uipc_usrreq.c revision 145312
1/*- 2 * Copyright 2004-2005 Robert N. M. Watson 3 * Copyright (c) 1982, 1986, 1989, 1991, 1993 4 * The Regents of the University of California. All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 4. Neither the name of the University nor the names of its contributors 15 * may be used to endorse or promote products derived from this software 16 * without specific prior written permission. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 21 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 28 * SUCH DAMAGE. 29 * 30 * From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94 31 */ 32 33#include <sys/cdefs.h> 34__FBSDID("$FreeBSD: head/sys/kern/uipc_usrreq.c 145312 2005-04-20 02:57:56Z mdodd $"); 35 36#include "opt_mac.h" 37 38#include <sys/param.h> 39#include <sys/domain.h> 40#include <sys/fcntl.h> 41#include <sys/malloc.h> /* XXX must be before <sys/file.h> */ 42#include <sys/file.h> 43#include <sys/filedesc.h> 44#include <sys/jail.h> 45#include <sys/kernel.h> 46#include <sys/lock.h> 47#include <sys/mac.h> 48#include <sys/mbuf.h> 49#include <sys/mutex.h> 50#include <sys/namei.h> 51#include <sys/proc.h> 52#include <sys/protosw.h> 53#include <sys/resourcevar.h> 54#include <sys/socket.h> 55#include <sys/socketvar.h> 56#include <sys/signalvar.h> 57#include <sys/stat.h> 58#include <sys/sx.h> 59#include <sys/sysctl.h> 60#include <sys/systm.h> 61#include <sys/un.h> 62#include <sys/unpcb.h> 63#include <sys/vnode.h> 64 65#include <vm/uma.h> 66 67static uma_zone_t unp_zone; 68static unp_gen_t unp_gencnt; 69static u_int unp_count; 70 71static struct unp_head unp_shead, unp_dhead; 72 73/* 74 * Unix communications domain. 75 * 76 * TODO: 77 * SEQPACKET, RDM 78 * rethink name space problems 79 * need a proper out-of-band 80 * lock pushdown 81 */ 82static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL }; 83static ino_t unp_ino; /* prototype for fake inode numbers */ 84struct mbuf *unp_addsockcred(struct thread *, struct mbuf *); 85 86/* 87 * Currently, UNIX domain sockets are protected by a single subsystem lock, 88 * which covers global data structures and variables, the contents of each 89 * per-socket unpcb structure, and the so_pcb field in sockets attached to 90 * the UNIX domain. This provides for a moderate degree of paralellism, as 91 * receive operations on UNIX domain sockets do not need to acquire the 92 * subsystem lock. Finer grained locking to permit send() without acquiring 93 * a global lock would be a logical next step. 94 * 95 * The UNIX domain socket lock preceds all socket layer locks, including the 96 * socket lock and socket buffer lock, permitting UNIX domain socket code to 97 * call into socket support routines without releasing its locks. 98 * 99 * Some caution is required in areas where the UNIX domain socket code enters 100 * VFS in order to create or find rendezvous points. This results in 101 * dropping of the UNIX domain socket subsystem lock, acquisition of the 102 * Giant lock, and potential sleeping. This increases the chances of races, 103 * and exposes weaknesses in the socket->protocol API by offering poor 104 * failure modes. 105 */ 106static struct mtx unp_mtx; 107#define UNP_LOCK_INIT() \ 108 mtx_init(&unp_mtx, "unp", NULL, MTX_DEF) 109#define UNP_LOCK() mtx_lock(&unp_mtx) 110#define UNP_UNLOCK() mtx_unlock(&unp_mtx) 111#define UNP_LOCK_ASSERT() mtx_assert(&unp_mtx, MA_OWNED) 112#define UNP_UNLOCK_ASSERT() mtx_assert(&unp_mtx, MA_NOTOWNED) 113 114static int unp_attach(struct socket *); 115static void unp_detach(struct unpcb *); 116static int unp_bind(struct unpcb *,struct sockaddr *, struct thread *); 117static int unp_connect(struct socket *,struct sockaddr *, struct thread *); 118static int unp_connect2(struct socket *so, struct socket *so2, int); 119static void unp_disconnect(struct unpcb *); 120static void unp_shutdown(struct unpcb *); 121static void unp_drop(struct unpcb *, int); 122static void unp_gc(void); 123static void unp_scan(struct mbuf *, void (*)(struct file *)); 124static void unp_mark(struct file *); 125static void unp_discard(struct file *); 126static void unp_freerights(struct file **, int); 127static int unp_internalize(struct mbuf **, struct thread *); 128static int unp_listen(struct socket *, struct unpcb *, struct thread *); 129 130static int 131uipc_abort(struct socket *so) 132{ 133 struct unpcb *unp; 134 135 UNP_LOCK(); 136 unp = sotounpcb(so); 137 if (unp == NULL) { 138 UNP_UNLOCK(); 139 return (EINVAL); 140 } 141 unp_drop(unp, ECONNABORTED); 142 unp_detach(unp); 143 UNP_UNLOCK_ASSERT(); 144 ACCEPT_LOCK(); 145 SOCK_LOCK(so); 146 sotryfree(so); 147 return (0); 148} 149 150static int 151uipc_accept(struct socket *so, struct sockaddr **nam) 152{ 153 struct unpcb *unp; 154 const struct sockaddr *sa; 155 156 /* 157 * Pass back name of connected socket, 158 * if it was bound and we are still connected 159 * (our peer may have closed already!). 160 */ 161 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 162 UNP_LOCK(); 163 unp = sotounpcb(so); 164 if (unp == NULL) { 165 UNP_UNLOCK(); 166 free(*nam, M_SONAME); 167 *nam = NULL; 168 return (EINVAL); 169 } 170 if (unp->unp_conn != NULL && unp->unp_conn->unp_addr != NULL) 171 sa = (struct sockaddr *) unp->unp_conn->unp_addr; 172 else 173 sa = &sun_noname; 174 bcopy(sa, *nam, sa->sa_len); 175 UNP_UNLOCK(); 176 return (0); 177} 178 179static int 180uipc_attach(struct socket *so, int proto, struct thread *td) 181{ 182 struct unpcb *unp = sotounpcb(so); 183 184 if (unp != NULL) 185 return (EISCONN); 186 return (unp_attach(so)); 187} 188 189static int 190uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 191{ 192 struct unpcb *unp; 193 int error; 194 195 UNP_LOCK(); 196 unp = sotounpcb(so); 197 if (unp == NULL) { 198 UNP_UNLOCK(); 199 return (EINVAL); 200 } 201 error = unp_bind(unp, nam, td); 202 UNP_UNLOCK(); 203 return (error); 204} 205 206static int 207uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 208{ 209 struct unpcb *unp; 210 int error; 211 212 KASSERT(td == curthread, ("uipc_connect: