1 /* $NetBSD: uipc_usrreq.c,v 1.186 2018/05/11 09:43:59 roy 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 1.186 2018/05/11 09:43:59 roy Exp $"); 100 101 #ifdef _KERNEL_OPT 102 #include "opt_compat_netbsd.h" 103 #endif 104 105 #include <sys/param.h> 106 #include <sys/systm.h> 107 #include <sys/proc.h> 108 #include <sys/filedesc.h> 109 #include <sys/domain.h> 110 #include <sys/protosw.h> 111 #include <sys/socket.h> 112 #include <sys/socketvar.h> 113 #include <sys/unpcb.h> 114 #include <sys/un.h> 115 #include <sys/namei.h> 116 #include <sys/vnode.h> 117 #include <sys/file.h> 118 #include <sys/stat.h> 119 #include <sys/mbuf.h> 120 #include <sys/kauth.h> 121 #include <sys/kmem.h> 122 #include <sys/atomic.h> 123 #include <sys/uidinfo.h> 124 #include <sys/kernel.h> 125 #include <sys/kthread.h> 126 127 #ifdef COMPAT_70 128 #include <compat/sys/socket.h> 129 #endif 130 131 /* 132 * Unix communications domain. 133 * 134 * TODO: 135 * RDM 136 * rethink name space problems 137 * need a proper out-of-band 138 * 139 * Notes on locking: 140 * 141 * The generic rules noted in uipc_socket2.c apply. In addition: 142 * 143 * o We have a global lock, uipc_lock. 144 * 145 * o All datagram sockets are locked by uipc_lock. 146 * 147 * o For stream socketpairs, the two endpoints are created sharing the same 148 * independent lock. Sockets presented to PRU_CONNECT2 must already have 149 * matching locks. 150 * 151 * o Stream sockets created via socket() start life with their own 152 * independent lock. 153 * 154 * o Stream connections to a named endpoint are slightly more complicated. 155 * Sockets that have called listen() have their lock pointer mutated to 156 * the global uipc_lock. When establishing a connection, the connecting 157 * socket also has its lock mutated to uipc_lock, which matches the head 158 * (listening socket). We create a new socket for accept() to return, and 159 * that also shares the head's lock. Until the connection is completely 160 * done on both ends, all three sockets are locked by uipc_lock. Once the 161 * connection is complete, the association with the head's lock is broken. 162 * The connecting socket and the socket returned from accept() have their 163 * lock pointers mutated away from uipc_lock, and back to the connecting 164 * socket's original, independent lock. The head continues to be locked 165 * by uipc_lock. 166 * 167 * o If uipc_lock is determined to be a significant source of contention, 168 * it could easily be hashed out. It is difficult to simply make it an 169 * independent lock because of visibility / garbage collection issues: 170 * if a socket has been associated with a lock at any point, that lock 171 * must remain valid until the socket is no longer visible in the system. 172 * The lock must not be freed or otherwise destroyed until any sockets 173 * that had referenced it have also been destroyed. 174 */ 175 const struct sockaddr_un sun_noname = { 176 .sun_len = offsetof(struct sockaddr_un, sun_path), 177 .sun_family = AF_LOCAL, 178 }; 179 ino_t unp_ino; /* prototype for fake inode numbers */ 180 181 static struct mbuf * unp_addsockcred(struct lwp *, struct mbuf *); 182 static void unp_discard_later(file_t *); 183 static void unp_discard_now(file_t *); 184 static void unp_disconnect1(struct unpcb *); 185 static bool unp_drop(struct unpcb *, int); 186 static int unp_internalize(struct mbuf **); 187 static void unp_mark(file_t *); 188 static void unp_scan(struct mbuf *, void (*)(file_t *), int); 189 static void unp_shutdown1(struct unpcb *); 190 static void unp_thread(void *); 191 static void unp_thread_kick(void); 192 193 static kmutex_t *uipc_lock; 194 195 static kcondvar_t unp_thread_cv; 196 static lwp_t *unp_thread_lwp; 197 static SLIST_HEAD(,file) unp_thread_discard; 198 static int unp_defer; 199 200 /* 201 * Initialize Unix protocols. 202 */ 203 void 204 uipc_init(void) 205 { 206 int error; 207 208 uipc_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 209 cv_init(&unp_thread_cv, "unpgc"); 210 211 error = kthread_create(PRI_NONE, KTHREAD_MPSAFE, NULL, unp_thread, 212 NULL, &unp_thread_lwp, "unpgc"); 213 if (error != 0) 214 panic("uipc_init %d", error); 215 } 216 217 static void 218 unp_connid(struct lwp *l, struct unpcb *unp, int flags) 219 { 220 unp->unp_connid.unp_pid = l->l_proc->p_pid; 221 unp->unp_connid.unp_euid = kauth_cred_geteuid(l->l_cred); 222 unp->unp_connid.unp_egid = kauth_cred_getegid(l->l_cred); 223 unp->unp_flags |= flags; 224 } 225 226 /* 227 * A connection succeeded: disassociate both endpoints from the head's 228 * lock, and make them share their own lock. There is a race here: for 229 * a very brief time one endpoint will be locked by a different lock 230 * than the other end. However, since the current thread holds the old 231 * lock (the listening socket's lock, the head) access can still only be 232 * made to one side of the connection. 233 */ 234 static void 235 unp_setpeerlocks(struct socket *so, struct socket *so2) 236 { 237 struct unpcb *unp; 238 kmutex_t *lock; 239 240 KASSERT(solocked2(so, so2)); 241 242 /* 243 * Bail out if either end of the socket is not yet fully 244 * connected or accepted. We only break the lock association 245 * with the head when the pair of sockets stand completely 246 * on their own. 247 */ 248 KASSERT(so->so_head == NULL); 249 if (so2->so_head != NULL) 250 return; 251 252 /* 253 * Drop references to old lock. A third reference (from the 254 * queue head) must be held as we still hold its lock. Bonus: 255 * we don't need to worry about garbage collecting the lock. 256 */ 257 lock = so->so_lock; 258 KASSERT(lock == uipc_lock); 259 mutex_obj_free(lock); 260 mutex_obj_free(lock); 261 262 /* 263 * Grab stream lock from the initiator and share between the two 264 * endpoints. Issue memory barrier to ensure all modifications 265 * become globally visible before the lock change. so2 is 266 * assumed not to have a stream lock, because it was created 267 * purely for the server side to accept this connection and 268 * started out life using the domain-wide lock. 269 */ 270 unp = sotounpcb(so); 271 KASSERT(unp->unp_streamlock != NULL); 272 KASSERT(sotounpcb(so2)->unp_streamlock == NULL); 273 lock = unp->unp_streamlock; 274 unp->unp_streamlock = NULL; 275 mutex_obj_hold(lock); 276 membar_exit(); 277 /* 278 * possible race if lock is not held - see comment in 279 * uipc_usrreq(PRU_ACCEPT). 