uipc_usrreq.c revision 156807
1/*- 2 * Copyright (c) 1982, 1986, 1989, 1991, 1993 3 * The Regents of the University of California. 4 * Copyright 2004-2006 Robert N. M. Watson 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 4. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94 32 */ 33 34#include <sys/cdefs.h> 35__FBSDID("$FreeBSD: head/sys/kern/uipc_usrreq.c 156807 2006-03-17 13:52:57Z rwatson $"); 36 37#include "opt_mac.h" 38 39#include <sys/param.h> 40#include <sys/domain.h> 41#include <sys/fcntl.h> 42#include <sys/malloc.h> /* XXX must be before <sys/file.h> */ 43#include <sys/file.h> 44#include <sys/filedesc.h> 45#include <sys/jail.h> 46#include <sys/kernel.h> 47#include <sys/lock.h> 48#include <sys/mac.h> 49#include <sys/mbuf.h> 50#include <sys/mount.h> 51#include <sys/mutex.h> 52#include <sys/namei.h> 53#include <sys/proc.h> 54#include <sys/protosw.h> 55#include <sys/resourcevar.h> 56#include <sys/socket.h> 57#include <sys/socketvar.h> 58#include <sys/signalvar.h> 59#include <sys/stat.h> 60#include <sys/sx.h> 61#include <sys/sysctl.h> 62#include <sys/systm.h> 63#include <sys/taskqueue.h> 64#include <sys/un.h> 65#include <sys/unpcb.h> 66#include <sys/vnode.h> 67 68#include <vm/uma.h> 69 70static uma_zone_t unp_zone; 71static unp_gen_t unp_gencnt; 72static u_int unp_count; 73 74static struct unp_head unp_shead, unp_dhead; 75 76/* 77 * Unix communications domain. 78 * 79 * TODO: 80 * SEQPACKET, RDM 81 * rethink name space problems 82 * need a proper out-of-band 83 * lock pushdown 84 */ 85static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL }; 86static ino_t unp_ino; /* prototype for fake inode numbers */ 87struct mbuf *unp_addsockcred(struct thread *, struct mbuf *); 88 89/* 90 * Currently, UNIX domain sockets are protected by a single subsystem lock, 91 * which covers global data structures and variables, the contents of each 92 * per-socket unpcb structure, and the so_pcb field in sockets attached to 93 * the UNIX domain. This provides for a moderate degree of paralellism, as 94 * receive operations on UNIX domain sockets do not need to acquire the 95 * subsystem lock. Finer grained locking to permit send() without acquiring 96 * a global lock would be a logical next step. 97 * 98 * The UNIX domain socket lock preceds all socket layer locks, including the 99 * socket lock and socket buffer lock, permitting UNIX domain socket code to 100 * call into socket support routines without releasing its locks. 101 * 102 * Some caution is required in areas where the UNIX domain socket code enters 103 * VFS in order to create or find rendezvous points. This results in 104 * dropping of the UNIX domain socket subsystem lock, acquisition of the 105 * Giant lock, and potential sleeping. This increases the chances of races, 106 * and exposes weaknesses in the socket->protocol API by offering poor 107 * failure modes. 108 */ 109static struct mtx unp_mtx; 110#define UNP_LOCK_INIT() \ 111 mtx_init(&unp_mtx, "unp", NULL, MTX_DEF) 112#define UNP_LOCK() mtx_lock(&unp_mtx) 113#define UNP_UNLOCK() mtx_unlock(&unp_mtx) 114#define UNP_LOCK_ASSERT() mtx_assert(&unp_mtx, MA_OWNED) 115#define UNP_UNLOCK_ASSERT() mtx_assert(&unp_mtx, MA_NOTOWNED) 116 117/* 118 * Garbage collection of cyclic file descriptor/socket references occurs 119 * asynchronously in a taskqueue context in order to avoid recursion and 120 * reentrance in the UNIX domain socket, file descriptor, and socket layer 121 * code. See unp_gc() for a full description. 122 */ 123static struct task unp_gc_task; 124 125static int unp_attach(struct socket *); 126static void unp_detach(struct unpcb *); 127static int unp_bind(struct unpcb *,struct sockaddr *, struct thread *); 128static int unp_connect(struct socket *,struct sockaddr *, struct thread *); 129static int unp_connect2(struct socket *so, struct socket *so2, int); 130static void unp_disconnect(struct unpcb *); 131static void unp_shutdown(struct unpcb *); 132static void unp_drop(struct unpcb *, int); 133static void unp_gc(__unused void *, int); 134static void unp_scan(struct mbuf *, void (*)(struct file *)); 135static void unp_mark(struct file *); 136static void unp_discard(struct file *); 137static void unp_freerights(struct file **, int); 138static int unp_internalize(struct mbuf **, struct thread *); 139static int unp_listen(struct socket *, struct unpcb *, int, 140 struct thread *); 141 142static int 143uipc_abort(struct socket *so) 144{ 145 struct unpcb *unp; 146 147 unp = sotounpcb(so); 148 KASSERT(unp != NULL, ("uipc_abort: unp == NULL")); 149 UNP_LOCK(); 150 unp_drop(unp, ECONNABORTED); 151 unp_detach(unp); 152 UNP_UNLOCK_ASSERT(); 153 ACCEPT_LOCK(); 154 SOCK_LOCK(so); 155 sotryfree(so); 156 return (0); 157} 158 159static int 160uipc_accept(struct socket *so, struct sockaddr **nam) 161{ 162 struct unpcb *unp; 163 const struct sockaddr *sa; 164 165 /* 166 * Pass back name of connected socket, 167 * if it was bound and we are still connected 168 * (our peer may have closed already!). 169 */ 170 unp = sotounpcb(so); 171 KASSERT(unp != NULL, ("uipc_accept: unp == NULL")); 172 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 173 UNP_LOCK(); 174 if (unp->unp_conn != NULL && unp->unp_conn->unp_addr != NULL) 175 sa = (struct sockaddr *) unp->unp_conn->unp_addr; 176 else 177 sa = &sun_noname; 178 bcopy(sa, *nam, sa->sa_len); 179 UNP_UNLOCK(); 180 return (0); 181} 182 183static int 184uipc_attach(struct socket *so, int proto, struct thread *td) 185{ 186 187 return (unp_attach(so)); 188} 189 190static int 191uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 192{ 193 struct unpcb *unp; 194 int error; 195 196 unp = sotounpcb(so); 197 KASSERT(unp != NULL, ("uipc_bind: unp == NULL")); 198 UNP_LOCK(); 199 error = unp_bind(unp, nam, td); 200 UNP_UNLOCK(); 201 return (error); 202} 203 204static int 205uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 206{ 207 struct unpcb *unp; 208 int error; 209 210 KASSERT(td == curthread, ("uipc_connect: td != curthread")); 211 unp = sotounpcb(so); 212 