td != curthread")); 213 214 UNP_LOCK(); 215 unp = sotounpcb(so); 216 if (unp == NULL) { 217 UNP_UNLOCK(); 218 return (EINVAL); 219 } 220 error = unp_connect(so, nam, td); 221 UNP_UNLOCK(); 222 return (error); 223} 224 225int 226uipc_connect2(struct socket *so1, struct socket *so2) 227{ 228 struct unpcb *unp; 229 int error; 230 231 UNP_LOCK(); 232 unp = sotounpcb(so1); 233 if (unp == NULL) { 234 UNP_UNLOCK(); 235 return (EINVAL); 236 } 237 error = unp_connect2(so1, so2, PRU_CONNECT2); 238 UNP_UNLOCK(); 239 return (error); 240} 241 242/* control is EOPNOTSUPP */ 243 244static int 245uipc_detach(struct socket *so) 246{ 247 struct unpcb *unp; 248 249 UNP_LOCK(); 250 unp = sotounpcb(so); 251 if (unp == NULL) { 252 UNP_UNLOCK(); 253 return (EINVAL); 254 } 255 unp_detach(unp); 256 UNP_UNLOCK_ASSERT(); 257 return (0); 258} 259 260static int 261uipc_disconnect(struct socket *so) 262{ 263 struct unpcb *unp; 264 265 UNP_LOCK(); 266 unp = sotounpcb(so); 267 if (unp == NULL) { 268 UNP_UNLOCK(); 269 return (EINVAL); 270 } 271 unp_disconnect(unp); 272 UNP_UNLOCK(); 273 return (0); 274} 275 276static int 277uipc_listen(struct socket *so, struct thread *td) 278{ 279 struct unpcb *unp; 280 int error; 281 282 UNP_LOCK(); 283 unp = sotounpcb(so); 284 if (unp == NULL || unp->unp_vnode == NULL) { 285 UNP_UNLOCK(); 286 return (EINVAL); 287 } 288 error = unp_listen(so, unp, td); 289 UNP_UNLOCK(); 290 return (error); 291} 292 293static int 294uipc_peeraddr(struct socket *so, struct sockaddr **nam) 295{ 296 struct unpcb *unp; 297 const struct sockaddr *sa; 298 299 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 300 UNP_LOCK(); 301 unp = sotounpcb(so); 302 if (unp == NULL) { 303 UNP_UNLOCK(); 304 free(*nam, M_SONAME); 305 *nam = NULL; 306 return (EINVAL); 307 } 308 if (unp->unp_conn != NULL && unp->unp_conn->unp_addr!= NULL) 309 sa = (struct sockaddr *) unp->unp_conn->unp_addr; 310 else { 311 /* 312 * XXX: It seems that this test always fails even when 313 * connection is established. So, this else clause is 314 * added as workaround to return PF_LOCAL sockaddr. 315 */ 316 sa = &sun_noname; 317 } 318 bcopy(sa, *nam, sa->sa_len); 319 UNP_UNLOCK(); 320 return (0); 321} 322 323static int 324uipc_rcvd(struct socket *so, int flags) 325{ 326 struct unpcb *unp; 327 struct socket *so2; 328 u_long newhiwat; 329 330 UNP_LOCK(); 331 unp = sotounpcb(so); 332 if (unp == NULL) { 333 UNP_UNLOCK(); 334 return (EINVAL); 335 } 336 switch (so->so_type) { 337 case SOCK_DGRAM: 338 panic("uipc_rcvd DGRAM?"); 339 /*NOTREACHED*/ 340 341 case SOCK_STREAM: 342 if (unp->unp_conn == NULL) 343 break; 344 so2 = unp->unp_conn->unp_socket; 345 SOCKBUF_LOCK(&so2->so_snd); 346 SOCKBUF_LOCK(&so->so_rcv); 347 /* 348 * Adjust backpressure on sender 349 * and wakeup any waiting to write. 350 */ 351 so2->so_snd.sb_mbmax += unp->unp_mbcnt - so->so_rcv.sb_mbcnt; 352 unp->unp_mbcnt = so->so_rcv.sb_mbcnt; 353 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - 354 so->so_rcv.sb_cc; 355 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat, 356 newhiwat, RLIM_INFINITY); 357 unp->unp_cc = so->so_rcv.sb_cc; 358 SOCKBUF_UNLOCK(&so->so_rcv); 359 sowwakeup_locked(so2); 360 break; 361 362 default: 363 panic("uipc_rcvd unknown socktype"); 364 } 365 UNP_UNLOCK(); 366 return (0); 367} 368 369/* pru_rcvoob is EOPNOTSUPP */ 370 371static int 372uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 373 struct mbuf *control, struct thread *td) 374{ 375 int error = 0; 376 struct unpcb *unp; 377 struct socket *so2; 378 u_long newhiwat; 379 380 unp = sotounpcb(so); 381 if (unp == NULL) { 382 error = EINVAL; 383 goto release; 384 } 385 if (flags & PRUS_OOB) { 386 error = EOPNOTSUPP; 387 goto release; 388 } 389 390 if (control != NULL && (error = unp_internalize(&control, td))) 391 goto release; 392 393 UNP_LOCK(); 394 unp = sotounpcb(so); 395 if (unp == NULL) { 396 UNP_UNLOCK(); 397 error = EINVAL; 398 goto dispose_release; 399 } 400 401 switch (so->so_type) { 402 case SOCK_DGRAM: 403 { 404 const struct sockaddr *from; 405 406 if (nam != NULL) { 407 if (unp->unp_conn != NULL) { 408 error = EISCONN; 409 break; 410 } 411 error = unp_connect(so, nam, td); 412 if (error) 413 break; 414 } else { 415 if (unp->unp_conn == NULL) { 416 error = ENOTCONN; 417 break; 418 } 419 } 420 so2 = unp->unp_conn->unp_socket; 421 if (unp->unp_addr != NULL) 422 from = (struct sockaddr *)unp->unp_addr; 423 else 424 from = &sun_noname; 425 if (unp->unp_conn->unp_flags & UNP_WANTCRED) 426 control = unp_addsockcred(td, control); 427 SOCKBUF_LOCK(&so2->so_rcv); 428 if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) { 429 sorwakeup_locked(so2); 430 m = NULL; 431 control = NULL; 432 } else { 433 SOCKBUF_UNLOCK(&so2->so_rcv); 434 error = ENOBUFS; 435 } 436 if (nam != NULL) 437 unp_disconnect(unp); 438 break; 439 } 440 441 case SOCK_STREAM: 442 /* Connect if not connected yet. */ 443 /* 444 * Note: A better implementation would complain 445 * if not equal to the peer's address. 446 */ 447 if ((so->so_state & SS_ISCONNECTED) == 0) { 448 if (nam != NULL) { 449 error = unp_connect(so, nam, td); 450 if (error) 451 break; /* XXX */ 452 } else { 453 error = ENOTCONN; 454 break; 455 } 456 } 457 458 SOCKBUF_LOCK(&so->so_snd); 459 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 460 SOCKBUF_UNLOCK(&so->so_snd); 461 error = EPIPE; 462 break; 463 } 464 if (unp->unp_conn == NULL) 465 panic("uipc_send connected but no connection?"); 466 so2 = unp->unp_conn->unp_socket; 467 SOCKBUF_LOCK(&so2->so_rcv); 468 if (unp->unp_conn->unp_flags & UNP_WANTCRED) { 469 /* 470 * Credentials are passed only once on 471 * SOCK_STREAM. 472 */ 473 unp->unp_conn->unp_flags &= ~UNP_WANTCRED; 474 control = unp_addsockcred(td, control); 475 } 476 /* 477 * Send to paired receive port, and then reduce 478 * send buffer hiwater marks to maintain backpressure. 479 * Wake up readers. 