uipc_usrreq.c revision 160584
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 160584 2006-07-22 18:41:42Z 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/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_UNLOCK(); 153} 154 155static int 156uipc_accept(struct socket *so, struct sockaddr **nam) 157{ 158 struct unpcb *unp; 159 const struct sockaddr *sa; 160 161 /* 162 * Pass back name of connected socket, if it was bound and we are 163 * still connected (our peer may have closed already!). 164 */ 165 unp = sotounpcb(so); 166 KASSERT(unp != NULL, ("uipc_accept: unp == NULL")); 167 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 168 UNP_LOCK(); 169 if (unp->unp_conn != NULL && unp->unp_conn->unp_addr != NULL) 170 sa = (struct sockaddr *) unp->unp_conn->unp_addr; 171 else 172 sa = &sun_noname; 173 bcopy(sa, *nam, sa->sa_len); 174 UNP_UNLOCK(); 175 return (0); 176} 177 178static int 179uipc_attach(struct socket *so, int proto, struct thread *td) 180{ 181 182 return (unp_attach(so)); 183} 184 185static int 186uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 187{ 188 struct unpcb *unp; 189 int error; 190 191 unp = sotounpcb(so); 192 KASSERT(unp != NULL, ("uipc_bind: unp == NULL")); 193 UNP_LOCK(); 194 error = unp_bind(unp, nam, td); 195 UNP_UNLOCK(); 196 return (error); 197} 198 199static int 200uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 201{ 202 int error; 203 204 KASSERT(td == curthread, ("uipc_connect: td != curthread")); 205 UNP_LOCK(); 206 error = unp_connect(so, nam, td); 207 UNP_UNLOCK(); 208 return (error); 209} 210 211/* 212 * XXXRW: Should also unbind? 213 */ 214static void 215uipc_close(struct socket *so) 216{ 217 struct unpcb *unp; 218 219 unp = sotounpcb(so); 220 KASSERT(unp != NULL, ("uipc_close: unp == NULL")); 221 UNP_LOCK(); 222 unp_disconnect(unp); 223 UNP_UNLOCK(); 224} 225 226int 227uipc_connect2(struct socket *so1, struct socket *so2) 228{ 229 struct unpcb *unp; 230 int error; 231 232 unp = sotounpcb(so1); 233 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL")); 234 UNP_LOCK(); 235 error = unp_connect2(so1, so2, PRU_CONNECT2); 236 UNP_UNLOCK(); 237 return (error); 238} 239 240/* control is EOPNOTSUPP */ 241 242static void 243uipc_detach(struct socket *so) 244{ 245 struct unpcb *unp; 246 247 unp = sotounpcb(so); 248 KASSERT(unp != NULL, ("uipc_detach: unp == NULL")); 249 UNP_LOCK(); 250 unp_detach(unp); 251 UNP_UNLOCK_ASSERT(); 252} 253 254static int 255uipc_disconnect(struct socket *so) 256{ 257 struct unpcb *unp; 258 259 unp = sotounpcb(so); 260 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL")); 261 UNP_LOCK(); 262 unp_disconnect(unp); 263 UNP_UNLOCK(); 264 return (0); 265} 266 267static int 268uipc_listen(struct socket *so, int backlog, struct thread *td) 269{ 270 struct unpcb *unp; 271 int error; 272 273 unp = sotounpcb(so); 274 KASSERT(unp != NULL, ("uipc_listen: unp == NULL")); 275 UNP_LOCK(); 276 if (unp->unp_vnode == NULL) { 277 UNP_UNLOCK(); 278 return (EINVAL); 279 } 280 error = unp_listen(so, unp, backlog, td); 281 UNP_UNLOCK(); 282 return (error); 283} 284 285static int 286uipc_peeraddr(struct socket *so, struct sockaddr **nam) 287{ 288 struct unpcb *unp; 289 const struct sockaddr *sa; 290 291 unp = sotounpcb(so); 292 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL")); 293 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 294 UNP_LOCK(); 295 if (unp->unp_conn != NULL && unp->unp_conn->unp_addr!= NULL) 296 sa = (struct sockaddr *) unp->unp_conn->unp_addr; 297 else { 298 /* 299 * XXX: It seems that this test always fails even when 300 * connection is established. So, this else clause is 301 * added as workaround to return PF_LOCAL sockaddr. 302 */ 303 sa = &sun_noname; 304 } 305 bcopy(sa, *nam, sa->sa_len); 306 UNP_UNLOCK(); 307 return (0); 308} 309 310static int 311uipc_rcvd(struct socket *so, int flags) 312{ 313 struct unpcb *unp; 314 struct socket *so2; 315 u_int mbcnt, sbcc; 316 u_long newhiwat; 317 318 unp = sotounpcb(so); 319 KASSERT(unp != NULL, ("uipc_rcvd: unp == NULL")); 320 switch (so->so_type) { 321 case SOCK_DGRAM: 322 panic("uipc_rcvd DGRAM?"); 323 /*NOTREACHED*/ 324 325 case SOCK_STREAM: 326 /* 327 * Adjust backpressure on sender and wakeup any waiting to 328 * write. 329 */ 330 SOCKBUF_LOCK(&so->so_rcv); 331 mbcnt = so->so_rcv.sb_mbcnt; 332 sbcc = so->so_rcv.sb_cc; 333 SOCKBUF_UNLOCK(&so->so_rcv); 334 UNP_LOCK(); 335 if (unp->unp_conn == NULL) { 336 UNP_UNLOCK(); 337 break; 338 } 339 so2 = unp->unp_conn->unp_socket; 340 SOCKBUF_LOCK(&so2->so_snd); 341 so2->so_snd.sb_mbmax += unp->unp_mbcnt - mbcnt; 342 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - sbcc; 343 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat, 344 newhiwat, RLIM_INFINITY); 345 sowwakeup_locked(so2); 346 unp->unp_mbcnt = mbcnt; 347 unp->unp_cc = sbcc; 348 UNP_UNLOCK(); 349 break; 350 351 default: 352 panic("uipc_rcvd unknown socktype"); 353 } 354 return (0); 355} 356 357/* pru_rcvoob is EOPNOTSUPP */ 358 359static int 360uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 361 struct mbuf *control, struct thread *td) 362{ 363 struct unpcb *unp, *unp2; 364 struct socket *so2; 365 u_int mbcnt, sbcc; 366 u_long newhiwat; 367 int error = 0; 368 369 unp = sotounpcb(so); 370 KASSERT(unp != NULL, ("uipc_send: unp == NULL")); 371 if (flags & PRUS_OOB) { 372 error = EOPNOTSUPP; 373 goto release; 374 } 375 376 if (control != NULL && (error = unp_internalize(&control, td))) 377 goto release; 378 379 UNP_LOCK(); 380 switch (so->so_type) { 381 case SOCK_DGRAM: 382 { 383 const struct