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