KASSERT(unp != NULL, ("uipc_connect: unp == NULL")); 213 UNP_LOCK(); 214 error = unp_connect(so, nam, td); 215 UNP_UNLOCK(); 216 return (error); 217} 218 219int 220uipc_connect2(struct socket *so1, struct socket *so2) 221{ 222 struct unpcb *unp; 223 int error; 224 225 unp = sotounpcb(so1); 226 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL")); 227 UNP_LOCK(); 228 error = unp_connect2(so1, so2, PRU_CONNECT2); 229 UNP_UNLOCK(); 230 return (error); 231} 232 233/* control is EOPNOTSUPP */ 234 235static int 236uipc_detach(struct socket *so) 237{ 238 struct unpcb *unp; 239 240 unp = sotounpcb(so); 241 KASSERT(unp != NULL, ("uipc_detach: unp == NULL")); 242 UNP_LOCK(); 243 unp_detach(unp); 244 UNP_UNLOCK_ASSERT(); 245 return (0); 246} 247 248static int 249uipc_disconnect(struct socket *so) 250{ 251 struct unpcb *unp; 252 253 unp = sotounpcb(so); 254 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL")); 255 UNP_LOCK(); 256 unp_disconnect(unp); 257 UNP_UNLOCK(); 258 return (0); 259} 260 261static int 262uipc_listen(struct socket *so, int backlog, struct thread *td) 263{ 264 struct unpcb *unp; 265 int error; 266 267 unp = sotounpcb(so); 268 KASSERT(unp != NULL, ("uipc_listen: unp == NULL")); 269 UNP_LOCK(); 270 if (unp->unp_vnode == NULL) { 271 UNP_UNLOCK(); 272 return (EINVAL); 273 } 274 error = unp_listen(so, unp, backlog, td); 275 UNP_UNLOCK(); 276 return (error); 277} 278 279static int 280uipc_peeraddr(struct socket *so, struct sockaddr **nam) 281{ 282 struct unpcb *unp; 283 const struct sockaddr *sa; 284 285 unp = sotounpcb(so); 286 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL")); 287 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 288 UNP_LOCK(); 289 if (unp->unp_conn != NULL && unp->unp_conn->unp_addr!= NULL) 290 sa = (struct sockaddr *) unp->unp_conn->unp_addr; 291 else { 292 /* 293 * XXX: It seems that this test always fails even when 294 * connection is established. So, this else clause is 295 * added as workaround to return PF_LOCAL sockaddr. 296 */ 297 sa = &sun_noname; 298 } 299 bcopy(sa, *nam, sa->sa_len); 300 UNP_UNLOCK(); 301 return (0); 302} 303 304static int 305uipc_rcvd(struct socket *so, int flags) 306{ 307 struct unpcb *unp; 308 struct socket *so2; 309 u_long newhiwat; 310 311 unp = sotounpcb(so); 312 KASSERT(unp != NULL, ("uipc_rcvd: unp == NULL")); 313 UNP_LOCK(); 314 switch (so->so_type) { 315 case SOCK_DGRAM: 316 panic("uipc_rcvd DGRAM?"); 317 /*NOTREACHED*/ 318 319 case SOCK_STREAM: 320 if (unp->unp_conn == NULL) 321 break; 322 so2 = unp->unp_conn->unp_socket; 323 SOCKBUF_LOCK(&so2->so_snd); 324 SOCKBUF_LOCK(&so->so_rcv); 325 /* 326 * Adjust backpressure on sender 327 * and wakeup any waiting to write. 328 */ 329 so2->so_snd.sb_mbmax += unp->unp_mbcnt - so->so_rcv.sb_mbcnt; 330 unp->unp_mbcnt = so->so_rcv.sb_mbcnt; 331 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - 332 so->so_rcv.sb_cc; 333 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat, 334 newhiwat, RLIM_INFINITY); 335 unp->unp_cc = so->so_rcv.sb_cc; 336 SOCKBUF_UNLOCK(&so->so_rcv); 337 sowwakeup_locked(so2); 338 break; 339 340 default: 341 panic("uipc_rcvd unknown socktype"); 342 } 343 UNP_UNLOCK(); 344 return (0); 345} 346 347/* pru_rcvoob is EOPNOTSUPP */ 348 349static int 350uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 351 struct mbuf *control, struct thread *td) 352{ 353 int error = 0; 354 struct unpcb *unp; 355 struct socket *so2; 356 u_long newhiwat; 357 358 unp = sotounpcb(so); 359 KASSERT(unp != NULL, ("uipc_send: unp == NULL")); 360 if (flags & PRUS_OOB) { 361 error = EOPNOTSUPP; 362 goto release; 363 } 364 365 if (control != NULL && (error = unp_internalize(&control, td))) 366 goto release; 367 368 UNP_LOCK(); 369 switch (so->so_type) { 370 case SOCK_DGRAM: 371 { 372 const struct sockaddr *from; 373 374 if (nam != NULL) { 375 if (unp->unp_conn != NULL) { 376 error = EISCONN; 377 break; 378 } 379 error = unp_connect(so, nam, td); 380 if (error) 381 break; 382 } else { 383 if (unp->unp_conn == NULL) { 384 error = ENOTCONN; 385 break; 386 } 387 } 388 so2 = unp->unp_conn->unp_socket; 389 if (unp->unp_addr != NULL) 390 from = (struct sockaddr *)unp->unp_addr; 391 else 392 from = &sun_noname; 393 if (unp->unp_conn->unp_flags & UNP_WANTCRED) 394 control = unp_addsockcred(td, control); 395 SOCKBUF_LOCK(&so2->so_rcv); 396 if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) { 397 sorwakeup_locked(so2); 398 m = NULL; 399 control = NULL; 400 } else { 401 SOCKBUF_UNLOCK(&so2->so_rcv); 402 error = ENOBUFS; 403 } 404 if (nam != NULL) 405 unp_disconnect(unp); 406 break; 407 } 408 409 case SOCK_STREAM: 410 /* Connect if not connected yet. */ 411 /* 412 * Note: A better implementation would complain 413 * if not equal to the peer's address. 414 */ 415 if ((so->so_state & SS_ISCONNECTED) == 0) { 416 if (nam != NULL) { 417 error = unp_connect(so, nam, td); 418 if (error) 419 break; /* XXX */ 420 } else { 421 error = ENOTCONN; 422 break; 423 } 424 } 425 426 SOCKBUF_LOCK(&so->so_snd); 427 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 428 SOCKBUF_UNLOCK(&so->so_snd); 429 error = EPIPE; 430 break; 431 } 432 if (unp->unp_conn == NULL) 433 panic("uipc_send connected but no connection?"); 434 so2 = unp->unp_conn->unp_socket; 435 SOCKBUF_LOCK(&so2->so_rcv); 436 if (unp->unp_conn->unp_flags & UNP_WANTCRED) { 437 /* 438 * Credentials are passed only once on 439 * SOCK_STREAM. 440 */ 441 unp->unp_conn->unp_flags &= ~UNP_WANTCRED; 442 control = unp_addsockcred(td, control); 443 } 444 /* 445 * Send to paired receive port, and then reduce 446 * send buffer hiwater marks to maintain backpressure. 447 * Wake up readers. 448 */ 449 if (control != NULL) { 450 if (sbappendcontrol_locked(&so2->so_rcv, m, control)) 451 control = NULL; 452 } else { 453 sbappend_locked(&so2->so_rcv, m); 454 } 455 so->so_snd.sb_mbmax -= 456 so2->so_rcv.sb_mbcnt - unp->unp_conn->unp_mbcnt; 457 unp->unp_conn->unp_mbcnt = so2->so_rcv.sb_mbcnt; 458 newhiwat = so->so_snd.sb_hiwat - 459 (so2->so_rcv.sb_cc - unp->unp_conn->unp_cc); 460 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat, 461 newhiwat, RLIM_INFINITY); 462 SOCKBUF_UNLOCK(&so->so_snd); 463 unp->unp_conn->unp_cc = so2->so_rcv.sb_cc; 464 sorwakeup_locked(so2); 465 m = NULL; 466 break; 467 468 default: 469 panic("uipc_send unknown socktype"); 470 } 471 472 /* 473 * SEND_EOF is equivalent to a SEND followed by 474 * a SHUTDOWN. 