uipc_usrreq.c revision 161019
1/*- 2 * Copyright (c) 1982, 1986, 1989, 1991, 1993 3 * The Regents of the University of California. 4 * Copyright 2004-2006 Robert N. M. Watson 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 4. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94 32 */ 33 34/* 35 * UNIX Domain (Local) Sockets 36 * 37 * This is an implementation of UNIX (local) domain sockets. Each socket has 38 * an associated struct unpcb (UNIX protocol control block). Stream sockets 39 * may be connected to 0 or 1 other socket. Datagram sockets may be 40 * connected to 0, 1, or many other sockets. Sockets may be created and 41 * connected in pairs (socketpair(2)), or bound/connected to using the file 42 * system name space. For most purposes, only the receive socket buffer is 43 * used, as sending on one socket delivers directly to the receive socket 44 * buffer of a second socket. The implementation is substantially 45 * complicated by the fact that "ancillary data", such as file descriptors or 46 * credentials, may be passed across UNIX domain sockets. The potential for 47 * passing UNIX domain sockets over other UNIX domain sockets requires the 48 * implementation of a simple garbage collector to find and tear down cycles 49 * of disconnected sockets. 50 */ 51 52#include <sys/cdefs.h> 53__FBSDID("$FreeBSD: head/sys/kern/uipc_usrreq.c 161019 2006-08-06 10:39:21Z rwatson $"); 54 55#include "opt_mac.h" 56 57#include <sys/param.h> 58#include <sys/domain.h> 59#include <sys/fcntl.h> 60#include <sys/malloc.h> /* XXX must be before <sys/file.h> */ 61#include <sys/eventhandler.h> 62#include <sys/file.h> 63#include <sys/filedesc.h> 64#include <sys/jail.h> 65#include <sys/kernel.h> 66#include <sys/lock.h> 67#include <sys/mac.h> 68#include <sys/mbuf.h> 69#include <sys/mount.h> 70#include <sys/mutex.h> 71#include <sys/namei.h> 72#include <sys/proc.h> 73#include <sys/protosw.h> 74#include <sys/resourcevar.h> 75#include <sys/socket.h> 76#include <sys/socketvar.h> 77#include <sys/signalvar.h> 78#include <sys/stat.h> 79#include <sys/sx.h> 80#include <sys/sysctl.h> 81#include <sys/systm.h> 82#include <sys/taskqueue.h> 83#include <sys/un.h> 84#include <sys/unpcb.h> 85#include <sys/vnode.h> 86 87#include <vm/uma.h> 88 89static uma_zone_t unp_zone; 90static unp_gen_t unp_gencnt; 91static u_int unp_count; 92 93static struct unp_head unp_shead, unp_dhead; 94 95/* 96 * Unix communications domain. 97 * 98 * TODO: 99 * SEQPACKET, RDM 100 * rethink name space problems 101 * need a proper out-of-band 102 * lock pushdown 103 */ 104static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL }; 105static ino_t unp_ino; /* prototype for fake inode numbers */ 106struct mbuf *unp_addsockcred(struct thread *, struct mbuf *); 107 108/* 109 * Both send and receive buffers are allocated PIPSIZ bytes of buffering for 110 * stream sockets, although the total for sender and receiver is actually 111 * only PIPSIZ. 112 * 113 * Datagram sockets really use the sendspace as the maximum datagram size, 114 * and don't really want to reserve the sendspace. Their recvspace should be 115 * large enough for at least one max-size datagram plus address. 116 */ 117#ifndef PIPSIZ 118#define PIPSIZ 8192 119#endif 120static u_long unpst_sendspace = PIPSIZ; 121static u_long unpst_recvspace = PIPSIZ; 122static u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 123static u_long unpdg_recvspace = 4*1024; 124 125static int unp_rights; /* file descriptors in flight */ 126 127SYSCTL_DECL(_net_local_stream); 128SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 129 &unpst_sendspace, 0, ""); 130SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 131 &unpst_recvspace, 0, ""); 132SYSCTL_DECL(_net_local_dgram); 133SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 134 &unpdg_sendspace, 0, ""); 135SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 136 &unpdg_recvspace, 0, ""); 137SYSCTL_DECL(_net_local); 138SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, ""); 139 140/* 141 * Currently, UNIX domain sockets are protected by a single subsystem lock, 142 * which covers global data structures and variables, the contents of each 143 * per-socket unpcb structure, and the so_pcb field in sockets attached to 144 * the UNIX domain. This provides for a moderate degree of paralellism, as 145 * receive operations on UNIX domain sockets do not need to acquire the 146 * subsystem lock. Finer grained locking to permit send() without acquiring 147 * a global lock would be a logical next step. 148 * 149 * The UNIX domain socket lock preceds all socket layer locks, including the 150 * socket lock and socket buffer lock, permitting UNIX domain socket code to 151 * call into socket support routines without releasing its locks. 152 * 153 * Some caution is required in areas where the UNIX domain socket code enters 154 * VFS in order to create or find rendezvous points. This results in 155 * dropping of the UNIX domain socket subsystem lock, acquisition of the 156 * Giant lock, and potential sleeping. This increases the chances of races, 157 * and exposes weaknesses in the socket->protocol API by offering poor 158 * failure modes. 159 */ 160static struct mtx unp_mtx; 161#define UNP_LOCK_INIT() \ 162 mtx_init(&unp_mtx, "unp", NULL, MTX_DEF) 163#define UNP_LOCK() mtx_lock(&unp_mtx) 164#define UNP_UNLOCK() mtx_unlock(&unp_mtx) 165#define UNP_LOCK_ASSERT() mtx_assert(&unp_mtx, MA_OWNED) 166#define UNP_UNLOCK_ASSERT() mtx_assert(&unp_mtx, MA_NOTOWNED) 167 168/* 169 * Garbage collection of cyclic file descriptor/socket references occurs 170 * asynchronously in a taskqueue context in order to avoid recursion and 171 * reentrance in the UNIX domain socket, file descriptor, and socket layer 172 * code. See unp_gc() for a full description. 