uipc_usrreq.c revision 157927
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 157927 2006-04-21 09:25:40Z ps $"); 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/eventhandler.h> 44#include <sys/file.h> 45#include <sys/filedesc.h> 46#include <sys/jail.h> 47#include <sys/kernel.h> 48#include <sys/lock.h> 49#include <sys/mac.h> 50#include <sys/mbuf.h> 51#include <sys/mount.h> 52#include <sys/mutex.h> 53#include <sys/namei.h> 54#include <sys/proc.h> 55#include <sys/protosw.h> 56#include <sys/resourcevar.h> 57#include <sys/socket.h> 58#include <sys/socketvar.h> 59#include <sys/signalvar.h> 60#include <sys/stat.h> 61#include <sys/sx.h> 62#include <sys/sysctl.h> 63#include <sys/systm.h> 64#include <sys/taskqueue.h> 65#include <sys/un.h> 66#include <sys/unpcb.h> 67#include <sys/vnode.h> 68 69#include <vm/uma.h> 70 71static uma_zone_t unp_zone; 72static unp_gen_t unp_gencnt; 73static u_int unp_count; 74 75static struct unp_head unp_shead, unp_dhead; 76 77/* 78 * Unix communications domain. 79 * 80 * TODO: 81 * SEQPACKET, RDM 82 * rethink name space problems 83 * need a proper out-of-band 84 * lock pushdown 85 */ 86static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL }; 87static ino_t unp_ino; /* prototype for fake inode numbers */ 88struct mbuf *unp_addsockcred(struct thread *, struct mbuf *); 89 90/* 91 * Currently, UNIX domain sockets are protected by a single subsystem lock, 92 * which covers global data structures and variables, the contents of each 93 * per-socket unpcb structure, and the so_pcb field in sockets attached to 94 * the UNIX domain. This provides for a moderate degree of paralellism, as 95 * receive operations on UNIX domain sockets do not need to acquire the 96 * subsystem lock. Finer grained locking to permit send() without acquiring 97 * a global lock would be a logical next step. 98 * 99 * The UNIX domain socket lock preceds all socket layer locks, including the 100 * socket lock and socket buffer lock, permitting UNIX domain socket code to 101 * call into socket support routines without releasing its locks. 102 * 103 * Some caution is required in areas where the UNIX domain socket code enters 104 * VFS in order to create or find rendezvous points. This results in 105 * dropping of the UNIX domain socket subsystem lock, acquisition of the 106 * Giant lock, and potential sleeping. This increases the chances of races, 107 * and exposes weaknesses in the socket->protocol API by offering poor 108 * failure modes. 109 */ 110static struct mtx unp_mtx; 111#define UNP_LOCK_INIT() \ 112 mtx_init(&unp_mtx, "unp", NULL, MTX_DEF) 113#define UNP_LOCK() mtx_lock(&unp_mtx) 114#define UNP_UNLOCK() mtx_unlock(&unp_mtx) 115#define UNP_LOCK_ASSERT() mtx_assert(&unp_mtx, MA_OWNED) 116#define UNP_UNLOCK_ASSERT() mtx_assert(&unp_mtx, MA_NOTOWNED) 117 118/* 119 * Garbage collection of cyclic file descriptor/socket references occurs 120 * asynchronously in a taskqueue context in order to avoid recursion and 121 * reentrance in the UNIX domain socket, file descriptor, and socket layer 122 * code. See unp_gc() for a full description. 123 */ 124static struct task unp_gc_task; 125 126static int unp_attach(struct socket *); 127static void unp_detach(struct unpcb *); 128static int unp_bind(struct unpcb *,struct sockaddr *, struct thread *); 129static int unp_connect(struct socket *,struct sockaddr *, struct thread *); 130static int unp_connect2(struct socket *so, struct socket *so2, int); 131static void unp_disconnect(struct unpcb *); 132static void unp_shutdown(struct unpcb *); 133static void unp_drop(struct unpcb *, int); 134static void unp_gc(__unused void *, int); 135static void unp_scan(struct mbuf *, void (*)(struct file *)); 136static void unp_mark(struct file *); 137static void unp_discard(struct file *); 138static void unp_freerights(struct file **, int); 139static int unp_internalize(struct mbuf **, struct thread *); 140static int unp_listen(struct socket *, struct unpcb *, int, 141 struct thread *); 142 143static void 144uipc_abort(struct socket *so) 145{ 146 struct unpcb *unp; 147 148 unp = sotounpcb(so); 149 KASSERT(unp != NULL, ("uipc_abort: unp == NULL")); 150 UNP_LOCK(); 151 unp_drop(unp, ECONNABORTED); 152 unp_detach(unp); 153 UNP_UNLOCK_ASSERT(); 154} 155 156static int 157uipc_accept(struct socket *so, struct sockaddr **nam) 158{ 159 struct unpcb *unp; 160 const struct sockaddr *sa; 161 162 /* 163 * Pass back name of connected socket, 164 * if it was bound and we are still connected 165 * (our peer may have closed already!). 166 */ 167 unp = sotounpcb(so); 168 KASSERT(unp != NULL, ("uipc_accept: unp == NULL")); 169 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 170 UNP_LOCK(); 171 if (unp->unp_conn != NULL && unp->unp_conn->unp_addr != NULL) 172 sa = (struct sockaddr *) unp->unp_conn->unp_addr; 173 else 174 sa = &sun_noname; 175 bcopy(sa, *nam, sa->sa_len); 176 UNP_UNLOCK(); 177 return (0); 178} 179 180static int 181uipc_attach(struct socket *so, int proto, struct thread *td) 182{ 183 184 return (unp_attach(so)); 185} 186 187static int 188uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 189{ 190 struct unpcb *unp; 191 int error; 192 193 unp = sotounpcb(so); 194 KASSERT(unp != NULL, ("uipc_bind: unp == NULL")); 195 UNP_LOCK(); 196 error = unp_bind(unp, nam, td); 197 UNP_UNLOCK(); 198 return (error); 199} 200 201static int 202uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 203{ 204 struct unpcb *unp; 205 int error; 206 207 KASSERT(td == curthread, ("uipc_connect: td != curthread")); 208 unp = sotounpcb(so); 209 KASSERT(unp != NULL, ("uipc_connect: unp == NULL")); 210 UNP_LOCK(); 211 error = unp_connect(so, nam, td); 212 UNP_UNLOCK(); 213 return (error); 214} 215 216int 217uipc_connect2(struct socket *so1, struct socket *so2) 218{ 219 struct unpcb *unp; 220 int error; 221 222 unp = sotounpcb(so1); 223 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL")); 224 UNP_LOCK(); 225 error = unp_connect2(so1, so2, PRU_CONNECT2); 226 UNP_UNLOCK(); 