475 */ 476 if (flags & PRUS_EOF) { 477 socantsendmore(so); 478 unp_shutdown(unp); 479 } 480 UNP_UNLOCK(); 481 482 if (control != NULL && error != 0) 483 unp_dispose(control); 484 485release: 486 if (control != NULL) 487 m_freem(control); 488 if (m != NULL) 489 m_freem(m); 490 return (error); 491} 492 493static int 494uipc_sense(struct socket *so, struct stat *sb) 495{ 496 struct unpcb *unp; 497 struct socket *so2; 498 499 unp = sotounpcb(so); 500 KASSERT(unp != NULL, ("uipc_sense: unp == NULL")); 501 UNP_LOCK(); 502 sb->st_blksize = so->so_snd.sb_hiwat; 503 if (so->so_type == SOCK_STREAM && unp->unp_conn != NULL) { 504 so2 = unp->unp_conn->unp_socket; 505 sb->st_blksize += so2->so_rcv.sb_cc; 506 } 507 sb->st_dev = NODEV; 508 if (unp->unp_ino == 0) 509 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino; 510 sb->st_ino = unp->unp_ino; 511 UNP_UNLOCK(); 512 return (0); 513} 514 515static int 516uipc_shutdown(struct socket *so) 517{ 518 struct unpcb *unp; 519 520 unp = sotounpcb(so); 521 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL")); 522 UNP_LOCK(); 523 socantsendmore(so); 524 unp_shutdown(unp); 525 UNP_UNLOCK(); 526 return (0); 527} 528 529static int 530uipc_sockaddr(struct socket *so, struct sockaddr **nam) 531{ 532 struct unpcb *unp; 533 const struct sockaddr *sa; 534 535 unp = sotounpcb(so); 536 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL")); 537 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 538 UNP_LOCK(); 539 if (unp->unp_addr != NULL) 540 sa = (struct sockaddr *) unp->unp_addr; 541 else 542 sa = &sun_noname; 543 bcopy(sa, *nam, sa->sa_len); 544 UNP_UNLOCK(); 545 return (0); 546} 547 548struct pr_usrreqs uipc_usrreqs = { 549 .pru_abort = uipc_abort, 550 .pru_accept = uipc_accept, 551 .pru_attach = uipc_attach, 552 .pru_bind = uipc_bind, 553 .pru_connect = uipc_connect, 554 .pru_connect2 = uipc_connect2, 555 .pru_detach = uipc_detach, 556 .pru_disconnect = uipc_disconnect, 557 .pru_listen = uipc_listen, 558 .pru_peeraddr = uipc_peeraddr, 559 .pru_rcvd = uipc_rcvd, 560 .pru_send = uipc_send, 561 .pru_sense = uipc_sense, 562 .pru_shutdown = uipc_shutdown, 563 .pru_sockaddr = uipc_sockaddr, 564 .pru_sosend = sosend, 565 .pru_soreceive = soreceive, 566 .pru_sopoll = sopoll, 567}; 568 569int 570uipc_ctloutput(struct socket *so, struct sockopt *sopt) 571{ 572 struct unpcb *unp; 573 struct xucred xu; 574 int error, optval; 575 576 if (sopt->sopt_level != 0) 577 return (EINVAL); 578 579 unp = sotounpcb(so); 580 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL")); 581 UNP_LOCK(); 582 error = 0; 583 switch (sopt->sopt_dir) { 584 case SOPT_GET: 585 switch (sopt->sopt_name) { 586 case LOCAL_PEERCRED: 587 if (unp->unp_flags & UNP_HAVEPC) 588 xu = unp->unp_peercred; 589 else { 590 if (so->so_type == SOCK_STREAM) 591 error = ENOTCONN; 592 else 593 error = EINVAL; 594 } 595 if (error == 0) 596 error = sooptcopyout(sopt, &xu, sizeof(xu)); 597 break; 598 case LOCAL_CREDS: 599 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0; 600 error = sooptcopyout(sopt, &optval, sizeof(optval)); 601 break; 602 case LOCAL_CONNWAIT: 603 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0; 604 error = sooptcopyout(sopt, &optval, sizeof(optval)); 605 break; 606 default: 607 error = EOPNOTSUPP; 608 break; 609 } 610 break; 611 case SOPT_SET: 612 switch (sopt->sopt_name) { 613 case LOCAL_CREDS: 614 case LOCAL_CONNWAIT: 615 error = sooptcopyin(sopt, &optval, sizeof(optval), 616 sizeof(optval)); 617 if (error) 618 break; 619 620#define OPTSET(bit) \ 621 if (optval) \ 622 unp->unp_flags |= bit; \ 623 else \ 624 unp->unp_flags &= ~bit; 625 626 switch (sopt->sopt_name) { 627 case LOCAL_CREDS: 628 OPTSET(UNP_WANTCRED); 629 break; 630 case LOCAL_CONNWAIT: 631 OPTSET(UNP_CONNWAIT); 632 break; 633 default: 634 break; 635 } 636 break; 637#undef OPTSET 638 default: 639 error = ENOPROTOOPT; 640 break; 641 } 642 break; 643 default: 644 error = EOPNOTSUPP; 645 break; 646 } 647 UNP_UNLOCK(); 648 return (error); 649} 650 651/* 652 * Both send and receive buffers are allocated PIPSIZ bytes of buffering 653 * for stream sockets, although the total for sender and receiver is 654 * actually only PIPSIZ. 655 * Datagram sockets really use the sendspace as the maximum datagram size, 656 * and don't really want to reserve the sendspace. Their recvspace should 657 * be large enough for at least one max-size datagram plus address. 658 */ 659#ifndef PIPSIZ 660#define PIPSIZ 8192 661#endif 662static u_long unpst_sendspace = PIPSIZ; 663static u_long unpst_recvspace = PIPSIZ; 664static u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 665static u_long unpdg_recvspace = 4*1024; 666 667static int unp_rights; /* file descriptors in flight */ 668 669SYSCTL_DECL(_net_local_stream); 670SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 671 &unpst_sendspace, 0, ""); 672SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 673 &unpst_recvspace, 0, ""); 674SYSCTL_DECL(_net_local_dgram); 675SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 676 &unpdg_sendspace, 0, ""); 677SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 678 &unpdg_recvspace, 0, ""); 679SYSCTL_DECL(_net_local); 680SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, ""); 681 682static int 683unp_attach(struct socket *so) 684{ 685 struct unpcb *unp; 686 int error; 687 688 KASSERT(so->so_pcb == NULL, ("unp_attach: so_pcb != NULL")); 689 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 690 switch (so->so_type) { 691 692 case SOCK_STREAM: 693 error = soreserve(so, unpst_sendspace, unpst_recvspace); 694 break; 695 696 case SOCK_DGRAM: 697 error = soreserve(so, unpdg_sendspace, unpdg_recvspace); 698 break; 699 700 default: 701 panic("unp_attach"); 702 } 703 if (error) 704 return (error); 705 } 706 unp = uma_zalloc(unp_zone, M_WAITOK | M_ZERO); 707 if (unp == NULL) 708 return (ENOBUFS); 709 LIST_INIT(&unp->unp_refs); 710 unp->unp_socket = so; 711 so->so_pcb = unp; 712 713 UNP_LOCK(); 714 unp->unp_gencnt = ++unp_gencnt; 715 unp_count++; 716 LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead 717 : &unp_shead, unp, unp_link); 718 UNP_UNLOCK(); 719 720 return (0); 721} 722 723static void 724unp_detach(struct unpcb *unp) 725{ 726 struct vnode *vp; 727 int local_unp_rights; 728 729 UNP_LOCK_ASSERT(); 730 731 LIST_REMOVE(unp, unp_link); 732 unp->unp_gencnt = ++unp_gencnt; 733 --unp_count; 734 if ((vp = unp->unp_vnode) != NULL) { 735 /* 736 * XXXRW: should v_socket be frobbed only while holding 737 * Giant? 