173 */ 174static struct task unp_gc_task; 175 176static int unp_connect(struct socket *,struct sockaddr *, struct thread *); 177static int unp_connect2(struct socket *so, struct socket *so2, int); 178static void unp_disconnect(struct unpcb *); 179static void unp_shutdown(struct unpcb *); 180static void unp_drop(struct unpcb *, int); 181static void unp_gc(__unused void *, int); 182static void unp_scan(struct mbuf *, void (*)(struct file *)); 183static void unp_mark(struct file *); 184static void unp_discard(struct file *); 185static void unp_freerights(struct file **, int); 186static int unp_internalize(struct mbuf **, struct thread *); 187static int unp_listen(struct socket *, struct unpcb *, int, 188 struct thread *); 189 190static void 191uipc_abort(struct socket *so) 192{ 193 struct unpcb *unp; 194 195 unp = sotounpcb(so); 196 KASSERT(unp != NULL, ("uipc_abort: unp == NULL")); 197 UNP_LOCK(); 198 unp_drop(unp, ECONNABORTED); 199 UNP_UNLOCK(); 200} 201 202static int 203uipc_accept(struct socket *so, struct sockaddr **nam) 204{ 205 struct unpcb *unp; 206 const struct sockaddr *sa; 207 208 /* 209 * Pass back name of connected socket, if it was bound and we are 210 * still connected (our peer may have closed already!). 211 */ 212 unp = sotounpcb(so); 213 KASSERT(unp != NULL, ("uipc_accept: unp == NULL")); 214 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 215 UNP_LOCK(); 216 if (unp->unp_conn != NULL && unp->unp_conn->unp_addr != NULL) 217 sa = (struct sockaddr *) unp->unp_conn->unp_addr; 218 else 219 sa = &sun_noname; 220 bcopy(sa, *nam, sa->sa_len); 221 UNP_UNLOCK(); 222 return (0); 223} 224 225static int 226uipc_attach(struct socket *so, int proto, struct thread *td) 227{ 228 struct unpcb *unp; 229 int error; 230 231 KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL")); 232 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 233 switch (so->so_type) { 234 235 case SOCK_STREAM: 236 error = soreserve(so, unpst_sendspace, unpst_recvspace); 237 break; 238 239 case SOCK_DGRAM: 240 error = soreserve(so, unpdg_sendspace, unpdg_recvspace); 241 break; 242 243 default: 244 panic("unp_attach"); 245 } 246 if (error) 247 return (error); 248 } 249 unp = uma_zalloc(unp_zone, M_WAITOK | M_ZERO); 250 if (unp == NULL) 251 return (ENOBUFS); 252 LIST_INIT(&unp->unp_refs); 253 unp->unp_socket = so; 254 so->so_pcb = unp; 255 256 UNP_LOCK(); 257 unp->unp_gencnt = ++unp_gencnt; 258 unp_count++; 259 LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead 260 : &unp_shead, unp, unp_link); 261 UNP_UNLOCK(); 262 263 return (0); 264} 265 266static int 267uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 268{ 269 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 270 struct vattr vattr; 271 int error, namelen; 272 struct nameidata nd; 273 struct unpcb *unp; 274 struct vnode *vp; 275 struct mount *mp; 276 char *buf; 277 278 unp = sotounpcb(so); 279 KASSERT(unp != NULL, ("uipc_bind: unp == NULL")); 280 281 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 282 if (namelen <= 0) 283 return (EINVAL); 284 285 /* 286 * We don't allow simultaneous bind() calls on a single UNIX domain 287 * socket, so flag in-progress operations, and return an error if an 288 * operation is already in progress. 289 * 290 * Historically, we have not allowed a socket to be rebound, so this 291 * also returns an error. Not allowing re-binding certainly 292 * simplifies the implementation and avoids a great many possible 293 * failure modes. 294 */ 295 UNP_LOCK(); 296 if (unp->unp_vnode != NULL) { 297 UNP_UNLOCK(); 298 return (EINVAL); 299 } 300 if (unp->unp_flags & UNP_BINDING) { 301 UNP_UNLOCK(); 302 return (EALREADY); 303 } 304 unp->unp_flags |= UNP_BINDING; 305 UNP_UNLOCK(); 306 307 buf = malloc(namelen + 1, M_TEMP, M_WAITOK); 308 strlcpy(buf, soun->sun_path, namelen + 1); 309 310 mtx_lock(&Giant); 311restart: 312 mtx_assert(&Giant, MA_OWNED); 313 NDINIT(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME, UIO_SYSSPACE, 314 buf, td); 315/* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 316 error = namei(&nd); 317 if (error) 318 goto error; 319 vp = nd.ni_vp; 320 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { 321 NDFREE(&nd, NDF_ONLY_PNBUF); 322 if (nd.ni_dvp == vp) 323 vrele(nd.ni_dvp); 324 else 325 vput(nd.ni_dvp); 326 if (vp != NULL) { 327 vrele(vp); 328 error = EADDRINUSE; 329 goto error; 330 } 331 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH); 332 if (error) 333 goto error; 334 goto restart; 335 } 336 VATTR_NULL(&vattr); 337 vattr.va_type = VSOCK; 338 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask); 339#ifdef MAC 340 error = mac_check_vnode_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, 341 &vattr); 342#endif 343 if (error == 0) { 344 VOP_LEASE(nd.ni_dvp, td, td->td_ucred, LEASE_WRITE); 345 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 346 } 347 NDFREE(&nd, NDF_ONLY_PNBUF); 348 vput(nd.ni_dvp); 349 if (error) { 350 vn_finished_write(mp); 351 goto error; 352 } 353 vp = nd.ni_vp; 354 ASSERT_VOP_LOCKED(vp, "uipc_bind"); 355 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK); 356 UNP_LOCK(); 357 vp->v_socket = unp->unp_socket; 358 unp->unp_vnode = vp; 359 unp->unp_addr = soun; 360 unp->unp_flags &= ~UNP_BINDING; 361 UNP_UNLOCK(); 362 VOP_UNLOCK(vp, 0, td); 363 vn_finished_write(mp); 364 mtx_unlock(&Giant); 365 free(buf, M_TEMP); 366 return (0); 367error: 368 UNP_LOCK(); 369 unp->unp_flags &= ~UNP_BINDING; 370 UNP_UNLOCK(); 371 mtx_unlock(&Giant); 372 free(buf, M_TEMP); 373 return (error); 374} 375 376static int 377uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 378{ 379 int error; 380 381 KASSERT(td == curthread, ("uipc_connect: td != curthread")); 382 UNP_LOCK(); 383 error = unp_connect(so, nam, td); 384 UNP_UNLOCK(); 385 return (error); 386} 387 388/* 389 * XXXRW: Should also unbind? 