227 return (error); 228} 229 230/* control is EOPNOTSUPP */ 231 232static void 233uipc_detach(struct socket *so) 234{ 235 struct unpcb *unp; 236 237 unp = sotounpcb(so); 238 KASSERT(unp != NULL, ("uipc_detach: unp == NULL")); 239 UNP_LOCK(); 240 unp_detach(unp); 241 UNP_UNLOCK_ASSERT(); 242} 243 244static int 245uipc_disconnect(struct socket *so) 246{ 247 struct unpcb *unp; 248 249 unp = sotounpcb(so); 250 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL")); 251 UNP_LOCK(); 252 unp_disconnect(unp); 253 UNP_UNLOCK(); 254 return (0); 255} 256 257static int 258uipc_listen(struct socket *so, int backlog, struct thread *td) 259{ 260 struct unpcb *unp; 261 int error; 262 263 unp = sotounpcb(so); 264 KASSERT(unp != NULL, ("uipc_listen: unp == NULL")); 265 UNP_LOCK(); 266 if (unp->unp_vnode == NULL) { 267 UNP_UNLOCK(); 268 return (EINVAL); 269 } 270 error = unp_listen(so, unp, backlog, td); 271 UNP_UNLOCK(); 272 return (error); 273} 274 275static int 276uipc_peeraddr(struct socket *so, struct sockaddr **nam) 277{ 278 struct unpcb *unp; 279 const struct sockaddr *sa; 280 281 unp = sotounpcb(so); 282 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL")); 283 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 284 UNP_LOCK(); 285 if (unp->unp_conn != NULL && unp->unp_conn->unp_addr!= NULL) 286 sa = (struct sockaddr *) unp->unp_conn->unp_addr; 287 else { 288 /* 289 * XXX: It seems that this test always fails even when 290 * connection is established. So, this else clause is 291 * added as workaround to return PF_LOCAL sockaddr. 292 */ 293 sa = &sun_noname; 294 } 295 bcopy(sa, *nam, sa->sa_len); 296 UNP_UNLOCK(); 297 return (0); 298} 299 300static int 301uipc_rcvd(struct socket *so, int flags) 302{ 303 struct unpcb *unp; 304 struct socket *so2; 305 u_long newhiwat; 306 307 unp = sotounpcb(so); 308 KASSERT(unp != NULL, ("uipc_rcvd: unp == NULL")); 309 UNP_LOCK(); 310 switch (so->so_type) { 311 case SOCK_DGRAM: 312 panic("uipc_rcvd DGRAM?"); 313 /*NOTREACHED*/ 314 315 case SOCK_STREAM: 316 if (unp->unp_conn == NULL) 317 break; 318 so2 = unp->unp_conn->unp_socket; 319 SOCKBUF_LOCK(&so2->so_snd); 320 SOCKBUF_LOCK(&so->so_rcv); 321 /* 322 * Adjust backpressure on sender 323 * and wakeup any waiting to write. 324 */ 325 so2->so_snd.sb_mbmax += unp->unp_mbcnt - so->so_rcv.sb_mbcnt; 326 unp->unp_mbcnt = so->so_rcv.sb_mbcnt; 327 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - 328 so->so_rcv.sb_cc; 329 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat, 330 newhiwat, RLIM_INFINITY); 331 unp->unp_cc = so->so_rcv.sb_cc; 332 SOCKBUF_UNLOCK(&so->so_rcv); 333 sowwakeup_locked(so2); 334 break; 335 336 default: 337 panic("uipc_rcvd unknown socktype"); 338 } 339 UNP_UNLOCK(); 340 return (0); 341} 342 343/* pru_rcvoob is EOPNOTSUPP */ 344 345static int 346uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 347 struct mbuf *control, struct thread *td) 348{ 349 int error = 0; 350 struct unpcb *unp; 351 struct socket *so2; 352 u_long newhiwat; 353 354 unp = sotounpcb(so); 355 KASSERT(unp != NULL, ("uipc_send: unp == NULL")); 356 if (flags & PRUS_OOB) { 357 error = EOPNOTSUPP; 358 goto release; 359 } 360 361 if (control != NULL && (error = unp_internalize(&control, td))) 362 goto release; 363 364 UNP_LOCK(); 365 switch (so->so_type) { 366 case SOCK_DGRAM: 367 { 368 const struct sockaddr *from; 369 370 if (nam != NULL) { 371 if (unp->unp_conn != NULL) { 372 error = EISCONN; 373 break; 374 } 375 error = unp_connect(so, nam, td); 376 if (error) 377 break; 378 } else { 379 if (unp->unp_conn == NULL) { 380 error = ENOTCONN; 381 break; 382 } 383 } 384 so2 = unp->unp_conn->unp_socket; 385 if (unp->unp_addr != NULL) 386 from = (struct sockaddr *)unp->unp_addr; 387 else 388 from = &sun_noname; 389 if (unp->unp_conn->unp_flags & UNP_WANTCRED) 390 control = unp_addsockcred(td, control); 391 SOCKBUF_LOCK(&so2->so_rcv); 392 if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) { 393 sorwakeup_locked(so2); 394 m = NULL; 395 control = NULL; 396 } else { 397 SOCKBUF_UNLOCK(&so2->so_rcv); 398 error = ENOBUFS; 399 } 400 if (nam != NULL) 401 unp_disconnect(unp); 402 break; 403 } 404 405 case SOCK_STREAM: 406 /* Connect if not connected yet. */ 407 /* 408 * Note: A better implementation would complain 409 * if not equal to the peer's address. 410 */ 411 if ((so->so_state & SS_ISCONNECTED) == 0) { 412 if (nam != NULL) { 413 error = unp_connect(so, nam, td); 414 if (error) 415 break; /* XXX */ 416 } else { 417 error = ENOTCONN; 418 break; 419 } 420 } 421 422 SOCKBUF_LOCK(&so->so_snd); 423 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 424 SOCKBUF_UNLOCK(&so->so_snd); 425 error = EPIPE; 426 break; 427 } 428 if (unp->unp_conn == NULL) 429 panic("uipc_send connected but no connection?"); 430 so2 = unp->unp_conn->unp_socket; 431 SOCKBUF_LOCK(&so2->so_rcv); 432 if (unp->unp_conn->unp_flags & UNP_WANTCRED) { 433 /* 434 * Credentials are passed only once on 435 * SOCK_STREAM. 436 */ 437 unp->unp_conn->unp_flags &= ~UNP_WANTCRED; 438 control = unp_addsockcred(td, control); 439 } 440 /* 441 * Send to paired receive port, and then reduce 442 * send buffer hiwater marks to maintain backpressure. 443 * Wake up readers. 444 */ 445 if (control != NULL) { 446 if (sbappendcontrol_locked(&so2->so_rcv, m, control)) 447 control = NULL; 448 } else { 449 sbappend_locked(&so2->so_rcv, m); 450 } 451 so->so_snd.sb_mbmax -= 452 so2->so_rcv.sb_mbcnt - unp->unp_conn->unp_mbcnt; 453 unp->unp_conn->unp_mbcnt = so2->so_rcv.sb_mbcnt; 454 newhiwat = so->so_snd.sb_hiwat - 455 (so2->so_rcv.sb_cc - unp->unp_conn->unp_cc); 456 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat, 457 newhiwat, RLIM_INFINITY); 458 SOCKBUF_UNLOCK(&so->so_snd); 459 unp->unp_conn->unp_cc = so2->so_rcv.sb_cc; 460 sorwakeup_locked(so2); 461 m = NULL; 462 break; 463 464 default: 465 panic("uipc_send unknown socktype"); 466 } 467 468 /* 469 * SEND_EOF is equivalent to a SEND followed by 470 * a SHUTDOWN. 