738 */ 739 unp->unp_vnode->v_socket = NULL; 740 unp->unp_vnode = NULL; 741 } 742 if (unp->unp_conn != NULL) 743 unp_disconnect(unp); 744 while (!LIST_EMPTY(&unp->unp_refs)) { 745 struct unpcb *ref = LIST_FIRST(&unp->unp_refs); 746 unp_drop(ref, ECONNRESET); 747 } 748 soisdisconnected(unp->unp_socket); 749 unp->unp_socket->so_pcb = NULL; 750 local_unp_rights = unp_rights; 751 UNP_UNLOCK(); 752 if (unp->unp_addr != NULL) 753 FREE(unp->unp_addr, M_SONAME); 754 uma_zfree(unp_zone, unp); 755 if (vp) { 756 int vfslocked; 757 758 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 759 vrele(vp); 760 VFS_UNLOCK_GIANT(vfslocked); 761 } 762 if (local_unp_rights) 763 taskqueue_enqueue(taskqueue_thread, &unp_gc_task); 764} 765 766static int 767unp_bind(struct unpcb *unp, struct sockaddr *nam, struct thread *td) 768{ 769 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 770 struct vnode *vp; 771 struct mount *mp; 772 struct vattr vattr; 773 int error, namelen; 774 struct nameidata nd; 775 char *buf; 776 777 UNP_LOCK_ASSERT(); 778 779 /* 780 * XXXRW: This test-and-set of unp_vnode is non-atomic; the 781 * unlocked read here is fine, but the value of unp_vnode needs 782 * to be tested again after we do all the lookups to see if the 783 * pcb is still unbound? 784 */ 785 if (unp->unp_vnode != NULL) 786 return (EINVAL); 787 788 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 789 if (namelen <= 0) 790 return (EINVAL); 791 792 UNP_UNLOCK(); 793 794 buf = malloc(namelen + 1, M_TEMP, M_WAITOK); 795 strlcpy(buf, soun->sun_path, namelen + 1); 796 797 mtx_lock(&Giant); 798restart: 799 mtx_assert(&Giant, MA_OWNED); 800 NDINIT(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME, UIO_SYSSPACE, 801 buf, td); 802/* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 803 error = namei(&nd); 804 if (error) 805 goto done; 806 vp = nd.ni_vp; 807 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { 808 NDFREE(&nd, NDF_ONLY_PNBUF); 809 if (nd.ni_dvp == vp) 810 vrele(nd.ni_dvp); 811 else 812 vput(nd.ni_dvp); 813 if (vp != NULL) { 814 vrele(vp); 815 error = EADDRINUSE; 816 goto done; 817 } 818 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH); 819 if (error) 820 goto done; 821 goto restart; 822 } 823 VATTR_NULL(&vattr); 824 vattr.va_type = VSOCK; 825 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask); 826#ifdef MAC 827 error = mac_check_vnode_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, 828 &vattr); 829#endif 830 if (error == 0) { 831 VOP_LEASE(nd.ni_dvp, td, td->td_ucred, LEASE_WRITE); 832 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 833 } 834 NDFREE(&nd, NDF_ONLY_PNBUF); 835 vput(nd.ni_dvp); 836 if (error) { 837 vn_finished_write(mp); 838 goto done; 839 } 840 vp = nd.ni_vp; 841 ASSERT_VOP_LOCKED(vp, "unp_bind"); 842 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK); 843 UNP_LOCK(); 844 vp->v_socket = unp->unp_socket; 845 unp->unp_vnode = vp; 846 unp->unp_addr = soun; 847 UNP_UNLOCK(); 848 VOP_UNLOCK(vp, 0, td); 849 vn_finished_write(mp); 850done: 851 mtx_unlock(&Giant); 852 free(buf, M_TEMP); 853 UNP_LOCK(); 854 return (error); 855} 856 857static int 858unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 859{ 860 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 861 struct vnode *vp; 862 struct socket *so2, *so3; 863 struct unpcb *unp, *unp2, *unp3; 864 int error, len; 865 struct nameidata nd; 866 char buf[SOCK_MAXADDRLEN]; 867 struct sockaddr *sa; 868 869 UNP_LOCK_ASSERT(); 870 871 unp = sotounpcb(so); 872 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 873 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 874 if (len <= 0) 875 return (EINVAL); 876 strlcpy(buf, soun->sun_path, len + 1); 877 UNP_UNLOCK(); 878 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 879 mtx_lock(&Giant); 880 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, td); 881 error = namei(&nd); 882 if (error) 883 vp = NULL; 884 else 885 vp = nd.ni_vp; 886 ASSERT_VOP_LOCKED(vp, "unp_connect"); 887 NDFREE(&nd, NDF_ONLY_PNBUF); 888 if (error) 889 goto bad; 890 891 if (vp->v_type != VSOCK) { 892 error = ENOTSOCK; 893 goto bad; 894 } 895 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); 896 if (error) 897 goto bad; 898 mtx_unlock(&Giant); 899 UNP_LOCK(); 900 unp = sotounpcb(so); 901 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 902 so2 = vp->v_socket; 903 if (so2 == NULL) { 904 error = ECONNREFUSED; 905 goto bad2; 906 } 907 if (so->so_type != so2->so_type) { 908 error = EPROTOTYPE; 909 goto bad2; 910 } 911 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 912 if (so2->so_options & SO_ACCEPTCONN) { 913 /* 914 * NB: drop locks here so unp_attach is entered 915 * w/o locks; this avoids a recursive lock 916 * of the head and holding sleep locks across 917 * a (potentially) blocking malloc. 918 */ 919 UNP_UNLOCK(); 920 so3 = sonewconn(so2, 0); 921 UNP_LOCK(); 922 } else 923 so3 = NULL; 924 if (so3 == NULL) { 925 error = ECONNREFUSED; 926 goto bad2; 927 } 928 unp = sotounpcb(so); 929 unp2 = sotounpcb(so2); 930 unp3 = sotounpcb(so3); 931 if (unp2->unp_addr != NULL) { 932 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); 933 unp3->unp_addr = (struct sockaddr_un *) sa; 934 sa = NULL; 935 } 936 /* 937 * unp_peercred management: 938 * 939 * The connecter's (client's) credentials are copied 940 * from its process structure at the time of connect() 941 * (which is now). 