280 */ 281 KASSERT(mutex_owned(lock)); 282 solockreset(so, lock); 283 solockreset(so2, lock); 284 } 285 286 /* 287 * Reset a socket's lock back to the domain-wide lock. 288 */ 289 static void 290 unp_resetlock(struct socket *so) 291 { 292 kmutex_t *olock, *nlock; 293 struct unpcb *unp; 294 295 KASSERT(solocked(so)); 296 297 olock = so->so_lock; 298 nlock = uipc_lock; 299 if (olock == nlock) 300 return; 301 unp = sotounpcb(so); 302 KASSERT(unp->unp_streamlock == NULL); 303 unp->unp_streamlock = olock; 304 mutex_obj_hold(nlock); 305 mutex_enter(nlock); 306 solockreset(so, nlock); 307 mutex_exit(olock); 308 } 309 310 static void 311 unp_free(struct unpcb *unp) 312 { 313 if (unp->unp_addr) 314 free(unp->unp_addr, M_SONAME); 315 if (unp->unp_streamlock != NULL) 316 mutex_obj_free(unp->unp_streamlock); 317 kmem_free(unp, sizeof(*unp)); 318 } 319 320 static int 321 unp_output(struct mbuf *m, struct mbuf *control, struct unpcb *unp) 322 { 323 struct socket *so2; 324 const struct sockaddr_un *sun; 325 326 /* XXX: server side closed the socket */ 327 if (unp->unp_conn == NULL) 328 return ECONNREFUSED; 329 so2 = unp->unp_conn->unp_socket; 330 331 KASSERT(solocked(so2)); 332 333 if (unp->unp_addr) 334 sun = unp->unp_addr; 335 else 336 sun = &sun_noname; 337 if (unp->unp_conn->unp_flags & UNP_WANTCRED) 338 control = unp_addsockcred(curlwp, control); 339 #ifdef COMPAT_SOCKCRED70 340 if (unp->unp_conn->unp_flags & UNP_OWANTCRED) 341 control = compat_70_unp_addsockcred(curlwp, control); 342 #endif 343 if (sbappendaddr(&so2->so_rcv, (const struct sockaddr *)sun, m, 344 control) == 0) { 345 unp_dispose(control); 346 m_freem(control); 347 m_freem(m); 348 soroverflow(so2); 349 return (ENOBUFS); 350 } else { 351 sorwakeup(so2); 352 return (0); 353 } 354 } 355 356 static void 357 unp_setaddr(struct socket *so, struct sockaddr *nam, bool peeraddr) 358 { 359 const struct sockaddr_un *sun = NULL; 360 struct unpcb *unp; 361 362 KASSERT(solocked(so)); 363 unp = sotounpcb(so); 364 365 if (peeraddr) { 366 if (unp->unp_conn && unp->unp_conn->unp_addr) 367 sun = unp->unp_conn->unp_addr; 368 } else { 369 if (unp->unp_addr) 370 sun = unp->unp_addr; 371 } 372 if (sun == NULL) 373 sun = &sun_noname; 374 375 memcpy(nam, sun, sun->sun_len); 376 } 377 378 static int 379 unp_rcvd(struct socket *so, int flags, struct lwp *l) 380 { 381 struct unpcb *unp = sotounpcb(so); 382 struct socket *so2; 383 u_int newhiwat; 384 385 KASSERT(solocked(so)); 386 KASSERT(unp != NULL); 387 388 switch (so->so_type) { 389 390 case SOCK_DGRAM: 391 panic("uipc 1"); 392 /*NOTREACHED*/ 393 394 case SOCK_SEQPACKET: /* FALLTHROUGH */ 395 case SOCK_STREAM: 396 #define rcv (&so->so_rcv) 397 #define snd (&so2->so_snd) 398 if (unp->unp_conn == 0) 399 break; 400 so2 = unp->unp_conn->unp_socket; 401 KASSERT(solocked2(so, so2)); 402 /* 403 * Adjust backpressure on sender 404 * and wakeup any waiting to write. 405 */ 406 snd->sb_mbmax += unp->unp_mbcnt - rcv->sb_mbcnt; 407 unp->unp_mbcnt = rcv->sb_mbcnt; 408 newhiwat = snd->sb_hiwat + unp->unp_cc - rcv->sb_cc; 409 (void)chgsbsize(so2->so_uidinfo, 410 &snd->sb_hiwat, newhiwat, RLIM_INFINITY); 411 unp->unp_cc = rcv->sb_cc; 412 sowwakeup(so2); 413 #undef snd 414 #undef rcv 415 break; 416 417 default: 418 panic("uipc 2"); 419 } 420 421 return 0; 422 } 423 424 static int 425 unp_recvoob(struct socket *so, struct mbuf *m, int flags) 426 { 427 KASSERT(solocked(so)); 428 429 return EOPNOTSUPP; 430 } 431 432 static int 433 unp_send(struct socket *so, struct mbuf *m, struct sockaddr *nam, 434 struct mbuf *control, struct lwp *l) 435 { 436 struct unpcb *unp = sotounpcb(so); 437 int error = 0; 438 u_int newhiwat; 439 struct socket *so2; 440 441 KASSERT(solocked(so)); 442 KASSERT(unp != NULL); 443 KASSERT(m != NULL); 444 445 /* 446 * Note: unp_internalize() rejects any control message 447 * other than SCM_RIGHTS, and only allows one. This 448 * has the side-effect of preventing a caller from 449 * forging SCM_CREDS. 450 */ 451 if (control) { 452 sounlock(so); 453 error = unp_internalize(&control); 454 solock(so); 455 if (error != 0) { 456 m_freem(control); 457 m_freem(m); 458 return error; 459 } 460 } 461 462 switch (so->so_type) { 463 464 case SOCK_DGRAM: { 465 KASSERT(so->so_lock == uipc_lock); 466 if (nam) { 467 if ((so->so_state & SS_ISCONNECTED) != 0) 468 error = EISCONN; 469 else { 470 /* 471 * Note: once connected, the 472 * socket's lock must not be 473 * dropped until we have sent 474 * the message and disconnected. 475 * This is necessary to prevent 476 * intervening control ops, like 477 * another connection. 478 */ 479 error = unp_connect(so, nam, l); 480 } 481 } else { 482 if ((so->so_state & SS_ISCONNECTED) == 0) 483 error = ENOTCONN; 484 } 485 if (error) { 486 unp_dispose(control); 487 m_freem(control); 488 m_freem(m); 489 return error; 490 } 491 error = unp_output(m, control, unp); 492 if (nam) 493 unp_disconnect1(unp); 494 break; 495 } 496 497 case SOCK_SEQPACKET: /* FALLTHROUGH */ 498 case SOCK_STREAM: 499 #define rcv (&so2->so_rcv) 500 #define snd (&so->so_snd) 501 if (unp->unp_conn == NULL) { 502 error = ENOTCONN; 503 break; 504 } 505 so2 = unp->unp_conn->unp_socket; 506 KASSERT(solocked2(so, so2)); 507 if (unp->unp_conn->unp_flags & UNP_WANTCRED) { 508 /* 509 * Credentials are passed only once on 510 * SOCK_STREAM and SOCK_SEQPACKET. 511 */ 512 unp->unp_conn->unp_flags &= ~UNP_WANTCRED; 513 control = unp_addsockcred(l, control); 514 } 515 #ifdef COMPAT_SOCKCRED70 516 if (unp->unp_conn->unp_flags & UNP_OWANTCRED) { 517 /* 518 * Credentials are passed only once on 519 * SOCK_STREAM and SOCK_SEQPACKET. 520 */ 521 unp->unp_conn->unp_flags &= ~UNP_OWANTCRED; 522 control = compat_70_unp_addsockcred(l, control); 523 } 524 #endif 525 /* 526 * Send to paired receive port, and then reduce 527 * send buffer hiwater marks to maintain backpressure. 528 * Wake up readers. 