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;
|
955#ifdef MAC 956 SOCK_LOCK(so); 957 mac_set_socket_peer_from_socket(so, so3); 958 mac_set_socket_peer_from_socket(so3, so); 959 SOCK_UNLOCK(so); 960#endif 961 962 so2 = so3; 963 } 964 error = unp_connect2(so, so2, PRU_CONNECT); 965bad2: 966 UNP_UNLOCK(); 967 mtx_lock(&Giant); 968bad: 969 mtx_assert(&Giant, MA_OWNED); 970 if (vp != NULL) 971 vput(vp); 972 mtx_unlock(&Giant); 973 free(sa, M_SONAME); 974 UNP_LOCK(); 975 return (error); 976} 977 978static int 979unp_connect2(struct socket *so, struct socket *so2, int req) 980{ 981 struct unpcb *unp = sotounpcb(so); 982 struct unpcb *unp2; 983 984 UNP_LOCK_ASSERT(); 985 986 if (so2->so_type != so->so_type) 987 return (EPROTOTYPE); 988 unp2 = sotounpcb(so2); 989 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL")); 990 unp->unp_conn = unp2; 991 switch (so->so_type) { 992 993 case SOCK_DGRAM: 994 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 995 soisconnected(so); 996 break; 997 998 case SOCK_STREAM: 999 unp2->unp_conn = unp; 1000 if (req == PRU_CONNECT && 1001 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 1002 soisconnecting(so); 1003 else 1004 soisconnected(so); 1005 soisconnected(so2); 1006 break; 1007 1008 default: 1009 panic("unp_connect2"); 1010 } 1011 return (0); 1012} 1013 1014static void 1015unp_disconnect(struct unpcb *unp) 1016{ 1017 struct unpcb *unp2 = unp->unp_conn; 1018 struct socket *so; 1019 1020 UNP_LOCK_ASSERT(); 1021 1022 if (unp2 == NULL) 1023 return; 1024 unp->unp_conn = NULL; 1025 switch (unp->unp_socket->so_type) { 1026 case SOCK_DGRAM: 1027 LIST_REMOVE(unp, unp_reflink); 1028 so = unp->unp_socket; 1029 SOCK_LOCK(so); 1030 so->so_state &= ~SS_ISCONNECTED; 1031 SOCK_UNLOCK(so); 1032 break; 1033 1034 case SOCK_STREAM: 1035 soisdisconnected(unp->unp_socket); 1036 unp2->unp_conn = NULL; 1037 soisdisconnected(unp2->unp_socket); 1038 break; 1039 } 1040} 1041 1042#ifdef notdef 1043void 1044unp_abort(struct unpcb *unp) 1045{ 1046 1047 unp_detach(unp); 1048 UNP_UNLOCK_ASSERT(); 1049} 1050#endif 1051 1052/* 1053 * unp_pcblist() assumes that UNIX domain socket memory is never reclaimed 1054 * by the zone (UMA_ZONE_NOFREE), and as such potentially stale pointers 1055 * are safe to reference. It first scans the list of struct unpcb's to 1056 * generate a pointer list, then it rescans its list one entry at a time to 1057 * externalize and copyout. It checks the generation number to see if a 1058 * struct unpcb has been reused, and will skip it if so. 1059 */ 1060static int 1061unp_pcblist(SYSCTL_HANDLER_ARGS) 1062{ 1063 int error, i, n; 1064 struct unpcb *unp, **unp_list; 1065 unp_gen_t gencnt; 1066 struct xunpgen *xug; 1067 struct unp_head *head; 1068 struct xunpcb *xu; 1069 1070 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead); 1071 1072 /* 1073 * The process of preparing the PCB list is too time-consuming and 1074 * resource-intensive to repeat twice on every request. 1075 */ 1076 if (req->oldptr == NULL) { 1077 n = unp_count; 1078 req->oldidx = 2 * (sizeof *xug) 1079 + (n + n/8) * sizeof(struct xunpcb); 1080 return (0); 1081 } 1082 1083 if (req->newptr != NULL) 1084 return (EPERM); 1085 1086 /* 1087 * OK, now we're committed to doing something. 