1 /* 2 * Copyright (c) 1982, 1986, 1989, 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94 30 * $FreeBSD: src/sys/kern/uipc_usrreq.c,v 1.54.2.10 2003/03/04 17:28:09 nectar Exp $ 31 */ 32 33 #include <sys/param.h> 34 #include <sys/systm.h> 35 #include <sys/kernel.h> 36 #include <sys/domain.h> 37 #include <sys/fcntl.h> 38 #include <sys/malloc.h> /* XXX must be before <sys/file.h> */ 39 #include <sys/proc.h> 40 #include <sys/file.h> 41 #include <sys/filedesc.h> 42 #include <sys/mbuf.h> 43 #include <sys/nlookup.h> 44 #include <sys/protosw.h> 45 #include <sys/socket.h> 46 #include <sys/socketvar.h> 47 #include <sys/resourcevar.h> 48 #include <sys/stat.h> 49 #include <sys/mount.h> 50 #include <sys/sysctl.h> 51 #include <sys/un.h> 52 #include <sys/unpcb.h> 53 #include <sys/vnode.h> 54 #include <sys/kern_syscall.h> 55 #include <sys/taskqueue.h> 56 57 #include <sys/file2.h> 58 #include <sys/spinlock2.h> 59 #include <sys/socketvar2.h> 60 #include <sys/msgport2.h> 61 62 /* 63 * Unix communications domain. 64 * 65 * TODO: 66 * RDM 67 * rethink name space problems 68 * need a proper out-of-band 69 * lock pushdown 70 * 71 * 72 * Unix domain sockets GC. 73 * 74 * It was originally designed to address following three cases: 75 * 1) Receiving unix domain socket can not accept the rights, e.g. 76 * when the so_rcv is full. 77 * 2) Caller of recvmsg(2) does not pass buffer to receive rights. 78 * 3) Unix domain sockets loop reference, e.g. s1 is on s2.so_rcv, 79 * while s2 on s1.so_rcv. 80 * 81 * Code under UNP_GC_ALLFILES is intended to address all above three 82 * cases. However, 1) was addressed a long time ago in uipc_send() 83 * (we inheritted the fix from FreeBSD when DragonFly forked). 2) 84 * was addressed in soreceive() by git-e62cfe62. 3) is the only 85 * case that needs GC. The new code (!UNP_GC_ALLFILES) addresses 86 * case 3) in the following way: 87 * - Record the struct file in unpcb, if the Unix domain socket is 88 * passed as one of the rights. 89 * - At GC time, only unpcbs are scanned, and only Unix domain sockets 90 * that are still used as rights are potential GC targets. 91 */ 92 93 #define UNP_DETACHED UNP_PRIVATE1 94 #define UNP_CONNECTING UNP_PRIVATE2 95 #define UNP_DROPPED UNP_PRIVATE3 96 #define UNP_MARKER UNP_PRIVATE4 97 98 #define UNPGC_REF 0x1 /* unpcb has external ref. */ 99 #define UNPGC_DEAD 0x2 /* unpcb might be dead. */ 100 #define UNPGC_SCANNED 0x4 /* Has been scanned. */ 101 102 #define UNP_GCFILE_MAX 256 103 104 /* For unp_internalize() and unp_externalize() */ 105 CTASSERT(sizeof(struct file *) >= sizeof(int)); 106 107 #define UNP_ISATTACHED(unp) \ 108 ((unp) != NULL && ((unp)->unp_flags & UNP_DETACHED) == 0) 109 110 #ifdef INVARIANTS 111 #define UNP_ASSERT_TOKEN_HELD(unp) \ 112 ASSERT_LWKT_TOKEN_HELD(lwkt_token_pool_lookup((unp))) 113 #else /* !INVARIANTS */ 114 #define UNP_ASSERT_TOKEN_HELD(unp) 115 #endif /* INVARIANTS */ 116 117 struct unp_defdiscard { 118 SLIST_ENTRY(unp_defdiscard) next; 119 struct file *fp; 120 }; 121 SLIST_HEAD(unp_defdiscard_list, unp_defdiscard); 122 123 TAILQ_HEAD(unpcb_qhead, unpcb); 124 struct unp_global_head { 125 struct unpcb_qhead list; 126 int count; 127 }; 128 129 static MALLOC_DEFINE(M_UNPCB, "unpcb", "unpcb struct"); 130 static unp_gen_t unp_gencnt; 131 132 static struct unp_global_head unp_stream_head; 133 static struct unp_global_head unp_dgram_head; 134 static struct unp_global_head unp_seqpkt_head; 135 136 static struct unp_global_head * const unp_heads[] = 137 { &unp_stream_head, &unp_dgram_head, &unp_seqpkt_head, NULL }; 138 139 static struct lwkt_token unp_token = LWKT_TOKEN_INITIALIZER(unp_token); 140 static struct taskqueue *unp_taskqueue; 141 142 static struct unp_defdiscard_list unp_defdiscard_head; 143 static struct spinlock unp_defdiscard_spin; 144 static struct task unp_defdiscard_task; 145 146 static struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL }; 147 148 static int unp_attach (struct socket *, struct pru_attach_info *); 149 static void unp_detach (struct unpcb *); 150 static int unp_bind (struct unpcb *,struct sockaddr *, struct thread *); 151 static int unp_connect (struct socket *,struct sockaddr *, 152 struct thread *); 153 static void unp_disconnect(struct unpcb *, int); 154 static void unp_shutdown (struct unpcb *); 155 static void unp_gc(void *, int); 156 #ifdef UNP_GC_ALLFILES 157 static int unp_gc_clearmarks(struct file *, void *); 158 static int unp_gc_checkmarks(struct file *, void *); 159 static int unp_gc_checkrefs(struct file *, void *); 160 static void unp_mark(struct file *, void *data); 161 #endif 162 static void unp_scan (struct mbuf *, void (*)(struct file *, void *), 163 void *data); 164 static void unp_discard (struct file *, void *); 165 static int unp_internalize (struct mbuf *, struct thread *); 166 static int unp_listen (struct unpcb *, struct thread *); 167 static void unp_fp_externalize(struct lwp *lp, struct file *fp, int fd, 168 int flags); 169 static int unp_find_lockref(struct sockaddr *nam, struct thread *td, 170 short type, struct unpcb **unp_ret); 171 static int unp_connect_pair(struct unpcb *unp, struct unpcb *unp2); 172 static void unp_drop(struct unpcb *unp, int error); 173 static void unp_defdiscard_taskfunc(void *, int); 174 175 static int unp_rights; /* file descriptors in flight */ 176 static struct lwkt_token unp_rights_token = 177 LWKT_TOKEN_INITIALIZER(unp_rights_token); 178 static struct task unp_gc_task; 179 static struct unpcb *unp_gc_marker; 180 181 SYSCTL_DECL(_net_local); 182 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, 183 "File descriptors in flight"); 184 185 /* 186 * SMP Considerations: 187 * 188 * Since unp_token will be automaticly released upon execution of 189 * blocking code, we need to reference unp_conn before any possible 190 * blocking code to prevent it from being ripped behind our back. 191 * 192 * Any adjustment to unp->unp_conn requires both the global unp_token 193 * AND the per-unp token (lwkt_token_pool_lookup(unp)) to be held. 194 * 195 * Any access to so_pcb to obtain unp requires the pool token for 196 * unp to be held. 197 */ 198 199 static __inline void 200 unp_reference(struct unpcb *unp) 201 { 202 /* 0->1 transition will not work */ 203 KKASSERT(unp->unp_refcnt > 0); 204 atomic_add_int(&unp->unp_refcnt, 1); 205 } 206 207 static __inline void 208 unp_free(struct unpcb *unp) 209 { 210 KKASSERT(unp->unp_refcnt > 0); 211 if (atomic_fetchadd_int(&unp->unp_refcnt, -1) == 1) 212 unp_detach(unp); 213 } 214 215 static __inline struct unpcb * 216 unp_getsocktoken(struct socket *so) 217 { 218 struct unpcb *unp; 219 220 /* 221 * The unp pointer is invalid until we verify that it is 222 * good by re-checking so_pcb AFTER obtaining the token. 223 */ 224 while ((unp = so->so_pcb) != NULL) { 225 lwkt_getpooltoken(unp); 226 if (unp == so->so_pcb) 227 break; 228 lwkt_relpooltoken(unp); 229 } 230 return unp; 231 } 232 233 static __inline void 234 unp_reltoken(struct unpcb *unp) 235 { 236 if (unp != NULL) 237 lwkt_relpooltoken(unp); 238 } 239 240 static __inline void 241 unp_setflags(struct unpcb *unp, int flags) 242 { 243 atomic_set_int(&unp->unp_flags, flags); 244 } 245 246 static __inline void 247 unp_clrflags(struct unpcb *unp, int flags) 248 { 249 atomic_clear_int(&unp->unp_flags, flags); 250 } 251 252 static __inline struct unp_global_head * 253 unp_globalhead(short type) 254 { 255 switch (type) { 256 case SOCK_STREAM: 257 return &unp_stream_head; 258 case SOCK_DGRAM: 259 return &unp_dgram_head; 260 case SOCK_SEQPACKET: 261 return &unp_seqpkt_head; 262 default: 263 panic("unknown socket type %d", type); 264 } 265 } 266 267 static __inline struct unpcb * 268 unp_fp2unpcb(struct file *fp) 269 { 270 struct socket *so; 271 272 if (fp->f_type != DTYPE_SOCKET) 273 return NULL; 274 275 so = fp->f_data; 276 if (so == NULL) 277 return NULL; 278 279 if (so->so_proto->pr_domain != &localdomain) 280 return NULL; 281 282 return so->so_pcb; 283 } 284 285 static __inline void 286 unp_add_right(struct file *fp) 287 { 288 struct unpcb *unp; 289 290 ASSERT_LWKT_TOKEN_HELD(&unp_rights_token); 291 KASSERT(fp->f_count > 0, ("invalid f_count %d", fp->f_count)); 292 293 unp = unp_fp2unpcb(fp); 294 if (unp != NULL) { 295 unp->unp_fp = fp; 296 unp->unp_msgcount++; 297 } 298 fp->f_msgcount++; 299 unp_rights++; 300 } 301 302 static __inline void 303 unp_del_right(struct file *fp) 304 { 305 struct unpcb *unp; 306 307 ASSERT_LWKT_TOKEN_HELD(&unp_rights_token); 308 KASSERT(fp->f_count > 0, ("invalid f_count %d", fp->f_count)); 309 310 unp = unp_fp2unpcb(fp); 311 if (unp != NULL) { 312 KASSERT(unp->unp_msgcount > 0, 313 ("invalid unp msgcount %d", unp->unp_msgcount)); 314 unp->unp_msgcount--; 315 if (unp->unp_msgcount == 0) 316 unp->unp_fp = NULL; 317 } 318 fp->f_msgcount--; 319 unp_rights--; 320 } 321 322 /* 323 * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort() 324 * will sofree() it when we return. 