529 */ 530 if (control) { 531 if (sbappendcontrol(rcv, m, control) != 0) 532 control = NULL; 533 } else { 534 switch(so->so_type) { 535 case SOCK_SEQPACKET: 536 sbappendrecord(rcv, m); 537 break; 538 case SOCK_STREAM: 539 sbappend(rcv, m); 540 break; 541 default: 542 panic("uipc_usrreq"); 543 break; 544 } 545 } 546 snd->sb_mbmax -= 547 rcv->sb_mbcnt - unp->unp_conn->unp_mbcnt; 548 unp->unp_conn->unp_mbcnt = rcv->sb_mbcnt; 549 newhiwat = snd->sb_hiwat - 550 (rcv->sb_cc - unp->unp_conn->unp_cc); 551 (void)chgsbsize(so->so_uidinfo, 552 &snd->sb_hiwat, newhiwat, RLIM_INFINITY); 553 unp->unp_conn->unp_cc = rcv->sb_cc; 554 sorwakeup(so2); 555 #undef snd 556 #undef rcv 557 if (control != NULL) { 558 unp_dispose(control); 559 m_freem(control); 560 } 561 break; 562 563 default: 564 panic("uipc 4"); 565 } 566 567 return error; 568 } 569 570 static int 571 unp_sendoob(struct socket *so, struct mbuf *m, struct mbuf * control) 572 { 573 KASSERT(solocked(so)); 574 575 m_freem(m); 576 m_freem(control); 577 578 return EOPNOTSUPP; 579 } 580 581 /* 582 * Unix domain socket option processing. 583 */ 584 int 585 uipc_ctloutput(int op, struct socket *so, struct sockopt *sopt) 586 { 587 struct unpcb *unp = sotounpcb(so); 588 int optval = 0, error = 0; 589 590 KASSERT(solocked(so)); 591 592 if (sopt->sopt_level != 0) { 593 error = ENOPROTOOPT; 594 } else switch (op) { 595 596 case PRCO_SETOPT: 597 switch (sopt->sopt_name) { 598 case LOCAL_CREDS: 599 case LOCAL_CONNWAIT: 600 #ifdef COMPAT_SOCKCRED70 601 case LOCAL_OCREDS: 602 #endif 603 error = sockopt_getint(sopt, &optval); 604 if (error) 605 break; 606 switch (sopt->sopt_name) { 607 #define OPTSET(bit) \ 608 if (optval) \ 609 unp->unp_flags |= (bit); \ 610 else \ 611 unp->unp_flags &= ~(bit); 612 613 case LOCAL_CREDS: 614 OPTSET(UNP_WANTCRED); 615 break; 616 case LOCAL_CONNWAIT: 617 OPTSET(UNP_CONNWAIT); 618 break; 619 #ifdef COMPAT_SOCKCRED70 620 case LOCAL_OCREDS: 621 OPTSET(UNP_OWANTCRED); 622 break; 623 #endif 624 } 625 break; 626 #undef OPTSET 627 628 default: 629 error = ENOPROTOOPT; 630 break; 631 } 632 break; 633 634 case PRCO_GETOPT: 635 sounlock(so); 636 switch (sopt->sopt_name) { 637 case LOCAL_PEEREID: 638 if (unp->unp_flags & UNP_EIDSVALID) { 639 error = sockopt_set(sopt, &unp->unp_connid, 640 sizeof(unp->unp_connid)); 641 } else { 642 error = EINVAL; 643 } 644 break; 645 case LOCAL_CREDS: 646 #define OPTBIT(bit) (unp->unp_flags & (bit) ? 1 : 0) 647 648 optval = OPTBIT(UNP_WANTCRED); 649 error = sockopt_setint(sopt, optval); 650 break; 651 #ifdef COMPAT_SOCKCRED70 652 case LOCAL_OCREDS: 653 optval = OPTBIT(UNP_OWANTCRED); 654 error = sockopt_setint(sopt, optval); 655 break; 656 #endif 657 #undef OPTBIT 658 659 default: 660 error = ENOPROTOOPT; 661 break; 662 } 663 solock(so); 664 break; 665 } 666 return (error); 667 } 668 669 /* 670 * Both send and receive buffers are allocated PIPSIZ bytes of buffering 671 * for stream sockets, although the total for sender and receiver is 672 * actually only PIPSIZ. 673 * Datagram sockets really use the sendspace as the maximum datagram size, 674 * and don't really want to reserve the sendspace. Their recvspace should 675 * be large enough for at least one max-size datagram plus address. 676 */ 677 #ifndef PIPSIZ 678 #define PIPSIZ 8192 679 #endif 680 u_long unpst_sendspace = PIPSIZ; 681 u_long unpst_recvspace = PIPSIZ; 682 u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 683 u_long unpdg_recvspace = 16*1024; 684 685 u_int unp_rights; /* files in flight */ 686 u_int unp_rights_ratio = 2; /* limit, fraction of maxfiles */ 687 688 static int 689 unp_attach(struct socket *so, int proto) 690 { 691 struct unpcb *unp = sotounpcb(so); 692 u_long sndspc, rcvspc; 693 int error; 694 695 KASSERT(unp == NULL); 696 697 switch (so->so_type) { 698 case SOCK_SEQPACKET: 699 /* FALLTHROUGH */ 700 case SOCK_STREAM: 701 if (so->so_lock == NULL) { 702 so->so_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 703 solock(so); 704 } 705 sndspc = unpst_sendspace; 706 rcvspc = unpst_recvspace; 707 break; 708 709 case SOCK_DGRAM: 710 if (so->so_lock == NULL) { 711 mutex_obj_hold(uipc_lock); 712 so->so_lock = uipc_lock; 713 solock(so); 714 } 715 sndspc = unpdg_sendspace; 716 rcvspc = unpdg_recvspace; 717 break; 718 719 default: 720 panic("unp_attach"); 721 } 722 723 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 724 error = soreserve(so, sndspc, rcvspc); 725 if (error) { 726 return error; 727 } 728 } 729 730 unp = kmem_zalloc(sizeof(*unp), KM_SLEEP); 731 nanotime(&unp->unp_ctime); 732 unp->unp_socket = so; 733 so->so_pcb = unp; 734 735 KASSERT(solocked(so)); 736 return 0; 737 } 738 739 static void 740 unp_detach(struct socket *so) 741 { 742 struct unpcb *unp; 743 vnode_t *vp; 744 745 unp = sotounpcb(so); 746 KASSERT(unp != NULL); 747 KASSERT(solocked(so)); 748 retry: 749 if ((vp = unp->unp_vnode) != NULL) { 750 sounlock(so); 751 /* Acquire v_interlock to protect against unp_connect(). */ 752 /* XXXAD racy */ 753 mutex_enter(vp->v_interlock); 754 vp->v_socket = NULL; 755 mutex_exit(vp->v_interlock); 756 vrele(vp); 757 solock(so); 758 unp->unp_vnode = NULL; 759 } 760 if (unp->unp_conn) 761 unp_disconnect1(unp); 762 while (unp->unp_refs) { 763 KASSERT(solocked2(so, unp->unp_refs->unp_socket)); 764 if (unp_drop(unp->unp_refs, ECONNRESET)) { 765 solock(so); 766 goto retry; 767 } 768 } 769 soisdisconnected(so); 770 so->so_pcb = NULL; 771 if (unp_rights) { 772 /* 773 * Normally the receive buffer is flushed later, in sofree, 774 * but if our receive buffer holds references to files that 775 * are now garbage, we will enqueue those file references to 776 * the garbage collector and kick it into action. 777 */ 778 sorflush(so); 779 unp_free(unp); 780 unp_thread_kick(); 781 } else 782 unp_free(unp); 783 } 784 785 static int 786 unp_accept(struct socket *so, struct sockaddr *nam) 787 { 788 struct unpcb *unp = sotounpcb(so); 789 struct socket *so2; 790 791 KASSERT(solocked(so)); 792 KASSERT(nam != NULL); 793 794 /* XXX code review required to determine if unp can ever be NULL */ 795 if (unp == NULL) 796 return EINVAL; 797 798 KASSERT(so->so_lock == uipc_lock); 799 /* 800 * Mark the initiating STREAM socket as connected *ONLY* 801 * after it's been accepted. This prevents a client from 802 * overrunning a server and receiving ECONNREFUSED. 803 */ 804 if (unp->unp_conn == NULL) { 805 /* 806 * This will use the empty socket and will not 807 * allocate. 808 */ 809 unp_setaddr(so, nam, true); 810 return 0; 811 } 812 so2 = unp->unp_conn->unp_socket; 813 if (so2->so_state & SS_ISCONNECTING) { 814 KASSERT(solocked2(so, so->so_head)); 815 KASSERT(solocked2(so2, so->so_head)); 816 soisconnected(so2); 817 } 818 /* 819 * If the connection is fully established, break the 820 * association with uipc_lock and give the connected 821 * pair a separate lock to share. 