471 */ 472 if (flags & PRUS_EOF) { 473 socantsendmore(so); 474 unp_shutdown(unp); 475 } 476 UNP_UNLOCK(); 477 478 if (control != NULL && error != 0) 479 unp_dispose(control); 480 481release: 482 if (control != NULL) 483 m_freem(control); 484 if (m != NULL) 485 m_freem(m); 486 return (error); 487} 488 489static int 490uipc_sense(struct socket *so, struct stat *sb) 491{ 492 struct unpcb *unp; 493 struct socket *so2; 494 495 unp = sotounpcb(so); 496 KASSERT(unp != NULL, ("uipc_sense: unp == NULL")); 497 UNP_LOCK(); 498 sb->st_blksize = so->so_snd.sb_hiwat; 499 if (so->so_type == SOCK_STREAM && unp->unp_conn != NULL) { 500 so2 = unp->unp_conn->unp_socket; 501 sb->st_blksize += so2->so_rcv.sb_cc; 502 } 503 sb->st_dev = NODEV; 504 if (unp->unp_ino == 0) 505 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino; 506 sb->st_ino = unp->unp_ino; 507 UNP_UNLOCK(); 508 return (0); 509} 510 511static int 512uipc_shutdown(struct socket *so) 513{ 514 struct unpcb *unp; 515 516 unp = sotounpcb(so); 517 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL")); 518 UNP_LOCK(); 519 socantsendmore(so); 520 unp_shutdown(unp); 521 UNP_UNLOCK(); 522 return (0); 523} 524 525static int 526uipc_sockaddr(struct socket *so, struct sockaddr **nam) 527{ 528 struct unpcb *unp; 529 const struct sockaddr *sa; 530 531 unp = sotounpcb(so); 532 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL")); 533 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 534 UNP_LOCK(); 535 if (unp->unp_addr != NULL) 536 sa = (struct sockaddr *) unp->unp_addr; 537 else 538 sa = &sun_noname; 539 bcopy(sa, *nam, sa->sa_len); 540 UNP_UNLOCK(); 541 return (0); 542} 543 544struct pr_usrreqs uipc_usrreqs = { 545 .pru_abort = uipc_abort, 546 .pru_accept = uipc_accept, 547 .pru_attach = uipc_attach, 548 .pru_bind = uipc_bind, 549 .pru_connect = uipc_connect, 550 .pru_connect2 = uipc_connect2, 551 .pru_detach = uipc_detach, 552 .pru_disconnect = uipc_disconnect, 553 .pru_listen = uipc_listen, 554 .pru_peeraddr = uipc_peeraddr, 555 .pru_rcvd = uipc_rcvd, 556 .pru_send = uipc_send, 557 .pru_sense = uipc_sense, 558 .pru_shutdown = uipc_shutdown, 559 .pru_sockaddr = uipc_sockaddr, 560 .pru_sosend = sosend, 561 .pru_soreceive = soreceive, 562 .pru_sopoll = sopoll, 563}; 564 565int 566uipc_ctloutput(struct socket *so, struct sockopt *sopt) 567{ 568 struct unpcb *unp; 569 struct xucred xu; 570 int error, optval; 571 572 if (sopt->sopt_level != 0) 573 return (EINVAL); 574 575 unp = sotounpcb(so); 576 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL")); 577 UNP_LOCK(); 578 error = 0; 579 switch (sopt->sopt_dir) { 580 case SOPT_GET: 581 switch (sopt->sopt_name) { 582 case LOCAL_PEERCRED: 583 if (unp->unp_flags & UNP_HAVEPC) 584 xu = unp->unp_peercred; 585 else { 586 if (so->so_type == SOCK_STREAM) 587 error = ENOTCONN; 588 else 589 error = EINVAL; 590 } 591 if (error == 0) 592 error = sooptcopyout(sopt, &xu, sizeof(xu)); 593 break; 594 case LOCAL_CREDS: 595 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0; 596 error = sooptcopyout(sopt, &optval, sizeof(optval)); 597 break; 598 case LOCAL_CONNWAIT: 599 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0; 600 error = sooptcopyout(sopt, &optval, sizeof(optval)); 601 break; 602 default: 603 error = EOPNOTSUPP; 604 break; 605 } 606 break; 607 case SOPT_SET: 608 switch (sopt->sopt_name) { 609 case LOCAL_CREDS: 610 case LOCAL_CONNWAIT: 611 error = sooptcopyin(sopt, &optval, sizeof(optval), 612 sizeof(optval)); 613 if (error) 614 break; 615 616#define OPTSET(bit) \ 617 if (optval) \ 618 unp->unp_flags |= bit; \ 619 else \ 620 unp->unp_flags &= ~bit; 621 622 switch (sopt->sopt_name) { 623 case LOCAL_CREDS: 624 OPTSET(UNP_WANTCRED); 625 break; 626 case LOCAL_CONNWAIT: 627 OPTSET(UNP_CONNWAIT); 628 break; 629 default: 630 break; 631 } 632 break; 633#undef OPTSET 634 default: 635 error = ENOPROTOOPT; 636 break; 637 } 638 break; 639 default: 640 error = EOPNOTSUPP; 641 break; 642 } 643 UNP_UNLOCK(); 644 return (error); 645} 646 647/* 648 * Both send and receive buffers are allocated PIPSIZ bytes of buffering 649 * for stream sockets, although the total for sender and receiver is 650 * actually only PIPSIZ. 651 * Datagram sockets really use the sendspace as the maximum datagram size, 652 * and don't really want to reserve the sendspace. Their recvspace should 653 * be large enough for at least one max-size datagram plus address. 654 */ 655#ifndef PIPSIZ 656#define PIPSIZ 8192 657#endif 658static u_long unpst_sendspace = PIPSIZ; 659static u_long unpst_recvspace = PIPSIZ; 660static u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 661static u_long unpdg_recvspace = 4*1024; 662 663static int unp_rights; /* file descriptors in flight */ 664 665SYSCTL_DECL(_net_local_stream); 666SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 667 &unpst_sendspace, 0, ""); 668SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 669 &unpst_recvspace, 0, ""); 670SYSCTL_DECL(_net_local_dgram); 671SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 672 &unpdg_sendspace, 0, ""); 673SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 674 &unpdg_recvspace, 0, ""); 675SYSCTL_DECL(_net_local); 676SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, ""); 677 678static int 679unp_attach(struct socket *so) 680{ 681 struct unpcb *unp; 682 int error; 683 684 KASSERT(so->so_pcb == NULL, ("unp_attach: so_pcb != NULL")); 685 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 686 switch (so->so_type) { 687 688 case SOCK_STREAM: 689 error = soreserve(so, unpst_sendspace, unpst_recvspace); 690 break; 691 692 case SOCK_DGRAM: 693 error = soreserve(so, unpdg_sendspace, unpdg_recvspace); 694 break; 695 696 default: 697 panic("unp_attach"); 698 } 699 if (error) 700 return (error); 701 } 702 unp = uma_zalloc(unp_zone, M_WAITOK | M_ZERO); 703 if (unp == NULL) 704 return (ENOBUFS); 705 LIST_INIT(&unp->unp_refs); 706 unp->unp_socket = so; 707 so->so_pcb = unp; 708 709 UNP_LOCK(); 710 unp->unp_gencnt = ++unp_gencnt; 711 unp_count++; 712 LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead 713 : &unp_shead, unp, unp_link); 714 UNP_UNLOCK(); 715 716 return (0); 717} 718 719static void 720unp_detach(struct unpcb *unp) 721{ 722 struct vnode *vp; 723 int local_unp_rights; 724 725 UNP_LOCK_ASSERT(); 726 727 LIST_REMOVE(unp, unp_link); 728 unp->unp_gencnt = ++unp_gencnt; 729 --unp_count; 730 if ((vp = unp->unp_vnode) != NULL) { 731 /* 732 * XXXRW: should v_socket be frobbed only while holding 733 * Giant? 