480 */ 481 if (control != NULL) { 482 if (sbappendcontrol_locked(&so2->so_rcv, m, control)) 483 control = NULL; 484 } else { 485 sbappend_locked(&so2->so_rcv, m); 486 } 487 so->so_snd.sb_mbmax -= 488 so2->so_rcv.sb_mbcnt - unp->unp_conn->unp_mbcnt; 489 unp->unp_conn->unp_mbcnt = so2->so_rcv.sb_mbcnt; 490 newhiwat = so->so_snd.sb_hiwat - 491 (so2->so_rcv.sb_cc - unp->unp_conn->unp_cc); 492 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat, 493 newhiwat, RLIM_INFINITY); 494 SOCKBUF_UNLOCK(&so->so_snd); 495 unp->unp_conn->unp_cc = so2->so_rcv.sb_cc; 496 sorwakeup_locked(so2); 497 m = NULL; 498 break; 499 500 default: 501 panic("uipc_send unknown socktype"); 502 } 503 504 /* 505 * SEND_EOF is equivalent to a SEND followed by 506 * a SHUTDOWN. 507 */ 508 if (flags & PRUS_EOF) { 509 socantsendmore(so); 510 unp_shutdown(unp); 511 } 512 UNP_UNLOCK(); 513 514dispose_release: 515 if (control != NULL && error != 0) 516 unp_dispose(control); 517 518release: 519 if (control != NULL) 520 m_freem(control); 521 if (m != NULL) 522 m_freem(m); 523 return (error); 524} 525 526static int 527uipc_sense(struct socket *so, struct stat *sb) 528{ 529 struct unpcb *unp; 530 struct socket *so2; 531 532 UNP_LOCK(); 533 unp = sotounpcb(so); 534 if (unp == NULL) { 535 UNP_UNLOCK(); 536 return (EINVAL); 537 } 538 sb->st_blksize = so->so_snd.sb_hiwat; 539 if (so->so_type == SOCK_STREAM && unp->unp_conn != NULL) { 540 so2 = unp->unp_conn->unp_socket; 541 sb->st_blksize += so2->so_rcv.sb_cc; 542 } 543 sb->st_dev = NODEV; 544 if (unp->unp_ino == 0) 545 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino; 546 sb->st_ino = unp->unp_ino; 547 UNP_UNLOCK(); 548 return (0); 549} 550 551static int 552uipc_shutdown(struct socket *so) 553{ 554 struct unpcb *unp; 555 556 UNP_LOCK(); 557 unp = sotounpcb(so); 558 if (unp == NULL) { 559 UNP_UNLOCK(); 560 return (EINVAL); 561 } 562 socantsendmore(so); 563 unp_shutdown(unp); 564 UNP_UNLOCK(); 565 return (0); 566} 567 568static int 569uipc_sockaddr(struct socket *so, struct sockaddr **nam) 570{ 571 struct unpcb *unp; 572 const struct sockaddr *sa; 573 574 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 575 UNP_LOCK(); 576 unp = sotounpcb(so); 577 if (unp == NULL) { 578 UNP_UNLOCK(); 579 free(*nam, M_SONAME); 580 *nam = NULL; 581 return (EINVAL); 582 } 583 if (unp->unp_addr != NULL) 584 sa = (struct sockaddr *) unp->unp_addr; 585 else 586 sa = &sun_noname; 587 bcopy(sa, *nam, sa->sa_len); 588 UNP_UNLOCK(); 589 return (0); 590} 591 592struct pr_usrreqs uipc_usrreqs = { 593 .pru_abort = uipc_abort, 594 .pru_accept = uipc_accept, 595 .pru_attach = uipc_attach, 596 .pru_bind = uipc_bind, 597 .pru_connect = uipc_connect, 598 .pru_connect2 = uipc_connect2, 599 .pru_detach = uipc_detach, 600 .pru_disconnect = uipc_disconnect, 601 .pru_listen = uipc_listen, 602 .pru_peeraddr = uipc_peeraddr, 603 .pru_rcvd = uipc_rcvd, 604 .pru_send = uipc_send, 605 .pru_sense = uipc_sense, 606 .pru_shutdown = uipc_shutdown, 607 .pru_sockaddr = uipc_sockaddr, 608 .pru_sosend = sosend, 609 .pru_soreceive = soreceive, 610 .pru_sopoll = sopoll, 611}; 612 613int 614uipc_ctloutput(struct socket *so, struct sockopt *sopt) 615{ 616 struct unpcb *unp; 617 struct xucred xu; 618 int error, optval; 619 620 if (sopt->sopt_level != 0) 621 return (EINVAL); 622 623 UNP_LOCK(); 624 unp = sotounpcb(so); 625 if (unp == NULL) { 626 UNP_UNLOCK(); 627 return (EINVAL); 628 } 629 error = 0; 630 631 switch (sopt->sopt_dir) { 632 case SOPT_GET: 633 switch (sopt->sopt_name) { 634 case LOCAL_PEERCRED: 635 if (unp->unp_flags & UNP_HAVEPC) 636 xu = unp->unp_peercred; 637 else { 638 if (so->so_type == SOCK_STREAM) 639 error = ENOTCONN; 640 else 641 error = EINVAL; 642 } 643 if (error == 0) 644 error = sooptcopyout(sopt, &xu, sizeof(xu)); 645 break; 646 case LOCAL_CREDS: 647 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0; 648 error = sooptcopyout(sopt, &optval, sizeof(optval)); 649 break; 650 case LOCAL_CONNWAIT: 651 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0; 652 error = sooptcopyout(sopt, &optval, sizeof(optval)); 653 break; 654 default: 655 error = EOPNOTSUPP; 656 break; 657 } 658 break; 659 case SOPT_SET: 660 switch (sopt->sopt_name) { 661 case LOCAL_CREDS: 662 case LOCAL_CONNWAIT: 663 error = sooptcopyin(sopt, &optval, sizeof(optval), 664 sizeof(optval)); 665 if (error) 666 break; 667 668#define OPTSET(bit) \ 669 if (optval) \ 670 unp->unp_flags |= bit; \ 671 else \ 672 unp->unp_flags &= ~bit; 673 674 switch (sopt->sopt_name) { 675 case LOCAL_CREDS: 676 OPTSET(UNP_WANTCRED); 677 break; 678 case LOCAL_CONNWAIT: 679 OPTSET(UNP_CONNWAIT); 680 break; 681 default: 682 break; 683 } 684 break; 685#undef OPTSET 686 default: 687 error = ENOPROTOOPT; 688 break; 689 } 690 default: 691 error = EOPNOTSUPP; 692 break; 693 } 694 UNP_UNLOCK(); 695 return (error); 696} 697 698/* 699 * Both send and receive buffers are allocated PIPSIZ bytes of buffering 700 * for stream sockets, although the total for sender and receiver is 701 * actually only PIPSIZ. 702 * Datagram sockets really use the sendspace as the maximum datagram size, 703 * and don't really want to reserve the sendspace. Their recvspace should 704 * be large enough for at least one max-size datagram plus address. 705 */ 706#ifndef PIPSIZ 707#define PIPSIZ 8192 708#endif 709static u_long unpst_sendspace = PIPSIZ; 710static u_long unpst_recvspace = PIPSIZ; 711static u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 712static u_long unpdg_recvspace = 4*1024; 713 714static int unp_rights; /* file descriptors in flight */ 715 716SYSCTL_DECL(_net_local_stream); 717SYSCTL_INT(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 718 &unpst_sendspace, 0, ""); 719SYSCTL_INT(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 720 &unpst_recvspace, 0, ""); 721SYSCTL_DECL(_net_local_dgram); 722SYSCTL_INT(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 723 &unpdg_sendspace, 0, ""); 724SYSCTL_INT(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 725 &unpdg_recvspace, 0, ""); 726SYSCTL_DECL(_net_local); 727SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, ""); 728 729static int 730unp_attach(struct socket *so) 731{ 732 struct unpcb *unp; 733 int error; 734 735 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 