1088 */ 1089 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK); 1090 UNP_LOCK(); 1091 gencnt = unp_gencnt; 1092 n = unp_count; 1093 UNP_UNLOCK(); 1094 1095 xug->xug_len = sizeof *xug; 1096 xug->xug_count = n; 1097 xug->xug_gen = gencnt; 1098 xug->xug_sogen = so_gencnt; 1099 error = SYSCTL_OUT(req, xug, sizeof *xug); 1100 if (error) { 1101 free(xug, M_TEMP); 1102 return (error); 1103 } 1104 1105 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 1106 1107 UNP_LOCK(); 1108 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 1109 unp = LIST_NEXT(unp, unp_link)) { 1110 if (unp->unp_gencnt <= gencnt) { 1111 if (cr_cansee(req->td->td_ucred, 1112 unp->unp_socket->so_cred)) 1113 continue; 1114 unp_list[i++] = unp; 1115 } 1116 } 1117 UNP_UNLOCK(); 1118 n = i; /* in case we lost some during malloc */ 1119 1120 error = 0; 1121 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO); 1122 for (i = 0; i < n; i++) { 1123 unp = unp_list[i]; 1124 if (unp->unp_gencnt <= gencnt) { 1125 xu->xu_len = sizeof *xu; 1126 xu->xu_unpp = unp; 1127 /* 1128 * XXX - need more locking here to protect against 1129 * connect/disconnect races for SMP. 1130 */ 1131 if (unp->unp_addr != NULL) 1132 bcopy(unp->unp_addr, &xu->xu_addr, 1133 unp->unp_addr->sun_len); 1134 if (unp->unp_conn != NULL && 1135 unp->unp_conn->unp_addr != NULL) 1136 bcopy(unp->unp_conn->unp_addr, 1137 &xu->xu_caddr, 1138 unp->unp_conn->unp_addr->sun_len); 1139 bcopy(unp, &xu->xu_unp, sizeof *unp); 1140 sotoxsocket(unp->unp_socket, &xu->xu_socket); 1141 error = SYSCTL_OUT(req, xu, sizeof *xu); 1142 } 1143 } 1144 free(xu, M_TEMP); 1145 if (!error) { 1146 /* 1147 * Give the user an updated idea of our state. 1148 * If the generation differs from what we told 1149 * her before, she knows that something happened 1150 * while we were processing this request, and it 1151 * might be necessary to retry. 1152 */ 1153 xug->xug_gen = unp_gencnt; 1154 xug->xug_sogen = so_gencnt; 1155 xug->xug_count = unp_count; 1156 error = SYSCTL_OUT(req, xug, sizeof *xug); 1157 } 1158 free(unp_list, M_TEMP); 1159 free(xug, M_TEMP); 1160 return (error); 1161} 1162 1163SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD, 1164 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 1165 "List of active local datagram sockets"); 1166SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD, 1167 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 1168 "List of active local stream sockets"); 1169 1170static void 1171unp_shutdown(struct unpcb *unp) 1172{ 1173 struct socket *so; 1174 1175 UNP_LOCK_ASSERT(); 1176 1177 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn && 1178 (so = unp->unp_conn->unp_socket)) 1179 socantrcvmore(so); 1180} 1181 1182static void 1183unp_drop(struct unpcb *unp, int errno) 1184{ 1185 struct socket *so = unp->unp_socket; 1186 1187 UNP_LOCK_ASSERT(); 1188 1189 so->so_error = errno; 1190 unp_disconnect(unp); 1191} 1192 1193#ifdef notdef 1194void 1195unp_drain(void) 1196{ 1197 1198} 1199#endif 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 1211 * unp_discard since it may end up 1212 * in unp_gc().. 