325 */ 326 static void 327 uipc_abort(netmsg_t msg) 328 { 329 struct unpcb *unp; 330 int error; 331 332 lwkt_gettoken(&unp_token); 333 unp = unp_getsocktoken(msg->base.nm_so); 334 335 if (UNP_ISATTACHED(unp)) { 336 unp_drop(unp, ECONNABORTED); 337 error = 0; 338 } else { 339 error = EINVAL; 340 } 341 342 unp_reltoken(unp); 343 lwkt_reltoken(&unp_token); 344 345 lwkt_replymsg(&msg->lmsg, error); 346 } 347 348 static void 349 uipc_accept(netmsg_t msg) 350 { 351 struct unpcb *unp; 352 int error; 353 354 lwkt_gettoken(&unp_token); 355 unp = unp_getsocktoken(msg->base.nm_so); 356 357 if (!UNP_ISATTACHED(unp)) { 358 error = EINVAL; 359 } else { 360 struct unpcb *unp2 = unp->unp_conn; 361 362 /* 363 * Pass back name of connected socket, 364 * if it was bound and we are still connected 365 * (our peer may have closed already!). 366 */ 367 if (unp2 && unp2->unp_addr) { 368 unp_reference(unp2); 369 *msg->accept.nm_nam = dup_sockaddr( 370 (struct sockaddr *)unp2->unp_addr); 371 unp_free(unp2); 372 } else { 373 *msg->accept.nm_nam = dup_sockaddr(&sun_noname); 374 } 375 error = 0; 376 } 377 378 unp_reltoken(unp); 379 lwkt_reltoken(&unp_token); 380 381 lwkt_replymsg(&msg->lmsg, error); 382 } 383 384 static void 385 uipc_attach(netmsg_t msg) 386 { 387 int error; 388 389 lwkt_gettoken(&unp_token); 390 391 KASSERT(msg->base.nm_so->so_pcb == NULL, ("double unp attach")); 392 error = unp_attach(msg->base.nm_so, msg->attach.nm_ai); 393 394 lwkt_reltoken(&unp_token); 395 lwkt_replymsg(&msg->lmsg, error); 396 } 397 398 static void 399 uipc_bind(netmsg_t msg) 400 { 401 struct unpcb *unp; 402 int error; 403 404 lwkt_gettoken(&unp_token); 405 unp = unp_getsocktoken(msg->base.nm_so); 406 407 if (UNP_ISATTACHED(unp)) 408 error = unp_bind(unp, msg->bind.nm_nam, msg->bind.nm_td); 409 else 410 error = EINVAL; 411 412 unp_reltoken(unp); 413 lwkt_reltoken(&unp_token); 414 415 lwkt_replymsg(&msg->lmsg, error); 416 } 417 418 static void 419 uipc_connect(netmsg_t msg) 420 { 421 int error; 422 423 error = unp_connect(msg->base.nm_so, msg->connect.nm_nam, 424 msg->connect.nm_td); 425 lwkt_replymsg(&msg->lmsg, error); 426 } 427 428 static void 429 uipc_connect2(netmsg_t msg) 430 { 431 int error; 432 433 error = unp_connect2(msg->connect2.nm_so1, msg->connect2.nm_so2, 434 msg->connect2.nm_cred); 435 lwkt_replymsg(&msg->lmsg, error); 436 } 437 438 /* control is EOPNOTSUPP */ 439 440 static void 441 uipc_detach(netmsg_t msg) 442 { 443 struct unpcb *unp; 444 int error; 445 446 lwkt_gettoken(&unp_token); 447 unp = unp_getsocktoken(msg->base.nm_so); 448 449 if (UNP_ISATTACHED(unp)) { 450 unp_drop(unp, 0); 451 error = 0; 452 } else { 453 error = EINVAL; 454 } 455 456 unp_reltoken(unp); 457 lwkt_reltoken(&unp_token); 458 459 lwkt_replymsg(&msg->lmsg, error); 460 } 461 462 static void 463 uipc_disconnect(netmsg_t msg) 464 { 465 struct unpcb *unp; 466 int error; 467 468 lwkt_gettoken(&unp_token); 469 unp = unp_getsocktoken(msg->base.nm_so); 470 471 if (UNP_ISATTACHED(unp)) { 472 unp_disconnect(unp, 0); 473 error = 0; 474 } else { 475 error = EINVAL; 476 } 477 478 unp_reltoken(unp); 479 lwkt_reltoken(&unp_token); 480 481 lwkt_replymsg(&msg->lmsg, error); 482 } 483 484 static void 485 uipc_listen(netmsg_t msg) 486 { 487 struct unpcb *unp; 488 int error; 489 490 lwkt_gettoken(&unp_token); 491 unp = unp_getsocktoken(msg->base.nm_so); 492 493 if (!UNP_ISATTACHED(unp) || unp->unp_vnode == NULL) 494 error = EINVAL; 495 else 496 error = unp_listen(unp, msg->listen.nm_td); 497 498 unp_reltoken(unp); 499 lwkt_reltoken(&unp_token); 500 501 lwkt_replymsg(&msg->lmsg, error); 502 } 503 504 static void 505 uipc_peeraddr(netmsg_t msg) 506 { 507 struct unpcb *unp; 508 int error; 509 510 lwkt_gettoken(&unp_token); 511 unp = unp_getsocktoken(msg->base.nm_so); 512 513 if (!UNP_ISATTACHED(unp)) { 514 error = EINVAL; 515 } else if (unp->unp_conn && unp->unp_conn->unp_addr) { 516 struct unpcb *unp2 = unp->unp_conn; 517 518 unp_reference(unp2); 519 *msg->peeraddr.nm_nam = dup_sockaddr( 520 (struct sockaddr *)unp2->unp_addr); 521 unp_free(unp2); 522 error = 0; 523 } else { 524 /* 525 * XXX: It seems that this test always fails even when 526 * connection is established. So, this else clause is 527 * added as workaround to return PF_LOCAL sockaddr. 528 */ 529 *msg->peeraddr.nm_nam = dup_sockaddr(&sun_noname); 530 error = 0; 531 } 532 533 unp_reltoken(unp); 534 lwkt_reltoken(&unp_token); 535 536 lwkt_replymsg(&msg->lmsg, error); 537 } 538 539 static void 540 uipc_rcvd(netmsg_t msg) 541 { 542 struct unpcb *unp, *unp2; 543 struct socket *so; 544 struct socket *so2; 545 int error; 546 547 /* 548 * so_pcb is only modified with both the global and the unp 549 * pool token held. 550 */ 551 so = msg->base.nm_so; 552 unp = unp_getsocktoken(so); 553 554 if (!UNP_ISATTACHED(unp)) { 555 error = EINVAL; 556 goto done; 557 } 558 559 switch (so->so_type) { 560 case SOCK_DGRAM: 561 panic("uipc_rcvd DGRAM?"); 562 /*NOTREACHED*/ 563 case SOCK_STREAM: 564 case SOCK_SEQPACKET: 565 if (unp->unp_conn == NULL) 566 break; 567 unp2 = unp->unp_conn; /* protected by pool token */ 568 569 /* 570 * Because we are transfering mbufs directly to the 571 * peer socket we have to use SSB_STOP on the sender 572 * to prevent it from building up infinite mbufs. 573 * 574 * As in several places in this module w ehave to ref unp2 575 * to ensure that it does not get ripped out from under us 576 * if we block on the so2 token or in sowwakeup(). 577 */ 578 so2 = unp2->unp_socket; 579 unp_reference(unp2); 580 lwkt_gettoken(&so2->so_rcv.ssb_token); 581 if (so->so_rcv.ssb_cc < so2->so_snd.ssb_hiwat && 582 so->so_rcv.ssb_mbcnt < so2->so_snd.ssb_mbmax 583 ) { 584 atomic_clear_int(&so2->so_snd.ssb_flags, SSB_STOP); 585 586 sowwakeup(so2); 587 } 588 lwkt_reltoken(&so2->so_rcv.ssb_token); 589 unp_free(unp2); 590 break; 591 default: 592 panic("uipc_rcvd unknown socktype"); 593 /*NOTREACHED*/ 594 } 595 error = 0; 596 done: 597 unp_reltoken(unp); 598 lwkt_replymsg(&msg->lmsg, error); 599 } 600 601 /* pru_rcvoob is EOPNOTSUPP */ 602 603 static void 604 uipc_send(netmsg_t msg) 605 { 606 struct unpcb *unp, *unp2; 607 struct socket *so; 608 struct socket *so2; 609 struct mbuf *control; 610 struct mbuf *m; 611 int error = 0; 612 613 so = msg->base.nm_so; 614 control = msg->send.nm_control; 615 m = msg->send.nm_m; 616 617 /* 618 * so_pcb is only modified with both the global and the unp 619 * pool token held. 620 */ 621 so = msg->base.nm_so; 622 unp = unp_getsocktoken(so); 623 624 if (!UNP_ISATTACHED(unp)) { 625 error = EINVAL; 626 goto release; 627 } 628 629 if (msg->send.nm_flags & PRUS_OOB) { 630 error = EOPNOTSUPP; 631 goto release; 632 } 633 634 wakeup_start_delayed(); 635 636 if (control && (error = unp_internalize(control, msg->send.nm_td))) 637 goto release; 638 639 switch (so->so_type) { 640 case SOCK_DGRAM: 641 { 642 struct sockaddr *from; 643 644 if (msg->send.nm_addr) { 645 if (unp->unp_conn) { 646 error = EISCONN; 647 break; 648 } 649 lwkt_gettoken(&unp_token); 650 error = unp_find_lockref(msg->send.nm_addr, 651 msg->send.nm_td, so->so_type, &unp2); 652 if (error) { 653 lwkt_reltoken(&unp_token); 654 break; 655 } 656 /* 657 * NOTE: 658 * unp2 is locked and referenced. 659 * 660 * We could unlock unp2 now, since it was checked 661 * and referenced. 662 */ 663 unp_reltoken(unp2); 664 lwkt_reltoken(&unp_token); 665 } else { 666 if (unp->unp_conn == NULL) { 667 error = ENOTCONN; 668 break; 669 } 670 unp2 = unp->unp_conn; 671 unp_reference(unp2); 672 } 673 /* NOTE: unp2 is referenced. */ 674 so2 = unp2->unp_socket; 675 676 /* 677 * Include creds if the receive side wants them, even if 678 * the send side did not send them. 679 */ 680 if (so2->so_options & SO_PASSCRED) { 681 struct mbuf **mp; 682 struct cmsghdr *cm; 683 struct cmsgcred cred; 684 struct mbuf *ncon; 685 686 mp = &control; 687 while ((ncon = *mp) != NULL) { 688 cm = mtod(ncon, struct cmsghdr *); 689 if (cm->cmsg_type == SCM_CREDS && 690 cm->cmsg_level == SOL_SOCKET) 691 break; 692 mp = &ncon->m_next; 693 } 694 if (ncon == NULL) { 695 ncon = sbcreatecontrol((caddr_t)&cred, 696 sizeof(cred), 697 SCM_CREDS, SOL_SOCKET); 698 unp_internalize(ncon, msg->send.nm_td); 699 *mp = ncon; 700 } 701 } 702 703 if (unp->unp_addr) 704 from = (struct sockaddr *)unp->unp_addr; 705 else 706 from = &sun_noname; 707 708 lwkt_gettoken(&so2->so_rcv.ssb_token); 709 if (ssb_appendaddr(&so2->so_rcv, from, m, control)) { 710 sorwakeup(so2); 711 m = NULL; 712 control = NULL; 713 } else { 714 error = ENOBUFS; 715 } 716 lwkt_reltoken(&so2->so_rcv.ssb_token); 717 718 unp_free(unp2); 719 break; 720 } 721 722 case SOCK_STREAM: 723 case SOCK_SEQPACKET: 724 /* Connect if not connected yet. */ 725 /* 726 * Note: A better implementation would complain 727 * if not equal to the peer's address. 