736 switch (so->so_type) { 737 738 case SOCK_STREAM: 739 error = soreserve(so, unpst_sendspace, unpst_recvspace); 740 break; 741 742 case SOCK_DGRAM: 743 error = soreserve(so, unpdg_sendspace, unpdg_recvspace); 744 break; 745 746 default: 747 panic("unp_attach"); 748 } 749 if (error) 750 return (error); 751 } 752 unp = uma_zalloc(unp_zone, M_WAITOK | M_ZERO); 753 if (unp == NULL) 754 return (ENOBUFS); 755 LIST_INIT(&unp->unp_refs); 756 unp->unp_socket = so; 757 so->so_pcb = unp; 758 759 UNP_LOCK(); 760 unp->unp_gencnt = ++unp_gencnt; 761 unp_count++; 762 LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead 763 : &unp_shead, unp, unp_link); 764 UNP_UNLOCK(); 765 766 return (0); 767} 768 769static void 770unp_detach(struct unpcb *unp) 771{ 772 struct vnode *vp; 773 774 UNP_LOCK_ASSERT(); 775 776 LIST_REMOVE(unp, unp_link); 777 unp->unp_gencnt = ++unp_gencnt; 778 --unp_count; 779 if ((vp = unp->unp_vnode) != NULL) { 780 /* 781 * XXXRW: should v_socket be frobbed only while holding 782 * Giant? 783 */ 784 unp->unp_vnode->v_socket = NULL; 785 unp->unp_vnode = NULL; 786 } 787 if (unp->unp_conn != NULL) 788 unp_disconnect(unp); 789 while (!LIST_EMPTY(&unp->unp_refs)) { 790 struct unpcb *ref = LIST_FIRST(&unp->unp_refs); 791 unp_drop(ref, ECONNRESET); 792 } 793 soisdisconnected(unp->unp_socket); 794 unp->unp_socket->so_pcb = NULL; 795 if (unp_rights) { 796 /* 797 * Normally the receive buffer is flushed later, 798 * in sofree, but if our receive buffer holds references 799 * to descriptors that are now garbage, we will dispose 800 * of those descriptor references after the garbage collector 801 * gets them (resulting in a "panic: closef: count < 0"). 802 */ 803 sorflush(unp->unp_socket); 804 unp_gc(); /* Will unlock UNP. */ 805 } else 806 UNP_UNLOCK(); 807 UNP_UNLOCK_ASSERT(); 808 if (unp->unp_addr != NULL) 809 FREE(unp->unp_addr, M_SONAME); 810 uma_zfree(unp_zone, unp); 811 if (vp) { 812 mtx_lock(&Giant); 813 vrele(vp); 814 mtx_unlock(&Giant); 815 } 816} 817 818static int 819unp_bind(struct unpcb *unp, struct sockaddr *nam, struct thread *td) 820{ 821 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 822 struct vnode *vp; 823 struct mount *mp; 824 struct vattr vattr; 825 int error, namelen; 826 struct nameidata nd; 827 char *buf; 828 829 UNP_LOCK_ASSERT(); 830 831 /* 832 * XXXRW: This test-and-set of unp_vnode is non-atomic; the 833 * unlocked read here is fine, but the value of unp_vnode needs 834 * to be tested again after we do all the lookups to see if the 835 * pcb is still unbound? 836 */ 837 if (unp->unp_vnode != NULL) 838 return (EINVAL); 839 840 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 841 if (namelen <= 0) 842 return (EINVAL); 843 844 UNP_UNLOCK(); 845 846 buf = malloc(namelen + 1, M_TEMP, M_WAITOK); 847 strlcpy(buf, soun->sun_path, namelen + 1); 848 849 mtx_lock(&Giant); 850restart: 851 mtx_assert(&Giant, MA_OWNED); 852 NDINIT(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME, UIO_SYSSPACE, 853 buf, td); 854/* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 855 error = namei(&nd); 856 if (error) 857 goto done; 858 vp = nd.ni_vp; 859 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { 860 NDFREE(&nd, NDF_ONLY_PNBUF); 861 if (nd.ni_dvp == vp) 862 vrele(nd.ni_dvp); 863 else 864 vput(nd.ni_dvp); 865 if (vp != NULL) { 866 vrele(vp); 867 error = EADDRINUSE; 868 goto done; 869 } 870 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH); 871 if (error) 872 goto done; 873 goto restart; 874 } 875 VATTR_NULL(&vattr); 876 vattr.va_type = VSOCK; 877 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask); 878#ifdef MAC 879 error = mac_check_vnode_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, 880 &vattr); 881#endif 882 if (error == 0) { 883 VOP_LEASE(nd.ni_dvp, td, td->td_ucred, LEASE_WRITE); 884 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 885 } 886 NDFREE(&nd, NDF_ONLY_PNBUF); 887 vput(nd.ni_dvp); 888 if (error) { 889 vn_finished_write(mp); 890 goto done; 891 } 892 vp = nd.ni_vp; 893 ASSERT_VOP_LOCKED(vp, "unp_bind"); 894 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK); 895 UNP_LOCK(); 896 vp->v_socket = unp->unp_socket; 897 unp->unp_vnode = vp; 898 unp->unp_addr = soun; 899 UNP_UNLOCK(); 900 VOP_UNLOCK(vp, 0, td); 901 vn_finished_write(mp); 902done: 903 mtx_unlock(&Giant); 904 free(buf, M_TEMP); 905 UNP_LOCK(); 906 return (error); 907} 908 909static int 910unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 911{ 912 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 913 struct vnode *vp; 914 struct socket *so2, *so3; 915 struct unpcb *unp, *unp2, *unp3; 916 int error, len; 917 struct nameidata nd; 918 char buf[SOCK_MAXADDRLEN]; 919 struct sockaddr *sa; 920 921 UNP_LOCK_ASSERT(); 922 unp = sotounpcb(so); 923 924 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 925 if (len <= 0) 926 return (EINVAL); 927 strlcpy(buf, soun->sun_path, len + 1); 928 UNP_UNLOCK(); 929 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 930 mtx_lock(&Giant); 931 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, td); 932 error = namei(&nd); 933 if (error) 934 vp = NULL; 935 else 936 vp = nd.ni_vp; 937 ASSERT_VOP_LOCKED(vp, "unp_connect"); 938 NDFREE(&nd, NDF_ONLY_PNBUF); 939 if (error) 940 goto bad; 941 942 if (vp->v_type != VSOCK) { 943 error = ENOTSOCK; 944 goto bad; 945 } 946 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); 947 if (error) 948 goto bad; 949 mtx_unlock(&Giant); 950 UNP_LOCK(); 951 unp = sotounpcb(so); 952 if (unp == NULL) { 953 error = EINVAL; 954 goto bad2; 955 } 956 so2 = vp->v_socket; 957 if (so2 == NULL) { 958 error = ECONNREFUSED; 959 goto bad2; 960 } 961 if (so->so_type != so2->so_type) { 962 error = EPROTOTYPE; 963 goto bad2; 964 } 965 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 966 if (so2->so_options & SO_ACCEPTCONN) { 967 /* 968 * NB: drop locks here so unp_attach is entered 969 * w/o locks; this avoids a recursive lock 970 * of the head and holding sleep locks across 971 * a (potentially) blocking malloc. 