1213 * 1214 * XXXRW: This is less true than it used to be. 1215 */ 1216 *rp++ = 0; 1217 unp_discard(fp); 1218 } 1219} 1220 1221int 1222unp_externalize(struct mbuf *control, struct mbuf **controlp) 1223{ 1224 struct thread *td = curthread; /* XXX */ 1225 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1226 int i; 1227 int *fdp; 1228 struct file **rp; 1229 struct file *fp; 1230 void *data; 1231 socklen_t clen = control->m_len, datalen; 1232 int error, newfds; 1233 int f; 1234 u_int newlen; 1235 1236 UNP_UNLOCK_ASSERT(); 1237 1238 error = 0; 1239 if (controlp != NULL) /* controlp == NULL => free control messages */ 1240 *controlp = NULL; 1241 1242 while (cm != NULL) { 1243 if (sizeof(*cm) > clen || cm->cmsg_len > clen) { 1244 error = EINVAL; 1245 break; 1246 } 1247 1248 data = CMSG_DATA(cm); 1249 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1250 1251 if (cm->cmsg_level == SOL_SOCKET 1252 && cm->cmsg_type == SCM_RIGHTS) { 1253 newfds = datalen / sizeof(struct file *); 1254 rp = data; 1255 1256 /* If we're not outputting the descriptors free them. */ 1257 if (error || controlp == NULL) { 1258 unp_freerights(rp, newfds); 1259 goto next; 1260 } 1261 FILEDESC_LOCK(td->td_proc->p_fd); 1262 /* if the new FD's will not fit free them. */ 1263 if (!fdavail(td, newfds)) { 1264 FILEDESC_UNLOCK(td->td_proc->p_fd); 1265 error = EMSGSIZE; 1266 unp_freerights(rp, newfds); 1267 goto next; 1268 } 1269 /* 1270 * now change each pointer to an fd in the global 1271 * table to an integer that is the index to the 1272 * local fd table entry that we set up to point 1273 * to the global one we are transferring. 1274 */ 1275 newlen = newfds * sizeof(int); 1276 *controlp = sbcreatecontrol(NULL, newlen, 1277 SCM_RIGHTS, SOL_SOCKET); 1278 if (*controlp == NULL) { 1279 FILEDESC_UNLOCK(td->td_proc->p_fd); 1280 error = E2BIG; 1281 unp_freerights(rp, newfds); 1282 goto next; 1283 } 1284 1285 fdp = (int *) 1286 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1287 for (i = 0; i < newfds; i++) { 1288 if (fdalloc(td, 0, &f)) 1289 panic("unp_externalize fdalloc failed"); 1290 fp = *rp++; 1291 td->td_proc->p_fd->fd_ofiles[f] = fp; 1292 FILE_LOCK(fp); 1293 fp->f_msgcount--; 1294 FILE_UNLOCK(fp); 1295 unp_rights--; 1296 *fdp++ = f; 1297 } 1298 FILEDESC_UNLOCK(td->td_proc->p_fd); 1299 } else { /* 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 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL, 1342 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 1343 if (unp_zone == NULL) 1344 panic("unp_init"); 1345 uma_zone_set_max(unp_zone, maxsockets); 1346 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change, 1347 NULL, EVENTHANDLER_PRI_ANY); 1348 LIST_INIT(&unp_dhead); 1349 LIST_INIT(&unp_shead); 1350 TASK_INIT(&unp_gc_task, 0, unp_gc, NULL); 1351 UNP_LOCK_INIT(); 1352} 1353 1354static int 1355unp_internalize(struct mbuf **controlp, struct thread *td) 1356{ 1357 struct mbuf *control = *controlp; 1358 struct proc *p = td->td_proc; 1359 struct filedesc *fdescp = p->p_fd; 1360 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1361 struct cmsgcred *cmcred; 1362 struct file **rp; 1363 struct file *fp; 1364 struct timeval *tv; 1365 int i, fd, *fdp; 1366 void *data; 1367 socklen_t clen = control->m_len, datalen; 1368 int error, oldfds; 1369 u_int newlen; 1370 1371 UNP_UNLOCK_ASSERT(); 1372 1373 error = 0; 1374 *controlp = NULL; 1375 1376 while (cm != NULL) { 1377 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET 1378 || cm->cmsg_len > clen) { 1379 error = EINVAL; 1380 goto out; 1381 } 1382 1383 data = CMSG_DATA(cm); 1384 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1385 1386 switch (cm->cmsg_type) { 1387 /* 1388 * Fill in credential information. 