728 */ 729 if (unp->unp_conn == NULL) { 730 if (msg->send.nm_addr) { 731 error = unp_connect(so, 732 msg->send.nm_addr, 733 msg->send.nm_td); 734 if (error) 735 break; /* XXX */ 736 } 737 /* 738 * NOTE: 739 * unp_conn still could be NULL, even if the 740 * above unp_connect() succeeds; since the 741 * current unp's token could be released due 742 * to blocking operations after unp_conn is 743 * assigned. 744 */ 745 if (unp->unp_conn == NULL) { 746 error = ENOTCONN; 747 break; 748 } 749 } 750 if (so->so_state & SS_CANTSENDMORE) { 751 error = EPIPE; 752 break; 753 } 754 755 unp2 = unp->unp_conn; 756 KASSERT(unp2 != NULL, ("unp is not connected")); 757 so2 = unp2->unp_socket; 758 759 unp_reference(unp2); 760 761 /* 762 * Send to paired receive port, and then reduce 763 * send buffer hiwater marks to maintain backpressure. 764 * Wake up readers. 765 */ 766 lwkt_gettoken(&so2->so_rcv.ssb_token); 767 if (control) { 768 if (ssb_appendcontrol(&so2->so_rcv, m, control)) { 769 control = NULL; 770 m = NULL; 771 } 772 } else if (so->so_type == SOCK_SEQPACKET) { 773 sbappendrecord(&so2->so_rcv.sb, m); 774 m = NULL; 775 } else { 776 sbappend(&so2->so_rcv.sb, m); 777 m = NULL; 778 } 779 780 /* 781 * Because we are transfering mbufs directly to the 782 * peer socket we have to use SSB_STOP on the sender 783 * to prevent it from building up infinite mbufs. 784 */ 785 if (so2->so_rcv.ssb_cc >= so->so_snd.ssb_hiwat || 786 so2->so_rcv.ssb_mbcnt >= so->so_snd.ssb_mbmax 787 ) { 788 atomic_set_int(&so->so_snd.ssb_flags, SSB_STOP); 789 } 790 lwkt_reltoken(&so2->so_rcv.ssb_token); 791 sorwakeup(so2); 792 793 unp_free(unp2); 794 break; 795 796 default: 797 panic("uipc_send unknown socktype"); 798 } 799 800 /* 801 * SEND_EOF is equivalent to a SEND followed by a SHUTDOWN. 802 */ 803 if (msg->send.nm_flags & PRUS_EOF) { 804 socantsendmore(so); 805 unp_shutdown(unp); 806 } 807 808 if (control && error != 0) 809 unp_dispose(control); 810 release: 811 unp_reltoken(unp); 812 wakeup_end_delayed(); 813 814 if (control) 815 m_freem(control); 816 if (m) 817 m_freem(m); 818 lwkt_replymsg(&msg->lmsg, error); 819 } 820 821 /* 822 * MPSAFE 823 */ 824 static void 825 uipc_sense(netmsg_t msg) 826 { 827 struct unpcb *unp; 828 struct socket *so; 829 struct stat *sb; 830 int error; 831 832 so = msg->base.nm_so; 833 sb = msg->sense.nm_stat; 834 835 /* 836 * so_pcb is only modified with both the global and the unp 837 * pool token held. 838 */ 839 unp = unp_getsocktoken(so); 840 841 if (!UNP_ISATTACHED(unp)) { 842 error = EINVAL; 843 goto done; 844 } 845 846 sb->st_blksize = so->so_snd.ssb_hiwat; 847 sb->st_dev = NOUDEV; 848 error = 0; 849 done: 850 unp_reltoken(unp); 851 lwkt_replymsg(&msg->lmsg, error); 852 } 853 854 static void 855 uipc_shutdown(netmsg_t msg) 856 { 857 struct socket *so; 858 struct unpcb *unp; 859 int error; 860 861 /* 862 * so_pcb is only modified with both the global and the unp 863 * pool token held. 864 */ 865 so = msg->base.nm_so; 866 unp = unp_getsocktoken(so); 867 868 if (UNP_ISATTACHED(unp)) { 869 socantsendmore(so); 870 unp_shutdown(unp); 871 error = 0; 872 } else { 873 error = EINVAL; 874 } 875 876 unp_reltoken(unp); 877 lwkt_replymsg(&msg->lmsg, error); 878 } 879 880 static void 881 uipc_sockaddr(netmsg_t msg) 882 { 883 struct unpcb *unp; 884 int error; 885 886 /* 887 * so_pcb is only modified with both the global and the unp 888 * pool token held. 889 */ 890 unp = unp_getsocktoken(msg->base.nm_so); 891 892 if (UNP_ISATTACHED(unp)) { 893 if (unp->unp_addr) { 894 *msg->sockaddr.nm_nam = 895 dup_sockaddr((struct sockaddr *)unp->unp_addr); 896 } 897 error = 0; 898 } else { 899 error = EINVAL; 900 } 901 902 unp_reltoken(unp); 903 lwkt_replymsg(&msg->lmsg, error); 904 } 905 906 struct pr_usrreqs uipc_usrreqs = { 907 .pru_abort = uipc_abort, 908 .pru_accept = uipc_accept, 909 .pru_attach = uipc_attach, 910 .pru_bind = uipc_bind, 911 .pru_connect = uipc_connect, 912 .pru_connect2 = uipc_connect2, 913 .pru_control = pr_generic_notsupp, 914 .pru_detach = uipc_detach, 915 .pru_disconnect = uipc_disconnect, 916 .pru_listen = uipc_listen, 917 .pru_peeraddr = uipc_peeraddr, 918 .pru_rcvd = uipc_rcvd, 919 .pru_rcvoob = pr_generic_notsupp, 920 .pru_send = uipc_send, 921 .pru_sense = uipc_sense, 922 .pru_shutdown = uipc_shutdown, 923 .pru_sockaddr = uipc_sockaddr, 924 .pru_sosend = sosend, 925 .pru_soreceive = soreceive 926 }; 927 928 void 929 uipc_ctloutput(netmsg_t msg) 930 { 931 struct socket *so; 932 struct sockopt *sopt; 933 struct unpcb *unp; 934 int error = 0; 935 936 so = msg->base.nm_so; 937 sopt = msg->ctloutput.nm_sopt; 938 939 lwkt_gettoken(&unp_token); 940 unp = unp_getsocktoken(so); 941 942 if (!UNP_ISATTACHED(unp)) { 943 error = EINVAL; 944 goto done; 945 } 946 947 switch (sopt->sopt_dir) { 948 case SOPT_GET: 949 switch (sopt->sopt_name) { 950 case LOCAL_PEERCRED: 951 if (unp->unp_flags & UNP_HAVEPC) 952 soopt_from_kbuf(sopt, &unp->unp_peercred, 953 sizeof(unp->unp_peercred)); 954 else { 955 if (so->so_type == SOCK_STREAM) 956 error = ENOTCONN; 957 else if (so->so_type == SOCK_SEQPACKET) 958 error = ENOTCONN; 959 else 960 error = EINVAL; 961 } 962 break; 963 default: 964 error = EOPNOTSUPP; 965 break; 966 } 967 break; 968 case SOPT_SET: 969 default: 970 error = EOPNOTSUPP; 971 break; 972 } 973 974 done: 975 unp_reltoken(unp); 976 lwkt_reltoken(&unp_token); 977 978 lwkt_replymsg(&msg->lmsg, error); 979 } 980 981 /* 982 * Both send and receive buffers are allocated PIPSIZ bytes of buffering 983 * for stream sockets, although the total for sender and receiver is 984 * actually only PIPSIZ. 985 * 986 * Datagram sockets really use the sendspace as the maximum datagram size, 987 * and don't really want to reserve the sendspace. Their recvspace should 988 * be large enough for at least one max-size datagram plus address. 989 * 990 * We want the local send/recv space to be significant larger then lo0's 991 * mtu of 16384. 992 * 993 * We no longer need to worry about avoiding the windows scaling option. 994 * Programs which use unix domain sockets expect larger defaults these days. 995 */ 996 #ifndef PIPSIZ 997 #define PIPSIZ 65536 998 #endif 999 static u_long unpst_sendspace = PIPSIZ; 1000 static u_long unpst_recvspace = PIPSIZ; 1001 static u_long unpdg_sendspace = PIPSIZ; /* really max datagram size */ 1002 static u_long unpdg_recvspace = PIPSIZ; 1003 static u_long unpsp_sendspace = PIPSIZ; /* really max datagram size */ 1004 static u_long unpsp_recvspace = PIPSIZ; 1005 1006 SYSCTL_DECL(_net_local_stream); 1007 SYSCTL_DECL(_net_local_dgram); 1008 SYSCTL_DECL(_net_local_seqpacket); 1009 1010 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 1011 &unpst_sendspace, 0, "Size of stream socket send buffer"); 1012 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 1013 &unpst_recvspace, 0, "Size of stream socket receive buffer"); 1014 1015 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 1016 &unpdg_sendspace, 0, "Max datagram socket size"); 1017 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 1018 &unpdg_recvspace, 0, "Size of datagram socket receive buffer"); 1019 1020 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW, 1021 &unpsp_sendspace, 0, "Default seqpacket send space."); 1022 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW, 1023 &unpsp_recvspace, 0, "Default seqpacket receive space."); 1024 1025 1026 static int 1027 unp_attach(struct socket *so, struct pru_attach_info *ai) 1028 { 1029 struct unp_global_head *head; 1030 struct unpcb *unp; 1031 int error; 1032 1033 lwkt_gettoken(&unp_token); 1034 1035 if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) { 1036 switch (so->so_type) { 1037 case SOCK_STREAM: 1038 error = soreserve(so, unpst_sendspace, unpst_recvspace, 1039 ai->sb_rlimit); 1040 break; 1041 case SOCK_DGRAM: 1042 error = soreserve(so, unpdg_sendspace, unpdg_recvspace, 1043 ai->sb_rlimit); 1044 break; 1045 case SOCK_SEQPACKET: 1046 error = soreserve(so, unpsp_sendspace, unpsp_recvspace, 1047 ai->sb_rlimit); 1048 break; 1049 default: 1050 panic("unp_attach"); 1051 } 1052 if (error) 1053 goto failed; 1054 } 1055 1056 /* 1057 * In order to support sendfile we have to set either SSB_STOPSUPP 1058 * or SSB_PREALLOC. Unix domain sockets use the SSB_STOP flow 1059 * control mechanism. 