734 */ 735 unp->unp_vnode->v_socket = NULL; 736 unp->unp_vnode = NULL; 737 } 738 if (unp->unp_conn != NULL) 739 unp_disconnect(unp); 740 while (!LIST_EMPTY(&unp->unp_refs)) { 741 struct unpcb *ref = LIST_FIRST(&unp->unp_refs); 742 unp_drop(ref, ECONNRESET); 743 } 744 soisdisconnected(unp->unp_socket); 745 unp->unp_socket->so_pcb = NULL; 746 local_unp_rights = unp_rights; 747 UNP_UNLOCK(); 748 if (unp->unp_addr != NULL) 749 FREE(unp->unp_addr, M_SONAME); 750 uma_zfree(unp_zone, unp); 751 if (vp) { 752 int vfslocked; 753 754 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 755 vrele(vp); 756 VFS_UNLOCK_GIANT(vfslocked); 757 } 758 if (local_unp_rights) 759 taskqueue_enqueue(taskqueue_thread, &unp_gc_task); 760} 761 762static int 763unp_bind(struct unpcb *unp, struct sockaddr *nam, struct thread *td) 764{ 765 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 766 struct vnode *vp; 767 struct mount *mp; 768 struct vattr vattr; 769 int error, namelen; 770 struct nameidata nd; 771 char *buf; 772 773 UNP_LOCK_ASSERT(); 774 775 /* 776 * XXXRW: This test-and-set of unp_vnode is non-atomic; the 777 * unlocked read here is fine, but the value of unp_vnode needs 778 * to be tested again after we do all the lookups to see if the 779 * pcb is still unbound? 780 */ 781 if (unp->unp_vnode != NULL) 782 return (EINVAL); 783 784 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 785 if (namelen <= 0) 786 return (EINVAL); 787 788 UNP_UNLOCK(); 789 790 buf = malloc(namelen + 1, M_TEMP, M_WAITOK); 791 strlcpy(buf, soun->sun_path, namelen + 1); 792 793 mtx_lock(&Giant); 794restart: 795 mtx_assert(&Giant, MA_OWNED); 796 NDINIT(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME, UIO_SYSSPACE, 797 buf, td); 798/* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 799 error = namei(&nd); 800 if (error) 801 goto done; 802 vp = nd.ni_vp; 803 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { 804 NDFREE(&nd, NDF_ONLY_PNBUF); 805 if (nd.ni_dvp == vp) 806 vrele(nd.ni_dvp); 807 else 808 vput(nd.ni_dvp); 809 if (vp != NULL) { 810 vrele(vp); 811 error = EADDRINUSE; 812 goto done; 813 } 814 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH); 815 if (error) 816 goto done; 817 goto restart; 818 } 819 VATTR_NULL(&vattr); 820 vattr.va_type = VSOCK; 821 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask); 822#ifdef MAC 823 error = mac_check_vnode_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, 824 &vattr); 825#endif 826 if (error == 0) { 827 VOP_LEASE(nd.ni_dvp, td, td->td_ucred, LEASE_WRITE); 828 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 829 } 830 NDFREE(&nd, NDF_ONLY_PNBUF); 831 vput(nd.ni_dvp); 832 if (error) { 833 vn_finished_write(mp); 834 goto done; 835 } 836 vp = nd.ni_vp; 837 ASSERT_VOP_LOCKED(vp, "unp_bind"); 838 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK); 839 UNP_LOCK(); 840 vp->v_socket = unp->unp_socket; 841 unp->unp_vnode = vp; 842 unp->unp_addr = soun; 843 UNP_UNLOCK(); 844 VOP_UNLOCK(vp, 0, td); 845 vn_finished_write(mp); 846done: 847 mtx_unlock(&Giant); 848 free(buf, M_TEMP); 849 UNP_LOCK(); 850 return (error); 851} 852 853static int 854unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 855{ 856 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 857 struct vnode *vp; 858 struct socket *so2, *so3; 859 struct unpcb *unp, *unp2, *unp3; 860 int error, len; 861 struct nameidata nd; 862 char buf[SOCK_MAXADDRLEN]; 863 struct sockaddr *sa; 864 865 UNP_LOCK_ASSERT(); 866 867 unp = sotounpcb(so); 868 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 869 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 870 if (len <= 0) 871 return (EINVAL); 872 strlcpy(buf, soun->sun_path, len + 1); 873 UNP_UNLOCK(); 874 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 875 mtx_lock(&Giant); 876 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, td); 877 error = namei(&nd); 878 if (error) 879 vp = NULL; 880 else 881 vp = nd.ni_vp; 882 ASSERT_VOP_LOCKED(vp, "unp_connect"); 883 NDFREE(&nd, NDF_ONLY_PNBUF); 884 if (error) 885 goto bad; 886 887 if (vp->v_type != VSOCK) { 888 error = ENOTSOCK; 889 goto bad; 890 } 891 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); 892 if (error) 893 goto bad; 894 mtx_unlock(&Giant); 895 UNP_LOCK(); 896 unp = sotounpcb(so); 897 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 898 so2 = vp->v_socket; 899 if (so2 == NULL) { 900 error = ECONNREFUSED; 901 goto bad2; 902 } 903 if (so->so_type != so2->so_type) { 904 error = EPROTOTYPE; 905 goto bad2; 906 } 907 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 908 if (so2->so_options & SO_ACCEPTCONN) { 909 /* 910 * NB: drop locks here so unp_attach is entered 911 * w/o locks; this avoids a recursive lock 912 * of the head and holding sleep locks across 913 * a (potentially) blocking malloc. 914 */ 915 UNP_UNLOCK(); 916 so3 = sonewconn(so2, 0); 917 UNP_LOCK(); 918 } else 919 so3 = NULL; 920 if (so3 == NULL) { 921 error = ECONNREFUSED; 922 goto bad2; 923 } 924 unp = sotounpcb(so); 925 unp2 = sotounpcb(so2); 926 unp3 = sotounpcb(so3); 927 if (unp2->unp_addr != NULL) { 928 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); 929 unp3->unp_addr = (struct sockaddr_un *) sa; 930 sa = NULL; 931 } 932 /* 933 * unp_peercred management: 934 * 935 * The connecter's (client's) credentials are copied 936 * from its process structure at the time of connect() 937 * (which is now). 