sockaddr *from; 384 385 if (nam != NULL) { 386 if (unp->unp_conn != NULL) { 387 error = EISCONN; 388 break; 389 } 390 error = unp_connect(so, nam, td); 391 if (error) 392 break; 393 } else { 394 if (unp->unp_conn == NULL) { 395 error = ENOTCONN; 396 break; 397 } 398 } 399 unp2 = unp->unp_conn; 400 so2 = unp2->unp_socket; 401 if (unp->unp_addr != NULL) 402 from = (struct sockaddr *)unp->unp_addr; 403 else 404 from = &sun_noname; 405 if (unp2->unp_flags & UNP_WANTCRED) 406 control = unp_addsockcred(td, control); 407 SOCKBUF_LOCK(&so2->so_rcv); 408 if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) { 409 sorwakeup_locked(so2); 410 m = NULL; 411 control = NULL; 412 } else { 413 SOCKBUF_UNLOCK(&so2->so_rcv); 414 error = ENOBUFS; 415 } 416 if (nam != NULL) 417 unp_disconnect(unp); 418 break; 419 } 420 421 case SOCK_STREAM: 422 /* 423 * Connect if not connected yet. 424 * 425 * Note: A better implementation would complain if not equal 426 * to the peer's address. 427 */ 428 if ((so->so_state & SS_ISCONNECTED) == 0) { 429 if (nam != NULL) { 430 error = unp_connect(so, nam, td); 431 if (error) 432 break; /* XXX */ 433 } else { 434 error = ENOTCONN; 435 break; 436 } 437 } 438 439 /* Lockless read. */ 440 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 441 error = EPIPE; 442 break; 443 } 444 unp2 = unp->unp_conn; 445 if (unp2 == NULL) 446 panic("uipc_send connected but no connection?"); 447 so2 = unp2->unp_socket; 448 SOCKBUF_LOCK(&so2->so_rcv); 449 if (unp2->unp_flags & UNP_WANTCRED) { 450 /* 451 * Credentials are passed only once on 452 * SOCK_STREAM. 453 */ 454 unp2->unp_flags &= ~UNP_WANTCRED; 455 control = unp_addsockcred(td, control); 456 } 457 /* 458 * Send to paired receive port, and then reduce send buffer 459 * hiwater marks to maintain backpressure. Wake up readers. 460 */ 461 if (control != NULL) { 462 if (sbappendcontrol_locked(&so2->so_rcv, m, control)) 463 control = NULL; 464 } else { 465 sbappend_locked(&so2->so_rcv, m); 466 } 467 mbcnt = so2->so_rcv.sb_mbcnt - unp2->unp_mbcnt; 468 unp2->unp_mbcnt = so2->so_rcv.sb_mbcnt; 469 sbcc = so2->so_rcv.sb_cc; 470 sorwakeup_locked(so2); 471 472 SOCKBUF_LOCK(&so->so_snd); 473 newhiwat = so->so_snd.sb_hiwat - (sbcc - unp2->unp_cc); 474 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat, 475 newhiwat, RLIM_INFINITY); 476 so->so_snd.sb_mbmax -= mbcnt; 477 SOCKBUF_UNLOCK(&so->so_snd); 478 479 unp2->unp_cc = sbcc; 480 m = NULL; 481 break; 482 483 default: 484 panic("uipc_send unknown socktype"); 485 } 486 487 /* 488 * SEND_EOF is equivalent to a SEND followed by 489 * a SHUTDOWN. 490 */ 491 if (flags & PRUS_EOF) { 492 socantsendmore(so); 493 unp_shutdown(unp); 494 } 495 UNP_UNLOCK(); 496 497 if (control != NULL && error != 0) 498 unp_dispose(control); 499 500release: 501 if (control != NULL) 502 m_freem(control); 503 if (m != NULL) 504 m_freem(m); 505 return (error); 506} 507 508static int 509uipc_sense(struct socket *so, struct stat *sb) 510{ 511 struct unpcb *unp; 512 struct socket *so2; 513 514 unp = sotounpcb(so); 515 KASSERT(unp != NULL, ("uipc_sense: unp == NULL")); 516 UNP_LOCK(); 517 sb->st_blksize = so->so_snd.sb_hiwat; 518 if (so->so_type == SOCK_STREAM && unp->unp_conn != NULL) { 519 so2 = unp->unp_conn->unp_socket; 520 sb->st_blksize += so2->so_rcv.sb_cc; 521 } 522 sb->st_dev = NODEV; 523 if (unp->unp_ino == 0) 524 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino; 525 sb->st_ino = unp->unp_ino; 526 UNP_UNLOCK(); 527 return (0); 528} 529 530static int 531uipc_shutdown(struct socket *so) 532{ 533 struct unpcb *unp; 534 535 unp = sotounpcb(so); 536 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL")); 537 UNP_LOCK(); 538 socantsendmore(so); 539 unp_shutdown(unp); 540 UNP_UNLOCK(); 541 return (0); 542} 543 544static int 545uipc_sockaddr(struct socket *so, struct sockaddr **nam) 546{ 547 struct unpcb *unp; 548 const struct sockaddr *sa; 549 550 unp = sotounpcb(so); 551 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL")); 552 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 553 UNP_LOCK(); 554 if (unp->unp_addr != NULL) 555 sa = (struct sockaddr *) unp->unp_addr; 556 else 557 sa = &sun_noname; 558 bcopy(sa, *nam, sa->sa_len); 559 UNP_UNLOCK(); 560 return (0); 561} 562 563struct pr_usrreqs uipc_usrreqs = { 564 .pru_abort = uipc_abort, 565 .pru_accept = uipc_accept, 566 .pru_attach = uipc_attach, 567 .pru_bind = uipc_bind, 568 .pru_connect = uipc_connect, 569 .pru_connect2 = uipc_connect2, 570 .pru_detach = uipc_detach, 571 .pru_disconnect = uipc_disconnect, 572 .pru_listen = uipc_listen, 573 .pru_peeraddr = uipc_peeraddr, 574 .pru_rcvd = uipc_rcvd, 575 .pru_send = uipc_send, 576 .pru_sense = uipc_sense, 577 .pru_shutdown = uipc_shutdown, 578 .pru_sockaddr = uipc_sockaddr, 579 .pru_sosend = sosend, 580 .pru_soreceive = soreceive, 581 .pru_sopoll = sopoll, 582 .pru_close = uipc_close, 583}; 584 585int 586uipc_ctloutput(struct socket *so, struct sockopt *sopt) 587{ 588 struct unpcb *unp; 589 struct xucred xu; 590 int error, optval; 591 592 if (sopt->sopt_level != 0) 593 return (EINVAL); 594 595 unp = sotounpcb(so); 596 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL")); 597 UNP_LOCK(); 598 error = 0; 599 switch (sopt->sopt_dir) { 600 case SOPT_GET: 601 switch (sopt->sopt_name) { 602 case LOCAL_PEERCRED: 603 if (unp->unp_flags & UNP_HAVEPC) 604 xu = unp->unp_peercred; 605 else { 606 if (so->so_type == SOCK_STREAM) 607 error = ENOTCONN; 608 else 609 error = EINVAL; 610 } 611 if (error == 0) 612 error = sooptcopyout(sopt, &xu, sizeof(xu)); 613 break; 614 case