1060 */ 1061 if (so->so_type == SOCK_STREAM) { 1062 atomic_set_int(&so->so_rcv.ssb_flags, SSB_STOPSUPP); 1063 atomic_set_int(&so->so_snd.ssb_flags, SSB_STOPSUPP); 1064 } 1065 1066 unp = kmalloc(sizeof(*unp), M_UNPCB, M_WAITOK | M_ZERO | M_NULLOK); 1067 if (unp == NULL) { 1068 error = ENOBUFS; 1069 goto failed; 1070 } 1071 unp->unp_refcnt = 1; 1072 unp->unp_gencnt = ++unp_gencnt; 1073 LIST_INIT(&unp->unp_refs); 1074 unp->unp_socket = so; 1075 unp->unp_rvnode = ai->fd_rdir; /* jail cruft XXX JH */ 1076 so->so_pcb = (caddr_t)unp; 1077 soreference(so); 1078 1079 head = unp_globalhead(so->so_type); 1080 TAILQ_INSERT_TAIL(&head->list, unp, unp_link); 1081 head->count++; 1082 error = 0; 1083 failed: 1084 lwkt_reltoken(&unp_token); 1085 return error; 1086 } 1087 1088 static void 1089 unp_detach(struct unpcb *unp) 1090 { 1091 struct socket *so; 1092 1093 lwkt_gettoken(&unp_token); 1094 lwkt_getpooltoken(unp); 1095 1096 so = unp->unp_socket; 1097 1098 unp->unp_gencnt = ++unp_gencnt; 1099 if (unp->unp_vnode) { 1100 unp->unp_vnode->v_socket = NULL; 1101 vrele(unp->unp_vnode); 1102 unp->unp_vnode = NULL; 1103 } 1104 soisdisconnected(so); 1105 KKASSERT(so->so_pcb == unp); 1106 so->so_pcb = NULL; /* both tokens required */ 1107 unp->unp_socket = NULL; 1108 1109 lwkt_relpooltoken(unp); 1110 lwkt_reltoken(&unp_token); 1111 1112 sofree(so); 1113 1114 KASSERT(unp->unp_conn == NULL, ("unp is still connected")); 1115 KASSERT(LIST_EMPTY(&unp->unp_refs), ("unp still has references")); 1116 1117 if (unp->unp_addr) 1118 kfree(unp->unp_addr, M_SONAME); 1119 kfree(unp, M_UNPCB); 1120 1121 if (unp_rights) 1122 taskqueue_enqueue(unp_taskqueue, &unp_gc_task); 1123 } 1124 1125 static int 1126 unp_bind(struct unpcb *unp, struct sockaddr *nam, struct thread *td) 1127 { 1128 struct proc *p = td->td_proc; 1129 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 1130 struct vnode *vp; 1131 struct vattr vattr; 1132 int error, namelen; 1133 struct nlookupdata nd; 1134 char buf[SOCK_MAXADDRLEN]; 1135 1136 ASSERT_LWKT_TOKEN_HELD(&unp_token); 1137 UNP_ASSERT_TOKEN_HELD(unp); 1138 1139 if (unp->unp_vnode != NULL) 1140 return EINVAL; 1141 1142 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 1143 if (namelen <= 0) 1144 return EINVAL; 1145 strncpy(buf, soun->sun_path, namelen); 1146 buf[namelen] = 0; /* null-terminate the string */ 1147 error = nlookup_init(&nd, buf, UIO_SYSSPACE, 1148 NLC_LOCKVP | NLC_CREATE | NLC_REFDVP); 1149 if (error == 0) 1150 error = nlookup(&nd); 1151 if (error == 0 && nd.nl_nch.ncp->nc_vp != NULL) 1152 error = EADDRINUSE; 1153 if (error == 0 && nd.nl_dvp == NULL) /* e.g. bind <mountpt> */ 1154 error = EINVAL; 1155 if (error) 1156 goto done; 1157 1158 VATTR_NULL(&vattr); 1159 vattr.va_type = VSOCK; 1160 vattr.va_mode = (ACCESSPERMS & ~p->p_fd->fd_cmask); 1161 error = VOP_NCREATE(&nd.nl_nch, nd.nl_dvp, &vp, nd.nl_cred, &vattr); 1162 if (error == 0) { 1163 if (unp->unp_vnode == NULL) { 1164 vp->v_socket = unp->unp_socket; 1165 unp->unp_vnode = vp; 1166 unp->unp_addr = (struct sockaddr_un *)dup_sockaddr(nam); 1167 vn_unlock(vp); 1168 } else { 1169 vput(vp); /* late race */ 1170 error = EINVAL; 1171 } 1172 } 1173 done: 1174 nlookup_done(&nd); 1175 return (error); 1176 } 1177 1178 static int 1179 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1180 { 1181 struct unpcb *unp, *unp2; 1182 int error, flags = 0; 1183 1184 lwkt_gettoken(&unp_token); 1185 1186 unp = unp_getsocktoken(so); 1187 if (!UNP_ISATTACHED(unp)) { 1188 error = EINVAL; 1189 goto failed; 1190 } 1191 1192 if ((unp->unp_flags & UNP_CONNECTING) || unp->unp_conn != NULL) { 1193 error = EISCONN; 1194 goto failed; 1195 } 1196 1197 flags = UNP_CONNECTING; 1198 unp_setflags(unp, flags); 1199 1200 error = unp_find_lockref(nam, td, so->so_type, &unp2); 1201 if (error) 1202 goto failed; 1203 /* 1204 * NOTE: 1205 * unp2 is locked and referenced. 1206 */ 1207 1208 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 1209 struct socket *so2, *so3; 1210 struct unpcb *unp3; 1211 1212 so2 = unp2->unp_socket; 1213 if (!(so2->so_options & SO_ACCEPTCONN) || 1214 /* listen is not completed yet */ 1215 !(unp2->unp_flags & UNP_HAVEPCCACHED) || 1216 (so3 = sonewconn_faddr(so2, 0, NULL, 1217 TRUE /* keep ref */)) == NULL) { 1218 error = ECONNREFUSED; 1219 goto done; 1220 } 1221 /* so3 has a socket reference. */ 1222 1223 unp3 = unp_getsocktoken(so3); 1224 if (!UNP_ISATTACHED(unp3)) { 1225 unp_reltoken(unp3); 1226 /* 1227 * Already aborted; we only need to drop the 1228 * socket reference held by sonewconn_faddr(). 1229 */ 1230 sofree(so3); 1231 error = ECONNREFUSED; 1232 goto done; 1233 } 1234 unp_reference(unp3); 1235 /* 1236 * NOTE: 1237 * unp3 is locked and referenced. 1238 */ 1239 1240 /* 1241 * Release so3 socket reference held by sonewconn_faddr(). 1242 * Since we have referenced unp3, neither unp3 nor so3 will 1243 * be destroyed here. 1244 */ 1245 sofree(so3); 1246 1247 if (unp2->unp_addr != NULL) { 1248 unp3->unp_addr = (struct sockaddr_un *) 1249 dup_sockaddr((struct sockaddr *)unp2->unp_addr); 1250 } 1251 1252 /* 1253 * unp_peercred management: 1254 * 1255 * The connecter's (client's) credentials are copied 1256 * from its process structure at the time of connect() 1257 * (which is now). 1258 */ 1259 cru2x(td->td_proc->p_ucred, &unp3->unp_peercred); 1260 unp_setflags(unp3, UNP_HAVEPC); 1261 /* 1262 * The receiver's (server's) credentials are copied 1263 * from the unp_peercred member of socket on which the 1264 * former called listen(); unp_listen() cached that 1265 * process's credentials at that time so we can use 1266 * them now. 1267 */ 1268 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED, 1269 ("unp_connect: listener without cached peercred")); 1270 memcpy(&unp->unp_peercred, &unp2->unp_peercred, 1271 sizeof(unp->unp_peercred)); 1272 unp_setflags(unp, UNP_HAVEPC); 1273 1274 error = unp_connect_pair(unp, unp3); 1275 if (error) 1276 soabort_direct(so3); 1277 1278 /* Done with unp3 */ 1279 unp_free(unp3); 1280 unp_reltoken(unp3); 1281 } else { 1282 error = unp_connect_pair(unp, unp2); 1283 } 1284 done: 1285 unp_free(unp2); 1286 unp_reltoken(unp2); 1287 failed: 1288 if (flags) 1289 unp_clrflags(unp, flags); 1290 unp_reltoken(unp); 1291 1292 lwkt_reltoken(&unp_token); 1293 return (error); 1294 } 1295 1296 /* 1297 * Connect two unix domain sockets together. 1298 * 1299 * NOTE: Semantics for any change to unp_conn requires that the per-unp 1300 * pool token also be held. 1301 */ 1302 int 1303 unp_connect2(struct socket *so, struct socket *so2, struct ucred *cred) 1304 { 1305 struct unpcb *unp, *unp2; 1306 int error; 1307 1308 lwkt_gettoken(&unp_token); 1309 if (so2->so_type != so->so_type) { 1310 lwkt_reltoken(&unp_token); 1311 return (EPROTOTYPE); 1312 } 1313 unp = unp_getsocktoken(so); 1314 unp2 = unp_getsocktoken(so2); 1315 1316 cru2x(cred, &unp->unp_peercred); 1317 cru2x(cred, &unp2->unp_peercred); 1318 unp_setflags(unp, UNP_HAVEPC); 1319 unp_setflags(unp2, UNP_HAVEPC); 1320 1321 if (!UNP_ISATTACHED(unp)) { 1322 error = EINVAL; 1323 goto done; 1324 } 1325 if (!UNP_ISATTACHED(unp2)) { 1326 error = ECONNREFUSED; 1327 goto done; 1328 } 1329 1330 if (unp->unp_conn != NULL) { 1331 error = EISCONN; 1332 goto done; 1333 } 1334 if ((so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET) && 1335 unp2->unp_conn != NULL) { 1336 error = EISCONN; 1337 goto done; 1338 } 1339 1340 error = unp_connect_pair(unp, unp2); 1341 done: 1342 unp_reltoken(unp2); 1343 unp_reltoken(unp); 1344 lwkt_reltoken(&unp_token); 1345 return (error); 1346 } 1347 1348 /* 1349 * Disconnect a unix domain socket pair. 1350 * 1351 * NOTE: Semantics for any change to unp_conn requires that the per-unp 1352 * pool token also be held. 1353 */ 1354 static void 1355 unp_disconnect(struct unpcb *unp, int error) 1356 { 1357 struct socket *so = unp->unp_socket; 1358 struct unpcb *unp2; 1359 1360 ASSERT_LWKT_TOKEN_HELD(&unp_token); 1361 UNP_ASSERT_TOKEN_HELD(unp); 1362 1363 if (error) 1364 so->so_error = error; 1365 1366 while ((unp2 = unp->unp_conn) != NULL) { 1367 lwkt_getpooltoken(unp2); 1368 if (unp2 == unp->unp_conn) 1369 break; 1370 lwkt_relpooltoken(unp2); 1371 } 1372 if (unp2 == NULL) 1373 return; 1374 /* unp2 is locked. */ 1375 1376 KASSERT((unp2->unp_flags & UNP_DROPPED) == 0, ("unp2 was dropped")); 1377 1378 unp->unp_conn = NULL; 1379 1380 switch (so->so_type) { 1381 case SOCK_DGRAM: 1382 LIST_REMOVE(unp, unp_reflink); 1383 soclrstate(so, SS_ISCONNECTED); 1384 break; 1385 1386 case SOCK_STREAM: 1387 case SOCK_SEQPACKET: 1388 /* 1389 * Keep a reference before clearing the unp_conn 1390 * to avoid racing uipc_detach()/uipc_abort() in 1391 * other thread. 