972 */ 973 UNP_UNLOCK(); 974 so3 = sonewconn(so2, 0); 975 UNP_LOCK(); 976 } else 977 so3 = NULL; 978 if (so3 == NULL) { 979 error = ECONNREFUSED; 980 goto bad2; 981 } 982 unp = sotounpcb(so); 983 unp2 = sotounpcb(so2); 984 unp3 = sotounpcb(so3); 985 if (unp2->unp_addr != NULL) { 986 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); 987 unp3->unp_addr = (struct sockaddr_un *) sa; 988 sa = NULL; 989 } 990 /* 991 * unp_peercred management: 992 * 993 * The connecter's (client's) credentials are copied 994 * from its process structure at the time of connect() 995 * (which is now). 996 */ 997 cru2x(td->td_ucred, &unp3->unp_peercred); 998 unp3->unp_flags |= UNP_HAVEPC; 999 /* 1000 * The receiver's (server's) credentials are copied 1001 * from the unp_peercred member of socket on which the 1002 * former called listen(); unp_listen() cached that 1003 * process's credentials at that time so we can use 1004 * them now. 1005 */ 1006 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED, 1007 ("unp_connect: listener without cached peercred")); 1008 memcpy(&unp->unp_peercred, &unp2->unp_peercred, 1009 sizeof(unp->unp_peercred)); 1010 unp->unp_flags |= UNP_HAVEPC; 1011#ifdef MAC 1012 SOCK_LOCK(so); 1013 mac_set_socket_peer_from_socket(so, so3); 1014 mac_set_socket_peer_from_socket(so3, so); 1015 SOCK_UNLOCK(so); 1016#endif 1017 1018 so2 = so3; 1019 } 1020 error = unp_connect2(so, so2, PRU_CONNECT); 1021bad2: 1022 UNP_UNLOCK(); 1023 mtx_lock(&Giant); 1024bad: 1025 mtx_assert(&Giant, MA_OWNED); 1026 if (vp != NULL) 1027 vput(vp); 1028 mtx_unlock(&Giant); 1029 free(sa, M_SONAME); 1030 UNP_LOCK(); 1031 return (error); 1032} 1033 1034static int 1035unp_connect2(struct socket *so, struct socket *so2, int req) 1036{ 1037 struct unpcb *unp = sotounpcb(so); 1038 struct unpcb *unp2; 1039 1040 UNP_LOCK_ASSERT(); 1041 1042 if (so2->so_type != so->so_type) 1043 return (EPROTOTYPE); 1044 unp2 = sotounpcb(so2); 1045 unp->unp_conn = unp2; 1046 switch (so->so_type) { 1047 1048 case SOCK_DGRAM: 1049 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 1050 soisconnected(so); 1051 break; 1052 1053 case SOCK_STREAM: 1054 unp2->unp_conn = unp; 1055 if (req == PRU_CONNECT && 1056 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 1057 soisconnecting(so); 1058 else 1059 soisconnected(so); 1060 soisconnected(so2); 1061 break; 1062 1063 default: 1064 panic("unp_connect2"); 1065 } 1066 return (0); 1067} 1068 1069static void 1070unp_disconnect(struct unpcb *unp) 1071{ 1072 struct unpcb *unp2 = unp->unp_conn; 1073 struct socket *so; 1074 1075 UNP_LOCK_ASSERT(); 1076 1077 if (unp2 == NULL) 1078 return; 1079 unp->unp_conn = NULL; 1080 switch (unp->unp_socket->so_type) { 1081 1082 case SOCK_DGRAM: 1083 LIST_REMOVE(unp, unp_reflink); 1084 so = unp->unp_socket; 1085 SOCK_LOCK(so); 1086 so->so_state &= ~SS_ISCONNECTED; 1087 SOCK_UNLOCK(so); 1088 break; 1089 1090 case SOCK_STREAM: 1091 soisdisconnected(unp->unp_socket); 1092 unp2->unp_conn = NULL; 1093 soisdisconnected(unp2->unp_socket); 1094 break; 1095 } 1096} 1097 1098#ifdef notdef 1099void 1100unp_abort(struct unpcb *unp) 1101{ 1102 1103 unp_detach(unp); 1104 UNP_UNLOCK_ASSERT(); 1105} 1106#endif 1107 1108/* 1109 * unp_pcblist() assumes that UNIX domain socket memory is never reclaimed 1110 * by the zone (UMA_ZONE_NOFREE), and as such potentially stale pointers 1111 * are safe to reference. It first scans the list of struct unpcb's to 1112 * generate a pointer list, then it rescans its list one entry at a time to 1113 * externalize and copyout. It checks the generation number to see if a 1114 * struct unpcb has been reused, and will skip it if so. 1115 */ 1116static int 1117unp_pcblist(SYSCTL_HANDLER_ARGS) 1118{ 1119 int error, i, n; 1120 struct unpcb *unp, **unp_list; 1121 unp_gen_t gencnt; 1122 struct xunpgen *xug; 1123 struct unp_head *head; 1124 struct xunpcb *xu; 1125 1126 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead); 1127 1128 /* 1129 * The process of preparing the PCB list is too time-consuming and 1130 * resource-intensive to repeat twice on every request. 1131 */ 1132 if (req->oldptr == NULL) { 1133 n = unp_count; 1134 req->oldidx = 2 * (sizeof *xug) 1135 + (n + n/8) * sizeof(struct xunpcb); 1136 return (0); 1137 } 1138 1139 if (req->newptr != NULL) 1140 return (EPERM); 1141 1142 /* 1143 * OK, now we're committed to doing something. 1144 */ 1145 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK); 1146 UNP_LOCK(); 1147 gencnt = unp_gencnt; 1148 n = unp_count; 1149 UNP_UNLOCK(); 1150 1151 xug->xug_len = sizeof *xug; 1152 xug->xug_count = n; 1153 xug->xug_gen = gencnt; 1154 xug->xug_sogen = so_gencnt; 1155 error = SYSCTL_OUT(req, xug, sizeof *xug); 1156 if (error) { 1157 free(xug, M_TEMP); 1158 return (error); 1159 } 1160 1161 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 1162 1163 UNP_LOCK(); 1164 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 1165 unp = LIST_NEXT(unp, unp_link)) { 1166 if (unp->unp_gencnt <= gencnt) { 1167 if (cr_cansee(req->td->td_ucred, 1168 unp->unp_socket->so_cred)) 1169 continue; 1170 unp_list[i++] = unp; 1171 } 1172 } 1173 UNP_UNLOCK(); 1174 n = i; /* in case we lost some during malloc */ 1175 1176 error = 0; 1177 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK); 1178 for (i = 0; i < n; i++) { 1179 unp = unp_list[i]; 1180 if (unp->unp_gencnt <= gencnt) { 1181 xu->xu_len = sizeof *xu; 1182 xu->xu_unpp = unp; 1183 /* 1184 * XXX - need more locking here to protect against 1185 * connect/disconnect races for SMP. 1186 */ 1187 if (unp->unp_addr != NULL) 1188 bcopy(unp->unp_addr, &xu->xu_addr, 1189 unp->unp_addr->sun_len); 1190 if (unp->unp_conn != NULL && 1191 unp->unp_conn->unp_addr != NULL) 1192 bcopy(unp->unp_conn->unp_addr, 1193 &xu->xu_caddr, 1194 unp->unp_conn->unp_addr->sun_len); 1195 bcopy(unp, &xu->xu_unp, sizeof *unp); 1196 sotoxsocket(unp->unp_socket, &xu->xu_socket); 1197 error = SYSCTL_OUT(req, xu, sizeof *xu); 1198 } 1199 } 1200 free(xu, M_TEMP); 1201 if (!error) { 1202 /* 1203 * Give the user an updated idea of our state. 