LOCAL_CREDS: 615 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0; 616 error = sooptcopyout(sopt, &optval, sizeof(optval)); 617 break; 618 case LOCAL_CONNWAIT: 619 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0; 620 error = sooptcopyout(sopt, &optval, sizeof(optval)); 621 break; 622 default: 623 error = EOPNOTSUPP; 624 break; 625 } 626 break; 627 case SOPT_SET: 628 switch (sopt->sopt_name) { 629 case LOCAL_CREDS: 630 case LOCAL_CONNWAIT: 631 error = sooptcopyin(sopt, &optval, sizeof(optval), 632 sizeof(optval)); 633 if (error) 634 break; 635 636#define OPTSET(bit) \ 637 if (optval) \ 638 unp->unp_flags |= bit; \ 639 else \ 640 unp->unp_flags &= ~bit; 641 642 switch (sopt->sopt_name) { 643 case LOCAL_CREDS: 644 OPTSET(UNP_WANTCRED); 645 break; 646 case LOCAL_CONNWAIT: 647 OPTSET(UNP_CONNWAIT); 648 break; 649 default: 650 break; 651 } 652 break; 653#undef OPTSET 654 default: 655 error = ENOPROTOOPT; 656 break; 657 } 658 break; 659 default: 660 error = EOPNOTSUPP; 661 break; 662 } 663 UNP_UNLOCK(); 664 return (error); 665} 666 667/* 668 * Both send and receive buffers are allocated PIPSIZ bytes of buffering 669 * for stream sockets, although the total for sender and receiver is 670 * actually only PIPSIZ. 671 * 672 * Datagram sockets really use the sendspace as the maximum datagram size, 673 * and don't really want to reserve the sendspace. Their recvspace should 674 * be large enough for at least one max-size datagram plus address. 675 */ 676#ifndef PIPSIZ 677#define PIPSIZ 8192 678#endif 679static u_long unpst_sendspace = PIPSIZ; 680static u_long unpst_recvspace = PIPSIZ; 681static u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 682static u_long unpdg_recvspace = 4*1024; 683 684static int unp_rights; /* file descriptors in flight */ 685 686SYSCTL_DECL(_net_local_stream); 687SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 688 &unpst_sendspace, 0, ""); 689SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 690 &unpst_recvspace, 0, ""); 691SYSCTL_DECL(_net_local_dgram); 692SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 693 &unpdg_sendspace, 0, ""); 694SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 695 &unpdg_recvspace, 0, ""); 696SYSCTL_DECL(_net_local); 697SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, ""); 698 699static int 700unp_attach(struct socket *so) 701{ 702 struct unpcb *unp; 703 int error; 704 705 KASSERT(so->so_pcb == NULL, ("unp_attach: so_pcb != NULL")); 706 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 707 switch (so->so_type) { 708 709 case SOCK_STREAM: 710 error = soreserve(so, unpst_sendspace, unpst_recvspace); 711 break; 712 713 case SOCK_DGRAM: 714 error = soreserve(so, unpdg_sendspace, unpdg_recvspace); 715 break; 716 717 default: 718 panic("unp_attach"); 719 } 720 if (error) 721 return (error); 722 } 723 unp = uma_zalloc(unp_zone, M_WAITOK | M_ZERO); 724 if (unp == NULL) 725 return (ENOBUFS); 726 LIST_INIT(&unp->unp_refs); 727 unp->unp_socket = so; 728 so->so_pcb = unp; 729 730 UNP_LOCK(); 731 unp->unp_gencnt = ++unp_gencnt; 732 unp_count++; 733 LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead 734 : &unp_shead, unp, unp_link); 735 UNP_UNLOCK(); 736 737 return (0); 738} 739 740static void 741unp_detach(struct unpcb *unp) 742{ 743 struct vnode *vp; 744 int local_unp_rights; 745 746 UNP_LOCK_ASSERT(); 747 748 LIST_REMOVE(unp, unp_link); 749 unp->unp_gencnt = ++unp_gencnt; 750 --unp_count; 751 if ((vp = unp->unp_vnode) != NULL) { 752 /* 753 * XXXRW: should v_socket be frobbed only while holding 754 * Giant? 755 */ 756 unp->unp_vnode->v_socket = NULL; 757 unp->unp_vnode = NULL; 758 } 759 if (unp->unp_conn != NULL) 760 unp_disconnect(unp); 761 while (!LIST_EMPTY(&unp->unp_refs)) { 762 struct unpcb *ref = LIST_FIRST(&unp->unp_refs); 763 unp_drop(ref, ECONNRESET); 764 } 765 soisdisconnected(unp->unp_socket); 766 unp->unp_socket->so_pcb = NULL; 767 local_unp_rights = unp_rights; 768 UNP_UNLOCK(); 769 if (unp->unp_addr != NULL) 770 FREE(unp->unp_addr, M_SONAME); 771 uma_zfree(unp_zone, unp); 772 if (vp) { 773 int vfslocked; 774 775 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 776 vrele(vp); 777 VFS_UNLOCK_GIANT(vfslocked); 778 } 779 if (local_unp_rights) 780 taskqueue_enqueue(taskqueue_thread, &unp_gc_task); 781} 782 783static int 784unp_bind(struct unpcb *unp, struct sockaddr *nam, struct thread *td) 785{ 786 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 787 struct vnode *vp; 788 struct mount *mp; 789 struct vattr vattr; 790 int error, namelen; 791 struct nameidata nd; 792 char *buf; 793 794 UNP_LOCK_ASSERT(); 795 796 /* 797 * XXXRW: This test-and-set of unp_vnode is non-atomic; the unlocked 798 * read here is fine, but the value of unp_vnode needs to be tested 799 * again after we do all the lookups to see if the pcb is still 800 * unbound? 