1392 */ 1393 unp_reference(unp2); 1394 KASSERT(unp2->unp_conn == unp, ("unp_conn mismatch")); 1395 unp2->unp_conn = NULL; 1396 1397 soisdisconnected(so); 1398 soisdisconnected(unp2->unp_socket); 1399 1400 unp_free(unp2); 1401 break; 1402 } 1403 1404 lwkt_relpooltoken(unp2); 1405 } 1406 1407 #ifdef notdef 1408 void 1409 unp_abort(struct unpcb *unp) 1410 { 1411 lwkt_gettoken(&unp_token); 1412 unp_free(unp); 1413 lwkt_reltoken(&unp_token); 1414 } 1415 #endif 1416 1417 static int 1418 prison_unpcb(struct thread *td, struct unpcb *unp) 1419 { 1420 struct proc *p; 1421 1422 if (td == NULL) 1423 return (0); 1424 if ((p = td->td_proc) == NULL) 1425 return (0); 1426 if (!p->p_ucred->cr_prison) 1427 return (0); 1428 if (p->p_fd->fd_rdir == unp->unp_rvnode) 1429 return (0); 1430 return (1); 1431 } 1432 1433 static int 1434 unp_pcblist(SYSCTL_HANDLER_ARGS) 1435 { 1436 struct unp_global_head *head = arg1; 1437 int error, i, n; 1438 struct unpcb *unp, *marker; 1439 1440 KKASSERT(curproc != NULL); 1441 1442 /* 1443 * The process of preparing the PCB list is too time-consuming and 1444 * resource-intensive to repeat twice on every request. 1445 */ 1446 if (req->oldptr == NULL) { 1447 n = head->count; 1448 req->oldidx = (n + n/8) * sizeof(struct xunpcb); 1449 return 0; 1450 } 1451 1452 if (req->newptr != NULL) 1453 return EPERM; 1454 1455 marker = kmalloc(sizeof(*marker), M_UNPCB, M_WAITOK | M_ZERO); 1456 marker->unp_flags |= UNP_MARKER; 1457 1458 lwkt_gettoken(&unp_token); 1459 1460 n = head->count; 1461 i = 0; 1462 error = 0; 1463 1464 TAILQ_INSERT_HEAD(&head->list, marker, unp_link); 1465 while ((unp = TAILQ_NEXT(marker, unp_link)) != NULL && i < n) { 1466 struct xunpcb xu; 1467 1468 TAILQ_REMOVE(&head->list, marker, unp_link); 1469 TAILQ_INSERT_AFTER(&head->list, unp, marker, unp_link); 1470 1471 if (unp->unp_flags & UNP_MARKER) 1472 continue; 1473 if (prison_unpcb(req->td, unp)) 1474 continue; 1475 1476 xu.xu_len = sizeof(xu); 1477 xu.xu_unpp = unp; 1478 1479 /* 1480 * NOTE: 1481 * unp->unp_addr and unp->unp_conn are protected by 1482 * unp_token. So if we want to get rid of unp_token 1483 * or reduce the coverage of unp_token, care must be 1484 * taken. 1485 */ 1486 if (unp->unp_addr) { 1487 bcopy(unp->unp_addr, &xu.xu_addr, 1488 unp->unp_addr->sun_len); 1489 } 1490 if (unp->unp_conn && unp->unp_conn->unp_addr) { 1491 bcopy(unp->unp_conn->unp_addr, 1492 &xu.xu_caddr, 1493 unp->unp_conn->unp_addr->sun_len); 1494 } 1495 bcopy(unp, &xu.xu_unp, sizeof(*unp)); 1496 sotoxsocket(unp->unp_socket, &xu.xu_socket); 1497 1498 /* NOTE: This could block and temporarily release unp_token */ 1499 error = SYSCTL_OUT(req, &xu, sizeof(xu)); 1500 if (error) 1501 break; 1502 ++i; 1503 } 1504 TAILQ_REMOVE(&head->list, marker, unp_link); 1505 1506 lwkt_reltoken(&unp_token); 1507 1508 kfree(marker, M_UNPCB); 1509 return error; 1510 } 1511 1512 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD, 1513 &unp_dgram_head, 0, unp_pcblist, "S,xunpcb", 1514 "List of active local datagram sockets"); 1515 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD, 1516 &unp_stream_head, 0, unp_pcblist, "S,xunpcb", 1517 "List of active local stream sockets"); 1518 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist, CTLFLAG_RD, 1519 &unp_seqpkt_head, 0, unp_pcblist, "S,xunpcb", 1520 "List of active local seqpacket sockets"); 1521 1522 static void 1523 unp_shutdown(struct unpcb *unp) 1524 { 1525 struct socket *so; 1526 1527 if ((unp->unp_socket->so_type == SOCK_STREAM || 1528 unp->unp_socket->so_type == SOCK_SEQPACKET) && 1529 unp->unp_conn != NULL && (so = unp->unp_conn->unp_socket)) { 1530 socantrcvmore(so); 1531 } 1532 } 1533 1534 #ifdef notdef 1535 void 1536 unp_drain(void) 1537 { 1538 lwkt_gettoken(&unp_token); 1539 lwkt_reltoken(&unp_token); 1540 } 1541 #endif 1542 1543 int 1544 unp_externalize(struct mbuf *rights, int flags) 1545 { 1546 struct thread *td = curthread; 1547 struct proc *p = td->td_proc; /* XXX */ 1548 struct lwp *lp = td->td_lwp; 1549 struct cmsghdr *cm = mtod(rights, struct cmsghdr *); 1550 int *fdp; 1551 int i; 1552 struct file **rp; 1553 struct file *fp; 1554 int newfds = (cm->cmsg_len - (CMSG_DATA(cm) - (u_char *)cm)) 1555 / sizeof(struct file *); 1556 int f; 1557 1558 lwkt_gettoken(&unp_rights_token); 1559 1560 /* 1561 * if the new FD's will not fit, then we free them all 1562 */ 1563 if (!fdavail(p, newfds)) { 1564 rp = (struct file **)CMSG_DATA(cm); 1565 for (i = 0; i < newfds; i++) { 1566 fp = *rp; 1567 /* 1568 * zero the pointer before calling unp_discard, 1569 * since it may end up in unp_gc().. 1570 */ 1571 *rp++ = NULL; 1572 unp_discard(fp, NULL); 1573 } 1574 lwkt_reltoken(&unp_rights_token); 1575 return (EMSGSIZE); 1576 } 1577 1578 /* 1579 * now change each pointer to an fd in the global table to 1580 * an integer that is the index to the local fd table entry 1581 * that we set up to point to the global one we are transferring. 1582 * Since the sizeof(struct file *) is bigger than or equal to 1583 * the sizeof(int), we do it in forward order. In that case, 1584 * an integer will always come in the same place or before its 1585 * corresponding struct file pointer. 1586 * 1587 * Hold revoke_token in 'shared' mode, so that we won't miss 1588 * the FREVOKED update on fps being externalized (fsetfd). 1589 */ 1590 lwkt_gettoken_shared(&revoke_token); 1591 fdp = (int *)CMSG_DATA(cm); 1592 rp = (struct file **)CMSG_DATA(cm); 1593 for (i = 0; i < newfds; i++) { 1594 if (fdalloc(p, 0, &f)) { 1595 int j; 1596 1597 /* 1598 * Previous fdavail() can't garantee 1599 * fdalloc() success due to SMP race. 1600 * Just clean up and return the same 1601 * error value as if fdavail() failed. 1602 */ 1603 lwkt_reltoken(&revoke_token); 1604 1605 /* Close externalized files */ 1606 for (j = 0; j < i; j++) 1607 kern_close(fdp[j]); 1608 /* Discard the rest of internal files */ 1609 for (; i < newfds; i++) 1610 unp_discard(rp[i], NULL); 1611 /* Wipe out the control message */ 1612 for (i = 0; i < newfds; i++) 1613 rp[i] = NULL; 1614 1615 lwkt_reltoken(&unp_rights_token); 1616 return (EMSGSIZE); 1617 } 1618 fp = rp[i]; 1619 unp_fp_externalize(lp, fp, f, flags); 1620 fdp[i] = f; 1621 } 1622 lwkt_reltoken(&revoke_token); 1623 1624 lwkt_reltoken(&unp_rights_token); 1625 1626 /* 1627 * Adjust length, in case sizeof(struct file *) and sizeof(int) 1628 * differs. 1629 */ 1630 cm->cmsg_len = CMSG_LEN(newfds * sizeof(int)); 1631 rights->m_len = cm->cmsg_len; 1632 1633 return (0); 1634 } 1635 1636 static void 1637 unp_fp_externalize(struct lwp *lp, struct file *fp, int fd, int flags) 1638 { 1639 if (lp) { 1640 struct filedesc *fdp = lp->lwp_proc->p_fd; 1641 1642 KKASSERT(fd >= 0); 1643 if (fp->f_flag & FREVOKED) { 1644 struct file *fx; 1645 int error; 1646 1647 kprintf("Warning: revoked fp exiting unix socket\n"); 1648 error = falloc(lp, &fx, NULL); 1649 if (error == 0) { 1650 if (flags & MSG_CMSG_CLOEXEC) 1651 fdp->fd_files[fd].fileflags |= UF_EXCLOSE; 1652 fsetfd(fdp, fx, fd); 1653 fdrop(fx); 1654 } else { 1655 fsetfd(fdp, NULL, fd); 1656 } 1657 } else { 1658 if (flags & MSG_CMSG_CLOEXEC) 1659 fdp->fd_files[fd].fileflags |= UF_EXCLOSE; 1660 fsetfd(fdp, fp, fd); 1661 } 1662 } 1663 unp_del_right(fp); 1664 fdrop(fp); 1665 } 1666 1667 void 1668 unp_init(void) 1669 { 1670 TAILQ_INIT(&unp_stream_head.list); 1671 TAILQ_INIT(&unp_dgram_head.list); 1672 TAILQ_INIT(&unp_seqpkt_head.list); 1673 1674 SLIST_INIT(&unp_defdiscard_head); 1675 spin_init(&unp_defdiscard_spin, "unpdisc"); 1676 TASK_INIT(&unp_defdiscard_task, 0, unp_defdiscard_taskfunc, NULL); 1677 1678 /* 1679 * This implies that only one gc can be in-progress at any 1680 * given moment. 1681 */ 1682 TASK_INIT(&unp_gc_task, 0, unp_gc, NULL); 1683 1684 unp_gc_marker = kmalloc(sizeof(*unp_gc_marker), M_UNPCB, 1685 M_WAITOK | M_ZERO); 1686 unp_gc_marker->unp_flags |= UNP_MARKER; 1687 1688 /* 1689 * Create taskqueue for defered discard, and stick it to 1690 * the last CPU. 1691 */ 1692 unp_taskqueue = taskqueue_create("unp_taskq", M_WAITOK, 1693 taskqueue_thread_enqueue, &unp_taskqueue); 1694 taskqueue_start_threads(&unp_taskqueue, 1, TDPRI_KERN_DAEMON, 1695 ncpus - 1, "unp taskq"); 1696 } 1697 1698 static int 1699 unp_internalize(struct mbuf *control, struct thread *td) 1700 { 1701 struct proc *p = td->td_proc; 1702 struct filedesc *fdescp; 1703 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1704 struct file **rp; 1705 struct file *fp; 1706 int i, fd, *fdp; 1707 struct cmsgcred *cmcred; 1708 int oldfds; 1709 u_int newlen; 1710 int error; 1711 1712 KKASSERT(p); 1713 1714 /* 1715 * Make sure the message is reasonable, and either CREDS or RIGHTS. 