390 */ 391static void 392uipc_close(struct socket *so) 393{ 394 struct unpcb *unp; 395 396 unp = sotounpcb(so); 397 KASSERT(unp != NULL, ("uipc_close: unp == NULL")); 398 UNP_LOCK(); 399 unp_disconnect(unp); 400 UNP_UNLOCK(); 401} 402 403int 404uipc_connect2(struct socket *so1, struct socket *so2) 405{ 406 struct unpcb *unp; 407 int error; 408 409 unp = sotounpcb(so1); 410 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL")); 411 UNP_LOCK(); 412 error = unp_connect2(so1, so2, PRU_CONNECT2); 413 UNP_UNLOCK(); 414 return (error); 415} 416 417/* control is EOPNOTSUPP */ 418 419static void 420uipc_detach(struct socket *so) 421{ 422 int local_unp_rights; 423 struct unpcb *unp; 424 struct vnode *vp; 425 426 unp = sotounpcb(so); 427 KASSERT(unp != NULL, ("uipc_detach: unp == NULL")); 428 UNP_LOCK(); 429 LIST_REMOVE(unp, unp_link); 430 unp->unp_gencnt = ++unp_gencnt; 431 --unp_count; 432 if ((vp = unp->unp_vnode) != NULL) { 433 /* 434 * XXXRW: should v_socket be frobbed only while holding 435 * Giant? 436 */ 437 unp->unp_vnode->v_socket = NULL; 438 unp->unp_vnode = NULL; 439 } 440 if (unp->unp_conn != NULL) 441 unp_disconnect(unp); 442 while (!LIST_EMPTY(&unp->unp_refs)) { 443 struct unpcb *ref = LIST_FIRST(&unp->unp_refs); 444 unp_drop(ref, ECONNRESET); 445 } 446 unp->unp_socket->so_pcb = NULL; 447 local_unp_rights = unp_rights; 448 UNP_UNLOCK(); 449 if (unp->unp_addr != NULL) 450 FREE(unp->unp_addr, M_SONAME); 451 uma_zfree(unp_zone, unp); 452 if (vp) { 453 int vfslocked; 454 455 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 456 vrele(vp); 457 VFS_UNLOCK_GIANT(vfslocked); 458 } 459 if (local_unp_rights) 460 taskqueue_enqueue(taskqueue_thread, &unp_gc_task); 461} 462 463static int 464uipc_disconnect(struct socket *so) 465{ 466 struct unpcb *unp; 467 468 unp = sotounpcb(so); 469 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL")); 470 UNP_LOCK(); 471 unp_disconnect(unp); 472 UNP_UNLOCK(); 473 return (0); 474} 475 476static int 477uipc_listen(struct socket *so, int backlog, struct thread *td) 478{ 479 struct unpcb *unp; 480 int error; 481 482 unp = sotounpcb(so); 483 KASSERT(unp != NULL, ("uipc_listen: unp == NULL")); 484 UNP_LOCK(); 485 if (unp->unp_vnode == NULL) { 486 UNP_UNLOCK(); 487 return (EINVAL); 488 } 489 error = unp_listen(so, unp, backlog, td); 490 UNP_UNLOCK(); 491 return (error); 492} 493 494static int 495uipc_peeraddr(struct socket *so, struct sockaddr **nam) 496{ 497 struct unpcb *unp; 498 const struct sockaddr *sa; 499 500 unp = sotounpcb(so); 501 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL")); 502 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 503 UNP_LOCK(); 504 if (unp->unp_conn != NULL && unp->unp_conn->unp_addr!= NULL) 505 sa = (struct sockaddr *) unp->unp_conn->unp_addr; 506 else { 507 /* 508 * XXX: It seems that this test always fails even when 509 * connection is established. So, this else clause is 510 * added as workaround to return PF_LOCAL sockaddr. 511 */ 512 sa = &sun_noname; 513 } 514 bcopy(sa, *nam, sa->sa_len); 515 UNP_UNLOCK(); 516 return (0); 517} 518 519static int 520uipc_rcvd(struct socket *so, int flags) 521{ 522 struct unpcb *unp; 523 struct socket *so2; 524 u_int mbcnt, sbcc; 525 u_long newhiwat; 526 527 unp = sotounpcb(so); 528 KASSERT(unp != NULL, ("uipc_rcvd: unp == NULL")); 529 switch (so->so_type) { 530 case SOCK_DGRAM: 531 panic("uipc_rcvd DGRAM?"); 532 /*NOTREACHED*/ 533 534 case SOCK_STREAM: 535 /* 536 * Adjust backpressure on sender and wakeup any waiting to 537 * write. 538 */ 539 SOCKBUF_LOCK(&so->so_rcv); 540 mbcnt = so->so_rcv.sb_mbcnt; 541 sbcc = so->so_rcv.sb_cc; 542 SOCKBUF_UNLOCK(&so->so_rcv); 543 UNP_LOCK(); 544 if (unp->unp_conn == NULL) { 545 UNP_UNLOCK(); 546 break; 547 } 548 so2 = unp->unp_conn->unp_socket; 549 SOCKBUF_LOCK(&so2->so_snd); 550 so2->so_snd.sb_mbmax += unp->unp_mbcnt - mbcnt; 551 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - sbcc; 552 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat, 553 newhiwat, RLIM_INFINITY); 554 sowwakeup_locked(so2); 555 unp->unp_mbcnt = mbcnt; 556 unp->unp_cc = sbcc; 557 UNP_UNLOCK(); 558 break; 559 560 default: 561 panic("uipc_rcvd unknown socktype"); 562 } 563 return (0); 564} 565 566/* pru_rcvoob is EOPNOTSUPP */ 567 568static int 569uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 570 struct mbuf *control, struct thread *td) 571{ 572 struct unpcb *unp, *unp2; 573 struct socket *so2; 574 u_int mbcnt, sbcc; 575 u_long newhiwat; 576 int error = 0; 577 578 unp = sotounpcb(so); 579 KASSERT(unp != NULL, ("uipc_send: unp == NULL")); 580 if (flags & PRUS_OOB) { 581 error = EOPNOTSUPP; 582 goto release; 583 } 584 585 if (control != NULL && (error = unp_internalize(&control, td))) 586 goto release; 587 588 UNP_LOCK(); 589 switch (so->so_type) { 590 case SOCK_DGRAM: 591 { 592 const struct sockaddr *from; 593 594 if (nam != NULL) { 595 if (unp->unp_conn != NULL) { 596 error = EISCONN; 597 break; 598 } 599 error = unp_connect(so, nam, td); 600 if (error) 601 break; 602 } 603 /* 604 * Because connect() and send() are non-atomic in a sendto() 605 * with a target address, it's possible that the socket will 606 * have disconnected before the send() can run. In that case 607 * return the slightly counter-intuitive but otherwise 608 * correct error that the socket is not connected. 609 */ 610 unp2 = unp->unp_conn; 611 if (unp2 == NULL) { 612 error = ENOTCONN; 613 break; 614 } 615 so2 = unp2->unp_socket; 616 if (unp->unp_addr != NULL) 617 from = (struct sockaddr *)unp->unp_addr; 618 else 619 from = &sun_noname; 620 if (unp2->unp_flags & UNP_WANTCRED) 621 control = unp_addsockcred(td, control); 622 SOCKBUF_LOCK(&so2->so_rcv); 623 if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) { 624 sorwakeup_locked(so2); 625 m = NULL; 626 control = NULL; 627 } else { 628 SOCKBUF_UNLOCK(&so2->so_rcv); 629 error = ENOBUFS; 630 } 631 if (nam != NULL) 632 unp_disconnect(unp); 633 break; 634 } 635 636 case SOCK_STREAM: 637 /* 638 * Connect if not connected yet. 639 * 640 * Note: A better implementation would complain if not equal 641 * to the peer's address. 