801 */ 802 if (unp->unp_vnode != NULL) 803 return (EINVAL); 804 805 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 806 if (namelen <= 0) 807 return (EINVAL); 808 809 UNP_UNLOCK(); 810 811 buf = malloc(namelen + 1, M_TEMP, M_WAITOK); 812 strlcpy(buf, soun->sun_path, namelen + 1); 813 814 mtx_lock(&Giant); 815restart: 816 mtx_assert(&Giant, MA_OWNED); 817 NDINIT(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME, UIO_SYSSPACE, 818 buf, td); 819/* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 820 error = namei(&nd); 821 if (error) 822 goto done; 823 vp = nd.ni_vp; 824 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { 825 NDFREE(&nd, NDF_ONLY_PNBUF); 826 if (nd.ni_dvp == vp) 827 vrele(nd.ni_dvp); 828 else 829 vput(nd.ni_dvp); 830 if (vp != NULL) { 831 vrele(vp); 832 error = EADDRINUSE; 833 goto done; 834 } 835 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH); 836 if (error) 837 goto done; 838 goto restart; 839 } 840 VATTR_NULL(&vattr); 841 vattr.va_type = VSOCK; 842 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask); 843#ifdef MAC 844 error = mac_check_vnode_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, 845 &vattr); 846#endif 847 if (error == 0) { 848 VOP_LEASE(nd.ni_dvp, td, td->td_ucred, LEASE_WRITE); 849 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 850 } 851 NDFREE(&nd, NDF_ONLY_PNBUF); 852 vput(nd.ni_dvp); 853 if (error) { 854 vn_finished_write(mp); 855 goto done; 856 } 857 vp = nd.ni_vp; 858 ASSERT_VOP_LOCKED(vp, "unp_bind"); 859 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK); 860 UNP_LOCK(); 861 vp->v_socket = unp->unp_socket; 862 unp->unp_vnode = vp; 863 unp->unp_addr = soun; 864 UNP_UNLOCK(); 865 VOP_UNLOCK(vp, 0, td); 866 vn_finished_write(mp); 867done: 868 mtx_unlock(&Giant); 869 free(buf, M_TEMP); 870 UNP_LOCK(); 871 return (error); 872} 873 874static int 875unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 876{ 877 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 878 struct vnode *vp; 879 struct socket *so2, *so3; 880 struct unpcb *unp, *unp2, *unp3; 881 int error, len; 882 struct nameidata nd; 883 char buf[SOCK_MAXADDRLEN]; 884 struct sockaddr *sa; 885 886 UNP_LOCK_ASSERT(); 887 888 unp = sotounpcb(so); 889 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 890 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 891 if (len <= 0) 892 return (EINVAL); 893 strlcpy(buf, soun->sun_path, len + 1); 894 UNP_UNLOCK(); 895 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 896 mtx_lock(&Giant); 897 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, td); 898 error = namei(&nd); 899 if (error) 900 vp = NULL; 901 else 902 vp = nd.ni_vp; 903 ASSERT_VOP_LOCKED(vp, "unp_connect"); 904 NDFREE(&nd, NDF_ONLY_PNBUF); 905 if (error) 906 goto bad; 907 908 if (vp->v_type != VSOCK) { 909 error = ENOTSOCK; 910 goto bad; 911 } 912 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); 913 if (error) 914 goto bad; 915 mtx_unlock(&Giant); 916 UNP_LOCK(); 917 unp = sotounpcb(so); 918 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 919 so2 = vp->v_socket; 920 if (so2 == NULL) { 921 error = ECONNREFUSED; 922 goto bad2; 923 } 924 if (so->so_type != so2->so_type) { 925 error = EPROTOTYPE; 926 goto bad2; 927 } 928 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 929 if (so2->so_options & SO_ACCEPTCONN) { 930 /* 931 * NB: drop locks here so unp_attach is entered w/o 932 * locks; this avoids a recursive lock of the head 933 * and holding sleep locks across a (potentially) 934 * blocking malloc. 935 */ 936 UNP_UNLOCK(); 937 so3 = sonewconn(so2, 0); 938 UNP_LOCK(); 939 } else 940 so3 = NULL; 941 if (so3 == NULL) { 942 error = ECONNREFUSED; 943 goto bad2; 944 } 945 unp = sotounpcb(so); 946 unp2 = sotounpcb(so2); 947 unp3 = sotounpcb(so3); 948 if (unp2->unp_addr != NULL) { 949 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); 950 unp3->unp_addr = (struct sockaddr_un *) sa; 951 sa = NULL; 952 } 953 /* 954 * unp_peercred management: 955 * 956 * The connecter's (client's) credentials are copied from its 957 * process structure at the time of connect() (which is now). 958 */ 959 cru2x(td->td_ucred, &unp3->unp_peercred); 960 unp3->unp_flags |= UNP_HAVEPC; 961 /* 962 * The receiver's (server's) credentials are copied from the 963 * unp_peercred member of socket on which the former called 964 * listen(); unp_listen() cached that process's credentials 965 * at that time so we can use them now. 966 */ 967 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED, 968 ("unp_connect: listener without cached peercred")); 969 memcpy(&unp->unp_peercred, &unp2->unp_peercred, 970 sizeof(unp->unp_peercred)); 971 unp->unp_flags |= UNP_HAVEPC; 972 if (unp2->unp_flags & UNP_WANTCRED) 973 unp3->unp_flags |= UNP_WANTCRED; 974#ifdef MAC 975 SOCK_LOCK(so); 976 mac_set_socket_peer_from_socket(so, so3); 977 mac_set_socket_peer_from_socket(so3, so); 978 SOCK_UNLOCK(so); 979#endif 980 981 so2 = so3; 982 } 983 error = unp_connect2(so, so2, PRU_CONNECT); 984bad2: 985 UNP_UNLOCK(); 986 mtx_lock(&Giant); 987bad: 988 mtx_assert(&Giant, MA_OWNED); 989 if (vp != NULL) 990 vput(vp); 991 mtx_unlock(&Giant); 992 free(sa, M_SONAME); 993 UNP_LOCK(); 994 return (error); 995} 996 997static int 998unp_connect2(struct socket *so, struct socket *so2, int req) 999{ 1000 struct unpcb *unp = sotounpcb(so); 1001 struct unpcb *unp2; 1002 1003 UNP_LOCK_ASSERT(); 1004 1005 if (so2->so_type != so->so_type) 1006 return (EPROTOTYPE); 1007 unp2 = sotounpcb(so2); 1008 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL")); 1009 unp->unp_conn = unp2; 1010 switch (so->so_type) { 1011 1012 case SOCK_DGRAM: 1013 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 1014 soisconnected(so); 1015 break; 1016 1017 case SOCK_STREAM: 1018 unp2->unp_conn = unp; 1019 if (req == PRU_CONNECT && 1020 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 1021 soisconnecting(so); 1022 else 1023 soisconnected(so); 1024 soisconnected(so2); 1025 break; 1026 1027 default: 1028 panic("unp_connect2"); 1029 } 1030 return (0); 1031} 1032 1033static void 1034unp_disconnect(struct unpcb *unp) 1035{ 1036 struct unpcb *unp2 = unp->unp_conn; 1037 struct socket *so; 1038 1039 UNP_LOCK_ASSERT(); 1040 1041 if (unp2 == NULL) 1042 return; 1043 unp->unp_conn = NULL; 1044 