1389 */ 1390 case SCM_CREDS: 1391 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred), 1392 SCM_CREDS, SOL_SOCKET); 1393 if (*controlp == NULL) { 1394 error = ENOBUFS; 1395 goto out; 1396 } 1397 1398 cmcred = (struct cmsgcred *) 1399 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1400 cmcred->cmcred_pid = p->p_pid; 1401 cmcred->cmcred_uid = td->td_ucred->cr_ruid; 1402 cmcred->cmcred_gid = td->td_ucred->cr_rgid; 1403 cmcred->cmcred_euid = td->td_ucred->cr_uid; 1404 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups, 1405 CMGROUP_MAX); 1406 for (i = 0; i < cmcred->cmcred_ngroups; i++) 1407 cmcred->cmcred_groups[i] = 1408 td->td_ucred->cr_groups[i]; 1409 break; 1410 1411 case SCM_RIGHTS: 1412 oldfds = datalen / sizeof (int); 1413 /* 1414 * check that all the FDs passed in refer to legal files 1415 * If not, reject the entire operation. 1416 */ 1417 fdp = data; 1418 FILEDESC_LOCK(fdescp); 1419 for (i = 0; i < oldfds; i++) { 1420 fd = *fdp++; 1421 if ((unsigned)fd >= fdescp->fd_nfiles || 1422 fdescp->fd_ofiles[fd] == NULL) { 1423 FILEDESC_UNLOCK(fdescp); 1424 error = EBADF; 1425 goto out; 1426 } 1427 fp = fdescp->fd_ofiles[fd]; 1428 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { 1429 FILEDESC_UNLOCK(fdescp); 1430 error = EOPNOTSUPP; 1431 goto out; 1432 } 1433 1434 } 1435 /* 1436 * Now replace the integer FDs with pointers to 1437 * the associated global file table entry.. 1438 */ 1439 newlen = oldfds * sizeof(struct file *); 1440 *controlp = sbcreatecontrol(NULL, newlen, 1441 SCM_RIGHTS, SOL_SOCKET); 1442 if (*controlp == NULL) { 1443 FILEDESC_UNLOCK(fdescp); 1444 error = E2BIG; 1445 goto out; 1446 } 1447 1448 fdp = data; 1449 rp = (struct file **) 1450 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1451 for (i = 0; i < oldfds; i++) { 1452 fp = fdescp->fd_ofiles[*fdp++]; 1453 *rp++ = fp; 1454 FILE_LOCK(fp); 1455 fp->f_count++; 1456 fp->f_msgcount++; 1457 FILE_UNLOCK(fp); 1458 unp_rights++; 1459 } 1460 FILEDESC_UNLOCK(fdescp); 1461 break; 1462 1463 case SCM_TIMESTAMP: 1464 *controlp = sbcreatecontrol(NULL, sizeof(*tv), 1465 SCM_TIMESTAMP, SOL_SOCKET); 1466 if (*controlp == NULL) { 1467 error = ENOBUFS; 1468 goto out; 1469 } 1470 tv = (struct timeval *) 1471 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1472 microtime(tv); 1473 break; 1474 1475 default: 1476 error = EINVAL; 1477 goto out; 1478 } 1479 1480 controlp = &(*controlp)->m_next; 1481 1482 if (CMSG_SPACE(datalen) < clen) { 1483 clen -= CMSG_SPACE(datalen); 1484 cm = (struct cmsghdr *) 1485 ((caddr_t)cm + CMSG_SPACE(datalen)); 1486 } else { 1487 clen = 0; 1488 cm = NULL; 1489 } 1490 } 1491 1492out: 1493 m_freem(control); 1494 1495 return (error); 1496} 1497 1498struct mbuf * 1499unp_addsockcred(struct thread *td, struct mbuf *control) 1500{ 1501 struct mbuf *m, *n; 1502 struct sockcred *sc; 1503 int ngroups; 1504 int i; 1505 1506 