642 */ 643 if ((so->so_state & SS_ISCONNECTED) == 0) { 644 if (nam != NULL) { 645 error = unp_connect(so, nam, td); 646 if (error) 647 break; /* XXX */ 648 } else { 649 error = ENOTCONN; 650 break; 651 } 652 } 653 654 /* Lockless read. */ 655 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 656 error = EPIPE; 657 break; 658 } 659 /* 660 * Because connect() and send() are non-atomic in a sendto() 661 * with a target address, it's possible that the socket will 662 * have disconnected before the send() can run. In that case 663 * return the slightly counter-intuitive but otherwise 664 * correct error that the socket is not connected. 665 */ 666 unp2 = unp->unp_conn; 667 if (unp2 == NULL) { 668 error = ENOTCONN; 669 break; 670 } 671 so2 = unp2->unp_socket; 672 SOCKBUF_LOCK(&so2->so_rcv); 673 if (unp2->unp_flags & UNP_WANTCRED) { 674 /* 675 * Credentials are passed only once on 676 * SOCK_STREAM. 677 */ 678 unp2->unp_flags &= ~UNP_WANTCRED; 679 control = unp_addsockcred(td, control); 680 } 681 /* 682 * Send to paired receive port, and then reduce send buffer 683 * hiwater marks to maintain backpressure. Wake up readers. 684 */ 685 if (control != NULL) { 686 if (sbappendcontrol_locked(&so2->so_rcv, m, control)) 687 control = NULL; 688 } else { 689 sbappend_locked(&so2->so_rcv, m); 690 } 691 mbcnt = so2->so_rcv.sb_mbcnt - unp2->unp_mbcnt; 692 unp2->unp_mbcnt = so2->so_rcv.sb_mbcnt; 693 sbcc = so2->so_rcv.sb_cc; 694 sorwakeup_locked(so2); 695 696 SOCKBUF_LOCK(&so->so_snd); 697 newhiwat = so->so_snd.sb_hiwat - (sbcc - unp2->unp_cc); 698 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat, 699 newhiwat, RLIM_INFINITY); 700 so->so_snd.sb_mbmax -= mbcnt; 701 SOCKBUF_UNLOCK(&so->so_snd); 702 703 unp2->unp_cc = sbcc; 704 m = NULL; 705 break; 706 707 default: 708 panic("uipc_send unknown socktype"); 709 } 710 711 /* 712 * SEND_EOF is equivalent to a SEND followed by 713 * a SHUTDOWN. 714 */ 715 if (flags & PRUS_EOF) { 716 socantsendmore(so); 717 unp_shutdown(unp); 718 } 719 UNP_UNLOCK(); 720 721 if (control != NULL && error != 0) 722 unp_dispose(control); 723 724release: 725 if (control != NULL) 726 m_freem(control); 727 if (m != NULL) 728 m_freem(m); 729 return (error); 730} 731 732static int 733uipc_sense(struct socket *so, struct stat *sb) 734{ 735 struct unpcb *unp; 736 struct socket *so2; 737 738 unp = sotounpcb(so); 739 KASSERT(unp != NULL, ("uipc_sense: unp == NULL")); 740 UNP_LOCK(); 741 sb->st_blksize = so->so_snd.sb_hiwat; 742 if (so->so_type == SOCK_STREAM && unp->unp_conn != NULL) { 743 so2 = unp->unp_conn->unp_socket; 744 sb->st_blksize += so2->so_rcv.sb_cc; 745 } 746 sb->st_dev = NODEV; 747 if (unp->unp_ino == 0) 748 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino; 749 sb->st_ino = unp->unp_ino; 750 UNP_UNLOCK(); 751 return (0); 752} 753 754static int 755uipc_shutdown(struct socket *so) 756{ 757 struct unpcb *unp; 758 759 unp = sotounpcb(so); 760 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL")); 761 UNP_LOCK(); 762 socantsendmore(so); 763 unp_shutdown(unp); 764 UNP_UNLOCK(); 765 return (0); 766} 767 768static int 769uipc_sockaddr(struct socket *so, struct sockaddr **nam) 770{ 771 struct unpcb *unp; 772 const struct sockaddr *sa; 773 774 unp = sotounpcb(so); 775 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL")); 776 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 777 UNP_LOCK(); 778 if (unp->unp_addr != NULL) 779 sa = (struct sockaddr *) unp->unp_addr; 780 else 781 sa = &sun_noname; 782 bcopy(sa, *nam, sa->sa_len); 783 UNP_UNLOCK(); 784 return (0); 785} 786 787struct pr_usrreqs uipc_usrreqs = { 788 .pru_abort = uipc_abort, 789 .pru_accept = uipc_accept, 790 .pru_attach = uipc_attach, 791 .pru_bind = uipc_bind, 792 .pru_connect = uipc_connect, 793 .pru_connect2 = uipc_connect2, 794 .pru_detach = uipc_detach, 795 .pru_disconnect = uipc_disconnect, 796 .pru_listen = uipc_listen, 797 .pru_peeraddr = uipc_peeraddr, 798 .pru_rcvd = uipc_rcvd, 799 .pru_send = uipc_send, 800 .pru_sense = uipc_sense, 801 .pru_shutdown = uipc_shutdown, 802 .pru_sockaddr = uipc_sockaddr, 803 .pru_close = uipc_close, 804}; 805 806int 807uipc_ctloutput(struct socket *so, struct sockopt *sopt) 808{ 809 struct unpcb *unp; 810 struct xucred xu; 811 int error, optval; 812 813 if (sopt->sopt_level != 0) 814 return (EINVAL); 815 816 unp = sotounpcb(so); 817 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL")); 818 UNP_LOCK(); 819 error = 0; 820 switch (sopt->sopt_dir) { 821 case SOPT_GET: 822 switch (sopt->sopt_name) { 823 case LOCAL_PEERCRED: 824 if (unp->unp_flags & UNP_HAVEPC) 825 xu = unp->unp_peercred; 826 else { 827 if (so->so_type == SOCK_STREAM) 828 error = ENOTCONN; 829 else 830 error = EINVAL; 831 } 832 if (error == 0) 833 error = sooptcopyout(sopt, &xu, sizeof(xu)); 834 break; 835 case LOCAL_CREDS: 836 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0; 837 error = sooptcopyout(sopt, &optval, sizeof(optval)); 838 break; 839 case LOCAL_CONNWAIT: 840 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0; 841 error = sooptcopyout(sopt, &optval, sizeof(optval)); 842 break; 843 default: 844 error = EOPNOTSUPP; 845 break; 846 } 847 break; 848 case SOPT_SET: 849 switch (sopt->sopt_name) { 850 case LOCAL_CREDS: 851 case LOCAL_CONNWAIT: 852 error = sooptcopyin(sopt, &optval, sizeof(optval), 853 sizeof(optval)); 854 if (error) 855 break; 856 857#define OPTSET(bit) \ 858 if (optval) \ 859 unp->unp_flags |= bit; \ 860 else \ 861 unp->unp_flags &= ~bit; 862 863 switch (sopt->sopt_name) { 864 case LOCAL_CREDS: 865 OPTSET(UNP_WANTCRED); 866 break; 867 case LOCAL_CONNWAIT: 868 OPTSET(UNP_CONNWAIT); 869 break; 870 default: 871 break; 872 } 873 break; 874#undef OPTSET 875 default: 876 error = ENOPROTOOPT; 877 break; 878 } 879 break; 880 default: 881 error = EOPNOTSUPP; 882 break; 883 } 884 UNP_UNLOCK(); 885 return (error); 886} 887 888static int 889unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 890{ 891 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 