1716 * 1717 * NOTE: overall message length does not have to be aligned, but the 1718 * data start does. 1719 */ 1720 if ((cm->cmsg_type != SCM_RIGHTS && cm->cmsg_type != SCM_CREDS) || 1721 cm->cmsg_level != SOL_SOCKET || 1722 control->m_len < sizeof(*cm) || /* control too small */ 1723 cm->cmsg_len < sizeof(*cm) || /* cmsg_len too small */ 1724 cm->cmsg_len > control->m_len) { /* cmsg_len too big */ 1725 return EINVAL; 1726 } 1727 1728 /* 1729 * Fill in credential information. 1730 */ 1731 if (cm->cmsg_type == SCM_CREDS) { 1732 cmcred = (struct cmsgcred *)CMSG_DATA(cm); 1733 cmcred->cmcred_pid = p->p_pid; 1734 cmcred->cmcred_uid = p->p_ucred->cr_ruid; 1735 cmcred->cmcred_gid = p->p_ucred->cr_rgid; 1736 cmcred->cmcred_euid = p->p_ucred->cr_uid; 1737 cmcred->cmcred_ngroups = MIN(p->p_ucred->cr_ngroups, 1738 CMGROUP_MAX); 1739 for (i = 0; i < cmcred->cmcred_ngroups; i++) 1740 cmcred->cmcred_groups[i] = p->p_ucred->cr_groups[i]; 1741 return 0; 1742 } 1743 1744 /* 1745 * cmsghdr may not be aligned, do not allow calculation(s) to 1746 * go negative. 1747 * 1748 * Data must be aligned but the data length does not have to be. 1749 * 1750 * If there are multiple headers (XXX not supported) then the 1751 * next header will be aligned after the end of the possibly 1752 * unaligned data. 1753 */ 1754 if (cm->cmsg_len < CMSG_LEN(0)) { 1755 return EINVAL; 1756 } 1757 1758 oldfds = (cm->cmsg_len - CMSG_LEN(0)) / sizeof(int); 1759 1760 /* 1761 * Now replace the integer FDs with pointers to 1762 * the associated global file table entry.. 1763 * Allocate a bigger buffer as necessary. But if an cluster is not 1764 * enough, return E2BIG. 1765 */ 1766 newlen = CMSG_LEN(oldfds * sizeof(struct file *)); 1767 if (newlen > MCLBYTES) 1768 return E2BIG; 1769 if (newlen - control->m_len > M_TRAILINGSPACE(control)) { 1770 if (control->m_flags & M_EXT) 1771 return E2BIG; 1772 MCLGET(control, M_WAITOK); 1773 if (!(control->m_flags & M_EXT)) 1774 return ENOBUFS; 1775 1776 /* copy the data to the cluster */ 1777 memcpy(mtod(control, char *), cm, cm->cmsg_len); 1778 cm = mtod(control, struct cmsghdr *); 1779 } 1780 1781 lwkt_gettoken(&unp_rights_token); 1782 1783 fdescp = p->p_fd; 1784 spin_lock_shared(&fdescp->fd_spin); 1785 1786 /* 1787 * check that all the FDs passed in refer to legal OPEN files 1788 * If not, reject the entire operation. 1789 */ 1790 fdp = (int *)CMSG_DATA(cm); 1791 for (i = 0; i < oldfds; i++) { 1792 fd = *fdp++; 1793 if ((unsigned)fd >= fdescp->fd_nfiles || 1794 fdescp->fd_files[fd].fp == NULL) { 1795 error = EBADF; 1796 goto done; 1797 } 1798 if (fdescp->fd_files[fd].fp->f_type == DTYPE_KQUEUE) { 1799 error = EOPNOTSUPP; 1800 goto done; 1801 } 1802 } 1803 1804 /* 1805 * Adjust length, in case sizeof(struct file *) and sizeof(int) 1806 * differs. 1807 */ 1808 cm->cmsg_len = newlen; 1809 control->m_len = CMSG_ALIGN(newlen); 1810 1811 /* 1812 * Transform the file descriptors into struct file pointers. 1813 * Since the sizeof(struct file *) is bigger than or equal to 1814 * the sizeof(int), we do it in reverse order so that the int 1815 * won't get trashed until we're done. 1816 */ 1817 fdp = (int *)CMSG_DATA(cm) + oldfds - 1; 1818 rp = (struct file **)CMSG_DATA(cm) + oldfds - 1; 1819 for (i = 0; i < oldfds; i++) { 1820 fp = fdescp->fd_files[*fdp--].fp; 1821 *rp-- = fp; 1822 fhold(fp); 1823 unp_add_right(fp); 1824 } 1825 error = 0; 1826 done: 1827 spin_unlock_shared(&fdescp->fd_spin); 1828 lwkt_reltoken(&unp_rights_token); 1829 return error; 1830 } 1831 1832 #ifdef UNP_GC_ALLFILES 1833 1834 /* 1835 * Garbage collect in-transit file descriptors that get lost due to 1836 * loops (i.e. when a socket is sent to another process over itself, 1837 * and more complex situations). 1838 * 1839 * NOT MPSAFE - TODO socket flush code and maybe fdrop. Rest is MPSAFE. 1840 */ 1841 1842 struct unp_gc_info { 1843 struct file **extra_ref; 1844 struct file *locked_fp; 1845 int defer; 1846 int index; 1847 int maxindex; 1848 }; 1849 1850 static void 1851 unp_gc(void *arg __unused, int pending __unused) 1852 { 1853 struct unp_gc_info info; 1854 struct file **fpp; 1855 int i; 1856 1857 lwkt_gettoken(&unp_rights_token); 1858 1859 /* 1860 * Before going through all this, set all FDs to be NOT defered 1861 * and NOT externally accessible (not marked). During the scan 1862 * a fd can be marked externally accessible but we may or may not 1863 * be able to immediately process it (controlled by FDEFER). 1864 * 1865 * If we loop sleep a bit. The complexity of the topology can cause 1866 * multiple loops. Also failure to acquire the socket's so_rcv 1867 * token can cause us to loop. 1868 */ 1869 allfiles_scan_exclusive(unp_gc_clearmarks, NULL); 1870 do { 1871 info.defer = 0; 1872 allfiles_scan_exclusive(unp_gc_checkmarks, &info); 1873 if (info.defer) 1874 tsleep(&info, 0, "gcagain", 1); 1875 } while (info.defer); 1876 1877 /* 1878 * We grab an extra reference to each of the file table entries 1879 * that are not otherwise accessible and then free the rights 1880 * that are stored in messages on them. 1881 * 1882 * The bug in the orginal code is a little tricky, so I'll describe 1883 * what's wrong with it here. 1884 * 1885 * It is incorrect to simply unp_discard each entry for f_msgcount 1886 * times -- consider the case of sockets A and B that contain 1887 * references to each other. On a last close of some other socket, 1888 * we trigger a gc since the number of outstanding rights (unp_rights) 1889 * is non-zero. If during the sweep phase the gc code unp_discards, 1890 * we end up doing a (full) fdrop on the descriptor. A fdrop on A 1891 * results in the following chain. Closef calls soo_close, which 1892 * calls soclose. Soclose calls first (through the switch 1893 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply 1894 * returns because the previous instance had set unp_gcing, and 1895 * we return all the way back to soclose, which marks the socket 1896 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush 1897 * to free up the rights that are queued in messages on the socket A, 1898 * i.e., the reference on B. The sorflush calls via the dom_dispose 1899 * switch unp_dispose, which unp_scans with unp_discard. This second 1900 * instance of unp_discard just calls fdrop on B. 1901 * 1902 * Well, a similar chain occurs on B, resulting in a sorflush on B, 1903 * which results in another fdrop on A. Unfortunately, A is already 1904 * being closed, and the descriptor has already been marked with 1905 * SS_NOFDREF, and soclose panics at this point. 1906 * 1907 * Here, we first take an extra reference to each inaccessible 1908 * descriptor. Then, we call sorflush ourself, since we know 1909 * it is a Unix domain socket anyhow. After we destroy all the 1910 * rights carried in messages, we do a last fdrop to get rid 1911 * of our extra reference. This is the last close, and the 1912 * unp_detach etc will shut down the socket. 1913 * 1914 * 91/09/19, bsy@cs.cmu.edu 1915 */ 1916 info.extra_ref = kmalloc(256 * sizeof(struct file *), M_FILE, M_WAITOK); 1917 info.maxindex = 256; 1918 1919 do { 1920 /* 1921 * Look for matches 1922 */ 1923 info.index = 0; 1924 allfiles_scan_exclusive(unp_gc_checkrefs, &info); 1925 1926 /* 1927 * For each FD on our hit list, do the following two things 1928 */ 1929 for (i = info.index, fpp = info.extra_ref; --i >= 0; ++fpp) { 1930 struct file *tfp = *fpp; 1931 if (tfp->f_type == DTYPE_SOCKET && tfp->f_data != NULL) 1932 sorflush((struct socket *)(tfp->f_data)); 1933 } 1934 for (i = info.index, fpp = info.extra_ref; --i >= 0; ++fpp) 1935 fdrop(*fpp); 1936 } while (info.index == info.maxindex); 1937 1938 kfree((caddr_t)info.extra_ref, M_FILE); 1939 1940 lwkt_reltoken(&unp_rights_token); 1941 } 1942 1943 /* 1944 * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry 1945 */ 1946 static int 1947 unp_gc_checkrefs(struct file *fp, void *data) 1948 { 1949 struct unp_gc_info *info = data; 1950 1951 if (fp->f_count == 0) 1952 return(0); 1953 if (info->index == info->maxindex) 1954 return(-1); 1955 1956 /* 1957 * If all refs are from msgs, and it's not marked accessible 1958 * then it must be referenced from some unreachable cycle 1959 * of (shut-down) FDs, so include it in our 1960 * list of FDs to remove 1961 */ 1962 if (fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) { 1963 info->extra_ref[info->index++] = fp; 1964 fhold(fp); 1965 } 1966 return(0); 1967 } 1968 1969 /* 1970 * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry 1971 */ 1972 static int 1973 unp_gc_clearmarks(struct file *fp, void *data __unused) 1974 { 1975 atomic_clear_int(&fp->f_flag, FMARK | FDEFER); 1976 return(0); 1977 } 1978 1979 /* 1980 * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry 1981 */ 1982 static int 1983 unp_gc_checkmarks(struct file *fp, void *data) 1984 { 1985 struct unp_gc_info *info = data; 1986 struct socket *so; 1987 1988 /* 1989 * If the file is not open, skip it. Make sure it isn't marked 1990 * defered or we could loop forever, in case we somehow race 1991 * something. 