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); 1507 1508 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET); 1509 if (m == NULL) 1510 return (control); 1511 m->m_next = NULL; 1512 1513 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *)); 1514 sc->sc_uid = td->td_ucred->cr_ruid; 1515 sc->sc_euid = td->td_ucred->cr_uid; 1516 sc->sc_gid = td->td_ucred->cr_rgid; 1517 sc->sc_egid = td->td_ucred->cr_gid; 1518 sc->sc_ngroups = ngroups; 1519 for (i = 0; i < sc->sc_ngroups; i++) 1520 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 1521 1522 /* 1523 * If a control message already exists, append us to the end. 1524 */ 1525 if (control != NULL) { 1526 for (n = control; n->m_next != NULL; n = n->m_next) 1527 ; 1528 n->m_next = m; 1529 } else 1530 control = m; 1531 1532 return (control); 1533} 1534 1535/* 1536 * unp_defer indicates whether additional work has been defered for a future 1537 * pass through unp_gc(). It is thread local and does not require explicit 1538 * synchronization. 1539 */ 1540static int unp_defer; 1541 1542static int unp_taskcount; 1543SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, ""); 1544 1545static int unp_recycled; 1546SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, ""); 1547 1548static void 1549unp_gc(__unused void *arg, int pending) 1550{ 1551 struct file *fp, *nextfp; 1552 struct socket *so; 1553 struct file **extra_ref, **fpp; 1554 int nunref, i; 1555 int nfiles_snap; 1556 int nfiles_slack = 20; 1557 1558 unp_taskcount++; 1559 unp_defer = 0; 1560 /* 1561 * before going through all this, set all FDs to 1562 * be NOT defered and NOT externally accessible 1563 */ 1564 sx_slock(&filelist_lock); 1565 LIST_FOREACH(fp, &filehead, f_list) 1566 fp->f_gcflag &= ~(FMARK|FDEFER); 1567 do { 1568 KASSERT(unp_defer >= 0, ("unp_gc: unp_defer %d", unp_defer)); 1569 LIST_FOREACH(fp, &filehead, f_list) { 1570 FILE_LOCK(fp); 1571 /* 1572 * If the file is not open, skip it -- could be a 1573 * file in the process of being opened, or in the 1574 * process of being closed. If the file is 1575 * "closing", it may have been marked for deferred 1576 * consideration. Clear the flag now if so. 1577 */ 1578 if (fp->f_count == 0) { 1579 if (fp->f_gcflag & FDEFER) 1580 unp_defer--; 1581 fp->f_gcflag &= ~(FMARK|FDEFER); 1582 FILE_UNLOCK(fp); 1583 continue; 1584 } 1585 /* 1586 * If we already marked it as 'defer' in a 1587 * previous pass, then try process it this time 1588 * and un-mark it 1589 */ 1590 if (fp->f_gcflag & FDEFER) { 1591 fp->f_gcflag &= ~FDEFER; 1592 unp_defer--; 1593 } else { 1594 /* 1595 * if it's not defered, then check if it's 1596 * already marked.. if so skip it 1597 */ 1598 if (fp->f_gcflag & FMARK) { 1599 FILE_UNLOCK(fp); 1600 continue; 1601 } 1602 /* 1603 * If all references are from messages 1604 * in transit, then skip it. it's not 1605 * externally accessible. 1606 */ 1607 if (fp->f_count == fp->f_msgcount) { 1608 FILE_UNLOCK(fp); 1609 continue; 1610 } 1611 /* 1612 * If it got this far then it must be 1613 * externally accessible. 1614 */ 1615 fp->f_gcflag |= FMARK; 1616 } 1617 /* 1618 * either it was defered, or it is externally 1619 * accessible and not already marked so. 1620 * Now check if it is possibly one of OUR sockets. 