1204 * If the generation differs from what we told 1205 * her before, she knows that something happened 1206 * while we were processing this request, and it 1207 * might be necessary to retry. 1208 */ 1209 xug->xug_gen = unp_gencnt; 1210 xug->xug_sogen = so_gencnt; 1211 xug->xug_count = unp_count; 1212 error = SYSCTL_OUT(req, xug, sizeof *xug); 1213 } 1214 free(unp_list, M_TEMP); 1215 free(xug, M_TEMP); 1216 return (error); 1217} 1218 1219SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD, 1220 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 1221 "List of active local datagram sockets"); 1222SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD, 1223 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 1224 "List of active local stream sockets"); 1225 1226static void 1227unp_shutdown(struct unpcb *unp) 1228{ 1229 struct socket *so; 1230 1231 UNP_LOCK_ASSERT(); 1232 1233 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn && 1234 (so = unp->unp_conn->unp_socket)) 1235 socantrcvmore(so); 1236} 1237 1238static void 1239unp_drop(struct unpcb *unp, int errno) 1240{ 1241 struct socket *so = unp->unp_socket; 1242 1243 UNP_LOCK_ASSERT(); 1244 1245 so->so_error = errno; 1246 unp_disconnect(unp); 1247} 1248 1249#ifdef notdef 1250void 1251unp_drain(void) 1252{ 1253 1254} 1255#endif 1256 1257static void 1258unp_freerights(struct file **rp, int fdcount) 1259{ 1260 int i; 1261 struct file *fp; 1262 1263 for (i = 0; i < fdcount; i++) { 1264 fp = *rp; 1265 /* 1266 * zero the pointer before calling 1267 * unp_discard since it may end up 1268 * in unp_gc().. 1269 */ 1270 *rp++ = 0; 1271 unp_discard(fp); 1272 } 1273} 1274 1275int 1276unp_externalize(struct mbuf *control, struct mbuf **controlp) 1277{ 1278 struct thread *td = curthread; /* XXX */ 1279 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1280 int i; 1281 int *fdp; 1282 struct file **rp; 1283 struct file *fp; 1284 void *data; 1285 socklen_t clen = control->m_len, datalen; 1286 int error, newfds; 1287 int f; 1288 u_int newlen; 1289 1290 UNP_UNLOCK_ASSERT(); 1291 1292 error = 0; 1293 if (controlp != NULL) /* controlp == NULL => free control messages */ 1294 *controlp = NULL; 1295 1296 while (cm != NULL) { 1297 if (sizeof(*cm) > clen || cm->cmsg_len > clen) { 1298 error = EINVAL; 1299 break; 1300 } 1301 1302 data = CMSG_DATA(cm); 1303 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1304 1305 if (cm->cmsg_level == SOL_SOCKET 1306 && cm->cmsg_type == SCM_RIGHTS) { 1307 newfds = datalen / sizeof(struct file *); 1308 rp = data; 1309 1310 /* If we're not outputting the descriptors free them. */ 1311 if (error || controlp == NULL) { 1312 unp_freerights(rp, newfds); 1313 goto next; 1314 } 1315 FILEDESC_LOCK(td->td_proc->p_fd); 1316 /* if the new FD's will not fit free them. */ 1317 if (!fdavail(td, newfds)) { 1318 FILEDESC_UNLOCK(td->td_proc->p_fd); 1319 error = EMSGSIZE; 1320 unp_freerights(rp, newfds); 1321 goto next; 1322 } 1323 /* 1324 * now change each pointer to an fd in the global 1325 * table to an integer that is the index to the 1326 * local fd table entry that we set up to point 1327 * to the global one we are transferring. 1328 */ 1329 newlen = newfds * sizeof(int); 1330 *controlp = sbcreatecontrol(NULL, newlen, 1331 SCM_RIGHTS, SOL_SOCKET); 1332 if (*controlp == NULL) { 1333 FILEDESC_UNLOCK(td->td_proc->p_fd); 1334 error = E2BIG; 1335 unp_freerights(rp, newfds); 1336 goto next; 1337 } 1338 1339 fdp = (int *) 1340 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1341 for (i = 0; i < newfds; i++) { 1342 if (fdalloc(td, 0, &f)) 1343 panic("unp_externalize fdalloc failed"); 1344 fp = *rp++; 1345 td->td_proc->p_fd->fd_ofiles[f] = fp; 1346 FILE_LOCK(fp); 1347 fp->f_msgcount--; 1348 FILE_UNLOCK(fp); 1349 unp_rights--; 1350 *fdp++ = f; 1351 } 1352 FILEDESC_UNLOCK(td->td_proc->p_fd); 1353 } else { /* We can just copy anything else across */ 1354 if (error || controlp == NULL) 1355 goto next; 1356 *controlp = sbcreatecontrol(NULL, datalen, 1357 cm->cmsg_type, cm->cmsg_level); 1358 if (*controlp == NULL) { 1359 error = ENOBUFS; 1360 goto next; 1361 } 1362 bcopy(data, 1363 CMSG_DATA(mtod(*controlp, struct cmsghdr *)), 1364 datalen); 1365 } 1366 1367 controlp = &(*controlp)->m_next; 1368 1369next: 1370 if (CMSG_SPACE(datalen) < clen) { 1371 clen -= CMSG_SPACE(datalen); 1372 cm = (struct cmsghdr *) 1373 ((caddr_t)cm + CMSG_SPACE(datalen)); 1374 } else { 1375 clen = 0; 1376 cm = NULL; 1377 } 1378 } 1379 1380 m_freem(control); 1381 1382 return (error); 1383} 1384 1385void 1386unp_init(void) 1387{ 1388 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL, 1389 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 1390 if (unp_zone == NULL) 1391 panic("unp_init"); 1392 uma_zone_set_max(unp_zone, nmbclusters); 1393 LIST_INIT(&unp_dhead); 1394 LIST_INIT(&unp_shead); 1395 1396 UNP_LOCK_INIT(); 1397} 1398 1399static int 1400unp_internalize(struct mbuf **controlp, struct thread *td) 1401{ 1402 struct mbuf *control = *controlp; 1403 struct proc *p = td->td_proc; 1404 struct filedesc *fdescp = p->p_fd; 1405 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1406 struct cmsgcred *cmcred; 1407 struct file **rp; 1408 struct file *fp; 1409 struct timeval *tv; 1410 int i, fd, *fdp; 1411 void *data; 1412 socklen_t clen = control->m_len, datalen; 1413 int error, oldfds; 1414 u_int newlen; 1415 1416 UNP_UNLOCK_ASSERT(); 1417 1418 error = 0; 1419 *controlp = NULL; 1420 1421 while (cm != NULL) { 1422 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET 1423 || cm->cmsg_len > clen) { 1424 error = EINVAL; 1425 goto out; 1426 } 1427 1428 data = CMSG_DATA(cm); 1429 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1430 1431 switch (cm->cmsg_type) { 1432 /* 1433 * Fill in credential information. 