switch (unp->unp_socket->so_type) { 1045 case SOCK_DGRAM: 1046 LIST_REMOVE(unp, unp_reflink); 1047 so = unp->unp_socket; 1048 SOCK_LOCK(so); 1049 so->so_state &= ~SS_ISCONNECTED; 1050 SOCK_UNLOCK(so); 1051 break; 1052 1053 case SOCK_STREAM: 1054 soisdisconnected(unp->unp_socket); 1055 unp2->unp_conn = NULL; 1056 soisdisconnected(unp2->unp_socket); 1057 break; 1058 } 1059} 1060 1061/* 1062 * unp_pcblist() assumes that UNIX domain socket memory is never reclaimed by 1063 * the zone (UMA_ZONE_NOFREE), and as such potentially stale pointers are 1064 * safe to reference. It first scans the list of struct unpcb's to generate 1065 * a pointer list, then it rescans its list one entry at a time to 1066 * externalize and copyout. It checks the generation number to see if a 1067 * struct unpcb has been reused, and will skip it if so. 1068 */ 1069static int 1070unp_pcblist(SYSCTL_HANDLER_ARGS) 1071{ 1072 int error, i, n; 1073 struct unpcb *unp, **unp_list; 1074 unp_gen_t gencnt; 1075 struct xunpgen *xug; 1076 struct unp_head *head; 1077 struct xunpcb *xu; 1078 1079 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead); 1080 1081 /* 1082 * The process of preparing the PCB list is too time-consuming and 1083 * resource-intensive to repeat twice on every request. 1084 */ 1085 if (req->oldptr == NULL) { 1086 n = unp_count; 1087 req->oldidx = 2 * (sizeof *xug) 1088 + (n + n/8) * sizeof(struct xunpcb); 1089 return (0); 1090 } 1091 1092 if (req->newptr != NULL) 1093 return (EPERM); 1094 1095 /* 1096 * OK, now we're committed to doing something. 1097 */ 1098 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK); 1099 UNP_LOCK(); 1100 gencnt = unp_gencnt; 1101 n = unp_count; 1102 UNP_UNLOCK(); 1103 1104 xug->xug_len = sizeof *xug; 1105 xug->xug_count = n; 1106 xug->xug_gen = gencnt; 1107 xug->xug_sogen = so_gencnt; 1108 error = SYSCTL_OUT(req, xug, sizeof *xug); 1109 if (error) { 1110 free(xug, M_TEMP); 1111 return (error); 1112 } 1113 1114 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 1115 1116 UNP_LOCK(); 1117 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 1118 unp = LIST_NEXT(unp, unp_link)) { 1119 if (unp->unp_gencnt <= gencnt) { 1120 if (cr_cansee(req->td->td_ucred, 1121 unp->unp_socket->so_cred)) 1122 continue; 1123 unp_list[i++] = unp; 1124 } 1125 } 1126 UNP_UNLOCK(); 1127 n = i; /* In case we lost some during malloc. */ 1128 1129 error = 0; 1130 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO); 1131 for (i = 0; i < n; i++) { 1132 unp = unp_list[i]; 1133 if (unp->unp_gencnt <= gencnt) { 1134 xu->xu_len = sizeof *xu; 1135 xu->xu_unpp = unp; 1136 /* 1137 * XXX - need more locking here to protect against 1138 * connect/disconnect races for SMP. 1139 */ 1140 if (unp->unp_addr != NULL) 1141 bcopy(unp->unp_addr, &xu->xu_addr, 1142 unp->unp_addr->sun_len); 1143 if (unp->unp_conn != NULL && 1144 unp->unp_conn->unp_addr != NULL) 1145 bcopy(unp->unp_conn->unp_addr, 1146 &xu->xu_caddr, 1147 unp->unp_conn->unp_addr->sun_len); 1148 bcopy(unp, &xu->xu_unp, sizeof *unp); 1149 sotoxsocket(unp->unp_socket, &xu->xu_socket); 1150 error = SYSCTL_OUT(req, xu, sizeof *xu); 1151 } 1152 } 1153 free(xu, M_TEMP); 1154 if (!error) { 1155 /* 1156 * Give the user an updated idea of our state. If the 1157 * generation differs from what we told her before, she knows 1158 * that something happened while we were processing this 1159 * request, and it might be necessary to retry. 1160 */ 1161 xug->xug_gen = unp_gencnt; 1162 xug->xug_sogen = so_gencnt; 1163 xug->xug_count = unp_count; 1164 error = SYSCTL_OUT(req, xug, sizeof *xug); 1165 } 1166 free(unp_list, M_TEMP); 1167 free(xug, M_TEMP); 1168 return (error); 1169} 1170 1171SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD, 1172 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 1173 "List of active local datagram sockets"); 1174SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD, 1175 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 1176 "List of active local stream sockets"); 1177 1178static void 1179unp_shutdown(struct unpcb *unp) 1180{ 1181 struct socket *so; 1182 1183 UNP_LOCK_ASSERT(); 1184 1185 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn && 1186 (so = unp->unp_conn->unp_socket)) 1187 socantrcvmore(so); 1188} 1189 1190static void 1191unp_drop(struct unpcb *unp, int errno) 1192{ 1193 struct socket *so = unp->unp_socket; 1194 1195 UNP_LOCK_ASSERT(); 1196 1197 so->so_error = errno; 1198 unp_disconnect(unp); 1199} 1200 1201static void 1202unp_freerights(struct file **rp, int fdcount) 1203{ 1204 int i; 1205 struct file *fp; 1206 1207 for (i = 0; i < fdcount; i++) { 1208 fp = *rp; 1209 /* 1210 * Zero the pointer before calling unp_discard since it may 1211 * end up in unp_gc().. 1212 * 1213 * XXXRW: This is less true than it used to be. 1214 */ 1215 *rp++ = 0; 1216 unp_discard(fp); 1217 } 1218} 1219 1220int 1221unp_externalize(struct mbuf *control, struct mbuf **controlp) 1222{ 1223 struct thread *td = curthread; /* XXX */ 1224 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1225 int i; 1226 int *fdp; 1227 struct file **rp; 1228 struct file *fp; 1229 void *data; 1230 socklen_t clen = control->m_len, datalen; 1231 int error, newfds; 1232 int f; 1233 u_int newlen; 1234 1235 UNP_UNLOCK_ASSERT(); 1236 1237 error = 0; 1238 if (controlp != NULL) /* controlp == NULL => free control messages */ 1239 *controlp = NULL; 1240 1241 while (cm != NULL) { 1242 if (sizeof(*cm) > clen || cm->cmsg_len > clen) { 1243 error = EINVAL; 1244 break; 1245 } 1246 1247 data = CMSG_DATA(cm); 1248 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1249 1250 if (cm->cmsg_level == SOL_SOCKET 1251 && cm->cmsg_type == SCM_RIGHTS) { 1252 newfds = datalen / sizeof(struct file *); 1253 rp = data; 1254 1255 /* If we're not outputting the descriptors free them. */ 1256 if (error || controlp == NULL) { 1257 unp_freerights(rp, newfds); 1258 goto next; 1259 } 1260 FILEDESC_LOCK(td->td_proc->p_fd); 1261 /* if the new FD's will not fit free them. */ 1262 if (!fdavail(td, newfds)) { 1263 FILEDESC_UNLOCK(td->td_proc->p_fd); 1264 error = EMSGSIZE; 1265 unp_freerights(rp, newfds); 1266 goto next; 1267 } 1268 /* 1269 * Now change each pointer to an fd in the global 1270 * table to an integer that is the index to the local 1271 * fd table entry that we set up to point to the 1272 * global one we are transferring. 