942 */ 943 cru2x(td->td_ucred, &unp3->unp_peercred); 944 unp3->unp_flags |= UNP_HAVEPC; 945 /* 946 * The receiver's (server's) credentials are copied 947 * from the unp_peercred member of socket on which the 948 * former called listen(); unp_listen() cached that 949 * process's credentials at that time so we can use 950 * them now. 951 */ 952 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED, 953 ("unp_connect: listener without cached peercred")); 954 memcpy(&unp->unp_peercred, &unp2->unp_peercred, 955 sizeof(unp->unp_peercred)); 956 unp->unp_flags |= UNP_HAVEPC; 957#ifdef MAC 958 SOCK_LOCK(so); 959 mac_set_socket_peer_from_socket(so, so3); 960 mac_set_socket_peer_from_socket(so3, so); 961 SOCK_UNLOCK(so); 962#endif 963 964 so2 = so3; 965 } 966 error = unp_connect2(so, so2, PRU_CONNECT); 967bad2: 968 UNP_UNLOCK(); 969 mtx_lock(&Giant); 970bad: 971 mtx_assert(&Giant, MA_OWNED); 972 if (vp != NULL) 973 vput(vp); 974 mtx_unlock(&Giant); 975 free(sa, M_SONAME); 976 UNP_LOCK(); 977 return (error); 978} 979 980static int 981unp_connect2(struct socket *so, struct socket *so2, int req) 982{ 983 struct unpcb *unp = sotounpcb(so); 984 struct unpcb *unp2; 985 986 UNP_LOCK_ASSERT(); 987 988 if (so2->so_type != so->so_type) 989 return (EPROTOTYPE); 990 unp2 = sotounpcb(so2); 991 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL")); 992 unp->unp_conn = unp2; 993 switch (so->so_type) { 994 995 case SOCK_DGRAM: 996 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 997 soisconnected(so); 998 break; 999 1000 case SOCK_STREAM: 1001 unp2->unp_conn = unp; 1002 if (req == PRU_CONNECT && 1003 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 1004 soisconnecting(so); 1005 else 1006 soisconnected(so); 1007 soisconnected(so2); 1008 break; 1009 1010 default: 1011 panic("unp_connect2"); 1012 } 1013 return (0); 1014} 1015 1016static void 1017unp_disconnect(struct unpcb *unp) 1018{ 1019 struct unpcb *unp2 = unp->unp_conn; 1020 struct socket *so; 1021 1022 UNP_LOCK_ASSERT(); 1023 1024 if (unp2 == NULL) 1025 return; 1026 unp->unp_conn = NULL; 1027 switch (unp->unp_socket->so_type) { 1028 case SOCK_DGRAM: 1029 LIST_REMOVE(unp, unp_reflink); 1030 so = unp->unp_socket; 1031 SOCK_LOCK(so); 1032 so->so_state &= ~SS_ISCONNECTED; 1033 SOCK_UNLOCK(so); 1034 break; 1035 1036 case SOCK_STREAM: 1037 soisdisconnected(unp->unp_socket); 1038 unp2->unp_conn = NULL; 1039 soisdisconnected(unp2->unp_socket); 1040 break; 1041 } 1042} 1043 1044#ifdef notdef 1045void 1046unp_abort(struct unpcb *unp) 1047{ 1048 1049 unp_detach(unp); 1050 UNP_UNLOCK_ASSERT(); 1051} 1052#endif 1053 1054/* 1055 * unp_pcblist() assumes that UNIX domain socket memory is never reclaimed 1056 * by the zone (UMA_ZONE_NOFREE), and as such potentially stale pointers 1057 * are safe to reference. It first scans the list of struct unpcb's to 1058 * generate a pointer list, then it rescans its list one entry at a time to 1059 * externalize and copyout. It checks the generation number to see if a 1060 * struct unpcb has been reused, and will skip it if so. 1061 */ 1062static int 1063unp_pcblist(SYSCTL_HANDLER_ARGS) 1064{ 1065 int error, i, n; 1066 struct unpcb *unp, **unp_list; 1067 unp_gen_t gencnt; 1068 struct xunpgen *xug; 1069 struct unp_head *head; 1070 struct xunpcb *xu; 1071 1072 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead); 1073 1074 /* 1075 * The process of preparing the PCB list is too time-consuming and 1076 * resource-intensive to repeat twice on every request. 1077 */ 1078 if (req->oldptr == NULL) { 1079 n = unp_count; 1080 req->oldidx = 2 * (sizeof *xug) 1081 + (n + n/8) * sizeof(struct xunpcb); 1082 return (0); 1083 } 1084 1085 if (req->newptr != NULL) 1086 return (EPERM); 1087 1088 /* 1089 * OK, now we're committed to doing something. 1090 */ 1091 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK); 1092 UNP_LOCK(); 1093 gencnt = unp_gencnt; 1094 n = unp_count; 1095 UNP_UNLOCK(); 1096 1097 xug->xug_len = sizeof *xug; 1098 xug->xug_count = n; 1099 xug->xug_gen = gencnt; 1100 xug->xug_sogen = so_gencnt; 1101 error = SYSCTL_OUT(req, xug, sizeof *xug); 1102 if (error) { 1103 free(xug, M_TEMP); 1104 return (error); 1105 } 1106 1107 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 1108 1109 UNP_LOCK(); 1110 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 1111 unp = LIST_NEXT(unp, unp_link)) { 1112 if (unp->unp_gencnt <= gencnt) { 1113 if (cr_cansee(req->td->td_ucred, 1114 unp->unp_socket->so_cred)) 1115 continue; 1116 unp_list[i++] = unp; 1117 } 1118 } 1119 UNP_UNLOCK(); 1120 n = i; /* in case we lost some during malloc */ 1121 1122 error = 0; 1123 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO); 1124 for (i = 0; i < n; i++) { 1125 unp = unp_list[i]; 1126 if (unp->unp_gencnt <= gencnt) { 1127 xu->xu_len = sizeof *xu; 1128 xu->xu_unpp = unp; 1129 /* 1130 * XXX - need more locking here to protect against 1131 * connect/disconnect races for SMP. 1132 */ 1133 if (unp->unp_addr != NULL) 1134 bcopy(unp->unp_addr, &xu->xu_addr, 1135 unp->unp_addr->sun_len); 1136 if (unp->unp_conn != NULL && 1137 unp->unp_conn->unp_addr != NULL) 1138 bcopy(unp->unp_conn->unp_addr, 1139 &xu->xu_caddr, 1140 unp->unp_conn->unp_addr->sun_len); 1141 bcopy(unp, &xu->xu_unp, sizeof *unp); 1142 sotoxsocket(unp->unp_socket, &xu->xu_socket); 1143 error = SYSCTL_OUT(req, xu, sizeof *xu); 1144 } 1145 } 1146 free(xu, M_TEMP); 1147 if (!error) { 1148 /* 1149 * Give the user an updated idea of our state. 1150 * If the generation differs from what we told 1151 * her before, she knows that something happened 1152 * while we were processing this request, and it 1153 * might be necessary to retry. 