892 struct vnode *vp; 893 struct socket *so2, *so3; 894 struct unpcb *unp, *unp2, *unp3; 895 int error, len; 896 struct nameidata nd; 897 char buf[SOCK_MAXADDRLEN]; 898 struct sockaddr *sa; 899 900 UNP_LOCK_ASSERT(); 901 902 unp = sotounpcb(so); 903 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 904 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 905 if (len <= 0) 906 return (EINVAL); 907 strlcpy(buf, soun->sun_path, len + 1); 908 if (unp->unp_flags & UNP_CONNECTING) { 909 UNP_UNLOCK(); 910 return (EALREADY); 911 } 912 UNP_UNLOCK(); 913 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 914 mtx_lock(&Giant); 915 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, td); 916 error = namei(&nd); 917 if (error) 918 vp = NULL; 919 else 920 vp = nd.ni_vp; 921 ASSERT_VOP_LOCKED(vp, "unp_connect"); 922 NDFREE(&nd, NDF_ONLY_PNBUF); 923 if (error) 924 goto bad; 925 926 if (vp->v_type != VSOCK) { 927 error = ENOTSOCK; 928 goto bad; 929 } 930 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); 931 if (error) 932 goto bad; 933 mtx_unlock(&Giant); 934 UNP_LOCK(); 935 unp = sotounpcb(so); 936 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 937 so2 = vp->v_socket; 938 if (so2 == NULL) { 939 error = ECONNREFUSED; 940 goto bad2; 941 } 942 if (so->so_type != so2->so_type) { 943 error = EPROTOTYPE; 944 goto bad2; 945 } 946 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 947 if (so2->so_options & SO_ACCEPTCONN) { 948 /* 949 * NB: drop locks here so unp_attach is entered w/o 950 * locks; this avoids a recursive lock of the head 951 * and holding sleep locks across a (potentially) 952 * blocking malloc. 953 */ 954 UNP_UNLOCK(); 955 so3 = sonewconn(so2, 0); 956 UNP_LOCK(); 957 } else 958 so3 = NULL; 959 if (so3 == NULL) { 960 error = ECONNREFUSED; 961 goto bad2; 962 } 963 unp = sotounpcb(so); 964 unp2 = sotounpcb(so2); 965 unp3 = sotounpcb(so3); 966 if (unp2->unp_addr != NULL) { 967 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); 968 unp3->unp_addr = (struct sockaddr_un *) sa; 969 sa = NULL; 970 } 971 /* 972 * unp_peercred management: 973 * 974 * The connecter's (client's) credentials are copied from its 975 * process structure at the time of connect() (which is now). 976 */ 977 cru2x(td->td_ucred, &unp3->unp_peercred); 978 unp3->unp_flags |= UNP_HAVEPC; 979 /* 980 * The receiver's (server's) credentials are copied from the 981 * unp_peercred member of socket on which the former called 982 * listen(); unp_listen() cached that process's credentials 983 * at that time so we can use them now. 984 */ 985 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED, 986 ("unp_connect: listener without cached peercred")); 987 memcpy(&unp->unp_peercred, &unp2->unp_peercred, 988 sizeof(unp->unp_peercred)); 989 unp->unp_flags |= UNP_HAVEPC; 990 if (unp2->unp_flags & UNP_WANTCRED) 991 unp3->unp_flags |= UNP_WANTCRED; 992#ifdef MAC 993 SOCK_LOCK(so); 994 mac_set_socket_peer_from_socket(so, so3); 995 mac_set_socket_peer_from_socket(so3, so); 996 SOCK_UNLOCK(so); 997#endif 998 999 so2 = so3; 1000 } 1001 error = unp_connect2(so, so2, PRU_CONNECT); 1002bad2: 1003 UNP_UNLOCK(); 1004 mtx_lock(&Giant); 1005bad: 1006 mtx_assert(&Giant, MA_OWNED); 1007 if (vp != NULL) 1008 vput(vp); 1009 mtx_unlock(&Giant); 1010 free(sa, M_SONAME); 1011 UNP_LOCK(); 1012 unp->unp_flags &= ~UNP_CONNECTING; 1013 return (error); 1014} 1015 1016static int 1017unp_connect2(struct socket *so, struct socket *so2, int req) 1018{ 1019 struct unpcb *unp = sotounpcb(so); 1020 struct unpcb *unp2; 1021 1022 UNP_LOCK_ASSERT(); 1023 1024 if (so2->so_type != so->so_type) 1025 return (EPROTOTYPE); 1026 unp2 = sotounpcb(so2); 1027 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL")); 1028 unp->unp_conn = unp2; 1029 switch (so->so_type) { 1030 1031 case SOCK_DGRAM: 1032 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 1033 soisconnected(so); 1034 break; 1035 1036 case SOCK_STREAM: 1037 unp2->unp_conn = unp; 1038 if (req == PRU_CONNECT && 1039 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 1040 soisconnecting(so); 1041 else 1042 soisconnected(so); 1043 soisconnected(so2); 1044 break; 1045 1046 default: 1047 panic("unp_connect2"); 1048 } 1049 return (0); 1050} 1051 1052static void 1053unp_disconnect(struct unpcb *unp) 1054{ 1055 struct unpcb *unp2 = unp->unp_conn; 1056 struct socket *so; 1057 1058 UNP_LOCK_ASSERT(); 1059 1060 if (unp2 == NULL) 1061 return; 1062 unp->unp_conn = NULL; 1063 switch (unp->unp_socket->so_type) { 1064 case SOCK_DGRAM: 1065 LIST_REMOVE(unp, unp_reflink); 1066 so = unp->unp_socket; 1067 SOCK_LOCK(so); 1068 so->so_state &= ~SS_ISCONNECTED; 1069 SOCK_UNLOCK(so); 1070 break; 1071 1072 case SOCK_STREAM: 1073 soisdisconnected(unp->unp_socket); 1074 unp2->unp_conn = NULL; 1075 soisdisconnected(unp2->unp_socket); 1076 break; 1077 } 1078} 1079 1080/* 1081 * unp_pcblist() assumes that UNIX domain socket memory is never reclaimed by 1082 * the zone (UMA_ZONE_NOFREE), and as such potentially stale pointers are 1083 * safe to reference. It first scans the list of struct unpcb's to generate 1084 * a pointer list, then it rescans its list one entry at a time to 1085 * externalize and copyout. It checks the generation number to see if a 1086 * struct unpcb has been reused, and will skip it if so. 1087 */ 1088static int 1089unp_pcblist(SYSCTL_HANDLER_ARGS) 1090{ 1091 int error, i, n; 1092 struct unpcb *unp, **unp_list; 1093 unp_gen_t gencnt; 1094 struct xunpgen *xug; 1095 struct unp_head *head; 1096 struct xunpcb *xu; 1097 1098 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead); 1099 1100 /* 1101 * The process of preparing the PCB list is too time-consuming and 1102 * resource-intensive to repeat twice on every request. 1103 */ 1104 if (req->oldptr == NULL) { 1105 n = unp_count; 1106 req->oldidx = 2 * (sizeof *xug) 1107 + (n + n/8) * sizeof(struct xunpcb); 1108 return (0); 1109 } 1110 1111 if (req->newptr != NULL) 1112 return (EPERM); 1113 1114 /* 1115 * OK, now we're committed to doing something. 