938 */ 939 cru2x(td->td_ucred, &unp3->unp_peercred); 940 unp3->unp_flags |= UNP_HAVEPC; 941 /* 942 * The receiver's (server's) credentials are copied 943 * from the unp_peercred member of socket on which the 944 * former called listen(); unp_listen() cached that 945 * process's credentials at that time so we can use 946 * them now. 947 */ 948 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED, 949 ("unp_connect: listener without cached peercred")); 950 memcpy(&unp->unp_peercred, &unp2->unp_peercred, 951 sizeof(unp->unp_peercred)); 952 unp->unp_flags |= UNP_HAVEPC; 953#ifdef MAC 954 SOCK_LOCK(so); 955 mac_set_socket_peer_from_socket(so, so3); 956 mac_set_socket_peer_from_socket(so3, so); 957 SOCK_UNLOCK(so); 958#endif 959 960 so2 = so3; 961 } 962 error = unp_connect2(so, so2, PRU_CONNECT); 963bad2: 964 UNP_UNLOCK(); 965 mtx_lock(&Giant); 966bad: 967 mtx_assert(&Giant, MA_OWNED); 968 if (vp != NULL) 969 vput(vp); 970 mtx_unlock(&Giant); 971 free(sa, M_SONAME); 972 UNP_LOCK(); 973 return (error); 974} 975 976static int 977unp_connect2(struct socket *so, struct socket *so2, int req) 978{ 979 struct unpcb *unp = sotounpcb(so); 980 struct unpcb *unp2; 981 982 UNP_LOCK_ASSERT(); 983 984 if (so2->so_type != so->so_type) 985 return (EPROTOTYPE); 986 unp2 = sotounpcb(so2); 987 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL")); 988 unp->unp_conn = unp2; 989 switch (so->so_type) { 990 991 case SOCK_DGRAM: 992 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 993 soisconnected(so); 994 break; 995 996 case SOCK_STREAM: 997 unp2->unp_conn = unp; 998 if (req == PRU_CONNECT && 999 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 1000 soisconnecting(so); 1001 else 1002 soisconnected(so); 1003 soisconnected(so2); 1004 break; 1005 1006 default: 1007 panic("unp_connect2"); 1008 } 1009 return (0); 1010} 1011 1012static void 1013unp_disconnect(struct unpcb *unp) 1014{ 1015 struct unpcb *unp2 = unp->unp_conn; 1016 struct socket *so; 1017 1018 UNP_LOCK_ASSERT(); 1019 1020 if (unp2 == NULL) 1021 return; 1022 unp->unp_conn = NULL; 1023 switch (unp->unp_socket->so_type) { 1024 case SOCK_DGRAM: 1025 LIST_REMOVE(unp, unp_reflink); 1026 so = unp->unp_socket; 1027 SOCK_LOCK(so); 1028 so->so_state &= ~SS_ISCONNECTED; 1029 SOCK_UNLOCK(so); 1030 break; 1031 1032 case SOCK_STREAM: 1033 soisdisconnected(unp->unp_socket); 1034 unp2->unp_conn = NULL; 1035 soisdisconnected(unp2->unp_socket); 1036 break; 1037 } 1038} 1039 1040#ifdef notdef 1041void 1042unp_abort(struct unpcb *unp) 1043{ 1044 1045 unp_detach(unp); 1046 UNP_UNLOCK_ASSERT(); 1047} 1048#endif 1049 1050/* 1051 * unp_pcblist() assumes that UNIX domain socket memory is never reclaimed 1052 * by the zone (UMA_ZONE_NOFREE), and as such potentially stale pointers 1053 * are safe to reference. It first scans the list of struct unpcb's to 1054 * generate a pointer list, then it rescans its list one entry at a time to 1055 * externalize and copyout. It checks the generation number to see if a 1056 * struct unpcb has been reused, and will skip it if so. 1057 */ 1058static int 1059unp_pcblist(SYSCTL_HANDLER_ARGS) 1060{ 1061 int error, i, n; 1062 struct unpcb *unp, **unp_list; 1063 unp_gen_t gencnt; 1064 struct xunpgen *xug; 1065 struct unp_head *head; 1066 struct xunpcb *xu; 1067 1068 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead); 1069 1070 /* 1071 * The process of preparing the PCB list is too time-consuming and 1072 * resource-intensive to repeat twice on every request. 1073 */ 1074 if (req->oldptr == NULL) { 1075 n = unp_count; 1076 req->oldidx = 2 * (sizeof *xug) 1077 + (n + n/8) * sizeof(struct xunpcb); 1078 return (0); 1079 } 1080 1081 if (req->newptr != NULL) 1082 return (EPERM); 1083 1084 /* 1085 * OK, now we're committed to doing something. 1086 */ 1087 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK); 1088 UNP_LOCK(); 1089 gencnt = unp_gencnt; 1090 n = unp_count; 1091 UNP_UNLOCK(); 1092 1093 xug->xug_len = sizeof *xug; 1094 xug->xug_count = n; 1095 xug->xug_gen = gencnt; 1096 xug->xug_sogen = so_gencnt; 1097 error = SYSCTL_OUT(req, xug, sizeof *xug); 1098 if (error) { 1099 free(xug, M_TEMP); 1100 return (error); 1101 } 1102 1103 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 1104 1105 UNP_LOCK(); 1106 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 1107 unp = LIST_NEXT(unp, unp_link)) { 1108 if (unp->unp_gencnt <= gencnt) { 1109 if (cr_cansee(req->td->td_ucred, 1110 unp->unp_socket->so_cred)) 1111 continue; 1112 unp_list[i++] = unp; 1113 } 1114 } 1115 UNP_UNLOCK(); 1116 n = i; /* in case we lost some during malloc */ 1117 1118 error = 0; 1119 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO); 1120 for (i = 0; i < n; i++) { 1121 unp = unp_list[i]; 1122 if (unp->unp_gencnt <= gencnt) { 1123 xu->xu_len = sizeof *xu; 1124 xu->xu_unpp = unp; 1125 /* 1126 * XXX - need more locking here to protect against 1127 * connect/disconnect races for SMP. 1128 */ 1129 if (unp->unp_addr != NULL) 1130 bcopy(unp->unp_addr, &xu->xu_addr, 1131 unp->unp_addr->sun_len); 1132 if (unp->unp_conn != NULL && 1133 unp->unp_conn->unp_addr != NULL) 1134 bcopy(unp->unp_conn->unp_addr, 1135 &xu->xu_caddr, 1136 unp->unp_conn->unp_addr->sun_len); 1137 bcopy(unp, &xu->xu_unp, sizeof *unp); 1138 sotoxsocket(unp->unp_socket, &xu->xu_socket); 1139 error = SYSCTL_OUT(req, xu, sizeof *xu); 1140 } 1141 } 1142 free(xu, M_TEMP); 1143 if (!error) { 1144 /* 1145 * Give the user an updated idea of our state. 1146 * If the generation differs from what we told 1147 * her before, she knows that something happened 1148 * while we were processing this request, and it 1149 * might be necessary to retry. 