822 * There is a race here: sotounpcb(so2)->unp_streamlock 823 * is not locked, so when changing so2->so_lock 824 * another thread can grab it while so->so_lock is still 825 * pointing to the (locked) uipc_lock. 826 * this should be harmless, except that this makes 827 * solocked2() and solocked() unreliable. 828 * Another problem is that unp_setaddr() expects the 829 * the socket locked. Grabing sotounpcb(so2)->unp_streamlock 830 * fixes both issues. 831 */ 832 mutex_enter(sotounpcb(so2)->unp_streamlock); 833 unp_setpeerlocks(so2, so); 834 /* 835 * Only now return peer's address, as we may need to 836 * block in order to allocate memory. 837 * 838 * XXX Minor race: connection can be broken while 839 * lock is dropped in unp_setaddr(). We will return 840 * error == 0 and sun_noname as the peer address. 841 */ 842 unp_setaddr(so, nam, true); 843 /* so_lock now points to unp_streamlock */ 844 mutex_exit(so2->so_lock); 845 return 0; 846 } 847 848 static int 849 unp_ioctl(struct socket *so, u_long cmd, void *nam, struct ifnet *ifp) 850 { 851 return EOPNOTSUPP; 852 } 853 854 static int 855 unp_stat(struct socket *so, struct stat *ub) 856 { 857 struct unpcb *unp; 858 struct socket *so2; 859 860 KASSERT(solocked(so)); 861 862 unp = sotounpcb(so); 863 if (unp == NULL) 864 return EINVAL; 865 866 ub->st_blksize = so->so_snd.sb_hiwat; 867 switch (so->so_type) { 868 case SOCK_SEQPACKET: /* FALLTHROUGH */ 869 case SOCK_STREAM: 870 if (unp->unp_conn == 0) 871 break; 872 873 so2 = unp->unp_conn->unp_socket; 874 KASSERT(solocked2(so, so2)); 875 ub->st_blksize += so2->so_rcv.sb_cc; 876 break; 877 default: 878 break; 879 } 880 ub->st_dev = NODEV; 881 if (unp->unp_ino == 0) 882 unp->unp_ino = unp_ino++; 883 ub->st_atimespec = ub->st_mtimespec = ub->st_ctimespec = unp->unp_ctime; 884 ub->st_ino = unp->unp_ino; 885 return (0); 886 } 887 888 static int 889 unp_peeraddr(struct socket *so, struct sockaddr *nam) 890 { 891 KASSERT(solocked(so)); 892 KASSERT(sotounpcb(so) != NULL); 893 KASSERT(nam != NULL); 894 895 unp_setaddr(so, nam, true); 896 return 0; 897 } 898 899 static int 900 unp_sockaddr(struct socket *so, struct sockaddr *nam) 901 { 902 KASSERT(solocked(so)); 903 KASSERT(sotounpcb(so) != NULL); 904 KASSERT(nam != NULL); 905 906 unp_setaddr(so, nam, false); 907 return 0; 908 } 909 910 /* 911 * we only need to perform this allocation until syscalls other than 912 * bind are adjusted to use sockaddr_big. 913 */ 914 static struct sockaddr_un * 915 makeun_sb(struct sockaddr *nam, size_t *addrlen) 916 { 917 struct sockaddr_un *sun; 918 919 *addrlen = nam->sa_len + 1; 920 sun = malloc(*addrlen, M_SONAME, M_WAITOK); 921 memcpy(sun, nam, nam->sa_len); 922 *(((char *)sun) + nam->sa_len) = '\0'; 923 return sun; 924 } 925 926 static int 927 unp_bind(struct socket *so, struct sockaddr *nam, struct lwp *l) 928 { 929 struct sockaddr_un *sun; 930 struct unpcb *unp; 931 vnode_t *vp; 932 struct vattr vattr; 933 size_t addrlen; 934 int error; 935 struct pathbuf *pb; 936 struct nameidata nd; 937 proc_t *p; 938 939 unp = sotounpcb(so); 940 941 KASSERT(solocked(so)); 942 KASSERT(unp != NULL); 943 KASSERT(nam != NULL); 944 945 if (unp->unp_vnode != NULL) 946 return (EINVAL); 947 if ((unp->unp_flags & UNP_BUSY) != 0) { 948 /* 949 * EALREADY may not be strictly accurate, but since this 950 * is a major application error it's hardly a big deal. 951 */ 952 return (EALREADY); 953 } 954 unp->unp_flags |= UNP_BUSY; 955 sounlock(so); 956 957 p = l->l_proc; 958 sun = makeun_sb(nam, &addrlen); 959 960 pb = pathbuf_create(sun->sun_path); 961 if (pb == NULL) { 962 error = ENOMEM; 963 goto bad; 964 } 965 NDINIT(&nd, CREATE, FOLLOW | LOCKPARENT | TRYEMULROOT, pb); 966 967 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 968 if ((error = namei(&nd)) != 0) { 969 pathbuf_destroy(pb); 970 goto bad; 971 } 972 vp = nd.ni_vp; 973 if (vp != NULL) { 974 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 975 if (nd.ni_dvp == vp) 976 vrele(nd.ni_dvp); 977 else 978 vput(nd.ni_dvp); 979 vrele(vp); 980 pathbuf_destroy(pb); 981 error = EADDRINUSE; 982 goto bad; 983 } 984 vattr_null(&vattr); 985 vattr.va_type = VSOCK; 986 vattr.va_mode = ACCESSPERMS & ~(p->p_cwdi->cwdi_cmask); 987 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 988 if (error) { 989 vput(nd.ni_dvp); 990 pathbuf_destroy(pb); 991 goto bad; 992 } 993 vp = nd.ni_vp; 994 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 995 solock(so); 996 vp->v_socket = unp->unp_socket; 997 unp->unp_vnode = vp; 998 unp->unp_addrlen = addrlen; 999 unp->unp_addr = sun; 1000 VOP_UNLOCK(vp); 1001 vput(nd.ni_dvp); 1002 unp->unp_flags &= ~UNP_BUSY; 1003 pathbuf_destroy(pb); 1004 return (0); 1005 1006 bad: 1007 free(sun, M_SONAME); 1008 solock(so); 1009 unp->unp_flags &= ~UNP_BUSY; 1010 return (error); 1011 } 1012 1013 static int 1014 unp_listen(struct socket *so, struct lwp *l) 1015 { 1016 struct unpcb *unp = sotounpcb(so); 1017 1018 KASSERT(solocked(so)); 1019 KASSERT(unp != NULL); 1020 1021 /* 1022 * If the socket can accept a connection, it must be 1023 * locked by uipc_lock. 1024 */ 1025 unp_resetlock(so); 1026 if (unp->unp_vnode == NULL) 1027 return EINVAL; 1028 1029 unp_connid(l, unp, UNP_EIDSBIND); 1030 return 0; 1031 } 1032 1033 static int 1034 unp_disconnect(struct socket *so) 1035 { 1036 KASSERT(solocked(so)); 1037 KASSERT(sotounpcb(so) != NULL); 1038 1039 unp_disconnect1(sotounpcb(so)); 1040 return 0; 1041 } 1042 1043 static int 1044 unp_shutdown(struct socket *so) 1045 { 1046 KASSERT(solocked(so)); 1047 KASSERT(sotounpcb(so) != NULL); 1048 1049 socantsendmore(so); 1050 unp_shutdown1(sotounpcb(so)); 1051 return 0; 1052 } 1053 1054 static int 1055 unp_abort(struct socket *so) 1056 { 1057 KASSERT(solocked(so)); 1058 KASSERT(sotounpcb(so) != NULL); 1059 1060 (void)unp_drop(sotounpcb(so), ECONNABORTED); 1061 KASSERT(so->so_head == NULL); 1062 KASSERT(so->so_pcb != NULL); 1063 unp_detach(so); 1064 return 0; 1065 } 1066 1067 static int 1068 unp_connect1(struct socket *so, struct socket *so2, struct lwp *l) 1069 { 1070 struct unpcb *unp = sotounpcb(so); 1071 struct unpcb *unp2; 1072 1073 if (so2->so_type != so->so_type) 1074 return EPROTOTYPE; 1075 1076 /* 1077 * All three sockets involved must be locked by same lock: 1078 * 1079 * local endpoint (so) 1080 * remote endpoint (so2) 1081 * queue head (so2->so_head, only if PR_CONNREQUIRED) 1082 */ 1083 KASSERT(solocked2(so, so2)); 1084 KASSERT(so->so_head == NULL); 1085 if (so2->so_head != NULL) { 1086 KASSERT(so2->so_lock == uipc_lock); 1087 KASSERT(solocked2(so2, so2->so_head)); 1088 } 1089 1090 unp2 = sotounpcb(so2); 1091 unp->unp_conn = unp2; 1092 1093 switch (so->so_type) { 1094 1095 case SOCK_DGRAM: 1096 unp->unp_nextref = unp2->unp_refs; 1097 unp2->unp_refs = unp; 1098 soisconnected(so); 1099 break; 1100 1101 case SOCK_SEQPACKET: /* FALLTHROUGH */ 1102 case SOCK_STREAM: 1103 1104 /* 1105 * SOCK_SEQPACKET and SOCK_STREAM cases are handled by callers 1106 * which are unp_connect() or unp_connect2(). 