1116 */ 1117 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK); 1118 UNP_LOCK(); 1119 gencnt = unp_gencnt; 1120 n = unp_count; 1121 UNP_UNLOCK(); 1122 1123 xug->xug_len = sizeof *xug; 1124 xug->xug_count = n; 1125 xug->xug_gen = gencnt; 1126 xug->xug_sogen = so_gencnt; 1127 error = SYSCTL_OUT(req, xug, sizeof *xug); 1128 if (error) { 1129 free(xug, M_TEMP); 1130 return (error); 1131 } 1132 1133 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 1134 1135 UNP_LOCK(); 1136 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 1137 unp = LIST_NEXT(unp, unp_link)) { 1138 if (unp->unp_gencnt <= gencnt) { 1139 if (cr_cansee(req->td->td_ucred, 1140 unp->unp_socket->so_cred)) 1141 continue; 1142 unp_list[i++] = unp; 1143 } 1144 } 1145 UNP_UNLOCK(); 1146 n = i; /* In case we lost some during malloc. */ 1147 1148 error = 0; 1149 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO); 1150 for (i = 0; i < n; i++) { 1151 unp = unp_list[i]; 1152 if (unp->unp_gencnt <= gencnt) { 1153 xu->xu_len = sizeof *xu; 1154 xu->xu_unpp = unp; 1155 /* 1156 * XXX - need more locking here to protect against 1157 * connect/disconnect races for SMP. 1158 */ 1159 if (unp->unp_addr != NULL) 1160 bcopy(unp->unp_addr, &xu->xu_addr, 1161 unp->unp_addr->sun_len); 1162 if (unp->unp_conn != NULL && 1163 unp->unp_conn->unp_addr != NULL) 1164 bcopy(unp->unp_conn->unp_addr, 1165 &xu->xu_caddr, 1166 unp->unp_conn->unp_addr->sun_len); 1167 bcopy(unp, &xu->xu_unp, sizeof *unp); 1168 sotoxsocket(unp->unp_socket, &xu->xu_socket); 1169 error = SYSCTL_OUT(req, xu, sizeof *xu); 1170 } 1171 } 1172 free(xu, M_TEMP); 1173 if (!error) { 1174 /* 1175 * Give the user an updated idea of our state. If the 1176 * generation differs from what we told her before, she knows 1177 * that something happened while we were processing this 1178 * request, and it might be necessary to retry. 1179 */ 1180 xug->xug_gen = unp_gencnt; 1181 xug->xug_sogen = so_gencnt; 1182 xug->xug_count = unp_count; 1183 error = SYSCTL_OUT(req, xug, sizeof *xug); 1184 } 1185 free(unp_list, M_TEMP); 1186 free(xug, M_TEMP); 1187 return (error); 1188} 1189 1190SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD, 1191 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 1192 "List of active local datagram sockets"); 1193SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD, 1194 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 1195 "List of active local stream sockets"); 1196 1197static void 1198unp_shutdown(struct unpcb *unp) 1199{ 1200 struct socket *so; 1201 1202 UNP_LOCK_ASSERT(); 1203 1204 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn && 1205 (so = unp->unp_conn->unp_socket)) 1206 socantrcvmore(so); 1207} 1208 1209static void 1210unp_drop(struct unpcb *unp, int errno) 1211{ 1212 struct socket *so = unp->unp_socket; 1213 1214 UNP_LOCK_ASSERT(); 1215 1216 so->so_error = errno; 1217 unp_disconnect(unp); 1218} 1219 1220static void 1221unp_freerights(struct file **rp, int fdcount) 1222{ 1223 int i; 1224 struct file *fp; 1225 1226 for (i = 0; i < fdcount; i++) { 1227 fp = *rp; 1228 /* 1229 * Zero the pointer before calling unp_discard since it may 1230 * end up in unp_gc().. 1231 * 1232 * XXXRW: This is less true than it used to be. 1233 */ 1234 *rp++ = 0; 1235 unp_discard(fp); 1236 } 1237} 1238 1239int 1240unp_externalize(struct mbuf *control, struct mbuf **controlp) 1241{ 1242 struct thread *td = curthread; /* XXX */ 1243 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1244 int i; 1245 int *fdp; 1246 struct file **rp; 1247 struct file *fp; 1248 void *data; 1249 socklen_t clen = control->m_len, datalen; 1250 int error, newfds; 1251 int f; 1252 u_int newlen; 1253 1254 UNP_UNLOCK_ASSERT(); 1255 1256 error = 0; 1257 if (controlp != NULL) /* controlp == NULL => free control messages */ 1258 *controlp = NULL; 1259 1260 while (cm != NULL) { 1261 if (sizeof(*cm) > clen || cm->cmsg_len > clen) { 1262 error = EINVAL; 1263 break; 1264 } 1265 1266 data = CMSG_DATA(cm); 1267 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1268 1269 if (cm->cmsg_level == SOL_SOCKET 1270 && cm->cmsg_type == SCM_RIGHTS) { 1271 newfds = datalen / sizeof(struct file *); 1272 rp = data; 1273 1274 /* If we're not outputting the descriptors free them. */ 1275 if (error || controlp == NULL) { 1276 unp_freerights(rp, newfds); 1277 goto next; 1278 } 1279 FILEDESC_LOCK(td->td_proc->p_fd); 1280 /* if the new FD's will not fit free them. */ 1281 if (!fdavail(td, newfds)) { 1282 FILEDESC_UNLOCK(td->td_proc->p_fd); 1283 error = EMSGSIZE; 1284 unp_freerights(rp, newfds); 1285 goto next; 1286 } 1287 /* 1288 * Now change each pointer to an fd in the global 1289 * table to an integer that is the index to the local 1290 * fd table entry that we set up to point to the 1291 * global one we are transferring. 1292 */ 1293 newlen = newfds * sizeof(int); 1294 *controlp = sbcreatecontrol(NULL, newlen, 1295 SCM_RIGHTS, SOL_SOCKET); 1296 if (*controlp == NULL) { 1297 FILEDESC_UNLOCK(td->td_proc->p_fd); 1298 error = E2BIG; 1299 unp_freerights(rp, newfds); 1300 goto next; 1301 } 1302 1303 fdp = (int *) 1304 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1305 for (i = 0; i < newfds; i++) { 1306 if (fdalloc(td, 0, &f)) 1307 panic("unp_externalize fdalloc failed"); 1308 fp = *rp++; 1309 td->td_proc->p_fd->fd_ofiles[f] = fp; 1310 FILE_LOCK(fp); 1311 fp->f_msgcount--; 1312 FILE_UNLOCK(fp); 1313 unp_rights--; 1314 *fdp++ = f; 1315 } 1316 FILEDESC_UNLOCK(td->td_proc->p_fd); 1317 } else { 1318 /* We can just copy anything else across. */ 1319 if (error || controlp == NULL) 1320 goto next; 1321 *controlp = sbcreatecontrol(NULL, datalen, 1322 cm->cmsg_type, cm->cmsg_level); 1323 if (*controlp == NULL) { 1324 error = ENOBUFS; 1325 goto next; 1326 } 1327 bcopy(data, 1328 CMSG_DATA(mtod(*controlp, struct cmsghdr *)), 1329 datalen); 1330 } 1331 1332 controlp = &(*controlp)->m_next; 1333 1334next: 1335 if (CMSG_SPACE(datalen) < clen) { 1336 clen -= CMSG_SPACE(datalen); 1337 cm = (struct cmsghdr *) 1338 ((caddr_t)cm + CMSG_SPACE(datalen)); 1339 } else { 1340 clen = 0; 1341 cm = NULL; 1342 } 1343 } 1344 1345 m_freem(control); 1346 1347 return (error); 1348} 1349 1350static void 1351unp_zone_change(void *tag) 1352{ 1353 1354 uma_zone_set_max(unp_zone, maxsockets); 1355} 1356 1357void 1358unp_init(void) 1359{ 1360 1361 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL, 1362 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 1363 if (unp_zone == NULL) 1364 panic("unp_init"); 1365 uma_zone_set_max(unp_zone, maxsockets); 1366 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change, 1367 NULL, EVENTHANDLER_PRI_ANY); 1368 LIST_INIT(&unp_dhead); 1369 LIST_INIT(&unp_shead); 1370 TASK_INIT(&unp_gc_task, 0, unp_gc, NULL); 1371 UNP_LOCK_INIT(); 1372} 1373 1374static int 1375unp_internalize(struct mbuf **controlp, struct thread *td) 1376{ 1377 struct mbuf *control = *controlp; 1378 struct proc *p = td->td_proc; 1379 struct filedesc *fdescp = p->p_fd; 1380 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1381 struct cmsgcred *cmcred; 1382 struct file **rp; 1383 struct file *fp; 1384 struct timeval *tv; 1385 int i, fd, *fdp; 1386 void *data; 1387 socklen_t clen = control->m_len, datalen; 1388 int error, oldfds; 1389 u_int newlen; 1390 1391 UNP_UNLOCK_ASSERT(); 1392 1393 error = 0; 1394 *controlp = NULL; 1395 1396 while (cm != NULL) { 1397 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET 1398 || cm->cmsg_len > clen) { 1399 error = EINVAL; 1400 goto out; 1401 } 1402 1403 data = CMSG_DATA(cm); 1404 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1405 1406 switch (cm->cmsg_type) { 1407 /* 1408 * Fill in credential information. 1409 */ 1410 case SCM_CREDS: 1411 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred), 1412 SCM_CREDS, SOL_SOCKET); 1413 if (*controlp == NULL) { 1414 error = ENOBUFS; 1415 goto out; 1416 } 1417 1418 cmcred = (struct cmsgcred *) 1419 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1420 cmcred->cmcred_pid = p->p_pid; 1421 cmcred->cmcred_uid = td->td_ucred->cr_ruid; 1422 cmcred->cmcred_gid = td->td_ucred->cr_rgid; 1423 cmcred->cmcred_euid = td->td_ucred->cr_uid; 1424 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups, 1425 CMGROUP_MAX); 1426 for (i = 0; i < cmcred->cmcred_ngroups; i++) 1427 cmcred->cmcred_groups[i] = 1428 td->td_ucred->cr_groups[i]; 1429 break; 1430 1431 case SCM_RIGHTS: 1432 oldfds = datalen / sizeof (int); 1433 /* 1434 * Check that all the FDs passed in refer to legal 1435 * files. If not, reject the entire operation. 1436 */ 1437 fdp = data; 1438 FILEDESC_LOCK(fdescp); 1439 for (i = 0; i < oldfds; i++) { 1440 fd = *fdp++; 1441 if ((unsigned)fd >= fdescp->fd_nfiles || 1442 fdescp->fd_ofiles[fd] == NULL) { 1443 FILEDESC_UNLOCK(fdescp); 1444 error = EBADF; 1445 goto out; 1446 } 1447 fp = fdescp->fd_ofiles[fd]; 1448 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { 1449 FILEDESC_UNLOCK(fdescp); 1450 error = EOPNOTSUPP; 1451 goto out; 1452 } 1453 1454 } 1455 /* 1456 * Now replace the integer FDs with pointers to the 1457 * associated global file table entry.. 1458 */ 1459 newlen = oldfds * sizeof(struct file *); 1460 *controlp = sbcreatecontrol(NULL, newlen, 1461 SCM_RIGHTS, SOL_SOCKET); 1462 if (*controlp == NULL) { 1463 FILEDESC_UNLOCK(fdescp); 1464 error = E2BIG; 1465 goto out; 1466 } 1467 1468 fdp = data; 1469 rp = (struct file **) 1470 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1471 for (i = 0; i < oldfds; i++) { 1472 fp = fdescp->fd_ofiles[*fdp++]; 1473 *rp++ = fp; 1474 FILE_LOCK(fp); 1475 fp->f_count++; 1476 fp->f_msgcount++; 1477 FILE_UNLOCK(fp); 1478 unp_rights++; 1479 } 1480 FILEDESC_UNLOCK(fdescp); 1481 break; 1482 1483 case SCM_TIMESTAMP: 1484 *controlp = sbcreatecontrol(NULL, sizeof(*tv), 1485 SCM_TIMESTAMP, SOL_SOCKET); 1486 if (*controlp == NULL) { 1487 error = ENOBUFS; 1488 goto out; 1489 } 1490 tv = (struct timeval *) 1491 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1492 microtime(tv); 1493 break; 1494 1495 default: 1496 error = EINVAL; 1497 goto out; 1498 } 1499 1500 controlp = &(*controlp)->m_next; 1501 1502 if (CMSG_SPACE(datalen) < clen) { 1503 clen -= CMSG_SPACE(datalen); 1504 cm = (struct cmsghdr *) 1505 ((caddr_t)cm + CMSG_SPACE(datalen)); 1506 } else { 1507 clen = 0; 1508 cm = NULL; 1509 } 1510 } 1511 1512out: 1513 m_freem(control); 1514 1515 return (error); 1516} 1517 1518struct mbuf * 1519unp_addsockcred(struct thread *td, struct mbuf *control) 1520{ 1521 struct mbuf *m, *n, *n_prev; 1522 struct sockcred *sc; 1523 const struct cmsghdr *cm; 1524 int ngroups; 1525 int i; 1526 1527 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); 1528 1529 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET); 1530 if (m == NULL) 1531 return (control); 1532 1533 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *)); 1534 sc->sc_uid = td->td_ucred->cr_ruid; 1535 sc->sc_euid = td->td_ucred->cr_uid; 1536 sc->sc_gid = td->td_ucred->cr_rgid; 1537 sc->sc_egid = td->td_ucred->cr_gid; 1538 sc->sc_ngroups = ngroups; 1539 for (i = 0; i < sc->sc_ngroups; i++) 1540 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 1541 1542 /* 1543 * Unlink SCM_CREDS control messages (struct cmsgcred), since just 1544 * created SCM_CREDS control message (struct sockcred) has another 1545 * format. 1546 */ 1547 if (control != NULL) 1548 for (n = control, n_prev = NULL; n != NULL;) { 1549 cm = mtod(n, struct cmsghdr *); 1550 if (cm->cmsg_level == SOL_SOCKET && 1551 cm->cmsg_type == SCM_CREDS) { 1552 if (n_prev == NULL) 1553 control = n->m_next; 1554 else 1555 n_prev->m_next = n->m_next; 1556 n = m_free(n); 1557 } else { 1558 n_prev = n; 1559 n = n->m_next; 1560 } 1561 } 1562 1563 /* Prepend it to the head. */ 1564 m->m_next = control; 1565 1566 return (m); 1567} 1568 1569/* 1570 * unp_defer indicates whether additional work has been defered for a future 1571 * pass through unp_gc(). It is thread local and does not require explicit 1572 * synchronization. 