1150 */ 1151 xug->xug_gen = unp_gencnt; 1152 xug->xug_sogen = so_gencnt; 1153 xug->xug_count = unp_count; 1154 error = SYSCTL_OUT(req, xug, sizeof *xug); 1155 } 1156 free(unp_list, M_TEMP); 1157 free(xug, M_TEMP); 1158 return (error); 1159} 1160 1161SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD, 1162 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 1163 "List of active local datagram sockets"); 1164SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD, 1165 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 1166 "List of active local stream sockets"); 1167 1168static void 1169unp_shutdown(struct unpcb *unp) 1170{ 1171 struct socket *so; 1172 1173 UNP_LOCK_ASSERT(); 1174 1175 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn && 1176 (so = unp->unp_conn->unp_socket)) 1177 socantrcvmore(so); 1178} 1179 1180static void 1181unp_drop(struct unpcb *unp, int errno) 1182{ 1183 struct socket *so = unp->unp_socket; 1184 1185 UNP_LOCK_ASSERT(); 1186 1187 so->so_error = errno; 1188 unp_disconnect(unp); 1189} 1190 1191#ifdef notdef 1192void 1193unp_drain(void) 1194{ 1195 1196} 1197#endif 1198 1199static void 1200unp_freerights(struct file **rp, int fdcount) 1201{ 1202 int i; 1203 struct file *fp; 1204 1205 for (i = 0; i < fdcount; i++) { 1206 fp = *rp; 1207 /* 1208 * zero the pointer before calling 1209 * unp_discard since it may end up 1210 * in unp_gc().. 1211 * 1212 * XXXRW: This is less true than it used to be. 1213 */ 1214 *rp++ = 0; 1215 unp_discard(fp); 1216 } 1217} 1218 1219int 1220unp_externalize(struct mbuf *control, struct mbuf **controlp) 1221{ 1222 struct thread *td = curthread; /* XXX */ 1223 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1224 int i; 1225 int *fdp; 1226 struct file **rp; 1227 struct file *fp; 1228 void *data; 1229 socklen_t clen = control->m_len, datalen; 1230 int error, newfds; 1231 int f; 1232 u_int newlen; 1233 1234 UNP_UNLOCK_ASSERT(); 1235 1236 error = 0; 1237 if (controlp != NULL) /* controlp == NULL => free control messages */ 1238 *controlp = NULL; 1239 1240 while (cm != NULL) { 1241 if (sizeof(*cm) > clen || cm->cmsg_len > clen) { 1242 error = EINVAL; 1243 break; 1244 } 1245 1246 data = CMSG_DATA(cm); 1247 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1248 1249 if (cm->cmsg_level == SOL_SOCKET 1250 && cm->cmsg_type == SCM_RIGHTS) { 1251 newfds = datalen / sizeof(struct file *); 1252 rp = data; 1253 1254 /* If we're not outputting the descriptors free them. */ 1255 if (error || controlp == NULL) { 1256 unp_freerights(rp, newfds); 1257 goto next; 1258 } 1259 FILEDESC_LOCK(td->td_proc->p_fd); 1260 /* if the new FD's will not fit free them. */ 1261 if (!fdavail(td, newfds)) { 1262 FILEDESC_UNLOCK(td->td_proc->p_fd); 1263 error = EMSGSIZE; 1264 unp_freerights(rp, newfds); 1265 goto next; 1266 } 1267 /* 1268 * now change each pointer to an fd in the global 1269 * table to an integer that is the index to the 1270 * local fd table entry that we set up to point 1271 * to the global one we are transferring. 1272 */ 1273 newlen = newfds * sizeof(int); 1274 *controlp = sbcreatecontrol(NULL, newlen, 1275 SCM_RIGHTS, SOL_SOCKET); 1276 if (*controlp == NULL) { 1277 FILEDESC_UNLOCK(td->td_proc->p_fd); 1278 error = E2BIG; 1279 unp_freerights(rp, newfds); 1280 goto next; 1281 } 1282 1283 fdp = (int *) 1284 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1285 for (i = 0; i < newfds; i++) { 1286 if (fdalloc(td, 0, &f)) 1287 panic("unp_externalize fdalloc failed"); 1288 fp = *rp++; 1289 td->td_proc->p_fd->fd_ofiles[f] = fp; 1290 FILE_LOCK(fp); 1291 fp->f_msgcount--; 1292 FILE_UNLOCK(fp); 1293 unp_rights--; 1294 *fdp++ = f; 1295 } 1296 FILEDESC_UNLOCK(td->td_proc->p_fd); 1297 } else { /* We can just copy anything else across */ 1298 if (error || controlp == NULL) 1299 goto next; 1300 *controlp = sbcreatecontrol(NULL, datalen, 1301 cm->cmsg_type, cm->cmsg_level); 1302 if (*controlp == NULL) { 1303 error = ENOBUFS; 1304 goto next; 1305 } 1306 bcopy(data, 1307 CMSG_DATA(mtod(*controlp, struct cmsghdr *)), 1308 datalen); 1309 } 1310 1311 controlp = &(*controlp)->m_next; 1312 1313next: 1314 if (CMSG_SPACE(datalen) < clen) { 1315 clen -= CMSG_SPACE(datalen); 1316 cm = (struct cmsghdr *) 1317 ((caddr_t)cm + CMSG_SPACE(datalen)); 1318 } else { 1319 clen = 0; 1320 cm = NULL; 1321 } 1322 } 1323 1324 m_freem(control); 1325 1326 return (error); 1327} 1328 1329static void 1330unp_zone_change(void *tag) 1331{ 1332 1333 uma_zone_set_max(unp_zone, maxsockets); 1334} 1335 1336void 1337unp_init(void) 1338{ 1339 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL, 1340 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 1341 if (unp_zone == NULL) 1342 panic("unp_init"); 1343 uma_zone_set_max(unp_zone, maxsockets); 1344 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change, 1345 NULL, EVENTHANDLER_PRI_ANY); 1346 LIST_INIT(&unp_dhead); 1347 LIST_INIT(&unp_shead); 1348 TASK_INIT(&unp_gc_task, 0, unp_gc, NULL); 1349 UNP_LOCK_INIT(); 1350} 1351 1352static int 1353unp_internalize(struct mbuf **controlp, struct thread *td) 1354{ 1355 struct mbuf *control = *controlp; 1356 struct proc *p = td->td_proc; 1357 struct filedesc *fdescp = p->p_fd; 1358 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1359 struct cmsgcred *cmcred; 1360 struct file **rp; 1361 struct file *fp; 1362 struct timeval *tv; 1363 int i, fd, *fdp; 1364 void *data; 1365 socklen_t clen = control->m_len, datalen; 1366 int error, oldfds; 1367 u_int newlen; 1368 1369 UNP_UNLOCK_ASSERT(); 1370 1371 error = 0; 1372 *controlp = NULL; 1373 1374 while (cm != NULL) { 1375 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET 1376 || cm->cmsg_len > clen) { 1377 error = EINVAL; 1378 goto out; 1379 } 1380 1381 data = CMSG_DATA(cm); 1382 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1383 1384 switch (cm->cmsg_type) { 1385 /* 1386 * Fill in credential information. 1387 */ 1388 case SCM_CREDS: 1389 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred), 1390 SCM_CREDS, SOL_SOCKET); 1391 if (*controlp == NULL) { 1392 error = ENOBUFS; 1393 goto out; 1394 } 1395 1396 cmcred = (struct cmsgcred *) 1397 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1398 cmcred->cmcred_pid = p->p_pid; 1399 cmcred->cmcred_uid = td->td_ucred->cr_ruid; 1400 cmcred->cmcred_gid = td->td_ucred->cr_rgid; 1401 cmcred->cmcred_euid = td->td_ucred->cr_uid; 1402 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups, 1403 CMGROUP_MAX); 1404 for (i = 0; i < cmcred->cmcred_ngroups; i++) 1405 cmcred->cmcred_groups[i] = 1406 td->td_ucred->cr_groups[i]; 1407 break; 1408 1409 case SCM_RIGHTS: 1410 oldfds = datalen / sizeof (int); 1411 /* 1412 * check that all the FDs passed in refer to legal files 1413 * If not, reject the entire operation. 