1107 */ 1108 1109 break; 1110 1111 default: 1112 panic("unp_connect1"); 1113 } 1114 1115 return 0; 1116 } 1117 1118 int 1119 unp_connect(struct socket *so, struct sockaddr *nam, struct lwp *l) 1120 { 1121 struct sockaddr_un *sun; 1122 vnode_t *vp; 1123 struct socket *so2, *so3; 1124 struct unpcb *unp, *unp2, *unp3; 1125 size_t addrlen; 1126 int error; 1127 struct pathbuf *pb; 1128 struct nameidata nd; 1129 1130 unp = sotounpcb(so); 1131 if ((unp->unp_flags & UNP_BUSY) != 0) { 1132 /* 1133 * EALREADY may not be strictly accurate, but since this 1134 * is a major application error it's hardly a big deal. 1135 */ 1136 return (EALREADY); 1137 } 1138 unp->unp_flags |= UNP_BUSY; 1139 sounlock(so); 1140 1141 sun = makeun_sb(nam, &addrlen); 1142 pb = pathbuf_create(sun->sun_path); 1143 if (pb == NULL) { 1144 error = ENOMEM; 1145 goto bad2; 1146 } 1147 1148 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb); 1149 1150 if ((error = namei(&nd)) != 0) { 1151 pathbuf_destroy(pb); 1152 goto bad2; 1153 } 1154 vp = nd.ni_vp; 1155 pathbuf_destroy(pb); 1156 if (vp->v_type != VSOCK) { 1157 error = ENOTSOCK; 1158 goto bad; 1159 } 1160 if ((error = VOP_ACCESS(vp, VWRITE, l->l_cred)) != 0) 1161 goto bad; 1162 /* Acquire v_interlock to protect against unp_detach(). */ 1163 mutex_enter(vp->v_interlock); 1164 so2 = vp->v_socket; 1165 if (so2 == NULL) { 1166 mutex_exit(vp->v_interlock); 1167 error = ECONNREFUSED; 1168 goto bad; 1169 } 1170 if (so->so_type != so2->so_type) { 1171 mutex_exit(vp->v_interlock); 1172 error = EPROTOTYPE; 1173 goto bad; 1174 } 1175 solock(so); 1176 unp_resetlock(so); 1177 mutex_exit(vp->v_interlock); 1178 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) { 1179 /* 1180 * This may seem somewhat fragile but is OK: if we can 1181 * see SO_ACCEPTCONN set on the endpoint, then it must 1182 * be locked by the domain-wide uipc_lock. 1183 */ 1184 KASSERT((so2->so_options & SO_ACCEPTCONN) == 0 || 1185 so2->so_lock == uipc_lock); 1186 if ((so2->so_options & SO_ACCEPTCONN) == 0 || 1187 (so3 = sonewconn(so2, false)) == NULL) { 1188 error = ECONNREFUSED; 1189 sounlock(so); 1190 goto bad; 1191 } 1192 unp2 = sotounpcb(so2); 1193 unp3 = sotounpcb(so3); 1194 if (unp2->unp_addr) { 1195 unp3->unp_addr = malloc(unp2->unp_addrlen, 1196 M_SONAME, M_WAITOK); 1197 memcpy(unp3->unp_addr, unp2->unp_addr, 1198 unp2->unp_addrlen); 1199 unp3->unp_addrlen = unp2->unp_addrlen; 1200 } 1201 unp3->unp_flags = unp2->unp_flags; 1202 so2 = so3; 1203 /* 1204 * The connector's (client's) credentials are copied from its 1205 * process structure at the time of connect() (which is now). 1206 */ 1207 unp_connid(l, unp3, UNP_EIDSVALID); 1208 /* 1209 * The receiver's (server's) credentials are copied from the 1210 * unp_peercred member of socket on which the former called 1211 * listen(); unp_listen() cached that process's credentials 1212 * at that time so we can use them now. 1213 */ 1214 if (unp2->unp_flags & UNP_EIDSBIND) { 1215 memcpy(&unp->unp_connid, &unp2->unp_connid, 1216 sizeof(unp->unp_connid)); 1217 unp->unp_flags |= UNP_EIDSVALID; 1218 } 1219 } 1220 error = unp_connect1(so, so2, l); 1221 if (error) { 1222 sounlock(so); 1223 goto bad; 1224 } 1225 unp2 = sotounpcb(so2); 1226 switch (so->so_type) { 1227 1228 /* 1229 * SOCK_DGRAM and default cases are handled in prior call to 1230 * unp_connect1(), do not add a default case without fixing 1231 * unp_connect1(). 1232 */ 1233 1234 case SOCK_SEQPACKET: /* FALLTHROUGH */ 1235 case SOCK_STREAM: 1236 unp2->unp_conn = unp; 1237 if ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT) 1238 soisconnecting(so); 1239 else 1240 soisconnected(so); 1241 soisconnected(so2); 1242 /* 1243 * If the connection is fully established, break the 1244 * association with uipc_lock and give the connected 1245 * pair a seperate lock to share. 1246 */ 1247 KASSERT(so2->so_head != NULL); 1248 unp_setpeerlocks(so, so2); 1249 break; 1250 1251 } 1252 sounlock(so); 1253 bad: 1254 vput(vp); 1255 bad2: 1256 free(sun, M_SONAME); 1257 solock(so); 1258 unp->unp_flags &= ~UNP_BUSY; 1259 return (error); 1260 } 1261 1262 int 1263 unp_connect2(struct socket *so, struct socket *so2) 1264 { 1265 struct unpcb *unp = sotounpcb(so); 1266 struct unpcb *unp2; 1267 int error = 0; 1268 1269 KASSERT(solocked2(so, so2)); 1270 1271 error = unp_connect1(so, so2, curlwp); 1272 if (error) 1273 return error; 1274 1275 unp2 = sotounpcb(so2); 1276 switch (so->so_type) { 1277 1278 /* 1279 * SOCK_DGRAM and default cases are handled in prior call to 1280 * unp_connect1(), do not add a default case without fixing 1281 * unp_connect1(). 1282 */ 1283 1284 case SOCK_SEQPACKET: /* FALLTHROUGH */ 1285 case SOCK_STREAM: 1286 unp2->unp_conn = unp; 1287 soisconnected(so); 1288 soisconnected(so2); 1289 break; 1290 1291 } 1292 return error; 1293 } 1294 1295 static void 1296 unp_disconnect1(struct unpcb *unp) 1297 { 1298 struct unpcb *unp2 = unp->unp_conn; 1299 struct socket *so; 1300 1301 if (unp2 == 0) 1302 return; 1303 unp->unp_conn = 0; 1304 so = unp->unp_socket; 1305 switch (so->so_type) { 1306 case SOCK_DGRAM: 1307 if (unp2->unp_refs == unp) 1308 unp2->unp_refs = unp->unp_nextref; 1309 else { 1310 unp2 = unp2->unp_refs; 1311 for (;;) { 1312 KASSERT(solocked2(so, unp2->unp_socket)); 1313 if (unp2 == 0) 1314 panic("unp_disconnect1"); 1315 if (unp2->unp_nextref == unp) 1316 break; 1317 unp2 = unp2->unp_nextref; 1318 } 1319 unp2->unp_nextref = unp->unp_nextref; 1320 } 1321 unp->unp_nextref = 0; 1322 so->so_state &= ~SS_ISCONNECTED; 1323 break; 1324 1325 case SOCK_SEQPACKET: /* FALLTHROUGH */ 1326 case SOCK_STREAM: 1327 KASSERT(solocked2(so, unp2->unp_socket)); 1328 soisdisconnected(so); 1329 unp2->unp_conn = 0; 1330 soisdisconnected(unp2->unp_socket); 1331 break; 1332 } 1333 } 1334 