1621 */ 1622 if (fp->f_type != DTYPE_SOCKET || 1623 (so = fp->f_data) == NULL) { 1624 FILE_UNLOCK(fp); 1625 continue; 1626 } 1627 FILE_UNLOCK(fp); 1628 if (so->so_proto->pr_domain != &localdomain || 1629 (so->so_proto->pr_flags&PR_RIGHTS) == 0) 1630 continue; 1631 /* 1632 * So, Ok, it's one of our sockets and it IS externally 1633 * accessible (or was defered). Now we look 1634 * to see if we hold any file descriptors in its 1635 * message buffers. Follow those links and mark them 1636 * as accessible too. 1637 */ 1638 SOCKBUF_LOCK(&so->so_rcv); 1639 unp_scan(so->so_rcv.sb_mb, unp_mark); 1640 SOCKBUF_UNLOCK(&so->so_rcv); 1641 } 1642 } while (unp_defer); 1643 sx_sunlock(&filelist_lock); 1644 /* 1645 * XXXRW: The following comments need updating for a post-SMPng and 1646 * deferred unp_gc() world, but are still generally accurate. 1647 * 1648 * We grab an extra reference to each of the file table entries 1649 * that are not otherwise accessible and then free the rights 1650 * that are stored in messages on them. 1651 * 1652 * The bug in the orginal code is a little tricky, so I'll describe 1653 * what's wrong with it here. 1654 * 1655 * It is incorrect to simply unp_discard each entry for f_msgcount 1656 * times -- consider the case of sockets A and B that contain 1657 * references to each other. On a last close of some other socket, 1658 * we trigger a gc since the number of outstanding rights (unp_rights) 1659 * is non-zero. If during the sweep phase the gc code unp_discards, 1660 * we end up doing a (full) closef on the descriptor. A closef on A 1661 * results in the following chain. Closef calls soo_close, which 1662 * calls soclose. Soclose calls first (through the switch 1663 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply 1664 * returns because the previous instance had set unp_gcing, and 1665 * we return all the way back to soclose, which marks the socket 1666 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush 1667 * to free up the rights that are queued in messages on the socket A, 1668 * i.e., the reference on B. The sorflush calls via the dom_dispose 1669 * switch unp_dispose, which unp_scans with unp_discard. This second 1670 * instance of unp_discard just calls closef on B. 1671 * 1672 * Well, a similar chain occurs on B, resulting in a sorflush on B, 1673 * which results in another closef on A. Unfortunately, A is already 1674 * being closed, and the descriptor has already been marked with 1675 * SS_NOFDREF, and soclose panics at this point. 1676 * 1677 * Here, we first take an extra reference to each inaccessible 1678 * descriptor. Then, we call sorflush ourself, since we know 1679 * it is a Unix domain socket anyhow. After we destroy all the 1680 * rights carried in messages, we do a last closef to get rid 1681 * of our extra reference. This is the last close, and the 1682 * unp_detach etc will shut down the socket. 