASSERT(mutex_owned(&filelist_lock)); 1731 np = LIST_NEXT(fp, f_list); 1732 mutex_enter(&fp->f_lock); 1733 1734 /* 1735 * Ignore non-sockets. 1736 * Ignore dead sockets, or sockets with pending close. 1737 * Ignore sockets obviously referenced elsewhere. 1738 * Ignore sockets marked as referenced by our scan. 1739 * Ignore new sockets that did not exist during the scan. 1740 */ 1741 if (fp->f_type != DTYPE_SOCKET || 1742 fp->f_count == 0 || fp->f_unpcount != 0 || 1743 fp->f_count != fp->f_msgcount || 1744 (fp->f_flag & (FMARK | FSCAN)) != FSCAN) { 1745 mutex_exit(&fp->f_lock); 1746 continue; 1747 } 1748 1749 /* Gain file ref, mark our position, and unlock. */ 1750 LIST_INSERT_AFTER(fp, dp, f_list); 1751 fp->f_count++; 1752 mutex_exit(&fp->f_lock); 1753 mutex_exit(&filelist_lock); 1754 1755 /* 1756 * Flush all data from the socket's receive buffer. 1757 * This will cause files referenced only by the 1758 * socket to be queued for close. 1759 */ 1760 so = fp->f_data; 1761 solock(so); 1762 sorflush(so); 1763 sounlock(so); 1764 1765 /* Re-lock and restart from where we left off. */ 1766 closef(fp); 1767 mutex_enter(&filelist_lock); 1768 np = LIST_NEXT(dp, f_list); 1769 LIST_REMOVE(dp, f_list); 1770 } 1771 } 1772 1773 /* 1774 * Garbage collector thread. While SCM_RIGHTS messages are in transit, 1775 * wake once per second to garbage collect. Run continually while we 1776 * have deferred closes to process. 1777 */ 1778 static void 1779 unp_thread(void *cookie) 1780 { 1781 file_t *dp; 1782 1783 /* Allocate a dummy file for our scans. */ 1784 if ((dp = fgetdummy()) == NULL) { 1785 panic("unp_thread"); 1786 } 1787 1788 mutex_enter(&filelist_lock); 1789 for (;;) { 1790 KASSERT(mutex_owned(&filelist_lock)); 1791 if (SLIST_EMPTY(&unp_thread_discard)) { 1792 if (unp_rights != 0) { 1793 (void)cv_timedwait(&unp_thread_cv, 1794 &filelist_lock, hz); 1795 } else { 1796 cv_wait(&unp_thread_cv, &filelist_lock); 1797 } 1798 } 1799 unp_gc(dp); 1800 } 1801 /* NOTREACHED */ 1802 } 1803 1804 /* 1805 * Kick the garbage collector into action if there is something for 1806 * it to process. 1807 */ 1808 static void 1809 unp_thread_kick(void) 1810 { 1811 1812 if (!SLIST_EMPTY(&unp_thread_discard) || unp_rights != 0) { 1813 mutex_enter(&filelist_lock); 1814 cv_signal(&unp_thread_cv); 1815 mutex_exit(&filelist_lock); 1816 } 1817 } 1818 1819 void 1820 unp_dispose(struct mbuf *m) 1821 { 1822 1823 if (m) 1824 unp_scan(m, unp_discard_later, 1); 1825 } 1826 1827 void 1828 unp_scan(struct mbuf *m0, void (*op)(file_t *), int discard) 1829 { 1830 struct mbuf *m; 1831 file_t **rp, *fp; 1832 struct cmsghdr *cm; 1833 int i, qfds; 1834 1835 while (m0) { 1836 for (m = m0; m; m = m->m_next) { 1837 if (m->m_type != MT_CONTROL || 1838 m->m_len < sizeof(*cm)) { 1839 continue; 1840 } 1841 cm = mtod(m, struct cmsghdr *); 1842 if (cm->cmsg_level != SOL_SOCKET || 1843 cm->cmsg_type != SCM_RIGHTS) 1844 continue; 1845 qfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) 1846 / sizeof(file_t *); 1847 rp = (file_t **)CMSG_DATA(cm); 1848 for (i = 0; i < qfds; i++) { 1849 fp = *rp; 1850 if (discard) { 1851 *rp = 0; 1852 } 1853 (*op)(fp); 1854 rp++; 1855 } 1856 } 1857 m0 = m0->m_nextpkt; 1858 } 1859 } 1860 1861 void 1862 unp_mark(file_t *fp) 1863 { 1864 1865 if (fp == NULL) 1866 return; 1867 1868 /* If we're already deferred, don't screw up the defer count */ 1869 mutex_enter(&fp->f_lock); 1870 if (fp->f_flag & (FMARK | FDEFER)) { 1871 mutex_exit(&fp->f_lock); 1872 return; 1873 } 1874 1875 /* 1876 * Minimize the number of deferrals... Sockets are the only type of 1877 * file which can hold references to another file, so just mark 1878 * other files, and defer unmarked sockets for the next pass. 1879 */ 1880 if (fp->f_type == DTYPE_SOCKET) { 1881 unp_defer++; 1882 KASSERT(fp->f_count != 0); 1883 atomic_or_uint(&fp->f_flag, FDEFER); 1884 } else { 1885 atomic_or_uint(&fp->f_flag, FMARK); 1886 } 1887 mutex_exit(&fp->f_lock); 1888 } 1889 1890 static void 1891 unp_discard_now(file_t *fp) 1892 { 1893 1894 if (fp == NULL) 1895 return; 1896 1897 KASSERT(fp->f_count > 0); 1898 KASSERT(fp->f_msgcount > 0); 1899 1900 mutex_enter(&fp->f_lock); 1901 fp->f_msgcount--; 1902 mutex_exit(&fp->f_lock); 1903 atomic_dec_uint(&unp_rights); 1904 (void)closef(fp); 1905 } 1906 1907 static void 1908 unp_discard_later(file_t *fp) 1909 { 1910 1911 if (fp == NULL) 1912 return; 1913 1914 KASSERT(fp->f_count > 0); 1915 KASSERT(fp->f_msgcount > 0); 1916 1917 mutex_enter(&filelist_lock); 1918 if (fp->f_unpcount++ == 0) { 1919 SLIST_INSERT_HEAD(&unp_thread_discard, fp, f_unplist); 1920 } 1921 mutex_exit(&filelist_lock); 1922 } 1923 1924 const struct pr_usrreqs unp_usrreqs = { 1925 .pr_attach = unp_attach, 1926 .pr_detach = unp_detach, 1927 .pr_accept = unp_accept, 1928 .pr_bind = unp_bind, 1929 .pr_listen = unp_listen, 1930 .pr_connect = unp_connect, 1931 .pr_disconnect = unp_disconnect, 1932 .pr_shutdown = unp_shutdown, 1933 .pr_abort = unp_abort, 1934 .pr_ioctl = unp_ioctl, 1935 .pr_stat = unp_stat, 1936 .pr_peeraddr = unp_peeraddr, 1937 .pr_sockaddr = unp_sockaddr, 1938 .pr_rcvd = unp_rcvd, 1939 .pr_recvoob = unp_recvoob, 1940 .pr_send = unp_send, 1941 .pr_sendoob = unp_sendoob, 1942 .pr_generic = unp_usrreq, 1943 }; 1944
Indexes created Sun Sep 21 20:09:37 GMT 2025