1434 */ 1435 case SCM_CREDS: 1436 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred), 1437 SCM_CREDS, SOL_SOCKET); 1438 if (*controlp == NULL) { 1439 error = ENOBUFS; 1440 goto out; 1441 } 1442 1443 cmcred = (struct cmsgcred *) 1444 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1445 cmcred->cmcred_pid = p->p_pid; 1446 cmcred->cmcred_uid = td->td_ucred->cr_ruid; 1447 cmcred->cmcred_gid = td->td_ucred->cr_rgid; 1448 cmcred->cmcred_euid = td->td_ucred->cr_uid; 1449 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups, 1450 CMGROUP_MAX); 1451 for (i = 0; i < cmcred->cmcred_ngroups; i++) 1452 cmcred->cmcred_groups[i] = 1453 td->td_ucred->cr_groups[i]; 1454 break; 1455 1456 case SCM_RIGHTS: 1457 oldfds = datalen / sizeof (int); 1458 /* 1459 * check that all the FDs passed in refer to legal files 1460 * If not, reject the entire operation. 1461 */ 1462 fdp = data; 1463 FILEDESC_LOCK(fdescp); 1464 for (i = 0; i < oldfds; i++) { 1465 fd = *fdp++; 1466 if ((unsigned)fd >= fdescp->fd_nfiles || 1467 fdescp->fd_ofiles[fd] == NULL) { 1468 FILEDESC_UNLOCK(fdescp); 1469 error = EBADF; 1470 goto out; 1471 } 1472 fp = fdescp->fd_ofiles[fd]; 1473 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { 1474 FILEDESC_UNLOCK(fdescp); 1475 error = EOPNOTSUPP; 1476 goto out; 1477 } 1478 1479 } 1480 /* 1481 * Now replace the integer FDs with pointers to 1482 * the associated global file table entry.. 1483 */ 1484 newlen = oldfds * sizeof(struct file *); 1485 *controlp = sbcreatecontrol(NULL, newlen, 1486 SCM_RIGHTS, SOL_SOCKET); 1487 if (*controlp == NULL) { 1488 FILEDESC_UNLOCK(fdescp); 1489 error = E2BIG; 1490 goto out; 1491 } 1492 1493 fdp = data; 1494 rp = (struct file **) 1495 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1496 for (i = 0; i < oldfds; i++) { 1497 fp = fdescp->fd_ofiles[*fdp++]; 1498 *rp++ = fp; 1499 FILE_LOCK(fp); 1500 fp->f_count++; 1501 fp->f_msgcount++; 1502 FILE_UNLOCK(fp); 1503 unp_rights++; 1504 } 1505 FILEDESC_UNLOCK(fdescp); 1506 break; 1507 1508 case SCM_TIMESTAMP: 1509 *controlp = sbcreatecontrol(NULL, sizeof(*tv), 1510 SCM_TIMESTAMP, SOL_SOCKET); 1511 if (*controlp == NULL) { 1512 error = ENOBUFS; 1513 goto out; 1514 } 1515 tv = (struct timeval *) 1516 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1517 microtime(tv); 1518 break; 1519 1520 default: 1521 error = EINVAL; 1522 goto out; 1523 } 1524 1525 controlp = &(*controlp)->m_next; 1526 1527 if (CMSG_SPACE(datalen) < clen) { 1528 clen -= CMSG_SPACE(datalen); 1529 cm = (struct cmsghdr *) 1530 ((caddr_t)cm + CMSG_SPACE(datalen)); 1531 } else { 1532 clen = 0; 1533 cm = NULL; 1534 } 1535 } 1536 1537out: 1538 m_freem(control); 1539 1540 return (error); 1541} 1542 1543struct mbuf * 1544unp_addsockcred(struct thread *td, struct mbuf *control) 1545{ 1546 struct mbuf *m, *n; 1547 struct sockcred *sc; 1548 int ngroups; 1549 int i; 1550 1551 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); 1552 1553 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET); 1554 if (m == NULL) 1555 return (control); 1556 m->m_next = NULL; 1557 1558 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *)); 1559 sc->sc_uid = td->td_ucred->cr_ruid; 1560 sc->sc_euid = td->td_ucred->cr_uid; 1561 sc->sc_gid = td->td_ucred->cr_rgid; 1562 sc->sc_egid = td->td_ucred->cr_gid; 1563 sc->sc_ngroups = ngroups; 1564 for (i = 0; i < sc->sc_ngroups; i++) 1565 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 1566 1567 /* 1568 * If a control message already exists, append us to the end. 1569 */ 1570 if (control != NULL) { 1571 for (n = control; n->m_next != NULL; n = n->m_next) 1572 ; 1573 n->m_next = m; 1574 } else 1575 control = m; 1576 1577 return (control); 1578} 1579 1580/* 1581 * unp_defer is thread-local during garbage collection, and does not require 1582 * explicit synchronization. unp_gcing prevents other threads from entering 1583 * garbage collection, and perhaps should be an sx lock instead. 1584 */ 1585static int unp_defer, unp_gcing; 1586 1587static void 1588unp_gc(void) 1589{ 1590 struct file *fp, *nextfp; 1591 struct socket *so; 1592 struct file **extra_ref, **fpp; 1593 int nunref, i; 1594 int nfiles_snap; 1595 int nfiles_slack = 20; 1596 1597 UNP_LOCK_ASSERT(); 1598 1599 if (unp_gcing) { 1600 UNP_UNLOCK(); 1601 return; 1602 } 1603 unp_gcing = 1; 1604 unp_defer = 0; 1605 UNP_UNLOCK(); 1606 /* 1607 * before going through all this, set all FDs to 1608 * be NOT defered and NOT externally accessible 1609 */ 1610 sx_slock(&filelist_lock); 1611 LIST_FOREACH(fp, &filehead, f_list) 1612 fp->f_gcflag &= ~(FMARK|FDEFER); 1613 do { 1614 LIST_FOREACH(fp, &filehead, f_list) { 1615 FILE_LOCK(fp); 1616 /* 1617 * If the file is not open, skip it 1618 */ 1619 if (fp->f_count == 0) { 1620 FILE_UNLOCK(fp); 1621 continue; 1622 } 1623 /* 1624 * If we already marked it as 'defer' in a 1625 * previous pass, then try process it this time 1626 * and un-mark it 1627 */ 1628 if (fp->f_gcflag & FDEFER) { 1629 fp->f_gcflag &= ~FDEFER; 1630 unp_defer--; 1631 } else { 1632 /* 1633 * if it's not defered, then check if it's 1634 * already marked.. if so skip it 1635 */ 1636 if (fp->f_gcflag & FMARK) { 1637 FILE_UNLOCK(fp); 1638 continue; 1639 } 1640 /* 1641 * If all references are from messages 1642 * in transit, then skip it. it's not 1643 * externally accessible. 1644 */ 1645 if (fp->f_count == fp->f_msgcount) { 1646 FILE_UNLOCK(fp); 1647 continue; 1648 } 1649 /* 1650 * If it got this far then it must be 1651 * externally accessible. 1652 */ 1653 fp->f_gcflag |= FMARK; 1654 } 1655 /* 1656 * either it was defered, or it is externally 1657 * accessible and not already marked so. 1658 * Now check if it is possibly one of OUR sockets. 