1154 */ 1155 xug->xug_gen = unp_gencnt; 1156 xug->xug_sogen = so_gencnt; 1157 xug->xug_count = unp_count; 1158 error = SYSCTL_OUT(req, xug, sizeof *xug); 1159 } 1160 free(unp_list, M_TEMP); 1161 free(xug, M_TEMP); 1162 return (error); 1163} 1164 1165SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD, 1166 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 1167 "List of active local datagram sockets"); 1168SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD, 1169 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 1170 "List of active local stream sockets"); 1171 1172static void 1173unp_shutdown(struct unpcb *unp) 1174{ 1175 struct socket *so; 1176 1177 UNP_LOCK_ASSERT(); 1178 1179 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn && 1180 (so = unp->unp_conn->unp_socket)) 1181 socantrcvmore(so); 1182} 1183 1184static void 1185unp_drop(struct unpcb *unp, int errno) 1186{ 1187 struct socket *so = unp->unp_socket; 1188 1189 UNP_LOCK_ASSERT(); 1190 1191 so->so_error = errno; 1192 unp_disconnect(unp); 1193} 1194 1195#ifdef notdef 1196void 1197unp_drain(void) 1198{ 1199 1200} 1201#endif 1202 1203static void 1204unp_freerights(struct file **rp, int fdcount) 1205{ 1206 int i; 1207 struct file *fp; 1208 1209 for (i = 0; i < fdcount; i++) { 1210 fp = *rp; 1211 /* 1212 * zero the pointer before calling 1213 * unp_discard since it may end up 1214 * in unp_gc().. 1215 * 1216 * XXXRW: This is less true than it used to be. 1217 */ 1218 *rp++ = 0; 1219 unp_discard(fp); 1220 } 1221} 1222 1223int 1224unp_externalize(struct mbuf *control, struct mbuf **controlp) 1225{ 1226 struct thread *td = curthread; /* XXX */ 1227 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1228 int i; 1229 int *fdp; 1230 struct file **rp; 1231 struct file *fp; 1232 void *data; 1233 socklen_t clen = control->m_len, datalen; 1234 int error, newfds; 1235 int f; 1236 u_int newlen; 1237 1238 UNP_UNLOCK_ASSERT(); 1239 1240 error = 0; 1241 if (controlp != NULL) /* controlp == NULL => free control messages */ 1242 *controlp = NULL; 1243 1244 while (cm != NULL) { 1245 if (sizeof(*cm) > clen || cm->cmsg_len > clen) { 1246 error = EINVAL; 1247 break; 1248 } 1249 1250 data = CMSG_DATA(cm); 1251 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1252 1253 if (cm->cmsg_level == SOL_SOCKET 1254 && cm->cmsg_type == SCM_RIGHTS) { 1255 newfds = datalen / sizeof(struct file *); 1256 rp = data; 1257 1258 /* If we're not outputting the descriptors free them. */ 1259 if (error || controlp == NULL) { 1260 unp_freerights(rp, newfds); 1261 goto next; 1262 } 1263 FILEDESC_LOCK(td->td_proc->p_fd); 1264 /* if the new FD's will not fit free them. */ 1265 if (!fdavail(td, newfds)) { 1266 FILEDESC_UNLOCK(td->td_proc->p_fd); 1267 error = EMSGSIZE; 1268 unp_freerights(rp, newfds); 1269 goto next; 1270 } 1271 /* 1272 * now change each pointer to an fd in the global 1273 * table to an integer that is the index to the 1274 * local fd table entry that we set up to point 1275 * to the global one we are transferring. 1276 */ 1277 newlen = newfds * sizeof(int); 1278 *controlp = sbcreatecontrol(NULL, newlen, 1279 SCM_RIGHTS, SOL_SOCKET); 1280 if (*controlp == NULL) { 1281 FILEDESC_UNLOCK(td->td_proc->p_fd); 1282 error = E2BIG; 1283 unp_freerights(rp, newfds); 1284 goto next; 1285 } 1286 1287 fdp = (int *) 1288 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1289 for (i = 0; i < newfds; i++) { 1290 if (fdalloc(td, 0, &f)) 1291 panic("unp_externalize fdalloc failed"); 1292 fp = *rp++; 1293 td->td_proc->p_fd->fd_ofiles[f] = fp; 1294 FILE_LOCK(fp); 1295 fp->f_msgcount--; 1296 FILE_UNLOCK(fp); 1297 unp_rights--; 1298 *fdp++ = f; 1299 } 1300 FILEDESC_UNLOCK(td->td_proc->p_fd); 1301 } else { /* We can just copy anything else across */ 1302 if (error || controlp == NULL) 1303 goto next; 1304 *controlp = sbcreatecontrol(NULL, datalen, 1305 cm->cmsg_type, cm->cmsg_level); 1306 if (*controlp == NULL) { 1307 error = ENOBUFS; 1308 goto next; 1309 } 1310 bcopy(data, 1311 CMSG_DATA(mtod(*controlp, struct cmsghdr *)), 1312 datalen); 1313 } 1314 1315 controlp = &(*controlp)->m_next; 1316 1317next: 1318 if (CMSG_SPACE(datalen) < clen) { 1319 clen -= CMSG_SPACE(datalen); 1320 cm = (struct cmsghdr *) 1321 ((caddr_t)cm + CMSG_SPACE(datalen)); 1322 } else { 1323 clen = 0; 1324 cm = NULL; 1325 } 1326 } 1327 1328 m_freem(control); 1329 1330 return (error); 1331} 1332 1333void 1334unp_init(void) 1335{ 1336 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL, 1337 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 1338 if (unp_zone == NULL) 1339 panic("unp_init"); 1340 uma_zone_set_max(unp_zone, nmbclusters); 1341 LIST_INIT(&unp_dhead); 1342 LIST_INIT(&unp_shead); 1343 TASK_INIT(&unp_gc_task, 0, unp_gc, NULL); 1344 UNP_LOCK_INIT(); 1345} 1346 1347static int 1348unp_internalize(struct mbuf **controlp, struct thread *td) 1349{ 1350 struct mbuf *control = *controlp; 1351 struct proc *p = td->td_proc; 1352 struct filedesc *fdescp = p->p_fd; 1353 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1354 struct cmsgcred *cmcred; 1355 struct file **rp; 1356 struct file *fp; 1357 struct timeval *tv; 1358 int i, fd, *fdp; 1359 void *data; 1360 socklen_t clen = control->m_len, datalen; 1361 int error, oldfds; 1362 u_int newlen; 1363 1364 UNP_UNLOCK_ASSERT(); 1365 1366 error = 0; 1367 *controlp = NULL; 1368 1369 while (cm != NULL) { 1370 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET 1371 || cm->cmsg_len > clen) { 1372 error = EINVAL; 1373 goto out; 1374 } 1375 1376 data = CMSG_DATA(cm); 1377 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1378 1379 switch (cm->cmsg_type) { 1380 /* 1381 * Fill in credential information. 1382 */ 1383 case SCM_CREDS: 1384 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred), 1385 SCM_CREDS, SOL_SOCKET); 1386 if (*controlp == NULL) { 1387 error = ENOBUFS; 1388 goto out; 1389 } 1390 1391 cmcred = (struct cmsgcred *) 1392 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1393 cmcred->cmcred_pid = p->p_pid; 1394 cmcred->cmcred_uid = td->td_ucred->cr_ruid; 1395 cmcred->cmcred_gid = td->td_ucred->cr_rgid; 1396 cmcred->cmcred_euid = td->td_ucred->cr_uid; 1397 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups, 1398 CMGROUP_MAX); 1399 for (i = 0; i < cmcred->cmcred_ngroups; i++) 1400 cmcred->cmcred_groups[i] = 1401 td->td_ucred->cr_groups[i]; 1402 break; 1403 1404 case SCM_RIGHTS: 1405 oldfds = datalen / sizeof (int); 1406 /* 1407 * check that all the FDs passed in refer to legal files 1408 * If not, reject the entire operation. 