1683 * 1684 * 91/09/19, bsy@cs.cmu.edu 1685 */ 1686again: 1687 nfiles_snap = openfiles + nfiles_slack; /* some slack */ 1688 extra_ref = malloc(nfiles_snap * sizeof(struct file *), M_TEMP, 1689 M_WAITOK); 1690 sx_slock(&filelist_lock); 1691 if (nfiles_snap < openfiles) { 1692 sx_sunlock(&filelist_lock); 1693 free(extra_ref, M_TEMP); 1694 nfiles_slack += 20; 1695 goto again; 1696 } 1697 for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref; 1698 fp != NULL; fp = nextfp) { 1699 nextfp = LIST_NEXT(fp, f_list); 1700 FILE_LOCK(fp); 1701 /* 1702 * If it's not open, skip it 1703 */ 1704 if (fp->f_count == 0) { 1705 FILE_UNLOCK(fp); 1706 continue; 1707 } 1708 /* 1709 * If all refs are from msgs, and it's not marked accessible 1710 * then it must be referenced from some unreachable cycle 1711 * of (shut-down) FDs, so include it in our 1712 * list of FDs to remove 1713 */ 1714 if (fp->f_count == fp->f_msgcount && !(fp->f_gcflag & FMARK)) { 1715 *fpp++ = fp; 1716 nunref++; 1717 fp->f_count++; 1718 } 1719 FILE_UNLOCK(fp); 1720 } 1721 sx_sunlock(&filelist_lock); 1722 /* 1723 * for each FD on our hit list, do the following two things 1724 */ 1725 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1726 struct file *tfp = *fpp; 1727 FILE_LOCK(tfp); 1728 if (tfp->f_type == DTYPE_SOCKET && 1729 tfp->f_data != NULL) { 1730 FILE_UNLOCK(tfp); 1731 sorflush(tfp->f_data); 1732 } else { 1733 FILE_UNLOCK(tfp); 1734 } 1735 } 1736 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1737 closef(*fpp, (struct thread *) NULL); 1738 unp_recycled++; 1739 } 1740 free(extra_ref, M_TEMP); 1741} 1742 1743void 1744unp_dispose(struct mbuf *m) 1745{ 1746 1747 if (m) 1748 unp_scan(m, unp_discard); 1749} 1750 1751static int 1752unp_listen(struct socket *so, struct unpcb *unp, int backlog, 1753 struct thread *td) 1754{ 1755 int error; 1756 1757 UNP_LOCK_ASSERT(); 1758 1759 SOCK_LOCK(so); 1760 error = solisten_proto_check(so); 1761 if (error == 0) { 1762 cru2x(td->td_ucred, &unp->unp_peercred); 1763 unp->unp_flags |= UNP_HAVEPCCACHED; 1764 solisten_proto(so, backlog); 1765 } 1766 SOCK_UNLOCK(so); 1767 return (error); 1768} 1769 1770static void 1771unp_scan(struct mbuf *m0, void (*op)(struct file *)) 1772{ 1773 struct mbuf *m; 1774 struct file **rp; 1775 struct cmsghdr *cm; 1776 void *data; 1777 int i; 1778 socklen_t clen, datalen; 1779 int qfds; 1780 1781 while (m0 != NULL) { 1782 for (m = m0; m; m = m->m_next) { 1783 if (m->m_type != MT_CONTROL) 1784 continue; 1785 1786 cm = mtod(m, struct cmsghdr *); 1787 clen = m->m_len; 1788 1789 while (cm != NULL) { 1790 if (sizeof(*cm) > clen || cm->cmsg_len > clen) 1791 break; 1792 1793 data = CMSG_DATA(cm); 1794 datalen = (caddr_t)cm + cm->cmsg_len 1795 - (caddr_t)data; 1796 1797 if (cm->cmsg_level == SOL_SOCKET && 1798 cm->cmsg_type == SCM_RIGHTS) { 1799 qfds = datalen / sizeof (struct file *); 1800 rp = data; 1801 for (i = 0; i < qfds; i++) 1802 (*op)(*rp++); 1803 } 1804 1805 if (CMSG_SPACE(datalen) < clen) { 1806 clen -= CMSG_SPACE(datalen); 1807 cm = (struct cmsghdr *) 1808 ((caddr_t)cm + CMSG_SPACE(datalen)); 1809 } else { 1810 clen = 0; 1811 cm = NULL; 1812 } 1813 } 1814 } 1815 m0 = m0->m_act; 1816 } 1817} 1818 1819static void 1820unp_mark(struct file *fp) 1821{ 1822 if (fp->f_gcflag & FMARK) 1823 return; 1824 unp_defer++; 1825 fp->f_gcflag |= (FMARK|FDEFER); 1826} 1827 1828static void 1829unp_discard(struct file *fp) 1830{ 1831 UNP_LOCK(); 1832 FILE_LOCK(fp); 1833 fp->f_msgcount--; 1834 unp_rights--; 1835 FILE_UNLOCK(fp); 1836 UNP_UNLOCK(); 1837 (void) closef(fp, (struct thread *)NULL); 1838}
|