1273 */ 1274 newlen = newfds * sizeof(int); 1275 *controlp = sbcreatecontrol(NULL, newlen, 1276 SCM_RIGHTS, SOL_SOCKET); 1277 if (*controlp == NULL) { 1278 FILEDESC_UNLOCK(td->td_proc->p_fd); 1279 error = E2BIG; 1280 unp_freerights(rp, newfds); 1281 goto next; 1282 } 1283 1284 fdp = (int *) 1285 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1286 for (i = 0; i < newfds; i++) { 1287 if (fdalloc(td, 0, &f)) 1288 panic("unp_externalize fdalloc failed"); 1289 fp = *rp++; 1290 td->td_proc->p_fd->fd_ofiles[f] = fp; 1291 FILE_LOCK(fp); 1292 fp->f_msgcount--; 1293 FILE_UNLOCK(fp); 1294 unp_rights--; 1295 *fdp++ = f; 1296 } 1297 FILEDESC_UNLOCK(td->td_proc->p_fd); 1298 } else { 1299 /* We can just copy anything else across. */ 1300 if (error || controlp == NULL) 1301 goto next; 1302 *controlp = sbcreatecontrol(NULL, datalen, 1303 cm->cmsg_type, cm->cmsg_level); 1304 if (*controlp == NULL) { 1305 error = ENOBUFS; 1306 goto next; 1307 } 1308 bcopy(data, 1309 CMSG_DATA(mtod(*controlp, struct cmsghdr *)), 1310 datalen); 1311 } 1312 1313 controlp = &(*controlp)->m_next; 1314 1315next: 1316 if (CMSG_SPACE(datalen) < clen) { 1317 clen -= CMSG_SPACE(datalen); 1318 cm = (struct cmsghdr *) 1319 ((caddr_t)cm + CMSG_SPACE(datalen)); 1320 } else { 1321 clen = 0; 1322 cm = NULL; 1323 } 1324 } 1325 1326 m_freem(control); 1327 1328 return (error); 1329} 1330 1331static void 1332unp_zone_change(void *tag) 1333{ 1334 1335 uma_zone_set_max(unp_zone, maxsockets); 1336} 1337 1338void 1339unp_init(void) 1340{ 1341 1342 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL, 1343 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 1344 if (unp_zone == NULL) 1345 panic("unp_init"); 1346 uma_zone_set_max(unp_zone, maxsockets); 1347 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change, 1348 NULL, EVENTHANDLER_PRI_ANY); 1349 LIST_INIT(&unp_dhead); 1350 LIST_INIT(&unp_shead); 1351 TASK_INIT(&unp_gc_task, 0, unp_gc, NULL); 1352 UNP_LOCK_INIT(); 1353} 1354 1355static int 1356unp_internalize(struct mbuf **controlp, struct thread *td) 1357{ 1358 struct mbuf *control = *controlp; 1359 struct proc *p = td->td_proc; 1360 struct filedesc *fdescp = p->p_fd; 1361 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1362 struct cmsgcred *cmcred; 1363 struct file **rp; 1364 struct file *fp; 1365 struct timeval *tv; 1366 int i, fd, *fdp; 1367 void *data; 1368 socklen_t clen = control->m_len, datalen; 1369 int error, oldfds; 1370 u_int newlen; 1371 1372 UNP_UNLOCK_ASSERT(); 1373 1374 error = 0; 1375 *controlp = NULL; 1376 1377 while (cm != NULL) { 1378 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET 1379 || cm->cmsg_len > clen) { 1380 error = EINVAL; 1381 goto out; 1382 } 1383 1384 data = CMSG_DATA(cm); 1385 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1386 1387 switch (cm->cmsg_type) { 1388 /* 1389 * Fill in credential information. 1390 */ 1391 case SCM_CREDS: 1392 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred), 1393 SCM_CREDS, SOL_SOCKET); 1394 if (*controlp == NULL) { 1395 error = ENOBUFS; 1396 goto out; 1397 } 1398 1399 cmcred = (struct cmsgcred *) 1400 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1401 cmcred->cmcred_pid = p->p_pid; 1402 cmcred->cmcred_uid = td->td_ucred->cr_ruid; 1403 cmcred->cmcred_gid = td->td_ucred->cr_rgid; 1404 cmcred->cmcred_euid = td->td_ucred->cr_uid; 1405 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups, 1406 CMGROUP_MAX); 1407 for (i = 0; i < cmcred->cmcred_ngroups; i++) 1408 cmcred->cmcred_groups[i] = 1409 td->td_ucred->cr_groups[i]; 1410 break; 1411 1412 case SCM_RIGHTS: 1413 oldfds = datalen / sizeof (int); 1414 /* 1415 * Check that all the FDs passed in refer to legal 1416 * files. If not, reject the entire operation. 1417 */ 1418 fdp = data; 1419 FILEDESC_LOCK(fdescp); 1420 for (i = 0; i < oldfds; i++) { 1421 fd = *fdp++; 1422 if ((unsigned)fd >= fdescp->fd_nfiles || 1423 fdescp->fd_ofiles[fd] == NULL) { 1424 FILEDESC_UNLOCK(fdescp); 1425 error = EBADF; 1426 goto out; 1427 } 1428 fp = fdescp->fd_ofiles[fd]; 1429 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { 1430 FILEDESC_UNLOCK(fdescp); 1431 error = EOPNOTSUPP; 1432 goto out; 1433 } 1434 1435 } 1436 /* 1437 * Now replace the integer FDs with pointers to the 1438 * associated global file table entry.. 