1659 */ 1660 if (fp->f_type != DTYPE_SOCKET || 1661 (so = fp->f_data) == NULL) { 1662 FILE_UNLOCK(fp); 1663 continue; 1664 } 1665 FILE_UNLOCK(fp); 1666 if (so->so_proto->pr_domain != &localdomain || 1667 (so->so_proto->pr_flags&PR_RIGHTS) == 0) 1668 continue; 1669#ifdef notdef 1670 if (so->so_rcv.sb_flags & SB_LOCK) { 1671 /* 1672 * This is problematical; it's not clear 1673 * we need to wait for the sockbuf to be 1674 * unlocked (on a uniprocessor, at least), 1675 * and it's also not clear what to do 1676 * if sbwait returns an error due to receipt 1677 * of a signal. If sbwait does return 1678 * an error, we'll go into an infinite 1679 * loop. Delete all of this for now. 1680 */ 1681 (void) sbwait(&so->so_rcv); 1682 goto restart; 1683 } 1684#endif 1685 /* 1686 * So, Ok, it's one of our sockets and it IS externally 1687 * accessible (or was defered). Now we look 1688 * to see if we hold any file descriptors in its 1689 * message buffers. Follow those links and mark them 1690 * as accessible too. 1691 */ 1692 SOCKBUF_LOCK(&so->so_rcv); 1693 unp_scan(so->so_rcv.sb_mb, unp_mark); 1694 SOCKBUF_UNLOCK(&so->so_rcv); 1695 } 1696 } while (unp_defer); 1697 sx_sunlock(&filelist_lock); 1698 /* 1699 * We grab an extra reference to each of the file table entries 1700 * that are not otherwise accessible and then free the rights 1701 * that are stored in messages on them. 1702 * 1703 * The bug in the orginal code is a little tricky, so I'll describe 1704 * what's wrong with it here. 1705 * 1706 * It is incorrect to simply unp_discard each entry for f_msgcount 1707 * times -- consider the case of sockets A and B that contain 1708 * references to each other. On a last close of some other socket, 1709 * we trigger a gc since the number of outstanding rights (unp_rights) 1710 * is non-zero. If during the sweep phase the gc code un_discards, 1711 * we end up doing a (full) closef on the descriptor. A closef on A 1712 * results in the following chain. Closef calls soo_close, which 1713 * calls soclose. Soclose calls first (through the switch 1714 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply 1715 * returns because the previous instance had set unp_gcing, and 1716 * we return all the way back to soclose, which marks the socket 1717 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush 1718 * to free up the rights that are queued in messages on the socket A, 1719 * i.e., the reference on B. The sorflush calls via the dom_dispose 1720 * switch unp_dispose, which unp_scans with unp_discard. This second 1721 * instance of unp_discard just calls closef on B. 1722 * 1723 * Well, a similar chain occurs on B, resulting in a sorflush on B, 1724 * which results in another closef on A. Unfortunately, A is already 1725 * being closed, and the descriptor has already been marked with 1726 * SS_NOFDREF, and soclose panics at this point. 1727 * 1728 * Here, we first take an extra reference to each inaccessible 1729 * descriptor. Then, we call sorflush ourself, since we know 1730 * it is a Unix domain socket anyhow. After we destroy all the 1731 * rights carried in messages, we do a last closef to get rid 1732 * of our extra reference. This is the last close, and the 1733 * unp_detach etc will shut down the socket. 1734 * 1735 * 91/09/19, bsy@cs.cmu.edu 1736 */ 1737again: 1738 nfiles_snap = openfiles + nfiles_slack; /* some slack */ 1739 extra_ref = malloc(nfiles_snap * sizeof(struct file *), M_TEMP, 1740 M_WAITOK); 1741 sx_slock(&filelist_lock); 1742 if (nfiles_snap < openfiles) { 1743 sx_sunlock(&filelist_lock); 1744 free(extra_ref, M_TEMP); 1745 nfiles_slack += 20; 1746 goto again; 1747 } 1748 for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref; 1749 fp != NULL; fp = nextfp) { 1750 nextfp = LIST_NEXT(fp, f_list); 1751 FILE_LOCK(fp); 1752 /* 1753 * If it's not open, skip it 1754 */ 1755 if (fp->f_count == 0) { 1756 FILE_UNLOCK(fp); 1757 continue; 1758 } 1759 /* 1760 * If all refs are from msgs, and it's not marked accessible 1761 * then it must be referenced from some unreachable cycle 1762 * of (shut-down) FDs, so include it in our 1763 * list of FDs to remove 1764 */ 1765 if (fp->f_count == fp->f_msgcount && !(fp->f_gcflag & FMARK)) { 1766 *fpp++ = fp; 1767 nunref++; 1768 fp->f_count++; 1769 } 1770 FILE_UNLOCK(fp); 1771 } 1772 sx_sunlock(&filelist_lock); 1773 /* 1774 * for each FD on our hit list, do the following two things 1775 */ 1776 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1777 struct file *tfp = *fpp; 1778 FILE_LOCK(tfp); 1779 if (tfp->f_type == DTYPE_SOCKET && 1780 tfp->f_data != NULL) { 1781 FILE_UNLOCK(tfp); 1782 sorflush(tfp->f_data); 1783 } else { 1784 FILE_UNLOCK(tfp); 1785 } 1786 } 1787 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) 1788 closef(*fpp, (struct thread *) NULL); 1789 free(extra_ref, M_TEMP); 1790 unp_gcing = 0; 1791 1792 UNP_UNLOCK_ASSERT(); 1793} 1794 1795void 1796unp_dispose(struct mbuf *m) 1797{ 1798 1799 if (m) 1800 unp_scan(m, unp_discard); 1801} 1802 1803static int 1804unp_listen(struct socket *so, struct unpcb *unp, struct thread *td) 1805{ 1806 int error; 1807 1808 UNP_LOCK_ASSERT(); 1809 1810 SOCK_LOCK(so); 1811 error = solisten_proto_check(so); 1812 if (error == 0) { 1813 cru2x(td->td_ucred, &unp->unp_peercred); 1814 unp->unp_flags |= UNP_HAVEPCCACHED; 1815 solisten_proto(so); 1816 } 1817 SOCK_UNLOCK(so); 1818 return (error); 1819} 1820 1821static void 1822unp_scan(struct mbuf *m0, void (*op)(struct file *)) 1823{ 1824 struct mbuf *m; 1825 struct file **rp; 1826 struct cmsghdr *cm; 1827 void *data; 1828 int i; 1829 socklen_t clen, datalen; 1830 int qfds; 1831 1832 while (m0 != NULL) { 1833 for (m = m0; m; m = m->m_next) { 1834 if (m->m_type != MT_CONTROL) 1835 continue; 1836 1837 cm = mtod(m, struct cmsghdr *); 1838 clen = m->m_len; 1839 1840 while (cm != NULL) { 1841 if (sizeof(*cm) > clen || cm->cmsg_len > clen) 1842 break; 1843 1844 data = CMSG_DATA(cm); 1845 datalen = (caddr_t)cm + cm->cmsg_len 1846 - (caddr_t)data; 1847 1848 if (cm->cmsg_level == SOL_SOCKET && 1849 cm->cmsg_type == SCM_RIGHTS) { 1850 qfds = datalen / sizeof (struct file *); 1851 rp = data; 1852 for (i = 0; i < qfds; i++) 1853 (*op)(*rp++); 1854 } 1855 1856 if (CMSG_SPACE(datalen) < clen) { 1857 clen -= CMSG_SPACE(datalen); 1858 cm = (struct cmsghdr *) 1859 ((caddr_t)cm + CMSG_SPACE(datalen)); 1860 } else { 1861 clen = 0; 1862 cm = NULL; 1863 } 1864 } 1865 } 1866 m0 = m0->m_act; 1867 } 1868} 1869 1870static void 1871unp_mark(struct file *fp) 1872{ 1873 if (fp->f_gcflag & FMARK) 1874 return; 1875 unp_defer++; 1876 fp->f_gcflag |= (FMARK|FDEFER); 1877} 1878 1879static void 1880unp_discard(struct file *fp) 1881{ 1882 FILE_LOCK(fp); 1883 fp->f_msgcount--; 1884 unp_rights--; 1885 FILE_UNLOCK(fp); 1886 (void) closef(fp, (struct thread *)NULL); 1887} 1888