1409 */ 1410 fdp = data; 1411 FILEDESC_LOCK(fdescp); 1412 for (i = 0; i < oldfds; i++) { 1413 fd = *fdp++; 1414 if ((unsigned)fd >= fdescp->fd_nfiles || 1415 fdescp->fd_ofiles[fd] == NULL) { 1416 FILEDESC_UNLOCK(fdescp); 1417 error = EBADF; 1418 goto out; 1419 } 1420 fp = fdescp->fd_ofiles[fd]; 1421 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { 1422 FILEDESC_UNLOCK(fdescp); 1423 error = EOPNOTSUPP; 1424 goto out; 1425 } 1426 1427 } 1428 /* 1429 * Now replace the integer FDs with pointers to 1430 * the associated global file table entry.. 1431 */ 1432 newlen = oldfds * sizeof(struct file *); 1433 *controlp = sbcreatecontrol(NULL, newlen, 1434 SCM_RIGHTS, SOL_SOCKET); 1435 if (*controlp == NULL) { 1436 FILEDESC_UNLOCK(fdescp); 1437 error = E2BIG; 1438 goto out; 1439 } 1440 1441 fdp = data; 1442 rp = (struct file **) 1443 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1444 for (i = 0; i < oldfds; i++) { 1445 fp = fdescp->fd_ofiles[*fdp++]; 1446 *rp++ = fp; 1447 FILE_LOCK(fp); 1448 fp->f_count++; 1449 fp->f_msgcount++; 1450 FILE_UNLOCK(fp); 1451 unp_rights++; 1452 } 1453 FILEDESC_UNLOCK(fdescp); 1454 break; 1455 1456 case SCM_TIMESTAMP: 1457 *controlp = sbcreatecontrol(NULL, sizeof(*tv), 1458 SCM_TIMESTAMP, SOL_SOCKET); 1459 if (*controlp == NULL) { 1460 error = ENOBUFS; 1461 goto out; 1462 } 1463 tv = (struct timeval *) 1464 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1465 microtime(tv); 1466 break; 1467 1468 default: 1469 error = EINVAL; 1470 goto out; 1471 } 1472 1473 controlp = &(*controlp)->m_next; 1474 1475 if (CMSG_SPACE(datalen) < clen) { 1476 clen -= CMSG_SPACE(datalen); 1477 cm = (struct cmsghdr *) 1478 ((caddr_t)cm + CMSG_SPACE(datalen)); 1479 } else { 1480 clen = 0; 1481 cm = NULL; 1482 } 1483 } 1484 1485out: 1486 m_freem(control); 1487 1488 return (error); 1489} 1490 1491struct mbuf * 1492unp_addsockcred(struct thread *td, struct mbuf *control) 1493{ 1494 struct mbuf *m, *n; 1495 struct sockcred *sc; 1496 int ngroups; 1497 int i; 1498 1499 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); 1500 1501 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET); 1502 if (m == NULL) 1503 return (control); 1504 m->m_next = NULL; 1505 1506 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *)); 1507 sc->sc_uid = td->td_ucred->cr_ruid; 1508 sc->sc_euid = td->td_ucred->cr_uid; 1509 sc->sc_gid = td->td_ucred->cr_rgid; 1510 sc->sc_egid = td->td_ucred->cr_gid; 1511 sc->sc_ngroups = ngroups; 1512 for (i = 0; i < sc->sc_ngroups; i++) 1513 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 1514 1515 /* 1516 * If a control message already exists, append us to the end. 1517 */ 1518 if (control != NULL) { 1519 for (n = control; n->m_next != NULL; n = n->m_next) 1520 ; 1521 n->m_next = m; 1522 } else 1523 control = m; 1524 1525 return (control); 1526} 1527 1528/* 1529 * unp_defer indicates whether additional work has been defered for a future 1530 * pass through unp_gc(). It is thread local and does not require explicit 1531 * synchronization. 1532 */ 1533static int unp_defer; 1534 1535static int unp_taskcount; 1536SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, ""); 1537 1538static int unp_recycled; 1539SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, ""); 1540 1541static void 1542unp_gc(__unused void *arg, int pending) 1543{ 1544 struct file *fp, *nextfp; 1545 struct socket *so; 1546 struct file **extra_ref, **fpp; 1547 int nunref, i; 1548 int nfiles_snap; 1549 int nfiles_slack = 20; 1550 1551 unp_taskcount++; 1552 unp_defer = 0; 1553 /* 1554 * before going through all this, set all FDs to 1555 * be NOT defered and NOT externally accessible 1556 */ 1557 sx_slock(&filelist_lock); 1558 LIST_FOREACH(fp, &filehead, f_list) 1559 fp->f_gcflag &= ~(FMARK|FDEFER); 1560 do { 1561 KASSERT(unp_defer >= 0, ("unp_gc: unp_defer %d", unp_defer)); 1562 LIST_FOREACH(fp, &filehead, f_list) { 1563 FILE_LOCK(fp); 1564 /* 1565 * If the file is not open, skip it -- could be a 1566 * file in the process of being opened, or in the 1567 * process of being closed. If the file is 1568 * "closing", it may have been marked for deferred 1569 * consideration. Clear the flag now if so. 1570 */ 1571 if (fp->f_count == 0) { 1572 if (fp->f_gcflag & FDEFER) 1573 unp_defer--; 1574 fp->f_gcflag &= ~(FMARK|FDEFER); 1575 FILE_UNLOCK(fp); 1576 continue; 1577 } 1578 /* 1579 * If we already marked it as 'defer' in a 1580 * previous pass, then try process it this time 1581 * and un-mark it 1582 */ 1583 if (fp->f_gcflag & FDEFER) { 1584 fp->f_gcflag &= ~FDEFER; 1585 unp_defer--; 1586 } else { 1587 /* 1588 * if it's not defered, then check if it's 1589 * already marked.. if so skip it 1590 */ 1591 if (fp->f_gcflag & FMARK) { 1592 FILE_UNLOCK(fp); 1593 continue; 1594 } 1595 /* 1596 * If all references are from messages 1597 * in transit, then skip it. it's not 1598 * externally accessible. 1599 */ 1600 if (fp->f_count == fp->f_msgcount) { 1601 FILE_UNLOCK(fp); 1602 continue; 1603 } 1604 /* 1605 * If it got this far then it must be 1606 * externally accessible. 1607 */ 1608 fp->f_gcflag |= FMARK; 1609 } 1610 /* 1611 * either it was defered, or it is externally 1612 * accessible and not already marked so. 1613 * Now check if it is possibly one of OUR sockets. 1614 */ 1615 if (fp->f_type != DTYPE_SOCKET || 1616 (so = fp->f_data) == NULL) { 1617 FILE_UNLOCK(fp); 1618 continue; 1619 } 1620 FILE_UNLOCK(fp); 1621 if (so->so_proto->pr_domain != &localdomain || 1622 (so->so_proto->pr_flags&PR_RIGHTS) == 0) 1623 continue; 1624 /* 1625 * So, Ok, it's one of our sockets and it IS externally 1626 * accessible (or was defered). Now we look 1627 * to see if we hold any file descriptors in its 1628 * message buffers. Follow those links and mark them 1629 * as accessible too. 1630 */ 1631 SOCKBUF_LOCK(&so->so_rcv); 1632 unp_scan(so->so_rcv.sb_mb, unp_mark); 1633 SOCKBUF_UNLOCK(&so->so_rcv); 1634 } 1635 } while (unp_defer); 1636 sx_sunlock(&filelist_lock); 1637 /* 1638 * XXXRW: The following comments need updating for a post-SMPng and 1639 * deferred unp_gc() world, but are still generally accurate. 