1439 */ 1440 newlen = oldfds * sizeof(struct file *); 1441 *controlp = sbcreatecontrol(NULL, newlen, 1442 SCM_RIGHTS, SOL_SOCKET); 1443 if (*controlp == NULL) { 1444 FILEDESC_UNLOCK(fdescp); 1445 error = E2BIG; 1446 goto out; 1447 } 1448 1449 fdp = data; 1450 rp = (struct file **) 1451 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1452 for (i = 0; i < oldfds; i++) { 1453 fp = fdescp->fd_ofiles[*fdp++]; 1454 *rp++ = fp; 1455 FILE_LOCK(fp); 1456 fp->f_count++; 1457 fp->f_msgcount++; 1458 FILE_UNLOCK(fp); 1459 unp_rights++; 1460 } 1461 FILEDESC_UNLOCK(fdescp); 1462 break; 1463 1464 case SCM_TIMESTAMP: 1465 *controlp = sbcreatecontrol(NULL, sizeof(*tv), 1466 SCM_TIMESTAMP, SOL_SOCKET); 1467 if (*controlp == NULL) { 1468 error = ENOBUFS; 1469 goto out; 1470 } 1471 tv = (struct timeval *) 1472 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1473 microtime(tv); 1474 break; 1475 1476 default: 1477 error = EINVAL; 1478 goto out; 1479 } 1480 1481 controlp = &(*controlp)->m_next; 1482 1483 if (CMSG_SPACE(datalen) < clen) { 1484 clen -= CMSG_SPACE(datalen); 1485 cm = (struct cmsghdr *) 1486 ((caddr_t)cm + CMSG_SPACE(datalen)); 1487 } else { 1488 clen = 0; 1489 cm = NULL; 1490 } 1491 } 1492 1493out: 1494 m_freem(control); 1495 1496 return (error); 1497} 1498 1499struct mbuf * 1500unp_addsockcred(struct thread *td, struct mbuf *control) 1501{ 1502 struct mbuf *m, *n, *n_prev; 1503 struct sockcred *sc; 1504 const struct cmsghdr *cm; 1505 int ngroups; 1506 int i; 1507 1508 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); 1509 1510 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET); 1511 if (m == NULL) 1512 return (control); 1513 1514 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *)); 1515 sc->sc_uid = td->td_ucred->cr_ruid; 1516 sc->sc_euid = td->td_ucred->cr_uid; 1517 sc->sc_gid = td->td_ucred->cr_rgid; 1518 sc->sc_egid = td->td_ucred->cr_gid; 1519 sc->sc_ngroups = ngroups; 1520 for (i = 0; i < sc->sc_ngroups; i++) 1521 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 1522 1523 /* 1524 * Unlink SCM_CREDS control messages (struct cmsgcred), since just 1525 * created SCM_CREDS control message (struct sockcred) has another 1526 * format. 1527 */ 1528 if (control != NULL) 1529 for (n = control, n_prev = NULL; n != NULL;) { 1530 cm = mtod(n, struct cmsghdr *); 1531 if (cm->cmsg_level == SOL_SOCKET && 1532 cm->cmsg_type == SCM_CREDS) { 1533 if (n_prev == NULL) 1534 control = n->m_next; 1535 else 1536 n_prev->m_next = n->m_next; 1537 n = m_free(n); 1538 } else { 1539 n_prev = n; 1540 n = n->m_next; 1541 } 1542 } 1543 1544 /* Prepend it to the head. */ 1545 m->m_next = control; 1546 1547 return (m); 1548} 1549 1550/* 1551 * unp_defer indicates whether additional work has been defered for a future 1552 * pass through unp_gc(). It is thread local and does not require explicit 1553 * synchronization. 1554 */ 1555static int unp_defer; 1556 1557static int unp_taskcount; 1558SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, ""); 1559 1560static int unp_recycled; 1561SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, ""); 1562 1563static void 1564unp_gc(__unused void *arg, int pending) 1565{ 1566 struct file *fp, *nextfp; 1567 struct socket *so; 1568 struct file **extra_ref, **fpp; 1569 int nunref, i; 1570 int nfiles_snap; 1571 int nfiles_slack = 20; 1572 1573 unp_taskcount++; 1574 unp_defer = 0; 1575 /* 1576 * Before going through all this, set all FDs to be NOT defered and 1577 * NOT externally accessible. 1578 */ 1579 sx_slock(&filelist_lock); 1580 LIST_FOREACH(fp, &filehead, f_list) 1581 fp->f_gcflag &= ~(FMARK|FDEFER); 1582 do { 1583 KASSERT(unp_defer >= 0, ("unp_gc: unp_defer %d", unp_defer)); 1584 LIST_FOREACH(fp, &filehead, f_list) { 1585 FILE_LOCK(fp); 1586 /* 1587 * If the file is not open, skip it -- could be a 1588 * file in the process of being opened, or in the 1589 * process of being closed. If the file is 1590 * "closing", it may have been marked for deferred 1591 * consideration. Clear the flag now if so. 1592 */ 1593 if (fp->f_count == 0) { 1594 if (fp->f_gcflag & FDEFER) 1595 unp_defer--; 1596 fp->f_gcflag &= ~(FMARK|FDEFER); 1597 FILE_UNLOCK(fp); 1598 continue; 1599 } 1600 /* 1601 * If we already marked it as 'defer' in a previous 1602 * pass, then try process it this time and un-mark 1603 * it. 1604 */ 1605 if (fp->f_gcflag & FDEFER) { 1606 fp->f_gcflag &= ~FDEFER; 1607 unp_defer--; 1608 } else { 1609 /* 1610 * if it's not defered, then check if it's 1611 * already marked.. if so skip it 1612 */ 1613 if (fp->f_gcflag & FMARK) { 1614 FILE_UNLOCK(fp); 1615 continue; 1616 } 1617 /* 1618 * If all references are from messages in 1619 * transit, then skip it. it's not externally 1620 * accessible. 1621 */ 1622 if (fp->f_count == fp->f_msgcount) { 1623 FILE_UNLOCK(fp); 1624 continue; 1625 } 1626 /* 1627 * If it got this far then it must be 1628 * externally accessible. 1629 */ 1630 fp->f_gcflag |= FMARK; 1631 } 1632 /* 1633 * Either it was defered, or it is externally 1634 * accessible and not already marked so. Now check 1635 * if it is possibly one of OUR sockets. 1636 */ 1637 if (fp->f_type != DTYPE_SOCKET || 1638 (so = fp->f_data) == NULL) { 1639 FILE_UNLOCK(fp); 1640 continue; 1641 } 1642 FILE_UNLOCK(fp); 1643 if (so->so_proto->pr_domain != &localdomain || 1644 (so->so_proto->pr_flags&PR_RIGHTS) == 0) 1645 continue; 1646 /* 1647 * So, Ok, it's one of our sockets and it IS 1648 * externally accessible (or was defered). Now we 1649 * look to see if we hold any file descriptors in its 1650 * message buffers. Follow those links and mark them 1651 * as accessible too. 1652 */ 1653 SOCKBUF_LOCK(&so->so_rcv); 1654 unp_scan(so->so_rcv.sb_mb, unp_mark); 1655 SOCKBUF_UNLOCK(&so->so_rcv); 1656 } 1657 } while (unp_defer); 1658 sx_sunlock(&filelist_lock); 1659 /* 1660 * XXXRW: The following comments need updating for a post-SMPng and 1661 * deferred unp_gc() world, but are still generally accurate. 