1992 */ 1993 if (fp->f_count == 0) { 1994 if (fp->f_flag & FDEFER) 1995 atomic_clear_int(&fp->f_flag, FDEFER); 1996 return(0); 1997 } 1998 /* 1999 * If we already marked it as 'defer' in a 2000 * previous pass, then try process it this time 2001 * and un-mark it 2002 */ 2003 if (fp->f_flag & FDEFER) { 2004 atomic_clear_int(&fp->f_flag, FDEFER); 2005 } else { 2006 /* 2007 * if it's not defered, then check if it's 2008 * already marked.. if so skip it 2009 */ 2010 if (fp->f_flag & FMARK) 2011 return(0); 2012 /* 2013 * If all references are from messages 2014 * in transit, then skip it. it's not 2015 * externally accessible. 2016 */ 2017 if (fp->f_count == fp->f_msgcount) 2018 return(0); 2019 /* 2020 * If it got this far then it must be 2021 * externally accessible. 2022 */ 2023 atomic_set_int(&fp->f_flag, FMARK); 2024 } 2025 2026 /* 2027 * either it was defered, or it is externally 2028 * accessible and not already marked so. 2029 * Now check if it is possibly one of OUR sockets. 2030 */ 2031 if (fp->f_type != DTYPE_SOCKET || 2032 (so = (struct socket *)fp->f_data) == NULL) { 2033 return(0); 2034 } 2035 if (so->so_proto->pr_domain != &localdomain || 2036 !(so->so_proto->pr_flags & PR_RIGHTS)) { 2037 return(0); 2038 } 2039 2040 /* 2041 * So, Ok, it's one of our sockets and it IS externally accessible 2042 * (or was defered). Now we look to see if we hold any file 2043 * descriptors in its message buffers. Follow those links and mark 2044 * them as accessible too. 2045 * 2046 * We are holding multiple spinlocks here, if we cannot get the 2047 * token non-blocking defer until the next loop. 2048 */ 2049 info->locked_fp = fp; 2050 if (lwkt_trytoken(&so->so_rcv.ssb_token)) { 2051 unp_scan(so->so_rcv.ssb_mb, unp_mark, info); 2052 lwkt_reltoken(&so->so_rcv.ssb_token); 2053 } else { 2054 atomic_set_int(&fp->f_flag, FDEFER); 2055 ++info->defer; 2056 } 2057 return (0); 2058 } 2059 2060 /* 2061 * Mark visibility. info->defer is recalculated on every pass. 2062 */ 2063 static void 2064 unp_mark(struct file *fp, void *data) 2065 { 2066 struct unp_gc_info *info = data; 2067 2068 if ((fp->f_flag & FMARK) == 0) { 2069 ++info->defer; 2070 atomic_set_int(&fp->f_flag, FMARK | FDEFER); 2071 } else if (fp->f_flag & FDEFER) { 2072 ++info->defer; 2073 } 2074 } 2075 2076 #else /* !UNP_GC_ALLFILES */ 2077 2078 /* 2079 * They are thread local and do not require explicit synchronization. 2080 */ 2081 static int unp_marked; 2082 static int unp_unreachable; 2083 2084 static void 2085 unp_accessable(struct file *fp, void *data __unused) 2086 { 2087 struct unpcb *unp; 2088 2089 if ((unp = unp_fp2unpcb(fp)) == NULL) 2090 return; 2091 if (unp->unp_gcflags & UNPGC_REF) 2092 return; 2093 unp->unp_gcflags &= ~UNPGC_DEAD; 2094 unp->unp_gcflags |= UNPGC_REF; 2095 unp_marked++; 2096 } 2097 2098 static void 2099 unp_gc_process(struct unpcb *unp) 2100 { 2101 struct file *fp; 2102 2103 /* Already processed. */ 2104 if (unp->unp_gcflags & UNPGC_SCANNED) 2105 return; 2106 fp = unp->unp_fp; 2107 2108 /* 2109 * Check for a socket potentially in a cycle. It must be in a 2110 * queue as indicated by msgcount, and this must equal the file 2111 * reference count. Note that when msgcount is 0 the file is NULL. 2112 */ 2113 if ((unp->unp_gcflags & UNPGC_REF) == 0 && fp && 2114 unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) { 2115 unp->unp_gcflags |= UNPGC_DEAD; 2116 unp_unreachable++; 2117 return; 2118 } 2119 2120 /* 2121 * Mark all sockets we reference with RIGHTS. 2122 */ 2123 if (UNP_ISATTACHED(unp)) { 2124 struct signalsockbuf *ssb = &unp->unp_socket->so_rcv; 2125 2126 unp_reference(unp); 2127 lwkt_gettoken(&ssb->ssb_token); 2128 /* 2129 * unp_token would be temporarily dropped, if getting 2130 * so_rcv token blocks, so we need to check unp state 2131 * here again. 2132 */ 2133 if (UNP_ISATTACHED(unp)) 2134 unp_scan(ssb->ssb_mb, unp_accessable, NULL); 2135 lwkt_reltoken(&ssb->ssb_token); 2136 unp->unp_gcflags |= UNPGC_SCANNED; 2137 unp_free(unp); 2138 } else { 2139 unp->unp_gcflags |= UNPGC_SCANNED; 2140 } 2141 } 2142 2143 static void 2144 unp_gc(void *arg __unused, int pending __unused) 2145 { 2146 struct unp_global_head *head; 2147 int h, filemax, fileidx, filetot; 2148 struct file **unref; 2149 struct unpcb *unp; 2150 2151 lwkt_gettoken(&unp_rights_token); 2152 lwkt_gettoken(&unp_token); 2153 2154 /* 2155 * First clear all gc flags from previous runs. 2156 */ 2157 for (h = 0; unp_heads[h] != NULL; ++h) { 2158 /* 2159 * NOTE: This loop does not block, so it is safe 2160 * to use TAILQ_FOREACH here. 2161 */ 2162 head = unp_heads[h]; 2163 TAILQ_FOREACH(unp, &head->list, unp_link) 2164 unp->unp_gcflags = 0; 2165 } 2166 2167 /* 2168 * Scan marking all reachable sockets with UNPGC_REF. Once a socket 2169 * is reachable all of the sockets it references are reachable. 2170 * Stop the scan once we do a complete loop without discovering 2171 * a new reachable socket. 2172 */ 2173 do { 2174 unp_unreachable = 0; 2175 unp_marked = 0; 2176 for (h = 0; unp_heads[h] != NULL; ++h) { 2177 head = unp_heads[h]; 2178 TAILQ_INSERT_HEAD(&head->list, unp_gc_marker, unp_link); 2179 while ((unp = TAILQ_NEXT(unp_gc_marker, unp_link)) 2180 != NULL) { 2181 TAILQ_REMOVE(&head->list, unp_gc_marker, 2182 unp_link); 2183 TAILQ_INSERT_AFTER(&head->list, unp, 2184 unp_gc_marker, unp_link); 2185 2186 if (unp->unp_flags & UNP_MARKER) 2187 continue; 2188 unp_gc_process(unp); 2189 } 2190 TAILQ_REMOVE(&head->list, unp_gc_marker, unp_link); 2191 } 2192 } while (unp_marked); 2193 2194 if (unp_unreachable == 0) 2195 goto done; 2196 2197 /* 2198 * We grab an extra reference to each of the file table entries 2199 * that are not otherwise accessible and then free the rights 2200 * that are stored in messages on them. 2201 * 2202 * The bug in the orginal code is a little tricky, so I'll describe 2203 * what's wrong with it here. 2204 * 2205 * It is incorrect to simply unp_discard each entry for f_msgcount 2206 * times -- consider the case of sockets A and B that contain 2207 * references to each other. On a last close of some other socket, 2208 * we trigger a gc since the number of outstanding rights (unp_rights) 2209 * is non-zero. If during the sweep phase the gc code unp_discards, 2210 * we end up doing a (full) fdrop on the descriptor. A fdrop on A 2211 * results in the following chain. Closef calls soo_close, which 2212 * calls soclose. Soclose calls first (through the switch 2213 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply 2214 * returns because the previous instance had set unp_gcing, and 2215 * we return all the way back to soclose, which marks the socket 2216 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush 2217 * to free up the rights that are queued in messages on the socket A, 2218 * i.e., the reference on B. The sorflush calls via the dom_dispose 2219 * switch unp_dispose, which unp_scans with unp_discard. This second 2220 * instance of unp_discard just calls fdrop on B. 2221 * 2222 * Well, a similar chain occurs on B, resulting in a sorflush on B, 2223 * which results in another fdrop on A. Unfortunately, A is already 2224 * being closed, and the descriptor has already been marked with 2225 * SS_NOFDREF, and soclose panics at this point. 2226 * 2227 * Here, we first take an extra reference to each inaccessible 2228 * descriptor. Then, we call sorflush ourself, since we know 2229 * it is a Unix domain socket anyhow. After we destroy all the 2230 * rights carried in messages, we do a last fdrop to get rid 2231 * of our extra reference. This is the last close, and the 2232 * unp_detach etc will shut down the socket. 2233 * 2234 * 91/09/19, bsy@cs.cmu.edu 2235 */ 2236 2237 filemax = unp_unreachable; 2238 if (filemax > UNP_GCFILE_MAX) 2239 filemax = UNP_GCFILE_MAX; 2240 unref = kmalloc(filemax * sizeof(struct file *), M_TEMP, M_WAITOK); 2241 2242 filetot = 0; 2243 do { 2244 int i; 2245 2246 /* 2247 * Iterate looking for sockets which have been specifically 2248 * marked as as unreachable and store them locally. 