1335 static void 1336 unp_shutdown1(struct unpcb *unp) 1337 { 1338 struct socket *so; 1339 1340 switch(unp->unp_socket->so_type) { 1341 case SOCK_SEQPACKET: /* FALLTHROUGH */ 1342 case SOCK_STREAM: 1343 if (unp->unp_conn && (so = unp->unp_conn->unp_socket)) 1344 socantrcvmore(so); 1345 break; 1346 default: 1347 break; 1348 } 1349 } 1350 1351 static bool 1352 unp_drop(struct unpcb *unp, int errno) 1353 { 1354 struct socket *so = unp->unp_socket; 1355 1356 KASSERT(solocked(so)); 1357 1358 so->so_error = errno; 1359 unp_disconnect1(unp); 1360 if (so->so_head) { 1361 so->so_pcb = NULL; 1362 /* sofree() drops the socket lock */ 1363 sofree(so); 1364 unp_free(unp); 1365 return true; 1366 } 1367 return false; 1368 } 1369 1370 #ifdef notdef 1371 unp_drain(void) 1372 { 1373 1374 } 1375 #endif 1376 1377 int 1378 unp_externalize(struct mbuf *rights, struct lwp *l, int flags) 1379 { 1380 struct cmsghdr * const cm = mtod(rights, struct cmsghdr *); 1381 struct proc * const p = l->l_proc; 1382 file_t **rp; 1383 int error = 0; 1384 1385 const size_t nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) / 1386 sizeof(file_t *); 1387 if (nfds == 0) 1388 goto noop; 1389 1390 int * const fdp = kmem_alloc(nfds * sizeof(int), KM_SLEEP); 1391 rw_enter(&p->p_cwdi->cwdi_lock, RW_READER); 1392 1393 /* Make sure the recipient should be able to see the files.. */ 1394 rp = (file_t **)CMSG_DATA(cm); 1395 for (size_t i = 0; i < nfds; i++) { 1396 file_t * const fp = *rp++; 1397 if (fp == NULL) { 1398 error = EINVAL; 1399 goto out; 1400 } 1401 /* 1402 * If we are in a chroot'ed directory, and 1403 * someone wants to pass us a directory, make 1404 * sure it's inside the subtree we're allowed 1405 * to access. 1406 */ 1407 if (p->p_cwdi->cwdi_rdir != NULL && fp->f_type == DTYPE_VNODE) { 1408 vnode_t *vp = fp->f_vnode; 1409 if ((vp->v_type == VDIR) && 1410 !vn_isunder(vp, p->p_cwdi->cwdi_rdir, l)) { 1411 error = EPERM; 1412 goto out; 1413 } 1414 } 1415 } 1416 1417 restart: 1418 /* 1419 * First loop -- allocate file descriptor table slots for the 1420 * new files. 1421 */ 1422 for (size_t i = 0; i < nfds; i++) { 1423 if ((error = fd_alloc(p, 0, &fdp[i])) != 0) { 1424 /* 1425 * Back out what we've done so far. 1426 */ 1427 while (i-- > 0) { 1428 fd_abort(p, NULL, fdp[i]); 1429 } 1430 if (error == ENOSPC) { 1431 fd_tryexpand(p); 1432 error = 0; 1433 goto restart; 1434 } 1435 /* 1436 * This is the error that has historically 1437 * been returned, and some callers may 1438 * expect it. 1439 */ 1440 error = EMSGSIZE; 1441 goto out; 1442 } 1443 } 1444 1445 /* 1446 * Now that adding them has succeeded, update all of the 1447 * file passing state and affix the descriptors. 1448 */ 1449 rp = (file_t **)CMSG_DATA(cm); 1450 int *ofdp = (int *)CMSG_DATA(cm); 1451 for (size_t i = 0; i < nfds; i++) { 1452 file_t * const fp = *rp++; 1453 const int fd = fdp[i]; 1454 atomic_dec_uint(&unp_rights); 1455 fd_set_exclose(l, fd, (flags & O_CLOEXEC) != 0); 1456 fd_affix(p, fp, fd); 1457 /* 1458 * Done with this file pointer, replace it with a fd; 1459 */ 1460 *ofdp++ = fd; 1461 mutex_enter(&fp->f_lock); 1462 fp->f_msgcount--; 1463 mutex_exit(&fp->f_lock); 1464 /* 1465 * Note that fd_affix() adds a reference to the file. 1466 * The file may already have been closed by another 1467 * LWP in the process, so we must drop the reference 1468 * added by unp_internalize() with closef(). 1469 */ 1470 closef(fp); 1471 } 1472 1473 /* 1474 * Adjust length, in case of transition from large file_t 1475 * pointers to ints. 1476 */ 1477 if (sizeof(file_t *) != sizeof(int)) { 1478 cm->cmsg_len = CMSG_LEN(nfds * sizeof(int)); 1479 rights->m_len = CMSG_SPACE(nfds * sizeof(int)); 1480 } 1481 out: 1482 if (__predict_false(error != 0)) { 1483 file_t **const fpp = (file_t **)CMSG_DATA(cm); 1484 for (size_t i = 0; i < nfds; i++) 1485 unp_discard_now(fpp[i]); 1486 /* 1487 * Truncate the array so that nobody will try to interpret 1488 * what is now garbage in it. 1489 */ 1490 cm->cmsg_len = CMSG_LEN(0); 1491 rights->m_len = CMSG_SPACE(0); 1492 } 1493 rw_exit(&p->p_cwdi->cwdi_lock); 1494 kmem_free(fdp, nfds * sizeof(int)); 1495 1496 noop: 1497 /* 1498 * Don't disclose kernel memory in the alignment space. 1499 */ 1500 KASSERT(cm->cmsg_len <= rights->m_len); 1501 memset(&mtod(rights, char *)[cm->cmsg_len], 0, rights->m_len - 1502 cm->cmsg_len); 1503 return error; 1504 } 1505 1506 static int 1507 unp_internalize(struct mbuf **controlp) 1508 { 1509 filedesc_t *fdescp = curlwp->l_fd; 1510 struct mbuf *control = *controlp; 1511 struct cmsghdr *newcm, *cm = mtod(control, struct cmsghdr *); 1512 file_t **rp, **files; 1513 file_t *fp; 1514 int i, fd, *fdp; 1515 int nfds, error; 1516 u_int maxmsg; 1517 1518 error = 0; 1519 newcm = NULL; 1520 1521 /* Sanity check the control message header. */ 1522 if (cm->cmsg_type != SCM_RIGHTS || cm->cmsg_level != SOL_SOCKET || 1523 cm->cmsg_len > control->m_len || 1524 cm->cmsg_len < CMSG_ALIGN(sizeof(*cm))) 1525 return (EINVAL); 1526 1527 /* 1528 * Verify that the file descriptors are valid, and acquire 1529 * a reference to each. 1530 */ 1531 nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) / sizeof(int); 1532 fdp = (int *)CMSG_DATA(cm); 1533 maxmsg = maxfiles / unp_rights_ratio; 1534 for (i = 0; i < nfds; i++) { 1535 fd = *fdp++; 1536 if (atomic_inc_uint_nv(&unp_rights) > maxmsg) { 1537 atomic_dec_uint(&unp_rights); 1538 nfds = i; 1539 error = EAGAIN; 1540 goto out; 1541 } 1542 if ((fp = fd_getfile(fd)) == NULL 1543 || fp->f_type == DTYPE_KQUEUE) { 1544 if (fp) 1545 fd_putfile(fd); 1546 atomic_dec_uint(&unp_rights); 1547 nfds = i; 1548 error = EBADF; 1549 goto out; 1550 } 1551 } 1552 1553 /* Allocate new space and copy header into it. */ 1554 newcm = malloc(CMSG_SPACE(nfds * sizeof(file_t *)), M_MBUF, M_WAITOK); 1555 if (newcm == NULL) { 1556 error = E2BIG; 1557 goto out; 1558 } 1559 memcpy(newcm, cm, sizeof(struct cmsghdr)); 1560 files = (file_t **)CMSG_DATA(newcm); 1561 1562 /* 1563 * Transform the file descriptors into file_t pointers, in 1564 * reverse order so that if pointers are bigger than ints, the 1565 * int won't get until we're done. No need to lock, as we have 1566 * already validated the descriptors with fd_getfile(). 