1662 * 1663 * We grab an extra reference to each of the file table entries that 1664 * are not otherwise accessible and then free the rights that are 1665 * stored in messages on them. 1666 * 1667 * The bug in the orginal code is a little tricky, so I'll describe 1668 * what's wrong with it here. 1669 * 1670 * It is incorrect to simply unp_discard each entry for f_msgcount 1671 * times -- consider the case of sockets A and B that contain 1672 * references to each other. On a last close of some other socket, 1673 * we trigger a gc since the number of outstanding rights (unp_rights) 1674 * is non-zero. If during the sweep phase the gc code unp_discards, 1675 * we end up doing a (full) closef on the descriptor. A closef on A 1676 * results in the following chain. Closef calls soo_close, which 1677 * calls soclose. Soclose calls first (through the switch 1678 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply 1679 * returns because the previous instance had set unp_gcing, and we 1680 * return all the way back to soclose, which marks the socket with 1681 * SS_NOFDREF, and then calls sofree. Sofree calls sorflush to free 1682 * up the rights that are queued in messages on the socket A, i.e., 1683 * the reference on B. The sorflush calls via the dom_dispose switch 1684 * unp_dispose, which unp_scans with unp_discard. This second 1685 * instance of unp_discard just calls closef on B. 1686 * 1687 * Well, a similar chain occurs on B, resulting in a sorflush on B, 1688 * which results in another closef on A. Unfortunately, A is already 1689 * being closed, and the descriptor has already been marked with 1690 * SS_NOFDREF, and soclose panics at this point. 1691 * 1692 * Here, we first take an extra reference to each inaccessible 1693 * descriptor. Then, we call sorflush ourself, since we know it is a 1694 * Unix domain socket anyhow. After we destroy all the rights 1695 * carried in messages, we do a last closef to get rid of our extra 1696 * reference. This is the last close, and the unp_detach etc will 1697 * shut down the socket. 1698 * 1699 * 91/09/19, bsy@cs.cmu.edu 1700 */ 1701again: 1702 nfiles_snap = openfiles + nfiles_slack; /* some slack */ 1703 extra_ref = malloc(nfiles_snap * sizeof(struct file *), M_TEMP, 1704 M_WAITOK); 1705 sx_slock(&filelist_lock); 1706 if (nfiles_snap < openfiles) { 1707 sx_sunlock(&filelist_lock); 1708 free(extra_ref, M_TEMP); 1709 nfiles_slack += 20; 1710 goto again; 1711 } 1712 for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref; 1713 fp != NULL; fp = nextfp) { 1714 nextfp = LIST_NEXT(fp, f_list); 1715 FILE_LOCK(fp); 1716 /* 1717 * If it's not open, skip it 1718 */ 1719 if (fp->f_count == 0) { 1720 FILE_UNLOCK(fp); 1721 continue; 1722 } 1723 /* 1724 * If all refs are from msgs, and it's not marked accessible 1725 * then it must be referenced from some unreachable cycle of 1726 * (shut-down) FDs, so include it in our list of FDs to 1727 * remove. 1728 */ 1729 if (fp->f_count == fp->f_msgcount && !(fp->f_gcflag & FMARK)) { 1730 *fpp++ = fp; 1731 nunref++; 1732 fp->f_count++; 1733 } 1734 FILE_UNLOCK(fp); 1735 } 1736 sx_sunlock(&filelist_lock); 1737 /* 1738 * For each FD on our hit list, do the following two things: 1739 */ 1740 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1741 struct file *tfp = *fpp; 1742 FILE_LOCK(tfp); 1743 if (tfp->f_type == DTYPE_SOCKET && 1744 tfp->f_data != NULL) { 1745 FILE_UNLOCK(tfp); 1746 sorflush(tfp->f_data); 1747 } else { 1748 FILE_UNLOCK(tfp); 1749 } 1750 } 1751 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1752 closef(*fpp, (struct thread *) NULL); 1753 unp_recycled++; 1754 } 1755 free(extra_ref, M_TEMP); 1756} 1757 1758void 1759unp_dispose(struct mbuf *m) 1760{ 1761 1762 if (m) 1763 unp_scan(m, unp_discard); 1764} 1765 1766static int 1767unp_listen(struct socket *so, struct unpcb *unp, int backlog, 1768 struct thread *td) 1769{ 1770 int error; 1771 1772 UNP_LOCK_ASSERT(); 1773 1774 SOCK_LOCK(so); 1775 error = solisten_proto_check(so); 1776 if (error == 0) { 1777 cru2x(td->td_ucred, &unp->unp_peercred); 1778 unp->unp_flags |= UNP_HAVEPCCACHED; 1779 solisten_proto(so, backlog); 1780 } 1781 SOCK_UNLOCK(so); 1782 return (error); 1783} 1784 1785static void 1786unp_scan(struct mbuf *m0, void (*op)(struct file *)) 1787{ 1788 struct mbuf *m; 1789 struct file **rp; 1790 struct cmsghdr *cm; 1791 void *data; 1792 int i; 1793 socklen_t clen, datalen; 1794 int qfds; 1795 1796 while (m0 != NULL) { 1797 for (m = m0; m; m = m->m_next) { 1798 if (m->m_type != MT_CONTROL) 1799 continue; 1800 1801 cm = mtod(m, struct cmsghdr *); 1802 clen = m->m_len; 1803 1804 while (cm != NULL) { 1805 if (sizeof(*cm) > clen || cm->cmsg_len > clen) 1806 break; 1807 1808 data = CMSG_DATA(cm); 1809 datalen = (caddr_t)cm + cm->cmsg_len 1810 - (caddr_t)data; 1811 1812 if (cm->cmsg_level == SOL_SOCKET && 1813 cm->cmsg_type == SCM_RIGHTS) { 1814 qfds = datalen / sizeof (struct file *); 1815 rp = data; 1816 for (i = 0; i < qfds; i++) 1817 (*op)(*rp++); 1818 } 1819 1820 if (CMSG_SPACE(datalen) < clen) { 1821 clen -= CMSG_SPACE(datalen); 1822 cm = (struct cmsghdr *) 1823 ((caddr_t)cm + CMSG_SPACE(datalen)); 1824 } else { 1825 clen = 0; 1826 cm = NULL; 1827 } 1828 } 1829 } 1830 m0 = m0->m_act; 1831 } 1832} 1833 1834static void 1835unp_mark(struct file *fp) 1836{ 1837 if (fp->f_gcflag & FMARK) 1838 return; 1839 unp_defer++; 1840 fp->f_gcflag |= (FMARK|FDEFER); 1841} 1842 1843static void 1844unp_discard(struct file *fp) 1845{ 1846 UNP_LOCK(); 1847 FILE_LOCK(fp); 1848 fp->f_msgcount--; 1849 unp_rights--; 1850 FILE_UNLOCK(fp); 1851 UNP_UNLOCK(); 1852 (void) closef(fp, (struct thread *)NULL); 1853} 1854