2249 */ 2250 fileidx = 0; 2251 for (h = 0; unp_heads[h] != NULL; ++h) { 2252 /* 2253 * NOTE: This loop does not block, so it is safe 2254 * to use TAILQ_FOREACH here. 2255 */ 2256 head = unp_heads[h]; 2257 TAILQ_FOREACH(unp, &head->list, unp_link) { 2258 struct file *fp; 2259 2260 if ((unp->unp_gcflags & UNPGC_DEAD) == 0) 2261 continue; 2262 unp->unp_gcflags &= ~UNPGC_DEAD; 2263 2264 fp = unp->unp_fp; 2265 if (unp->unp_msgcount == 0 || fp == NULL || 2266 fp->f_count != unp->unp_msgcount) 2267 continue; 2268 fhold(fp); 2269 2270 KASSERT(fileidx < filemax, 2271 ("invalid fileidx %d, filemax %d", 2272 fileidx, filemax)); 2273 unref[fileidx++] = fp; 2274 2275 KASSERT(filetot < unp_unreachable, 2276 ("invalid filetot %d and " 2277 "unp_unreachable %d", 2278 filetot, unp_unreachable)); 2279 ++filetot; 2280 2281 if (fileidx == filemax || 2282 filetot == unp_unreachable) 2283 goto dogc; 2284 } 2285 } 2286 dogc: 2287 /* 2288 * For each Unix domain socket on our hit list, do the 2289 * following two things. 2290 */ 2291 for (i = 0; i < fileidx; ++i) 2292 sorflush(unref[i]->f_data); 2293 for (i = 0; i < fileidx; ++i) 2294 fdrop(unref[i]); 2295 } while (fileidx == filemax && filetot < unp_unreachable); 2296 kfree(unref, M_TEMP); 2297 done: 2298 lwkt_reltoken(&unp_token); 2299 lwkt_reltoken(&unp_rights_token); 2300 } 2301 2302 #endif /* UNP_GC_ALLFILES */ 2303 2304 /* 2305 * Dispose of the fp's stored in a mbuf. 2306 * 2307 * The dds loop can cause additional fps to be entered onto the 2308 * list while it is running, flattening out the operation and avoiding 2309 * a deep kernel stack recursion. 2310 */ 2311 void 2312 unp_dispose(struct mbuf *m) 2313 { 2314 lwkt_gettoken(&unp_rights_token); 2315 if (m) 2316 unp_scan(m, unp_discard, NULL); 2317 lwkt_reltoken(&unp_rights_token); 2318 } 2319 2320 static int 2321 unp_listen(struct unpcb *unp, struct thread *td) 2322 { 2323 struct proc *p = td->td_proc; 2324 2325 ASSERT_LWKT_TOKEN_HELD(&unp_token); 2326 UNP_ASSERT_TOKEN_HELD(unp); 2327 2328 KKASSERT(p); 2329 cru2x(p->p_ucred, &unp->unp_peercred); 2330 unp_setflags(unp, UNP_HAVEPCCACHED); 2331 return (0); 2332 } 2333 2334 static void 2335 unp_scan(struct mbuf *m0, void (*op)(struct file *, void *), void *data) 2336 { 2337 struct mbuf *m; 2338 struct file **rp; 2339 struct cmsghdr *cm; 2340 int i; 2341 int qfds; 2342 2343 while (m0) { 2344 for (m = m0; m; m = m->m_next) { 2345 if (m->m_type == MT_CONTROL && 2346 m->m_len >= sizeof(*cm)) { 2347 cm = mtod(m, struct cmsghdr *); 2348 if (cm->cmsg_level != SOL_SOCKET || 2349 cm->cmsg_type != SCM_RIGHTS) 2350 continue; 2351 qfds = (cm->cmsg_len - CMSG_LEN(0)) / 2352 sizeof(void *); 2353 rp = (struct file **)CMSG_DATA(cm); 2354 for (i = 0; i < qfds; i++) 2355 (*op)(*rp++, data); 2356 break; /* XXX, but saves time */ 2357 } 2358 } 2359 m0 = m0->m_nextpkt; 2360 } 2361 } 2362 2363 /* 2364 * Discard a fp previously held in a unix domain socket mbuf. To 2365 * avoid blowing out the kernel stack due to contrived chain-reactions 2366 * we may have to defer the operation to a dedicated taskqueue. 2367 * 2368 * Caller holds unp_rights_token. 2369 */ 2370 static void 2371 unp_discard(struct file *fp, void *data __unused) 2372 { 2373 unp_del_right(fp); 2374 if (unp_fp2unpcb(fp) != NULL) { 2375 struct unp_defdiscard *d; 2376 2377 /* 2378 * This fp is a Unix domain socket itself and fdrop() 2379 * it here directly may cause deep unp_discard() 2380 * recursion, so the fdrop() is defered to the 2381 * dedicated taskqueue. 2382 */ 2383 d = kmalloc(sizeof(*d), M_UNPCB, M_WAITOK); 2384 d->fp = fp; 2385 2386 spin_lock(&unp_defdiscard_spin); 2387 SLIST_INSERT_HEAD(&unp_defdiscard_head, d, next); 2388 spin_unlock(&unp_defdiscard_spin); 2389 2390 taskqueue_enqueue(unp_taskqueue, &unp_defdiscard_task); 2391 } else { 2392 /* This fp is not a Unix domain socket */ 2393 fdrop(fp); 2394 } 2395 } 2396 2397 /* 2398 * NOTE: 2399 * unp_token must be held before calling this function to avoid name 2400 * resolution and v_socket accessing races, especially racing against 2401 * the unp_detach(). 2402 * 2403 * NOTE: 2404 * For anyone caring about unconnected Unix domain socket sending 2405 * performance, other approach could be taken... 2406 */ 2407 static int 2408 unp_find_lockref(struct sockaddr *nam, struct thread *td, short type, 2409 struct unpcb **unp_ret) 2410 { 2411 struct proc *p = td->td_proc; 2412 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 2413 struct vnode *vp = NULL; 2414 struct socket *so; 2415 struct unpcb *unp; 2416 int error, len; 2417 struct nlookupdata nd; 2418 char buf[SOCK_MAXADDRLEN]; 2419 2420 ASSERT_LWKT_TOKEN_HELD(&unp_token); 2421 2422 *unp_ret = NULL; 2423 2424 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 2425 if (len <= 0) { 2426 error = EINVAL; 2427 goto failed; 2428 } 2429 strncpy(buf, soun->sun_path, len); 2430 buf[len] = 0; 2431 2432 error = nlookup_init(&nd, buf, UIO_SYSSPACE, NLC_FOLLOW); 2433 if (error == 0) 2434 error = nlookup(&nd); 2435 if (error == 0) 2436 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp); 2437 nlookup_done(&nd); 2438 if (error) { 2439 vp = NULL; 2440 goto failed; 2441 } 2442 2443 if (vp->v_type != VSOCK) { 2444 error = ENOTSOCK; 2445 goto failed; 2446 } 2447 error = VOP_EACCESS(vp, VWRITE, p->p_ucred); 2448 if (error) 2449 goto failed; 2450 so = vp->v_socket; 2451 if (so == NULL) { 2452 error = ECONNREFUSED; 2453 goto failed; 2454 } 2455 if (so->so_type != type) { 2456 error = EPROTOTYPE; 2457 goto failed; 2458 } 2459 2460 /* Lock this unp. */ 2461 unp = unp_getsocktoken(so); 2462 if (!UNP_ISATTACHED(unp)) { 2463 unp_reltoken(unp); 2464 error = ECONNREFUSED; 2465 goto failed; 2466 } 2467 /* And keep this unp referenced. */ 2468 unp_reference(unp); 2469 2470 /* Done! */ 2471 *unp_ret = unp; 2472 error = 0; 2473 failed: 2474 if (vp != NULL) 2475 vput(vp); 2476 return error; 2477 } 2478 2479 static int 2480 unp_connect_pair(struct unpcb *unp, struct unpcb *unp2) 2481 { 2482 struct socket *so = unp->unp_socket; 2483 struct socket *so2 = unp2->unp_socket; 2484 2485 ASSERT_LWKT_TOKEN_HELD(&unp_token); 2486 UNP_ASSERT_TOKEN_HELD(unp); 2487 UNP_ASSERT_TOKEN_HELD(unp2); 2488 2489 KASSERT(so->so_type == so2->so_type, 2490 ("socket type mismatch, so %d, so2 %d", so->so_type, so2->so_type)); 2491 2492 if (!UNP_ISATTACHED(unp)) 2493 return EINVAL; 2494 if (!UNP_ISATTACHED(unp2)) 2495 return ECONNREFUSED; 2496 2497 KASSERT(unp->unp_conn == NULL, ("unp is already connected")); 2498 unp->unp_conn = unp2; 2499 2500 switch (so->so_type) { 2501 case SOCK_DGRAM: 2502 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 2503 soisconnected(so); 2504 break; 2505 2506 case SOCK_STREAM: 2507 case SOCK_SEQPACKET: 2508 KASSERT(unp2->unp_conn == NULL, ("unp2 is already connected")); 2509 unp2->unp_conn = unp; 2510 soisconnected(so); 2511 soisconnected(so2); 2512 break; 2513 2514 default: 2515 panic("unp_connect_pair: unknown socket type %d", so->so_type); 2516 } 2517 return 0; 2518 } 2519 2520 static void 2521 unp_drop(struct unpcb *unp, int error) 2522 { 2523 struct unp_global_head *head; 2524 struct unpcb *unp2; 2525 2526 ASSERT_LWKT_TOKEN_HELD(&unp_token); 2527 UNP_ASSERT_TOKEN_HELD(unp); 2528 2529 KASSERT((unp->unp_flags & (UNP_DETACHED | UNP_DROPPED)) == 0, 2530 ("unp is dropped")); 2531 2532 /* Mark this unp as detached. */ 2533 unp_setflags(unp, UNP_DETACHED); 2534 2535 /* Remove this unp from the global unp list. */ 2536 head = unp_globalhead(unp->unp_socket->so_type); 2537 KASSERT(head->count > 0, ("invalid unp count")); 2538 TAILQ_REMOVE(&head->list, unp, unp_link); 2539 head->count--; 2540 2541 /* Disconnect all. */ 2542 unp_disconnect(unp, error); 2543 while ((unp2 = LIST_FIRST(&unp->unp_refs)) != NULL) { 2544 lwkt_getpooltoken(unp2); 2545 unp_disconnect(unp2, ECONNRESET); 2546 lwkt_relpooltoken(unp2); 2547 } 2548 unp_setflags(unp, UNP_DROPPED); 2549 2550 /* Try freeing this unp. */ 2551 unp_free(unp); 2552 } 2553 2554 static void 2555 unp_defdiscard_taskfunc(void *arg __unused, int pending __unused) 2556 { 2557 struct unp_defdiscard *d; 2558 2559 spin_lock(&unp_defdiscard_spin); 2560 while ((d = SLIST_FIRST(&unp_defdiscard_head)) != NULL) { 2561 SLIST_REMOVE_HEAD(&unp_defdiscard_head, next); 2562 spin_unlock(&unp_defdiscard_spin); 2563 2564 fdrop(d->fp); 2565 kfree(d, M_UNPCB); 2566 2567 spin_lock(&unp_defdiscard_spin); 2568 } 2569 spin_unlock(&unp_defdiscard_spin); 2570 } 2571