1567 */ 1568 fdp = (int *)CMSG_DATA(cm) + nfds; 1569 rp = files + nfds; 1570 for (i = 0; i < nfds; i++) { 1571 fp = fdescp->fd_dt->dt_ff[*--fdp]->ff_file; 1572 KASSERT(fp != NULL); 1573 mutex_enter(&fp->f_lock); 1574 *--rp = fp; 1575 fp->f_count++; 1576 fp->f_msgcount++; 1577 mutex_exit(&fp->f_lock); 1578 } 1579 1580 out: 1581 /* Release descriptor references. */ 1582 fdp = (int *)CMSG_DATA(cm); 1583 for (i = 0; i < nfds; i++) { 1584 fd_putfile(*fdp++); 1585 if (error != 0) { 1586 atomic_dec_uint(&unp_rights); 1587 } 1588 } 1589 1590 if (error == 0) { 1591 if (control->m_flags & M_EXT) { 1592 m_freem(control); 1593 *controlp = control = m_get(M_WAIT, MT_CONTROL); 1594 } 1595 MEXTADD(control, newcm, CMSG_SPACE(nfds * sizeof(file_t *)), 1596 M_MBUF, NULL, NULL); 1597 cm = newcm; 1598 /* 1599 * Adjust message & mbuf to note amount of space 1600 * actually used. 1601 */ 1602 cm->cmsg_len = CMSG_LEN(nfds * sizeof(file_t *)); 1603 control->m_len = CMSG_SPACE(nfds * sizeof(file_t *)); 1604 } 1605 1606 return error; 1607 } 1608 1609 struct mbuf * 1610 unp_addsockcred(struct lwp *l, struct mbuf *control) 1611 { 1612 struct sockcred *sc; 1613 struct mbuf *m; 1614 void *p; 1615 1616 m = sbcreatecontrol1(&p, SOCKCREDSIZE(kauth_cred_ngroups(l->l_cred)), 1617 SCM_CREDS, SOL_SOCKET, M_WAITOK); 1618 if (m == NULL) 1619 return control; 1620 1621 sc = p; 1622 sc->sc_pid = l->l_proc->p_pid; 1623 sc->sc_uid = kauth_cred_getuid(l->l_cred); 1624 sc->sc_euid = kauth_cred_geteuid(l->l_cred); 1625 sc->sc_gid = kauth_cred_getgid(l->l_cred); 1626 sc->sc_egid = kauth_cred_getegid(l->l_cred); 1627 sc->sc_ngroups = kauth_cred_ngroups(l->l_cred); 1628 1629 for (int i = 0; i < sc->sc_ngroups; i++) 1630 sc->sc_groups[i] = kauth_cred_group(l->l_cred, i); 1631 1632 return m_add(control, m); 1633 } 1634 1635 /* 1636 * Do a mark-sweep GC of files in the system, to free up any which are 1637 * caught in flight to an about-to-be-closed socket. Additionally, 1638 * process deferred file closures. 1639 */ 1640 static void 1641 unp_gc(file_t *dp) 1642 { 1643 extern struct domain unixdomain; 1644 file_t *fp, *np; 1645 struct socket *so, *so1; 1646 u_int i, oflags, rflags; 1647 bool didwork; 1648 1649 KASSERT(curlwp == unp_thread_lwp); 1650 KASSERT(mutex_owned(&filelist_lock)); 1651 1652 /* 1653 * First, process deferred file closures. 1654 */ 1655 while (!SLIST_EMPTY(&unp_thread_discard)) { 1656 fp = SLIST_FIRST(&unp_thread_discard); 1657 KASSERT(fp->f_unpcount > 0); 1658 KASSERT(fp->f_count > 0); 1659 KASSERT(fp->f_msgcount > 0); 1660 KASSERT(fp->f_count >= fp->f_unpcount); 1661 KASSERT(fp->f_count >= fp->f_msgcount); 1662 KASSERT(fp->f_msgcount >= fp->f_unpcount); 1663 SLIST_REMOVE_HEAD(&unp_thread_discard, f_unplist); 1664 i = fp->f_unpcount; 1665 fp->f_unpcount = 0; 1666 mutex_exit(&filelist_lock); 1667 for (; i != 0; i--) { 1668 unp_discard_now(fp); 1669 } 1670 mutex_enter(&filelist_lock); 1671 } 1672 1673 /* 1674 * Clear mark bits. Ensure that we don't consider new files 1675 * entering the file table during this loop (they will not have 1676 * FSCAN set). 1677 */ 1678 unp_defer = 0; 1679 LIST_FOREACH(fp, &filehead, f_list) { 1680 for (oflags = fp->f_flag;; oflags = rflags) { 1681 rflags = atomic_cas_uint(&fp->f_flag, oflags, 1682 (oflags | FSCAN) & ~(FMARK|FDEFER)); 1683 if (__predict_true(oflags == rflags)) { 1684 break; 1685 } 1686 } 1687 } 1688 1689 /* 1690 * Iterate over the set of sockets, marking ones believed (based on 1691 * refcount) to be referenced from a process, and marking for rescan 1692 * sockets which are queued on a socket. Recan continues descending 1693 * and searching for sockets referenced by sockets (FDEFER), until 1694 * there are no more socket->socket references to be discovered. 1695 */ 1696 do { 1697 didwork = false; 1698 for (fp = LIST_FIRST(&filehead); fp != NULL; fp = np) { 1699 KASSERT(mutex_owned(&filelist_lock)); 1700 np = LIST_NEXT(fp, f_list); 1701 mutex_enter(&fp->f_lock); 1702 if ((fp->f_flag & FDEFER) != 0) { 1703 atomic_and_uint(&fp->f_flag, ~FDEFER); 1704 unp_defer--; 1705 if (fp->f_count == 0) { 1706 /* 1707 * XXX: closef() doesn't pay attention 1708 * to FDEFER 1709 */ 1710 mutex_exit(&fp->f_lock); 1711 continue; 1712 } 1713 } else { 1714 if (fp->f_count == 0 || 1715 (fp->f_flag & FMARK) != 0 || 1716 fp->f_count == fp->f_msgcount || 1717 fp->f_unpcount != 0) { 1718 mutex_exit(&fp->f_lock); 1719 continue; 1720 } 1721 } 1722 atomic_or_uint(&fp->f_flag, FMARK); 1723 1724 if (fp->f_type != DTYPE_SOCKET || 1725 (so = fp->f_socket) == NULL || 1726 so->so_proto->pr_domain != &unixdomain || 1727 (so->so_proto->pr_flags & PR_RIGHTS) == 0) { 1728 mutex_exit(&fp->f_lock); 1729 continue; 1730 } 1731 1732 /* Gain file ref, mark our position, and unlock. */ 1733 didwork = true; 1734 LIST_INSERT_AFTER(fp, dp, f_list); 1735 fp->f_count++; 1736 mutex_exit(&fp->f_lock); 1737 mutex_exit(&filelist_lock); 1738 1739 /* 1740 * Mark files referenced from sockets queued on the 1741 * accept queue as well. 1742 */ 1743 solock(so); 1744 unp_scan(so->so_rcv.sb_mb, unp_mark, 0); 1745 if ((so->so_options & SO_ACCEPTCONN) != 0) { 1746 TAILQ_FOREACH(so1, &so->so_q0, so_qe) { 1747 unp_scan(so1->so_rcv.sb_mb, unp_mark, 0); 1748 } 1749 TAILQ_FOREACH(so1, &so->so_q, so_qe) { 1750 unp_scan(so1->so_rcv.sb_mb, unp_mark, 0); 1751 } 1752 } 1753 sounlock(so); 1754 1755 /* Re-lock and restart from where we left off. */ 1756 closef(fp); 1757 mutex_enter(&filelist_lock); 1758 np = LIST_NEXT(dp, f_list); 1759 LIST_REMOVE(dp, f_list); 1760 } 1761 /* 1762 * Bail early if we did nothing in the loop above. Could 1763 * happen because of concurrent activity causing unp_defer 1764 * to get out of sync. 1765 */ 1766 } while (unp_defer != 0 && didwork); 1767 1768 /* 1769 * Sweep pass. 1770 * 1771 * We grab an extra reference to each of the files that are 1772 * not otherwise accessible and then free the rights that are 1773 * stored in messages on them. 1774 */ 1775 for (fp = LIST_FIRST(&filehead); fp != NULL; fp = np) { 1776 KASSERT(mutex_owned(&filelist_lock)); 1777 np = LIST_NEXT(fp, f_list); 1778 mutex_enter(&fp->f_lock); 1779 1780 /* 1781 * Ignore non-sockets. 1782 * Ignore dead sockets, or sockets with pending close. 1783 * Ignore sockets obviously referenced elsewhere. 