1640 * 1641 * We grab an extra reference to each of the file table entries 1642 * that are not otherwise accessible and then free the rights 1643 * that are stored in messages on them. 1644 * 1645 * The bug in the orginal code is a little tricky, so I'll describe 1646 * what's wrong with it here. 1647 * 1648 * It is incorrect to simply unp_discard each entry for f_msgcount 1649 * times -- consider the case of sockets A and B that contain 1650 * references to each other. On a last close of some other socket, 1651 * we trigger a gc since the number of outstanding rights (unp_rights) 1652 * is non-zero. If during the sweep phase the gc code unp_discards, 1653 * we end up doing a (full) closef on the descriptor. A closef on A 1654 * results in the following chain. Closef calls soo_close, which 1655 * calls soclose. Soclose calls first (through the switch 1656 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply 1657 * returns because the previous instance had set unp_gcing, and 1658 * we return all the way back to soclose, which marks the socket 1659 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush 1660 * to free up the rights that are queued in messages on the socket A, 1661 * i.e., the reference on B. The sorflush calls via the dom_dispose 1662 * switch unp_dispose, which unp_scans with unp_discard. This second 1663 * instance of unp_discard just calls closef on B. 1664 * 1665 * Well, a similar chain occurs on B, resulting in a sorflush on B, 1666 * which results in another closef on A. Unfortunately, A is already 1667 * being closed, and the descriptor has already been marked with 1668 * SS_NOFDREF, and soclose panics at this point. 1669 * 1670 * Here, we first take an extra reference to each inaccessible 1671 * descriptor. Then, we call sorflush ourself, since we know 1672 * it is a Unix domain socket anyhow. After we destroy all the 1673 * rights carried in messages, we do a last closef to get rid 1674 * of our extra reference. This is the last close, and the 1675 * unp_detach etc will shut down the socket. 1676 * 1677 * 91/09/19, bsy@cs.cmu.edu 1678 */ 1679again: 1680 nfiles_snap = openfiles + nfiles_slack; /* some slack */ 1681 extra_ref = malloc(nfiles_snap * sizeof(struct file *), M_TEMP, 1682 M_WAITOK); 1683 sx_slock(&filelist_lock); 1684 if (nfiles_snap < openfiles) { 1685 sx_sunlock(&filelist_lock); 1686 free(extra_ref, M_TEMP); 1687 nfiles_slack += 20; 1688 goto again; 1689 } 1690 for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref; 1691 fp != NULL; fp = nextfp) { 1692 nextfp = LIST_NEXT(fp, f_list); 1693 FILE_LOCK(fp); 1694 /* 1695 * If it's not open, skip it 1696 */ 1697 if (fp->f_count == 0) { 1698 FILE_UNLOCK(fp); 1699 continue; 1700 } 1701 /* 1702 * If all refs are from msgs, and it's not marked accessible 1703 * then it must be referenced from some unreachable cycle 1704 * of (shut-down) FDs, so include it in our 1705 * list of FDs to remove 1706 */ 1707 if (fp->f_count == fp->f_msgcount && !(fp->f_gcflag & FMARK)) { 1708 *fpp++ = fp; 1709 nunref++; 1710 fp->f_count++; 1711 } 1712 FILE_UNLOCK(fp); 1713 } 1714 sx_sunlock(&filelist_lock); 1715 /* 1716 * for each FD on our hit list, do the following two things 1717 */ 1718 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1719 struct file *tfp = *fpp; 1720 FILE_LOCK(tfp); 1721 if (tfp->f_type == DTYPE_SOCKET && 1722 tfp->f_data != NULL) { 1723 FILE_UNLOCK(tfp); 1724 sorflush(tfp->f_data); 1725 } else { 1726 FILE_UNLOCK(tfp); 1727 } 1728 } 1729 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1730 closef(*fpp, (struct thread *) NULL); 1731 unp_recycled++; 1732 } 1733 free(extra_ref, M_TEMP); 1734} 1735 1736void 1737unp_dispose(struct mbuf *m) 1738{ 1739 1740 if (m) 1741 unp_scan(m, unp_discard); 1742} 1743 1744static int 1745unp_listen(struct socket *so, struct unpcb *unp, int backlog, 1746 struct thread *td) 1747{ 1748 int error; 1749 1750 UNP_LOCK_ASSERT(); 1751 1752 SOCK_LOCK(so); 1753 error = solisten_proto_check(so); 1754 if (error == 0) { 1755 cru2x(td->td_ucred, &unp->unp_peercred); 1756 unp->unp_flags |= UNP_HAVEPCCACHED; 1757 solisten_proto(so, backlog); 1758 } 1759 SOCK_UNLOCK(so); 1760 return (error); 1761} 1762 1763static void 1764unp_scan(struct mbuf *m0, void (*op)(struct file *)) 1765{ 1766 struct mbuf *m; 1767 struct file **rp; 1768 struct cmsghdr *cm; 1769 void *data; 1770 int i; 1771 socklen_t clen, datalen; 1772 int qfds; 1773 1774 while (m0 != NULL) { 1775 for (m = m0; m; m = m->m_next) { 1776 if (m->m_type != MT_CONTROL) 1777 continue; 1778 1779 cm = mtod(m, struct cmsghdr *); 1780 clen = m->m_len; 1781 1782 while (cm != NULL) { 1783 if (sizeof(*cm) > clen || cm->cmsg_len > clen) 1784 break; 1785 1786 data = CMSG_DATA(cm); 1787 datalen = (caddr_t)cm + cm->cmsg_len 1788 - (caddr_t)data; 1789 1790 if (cm->cmsg_level == SOL_SOCKET && 1791 cm->cmsg_type == SCM_RIGHTS) { 1792 qfds = datalen / sizeof (struct file *); 1793 rp = data; 1794 for (i = 0; i < qfds; i++) 1795 (*op)(*rp++); 1796 } 1797 1798 if (CMSG_SPACE(datalen) < clen) { 1799 clen -= CMSG_SPACE(datalen); 1800 cm = (struct cmsghdr *) 1801 ((caddr_t)cm + CMSG_SPACE(datalen)); 1802 } else { 1803 clen = 0; 1804 cm = NULL; 1805 } 1806 } 1807 } 1808 m0 = m0->m_act; 1809 } 1810} 1811 1812static void 1813unp_mark(struct file *fp) 1814{ 1815 if (fp->f_gcflag & FMARK) 1816 return; 1817 unp_defer++; 1818 fp->f_gcflag |= (FMARK|FDEFER); 1819} 1820 1821static void 1822unp_discard(struct file *fp) 1823{ 1824 UNP_LOCK(); 1825 FILE_LOCK(fp); 1826 fp->f_msgcount--; 1827 unp_rights--; 1828 FILE_UNLOCK(fp); 1829 UNP_UNLOCK(); 1830 (void) closef(fp, (struct thread *)NULL); 1831} 1832