1414 */ 1415 fdp = data; 1416 FILEDESC_LOCK(fdescp); 1417 for (i = 0; i < oldfds; i++) { 1418 fd = *fdp++; 1419 if ((unsigned)fd >= fdescp->fd_nfiles || 1420 fdescp->fd_ofiles[fd] == NULL) { 1421 FILEDESC_UNLOCK(fdescp); 1422 error = EBADF; 1423 goto out; 1424 } 1425 fp = fdescp->fd_ofiles[fd]; 1426 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { 1427 FILEDESC_UNLOCK(fdescp); 1428 error = EOPNOTSUPP; 1429 goto out; 1430 } 1431 1432 } 1433 /* 1434 * Now replace the integer FDs with pointers to 1435 * the associated global file table entry.. 1436 */ 1437 newlen = oldfds * sizeof(struct file *); 1438 *controlp = sbcreatecontrol(NULL, newlen, 1439 SCM_RIGHTS, SOL_SOCKET); 1440 if (*controlp == NULL) { 1441 FILEDESC_UNLOCK(fdescp); 1442 error = E2BIG; 1443 goto out; 1444 } 1445 1446 fdp = data; 1447 rp = (struct file **) 1448 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1449 for (i = 0; i < oldfds; i++) { 1450 fp = fdescp->fd_ofiles[*fdp++]; 1451 *rp++ = fp; 1452 FILE_LOCK(fp); 1453 fp->f_count++; 1454 fp->f_msgcount++; 1455 FILE_UNLOCK(fp); 1456 unp_rights++; 1457 } 1458 FILEDESC_UNLOCK(fdescp); 1459 break; 1460 1461 case SCM_TIMESTAMP: 1462 *controlp = sbcreatecontrol(NULL, sizeof(*tv), 1463 SCM_TIMESTAMP, SOL_SOCKET); 1464 if (*controlp == NULL) { 1465 error = ENOBUFS; 1466 goto out; 1467 } 1468 tv = (struct timeval *) 1469 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1470 microtime(tv); 1471 break; 1472 1473 default: 1474 error = EINVAL; 1475 goto out; 1476 } 1477 1478 controlp = &(*controlp)->m_next; 1479 1480 if (CMSG_SPACE(datalen) < clen) { 1481 clen -= CMSG_SPACE(datalen); 1482 cm = (struct cmsghdr *) 1483 ((caddr_t)cm + CMSG_SPACE(datalen)); 1484 } else { 1485 clen = 0; 1486 cm = NULL; 1487 } 1488 } 1489 1490out: 1491 m_freem(control); 1492 1493 return (error); 1494} 1495 1496struct mbuf * 1497unp_addsockcred(struct thread *td, struct mbuf *control) 1498{ 1499 struct mbuf *m, *n; 1500 struct sockcred *sc; 1501 int ngroups; 1502 int i; 1503 1504 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); 1505 1506 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET); 1507 if (m == NULL) 1508 return (control); 1509 m->m_next = NULL; 1510 1511 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *)); 1512 sc->sc_uid = td->td_ucred->cr_ruid; 1513 sc->sc_euid = td->td_ucred->cr_uid; 1514 sc->sc_gid = td->td_ucred->cr_rgid; 1515 sc->sc_egid = td->td_ucred->cr_gid; 1516 sc->sc_ngroups = ngroups; 1517 for (i = 0; i < sc->sc_ngroups; i++) 1518 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 1519 1520 /* 1521 * If a control message already exists, append us to the end. 1522 */ 1523 if (control != NULL) { 1524 for (n = control; n->m_next != NULL; n = n->m_next) 1525 ; 1526 n->m_next = m; 1527 } else 1528 control = m; 1529 1530 return (control); 1531} 1532 1533/* 1534 * unp_defer indicates whether additional work has been defered for a future 1535 * pass through unp_gc(). It is thread local and does not require explicit 1536 * synchronization. 1537 */ 1538static int unp_defer; 1539 1540static int unp_taskcount; 1541SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, ""); 1542 1543static int unp_recycled; 1544SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, ""); 1545 1546static void 1547unp_gc(__unused void *arg, int pending) 1548{ 1549 struct file *fp, *nextfp; 1550 struct socket *so; 1551 struct file **extra_ref, **fpp; 1552 int nunref, i; 1553 int nfiles_snap; 1554 int nfiles_slack = 20; 1555 1556 unp_taskcount++; 1557 unp_defer = 0; 1558 /* 1559 * before going through all this, set all FDs to 1560 * be NOT defered and NOT externally accessible 1561 */ 1562 sx_slock(&filelist_lock); 1563 LIST_FOREACH(fp, &filehead, f_list) 1564 fp->f_gcflag &= ~(FMARK|FDEFER); 1565 do { 1566 KASSERT(unp_defer >= 0, ("unp_gc: unp_defer %d", unp_defer)); 1567 LIST_FOREACH(fp, &filehead, f_list) { 1568 FILE_LOCK(fp); 1569 /* 1570 * If the file is not open, skip it -- could be a 1571 * file in the process of being opened, or in the 1572 * process of being closed. If the file is 1573 * "closing", it may have been marked for deferred 1574 * consideration. Clear the flag now if so. 1575 */ 1576 if (fp->f_count == 0) { 1577 if (fp->f_gcflag & FDEFER) 1578 unp_defer--; 1579 fp->f_gcflag &= ~(FMARK|FDEFER); 1580 FILE_UNLOCK(fp); 1581 continue; 1582 } 1583 /* 1584 * If we already marked it as 'defer' in a 1585 * previous pass, then try process it this time 1586 * and un-mark it 1587 */ 1588 if (fp->f_gcflag & FDEFER) { 1589 fp->f_gcflag &= ~FDEFER; 1590 unp_defer--; 1591 } else { 1592 /* 1593 * if it's not defered, then check if it's 1594 * already marked.. if so skip it 1595 */ 1596 if (fp->f_gcflag & FMARK) { 1597 FILE_UNLOCK(fp); 1598 continue; 1599 } 1600 /* 1601 * If all references are from messages 1602 * in transit, then skip it. it's not 1603 * externally accessible. 1604 */ 1605 if (fp->f_count == fp->f_msgcount) { 1606 FILE_UNLOCK(fp); 1607 continue; 1608 } 1609 /* 1610 * If it got this far then it must be 1611 * externally accessible. 1612 */ 1613 fp->f_gcflag |= FMARK; 1614 } 1615 /* 1616 * either it was defered, or it is externally 1617 * accessible and not already marked so. 1618 * Now check if it is possibly one of OUR sockets. 1619 */ 1620 if (fp->f_type != DTYPE_SOCKET || 1621 (so = fp->f_data) == NULL) { 1622 FILE_UNLOCK(fp); 1623 continue; 1624 } 1625 FILE_UNLOCK(fp); 1626 if (so->so_proto->pr_domain != &localdomain || 1627 (so->so_proto->pr_flags&PR_RIGHTS) == 0) 1628 continue; 1629 /* 1630 * So, Ok, it's one of our sockets and it IS externally 1631 * accessible (or was defered). Now we look 1632 * to see if we hold any file descriptors in its 1633 * message buffers. Follow those links and mark them 1634 * as accessible too. 1635 */ 1636 SOCKBUF_LOCK(&so->so_rcv); 1637 unp_scan(so->so_rcv.sb_mb, unp_mark); 1638 SOCKBUF_UNLOCK(&so->so_rcv); 1639 } 1640 } while (unp_defer); 1641 sx_sunlock(&filelist_lock); 1642 /* 1643 * XXXRW: The following comments need updating for a post-SMPng and 1644 * deferred unp_gc() world, but are still generally accurate. 