1784 * Ignore sockets marked as referenced by our scan. 1785 * Ignore new sockets that did not exist during the scan. 1786 */ 1787 if (fp->f_type != DTYPE_SOCKET || 1788 fp->f_count == 0 || fp->f_unpcount != 0 || 1789 fp->f_count != fp->f_msgcount || 1790 (fp->f_flag & (FMARK | FSCAN)) != FSCAN) { 1791 mutex_exit(&fp->f_lock); 1792 continue; 1793 } 1794 1795 /* Gain file ref, mark our position, and unlock. */ 1796 LIST_INSERT_AFTER(fp, dp, f_list); 1797 fp->f_count++; 1798 mutex_exit(&fp->f_lock); 1799 mutex_exit(&filelist_lock); 1800 1801 /* 1802 * Flush all data from the socket's receive buffer. 1803 * This will cause files referenced only by the 1804 * socket to be queued for close. 1805 */ 1806 so = fp->f_socket; 1807 solock(so); 1808 sorflush(so); 1809 sounlock(so); 1810 1811 /* Re-lock and restart from where we left off. */ 1812 closef(fp); 1813 mutex_enter(&filelist_lock); 1814 np = LIST_NEXT(dp, f_list); 1815 LIST_REMOVE(dp, f_list); 1816 } 1817 } 1818 1819 /* 1820 * Garbage collector thread. While SCM_RIGHTS messages are in transit, 1821 * wake once per second to garbage collect. Run continually while we 1822 * have deferred closes to process. 1823 */ 1824 static void 1825 unp_thread(void *cookie) 1826 { 1827 file_t *dp; 1828 1829 /* Allocate a dummy file for our scans. */ 1830 if ((dp = fgetdummy()) == NULL) { 1831 panic("unp_thread"); 1832 } 1833 1834 mutex_enter(&filelist_lock); 1835 for (;;) { 1836 KASSERT(mutex_owned(&filelist_lock)); 1837 if (SLIST_EMPTY(&unp_thread_discard)) { 1838 if (unp_rights != 0) { 1839 (void)cv_timedwait(&unp_thread_cv, 1840 &filelist_lock, hz); 1841 } else { 1842 cv_wait(&unp_thread_cv, &filelist_lock); 1843 } 1844 } 1845 unp_gc(dp); 1846 } 1847 /* NOTREACHED */ 1848 } 1849 1850 /* 1851 * Kick the garbage collector into action if there is something for 1852 * it to process. 1853 */ 1854 static void 1855 unp_thread_kick(void) 1856 { 1857 1858 if (!SLIST_EMPTY(&unp_thread_discard) || unp_rights != 0) { 1859 mutex_enter(&filelist_lock); 1860 cv_signal(&unp_thread_cv); 1861 mutex_exit(&filelist_lock); 1862 } 1863 } 1864 1865 void 1866 unp_dispose(struct mbuf *m) 1867 { 1868 1869 if (m) 1870 unp_scan(m, unp_discard_later, 1); 1871 } 1872 1873 void 1874 unp_scan(struct mbuf *m0, void (*op)(file_t *), int discard) 1875 { 1876 struct mbuf *m; 1877 file_t **rp, *fp; 1878 struct cmsghdr *cm; 1879 int i, qfds; 1880 1881 while (m0) { 1882 for (m = m0; m; m = m->m_next) { 1883 if (m->m_type != MT_CONTROL || 1884 m->m_len < sizeof(*cm)) { 1885 continue; 1886 } 1887 cm = mtod(m, struct cmsghdr *); 1888 if (cm->cmsg_level != SOL_SOCKET || 1889 cm->cmsg_type != SCM_RIGHTS) 1890 continue; 1891 qfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) 1892 / sizeof(file_t *); 1893 rp = (file_t **)CMSG_DATA(cm); 1894 for (i = 0; i < qfds; i++) { 1895 fp = *rp; 1896 if (discard) { 1897 *rp = 0; 1898 } 1899 (*op)(fp); 1900 rp++; 1901 } 1902 } 1903 m0 = m0->m_nextpkt; 1904 } 1905 } 1906 1907 void 1908 unp_mark(file_t *fp) 1909 { 1910 1911 if (fp == NULL) 1912 return; 1913 1914 /* If we're already deferred, don't screw up the defer count */ 1915 mutex_enter(&fp->f_lock); 1916 if (fp->f_flag & (FMARK | FDEFER)) { 1917 mutex_exit(&fp->f_lock); 1918 return; 1919 } 1920 1921 /* 1922 * Minimize the number of deferrals... Sockets are the only type of 1923 * file which can hold references to another file, so just mark 1924 * other files, and defer unmarked sockets for the next pass. 1925 */ 1926 if (fp->f_type == DTYPE_SOCKET) { 1927 unp_defer++; 1928 KASSERT(fp->f_count != 0); 1929 atomic_or_uint(&fp->f_flag, FDEFER); 1930 } else { 1931 atomic_or_uint(&fp->f_flag, FMARK); 1932 } 1933 mutex_exit(&fp->f_lock); 1934 } 1935 1936 static void 1937 unp_discard_now(file_t *fp) 1938 { 1939 1940 if (fp == NULL) 1941 return; 1942 1943 KASSERT(fp->f_count > 0); 1944 KASSERT(fp->f_msgcount > 0); 1945 1946 mutex_enter(&fp->f_lock); 1947 fp->f_msgcount--; 1948 mutex_exit(&fp->f_lock); 1949 atomic_dec_uint(&unp_rights); 1950 (void)closef(fp); 1951 } 1952 1953 static void 1954 unp_discard_later(file_t *fp) 1955 { 1956 1957 if (fp == NULL) 1958 return; 1959 1960 KASSERT(fp->f_count > 0); 1961 KASSERT(fp->f_msgcount > 0); 1962 1963 mutex_enter(&filelist_lock); 1964 if (fp->f_unpcount++ == 0) { 1965 SLIST_INSERT_HEAD(&unp_thread_discard, fp, f_unplist); 1966 } 1967 mutex_exit(&filelist_lock); 1968 } 1969 1970 void 1971 unp_sysctl_create(struct sysctllog **clog) 1972 { 1973 sysctl_createv(clog, 0, NULL, NULL, 1974 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1975 CTLTYPE_LONG, "sendspace", 1976 SYSCTL_DESCR("Default stream send space"), 1977 NULL, 0, &unpst_sendspace, 0, 1978 CTL_NET, PF_LOCAL, SOCK_STREAM, CTL_CREATE, CTL_EOL); 1979 sysctl_createv(clog, 0, NULL, NULL, 1980 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1981 CTLTYPE_LONG, "recvspace", 1982 SYSCTL_DESCR("Default stream recv space"), 1983 NULL, 0, &unpst_recvspace, 0, 1984 CTL_NET, PF_LOCAL, SOCK_STREAM, CTL_CREATE, CTL_EOL); 1985 sysctl_createv(clog, 0, NULL, NULL, 1986 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1987 CTLTYPE_LONG, "sendspace", 1988 SYSCTL_DESCR("Default datagram send space"), 1989 NULL, 0, &unpdg_sendspace, 0, 1990 CTL_NET, PF_LOCAL, SOCK_DGRAM, CTL_CREATE, CTL_EOL); 1991 sysctl_createv(clog, 0, NULL, NULL, 1992 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1993 CTLTYPE_LONG, "recvspace", 1994 SYSCTL_DESCR("Default datagram recv space"), 1995 NULL, 0, &unpdg_recvspace, 0, 1996 CTL_NET, PF_LOCAL, SOCK_DGRAM, CTL_CREATE, CTL_EOL); 1997 sysctl_createv(clog, 0, NULL, NULL, 1998 CTLFLAG_PERMANENT|CTLFLAG_READONLY, 1999 CTLTYPE_INT, "inflight", 2000 SYSCTL_DESCR("File descriptors in flight"), 2001 NULL, 0, &unp_rights, 0, 2002 CTL_NET, PF_LOCAL, CTL_CREATE, CTL_EOL); 2003 sysctl_createv(clog, 0, NULL, NULL, 2004 CTLFLAG_PERMANENT|CTLFLAG_READONLY, 2005 CTLTYPE_INT, "deferred", 2006 SYSCTL_DESCR("File descriptors deferred for close"), 2007 NULL, 0, &unp_defer, 0, 2008 CTL_NET, PF_LOCAL, CTL_CREATE, CTL_EOL); 2009 } 2010 2011 const struct pr_usrreqs unp_usrreqs = { 2012 .pr_attach = unp_attach, 2013 .pr_detach = unp_detach, 2014 .pr_accept = unp_accept, 2015 .pr_bind = unp_bind, 2016 .pr_listen = unp_listen, 2017 .pr_connect = unp_connect, 2018 .pr_connect2 = unp_connect2, 2019 .pr_disconnect = unp_disconnect, 2020 .pr_shutdown = unp_shutdown, 2021 .pr_abort = unp_abort, 2022 .pr_ioctl = unp_ioctl, 2023 .pr_stat = unp_stat, 2024 .pr_peeraddr = unp_peeraddr, 2025 .pr_sockaddr = unp_sockaddr, 2026 .pr_rcvd = unp_rcvd, 2027 .pr_recvoob = unp_recvoob, 2028 .pr_send = unp_send, 2029 .pr_sendoob = unp_sendoob, 2030 }; 2031