1645 * 1646 * We grab an extra reference to each of the file table entries 1647 * that are not otherwise accessible and then free the rights 1648 * that are stored in messages on them. 1649 * 1650 * The bug in the orginal code is a little tricky, so I'll describe 1651 * what's wrong with it here. 1652 * 1653 * It is incorrect to simply unp_discard each entry for f_msgcount 1654 * times -- consider the case of sockets A and B that contain 1655 * references to each other. On a last close of some other socket, 1656 * we trigger a gc since the number of outstanding rights (unp_rights) 1657 * is non-zero. If during the sweep phase the gc code unp_discards, 1658 * we end up doing a (full) closef on the descriptor. A closef on A 1659 * results in the following chain. Closef calls soo_close, which 1660 * calls soclose. Soclose calls first (through the switch 1661 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply 1662 * returns because the previous instance had set unp_gcing, and 1663 * we return all the way back to soclose, which marks the socket 1664 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush 1665 * to free up the rights that are queued in messages on the socket A, 1666 * i.e., the reference on B. The sorflush calls via the dom_dispose 1667 * switch unp_dispose, which unp_scans with unp_discard. This second 1668 * instance of unp_discard just calls closef on B. 1669 * 1670 * Well, a similar chain occurs on B, resulting in a sorflush on B, 1671 * which results in another closef on A. Unfortunately, A is already 1672 * being closed, and the descriptor has already been marked with 1673 * SS_NOFDREF, and soclose panics at this point. 1674 * 1675 * Here, we first take an extra reference to each inaccessible 1676 * descriptor. Then, we call sorflush ourself, since we know 1677 * it is a Unix domain socket anyhow. After we destroy all the 1678 * rights carried in messages, we do a last closef to get rid 1679 * of our extra reference. This is the last close, and the 1680 * unp_detach etc will shut down the socket. 1681 * 1682 * 91/09/19, bsy@cs.cmu.edu 1683 */ 1684again: 1685 nfiles_snap = openfiles + nfiles_slack; /* some slack */ 1686 extra_ref = malloc(nfiles_snap * sizeof(struct file *), M_TEMP, 1687 M_WAITOK); 1688 sx_slock(&filelist_lock); 1689 if (nfiles_snap < openfiles) { 1690 sx_sunlock(&filelist_lock); 1691 free(extra_ref, M_TEMP); 1692 nfiles_slack += 20; 1693 goto again; 1694 } 1695 for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref; 1696 fp != NULL; fp = nextfp) { 1697 nextfp = LIST_NEXT(fp, f_list); 1698 FILE_LOCK(fp); 1699 /* 1700 * If it's not open, skip it 1701 */ 1702 if (fp->f_count == 0) { 1703 FILE_UNLOCK(fp); 1704 continue; 1705 } 1706 /* 1707 * If all refs are from msgs, and it's not marked accessible 1708 * then it must be referenced from some unreachable cycle 1709 * of (shut-down) FDs, so include it in our 1710 * list of FDs to remove 1711 */ 1712 if (fp->f_count == fp->f_msgcount && !(fp->f_gcflag & FMARK)) { 1713 *fpp++ = fp; 1714 nunref++; 1715 fp->f_count++; 1716 } 1717 FILE_UNLOCK(fp); 1718 } 1719 sx_sunlock(&filelist_lock); 1720 /* 1721 * for each FD on our hit list, do the following two things 1722 */ 1723 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1724 struct file *tfp = *fpp; 1725 FILE_LOCK(tfp); 1726 if (tfp->f_type == DTYPE_SOCKET && 1727 tfp->f_data != NULL) { 1728 FILE_UNLOCK(tfp); 1729 sorflush(tfp->f_data); 1730 } else { 1731 FILE_UNLOCK(tfp); 1732 } 1733 } 1734 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1735 closef(*fpp, (struct thread *) NULL); 1736 unp_recycled++; 1737 } 1738 free(extra_ref, M_TEMP); 1739} 1740 1741void 1742unp_dispose(struct mbuf *m) 1743{ 1744 1745 if (m) 1746 unp_scan(m, unp_discard); 1747} 1748 1749static int 1750unp_listen(struct socket *so, struct unpcb *unp, int backlog, 1751 struct thread *td) 1752{ 1753 int error; 1754 1755 UNP_LOCK_ASSERT(); 1756 1757 SOCK_LOCK(so); 1758 error = solisten_proto_check(so); 1759 if (error == 0) { 1760 cru2x(td->td_ucred, &unp->unp_peercred); 1761 unp->unp_flags |= UNP_HAVEPCCACHED; 1762 solisten_proto(so, backlog); 1763 } 1764 SOCK_UNLOCK(so); 1765 return (error); 1766} 1767 1768static void 1769unp_scan(struct mbuf *m0, void (*op)(struct file *)) 1770{ 1771 struct mbuf *m; 1772 struct file **rp; 1773 struct cmsghdr *cm; 1774 void *data; 1775 int i; 1776 socklen_t clen, datalen; 1777 int qfds; 1778 1779 while (m0 != NULL) { 1780 for (m = m0; m; m = m->m_next) { 1781 if (m->m_type != MT_CONTROL) 1782 continue; 1783 1784 cm = mtod(m, struct cmsghdr *); 1785 clen = m->m_len; 1786 1787 while (cm != NULL) { 1788 if (sizeof(*cm) > clen || cm->cmsg_len > clen) 1789 break; 1790 1791 data = CMSG_DATA(cm); 1792 datalen = (caddr_t)cm + cm->cmsg_len 1793 - (caddr_t)data; 1794 1795 if (cm->cmsg_level == SOL_SOCKET && 1796 cm->cmsg_type == SCM_RIGHTS) { 1797 qfds = datalen / sizeof (struct file *); 1798 rp = data; 1799 for (i = 0; i < qfds; i++) 1800 (*op)(*rp++); 1801 } 1802 1803 if (CMSG_SPACE(datalen) < clen) { 1804 clen -= CMSG_SPACE(datalen); 1805 cm = (struct cmsghdr *) 1806 ((caddr_t)cm + CMSG_SPACE(datalen)); 1807 } else { 1808 clen = 0; 1809 cm = NULL; 1810 } 1811 } 1812 } 1813 m0 = m0->m_act; 1814 } 1815} 1816 1817static void 1818unp_mark(struct file *fp) 1819{ 1820 if (fp->f_gcflag & FMARK) 1821 return; 1822 unp_defer++; 1823 fp->f_gcflag |= (FMARK|FDEFER); 1824} 1825 1826static void 1827unp_discard(struct file *fp) 1828{ 1829 UNP_LOCK(); 1830 FILE_LOCK(fp); 1831 fp->f_msgcount--; 1832 unp_rights--; 1833 FILE_UNLOCK(fp); 1834 UNP_UNLOCK(); 1835 (void) closef(fp, (struct thread *)NULL); 1836} 1837