1573 */ 1574static int unp_defer; 1575 1576static int unp_taskcount; 1577SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, ""); 1578 1579static int unp_recycled; 1580SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, ""); 1581 1582static void 1583unp_gc(__unused void *arg, int pending) 1584{ 1585 struct file *fp, *nextfp; 1586 struct socket *so; 1587 struct file **extra_ref, **fpp; 1588 int nunref, i; 1589 int nfiles_snap; 1590 int nfiles_slack = 20; 1591 1592 unp_taskcount++; 1593 unp_defer = 0; 1594 /* 1595 * Before going through all this, set all FDs to be NOT defered and 1596 * NOT externally accessible. 1597 */ 1598 sx_slock(&filelist_lock); 1599 LIST_FOREACH(fp, &filehead, f_list) 1600 fp->f_gcflag &= ~(FMARK|FDEFER); 1601 do { 1602 KASSERT(unp_defer >= 0, ("unp_gc: unp_defer %d", unp_defer)); 1603 LIST_FOREACH(fp, &filehead, f_list) { 1604 FILE_LOCK(fp); 1605 /* 1606 * If the file is not open, skip it -- could be a 1607 * file in the process of being opened, or in the 1608 * process of being closed. If the file is 1609 * "closing", it may have been marked for deferred 1610 * consideration. Clear the flag now if so. 1611 */ 1612 if (fp->f_count == 0) { 1613 if (fp->f_gcflag & FDEFER) 1614 unp_defer--; 1615 fp->f_gcflag &= ~(FMARK|FDEFER); 1616 FILE_UNLOCK(fp); 1617 continue; 1618 } 1619 /* 1620 * If we already marked it as 'defer' in a previous 1621 * pass, then try process it this time and un-mark 1622 * it. 1623 */ 1624 if (fp->f_gcflag & FDEFER) { 1625 fp->f_gcflag &= ~FDEFER; 1626 unp_defer--; 1627 } else { 1628 /* 1629 * if it's not defered, then check if it's 1630 * already marked.. if so skip it 1631 */ 1632 if (fp->f_gcflag & FMARK) { 1633 FILE_UNLOCK(fp); 1634 continue; 1635 } 1636 /* 1637 * If all references are from messages in 1638 * transit, then skip it. it's not externally 1639 * accessible. 1640 */ 1641 if (fp->f_count == fp->f_msgcount) { 1642 FILE_UNLOCK(fp); 1643 continue; 1644 } 1645 /* 1646 * If it got this far then it must be 1647 * externally accessible. 1648 */ 1649 fp->f_gcflag |= FMARK; 1650 } 1651 /* 1652 * Either it was defered, or it is externally 1653 * accessible and not already marked so. Now check 1654 * if it is possibly one of OUR sockets. 1655 */ 1656 if (fp->f_type != DTYPE_SOCKET || 1657 (so = fp->f_data) == NULL) { 1658 FILE_UNLOCK(fp); 1659 continue; 1660 } 1661 FILE_UNLOCK(fp); 1662 if (so->so_proto->pr_domain != &localdomain || 1663 (so->so_proto->pr_flags&PR_RIGHTS) == 0) 1664 continue; 1665 /* 1666 * So, Ok, it's one of our sockets and it IS 1667 * externally accessible (or was defered). Now we 1668 * look to see if we hold any file descriptors in its 1669 * message buffers. Follow those links and mark them 1670 * as accessible too. 1671 */ 1672 SOCKBUF_LOCK(&so->so_rcv); 1673 unp_scan(so->so_rcv.sb_mb, unp_mark); 1674 SOCKBUF_UNLOCK(&so->so_rcv); 1675 } 1676 } while (unp_defer); 1677 sx_sunlock(&filelist_lock); 1678 /* 1679 * XXXRW: The following comments need updating for a post-SMPng and 1680 * deferred unp_gc() world, but are still generally accurate. 1681 * 1682 * We grab an extra reference to each of the file table entries that 1683 * are not otherwise accessible and then free the rights that are 1684 * stored in messages on them. 1685 * 1686 * The bug in the orginal code is a little tricky, so I'll describe 1687 * what's wrong with it here. 1688 * 1689 * It is incorrect to simply unp_discard each entry for f_msgcount 1690 * times -- consider the case of sockets A and B that contain 1691 * references to each other. On a last close of some other socket, 1692 * we trigger a gc since the number of outstanding rights (unp_rights) 1693 * is non-zero. If during the sweep phase the gc code unp_discards, 1694 * we end up doing a (full) closef on the descriptor. A closef on A 1695 * results in the following chain. Closef calls soo_close, which 1696 * calls soclose. Soclose calls first (through the switch 1697 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply 1698 * returns because the previous instance had set unp_gcing, and we 1699 * return all the way back to soclose, which marks the socket with 1700 * SS_NOFDREF, and then calls sofree. Sofree calls sorflush to free 1701 * up the rights that are queued in messages on the socket A, i.e., 1702 * the reference on B. The sorflush calls via the dom_dispose switch 1703 * unp_dispose, which unp_scans with unp_discard. This second 1704 * instance of unp_discard just calls closef on B. 1705 * 1706 * Well, a similar chain occurs on B, resulting in a sorflush on B, 1707 * which results in another closef on A. Unfortunately, A is already 1708 * being closed, and the descriptor has already been marked with 1709 * SS_NOFDREF, and soclose panics at this point. 1710 * 1711 * Here, we first take an extra reference to each inaccessible 1712 * descriptor. Then, we call sorflush ourself, since we know it is a 1713 * Unix domain socket anyhow. After we destroy all the rights 1714 * carried in messages, we do a last closef to get rid of our extra 1715 * reference. This is the last close, and the unp_detach etc will 1716 * shut down the socket. 1717 * 1718 * 91/09/19, bsy@cs.cmu.edu 1719 */ 1720again: 1721 nfiles_snap = openfiles + nfiles_slack; /* some slack */ 1722 extra_ref = malloc(nfiles_snap * sizeof(struct file *), M_TEMP, 1723 M_WAITOK); 1724 sx_slock(&filelist_lock); 1725 if (nfiles_snap < openfiles) { 1726 sx_sunlock(&filelist_lock); 1727 free(extra_ref, M_TEMP); 1728 nfiles_slack += 20; 1729 goto again; 1730 } 1731 for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref; 1732 fp != NULL; fp = nextfp) { 1733 nextfp = LIST_NEXT(fp, f_list); 1734 FILE_LOCK(fp); 1735 /* 1736 * If it's not open, skip it 1737 */ 1738 if (fp->f_count == 0) { 1739 FILE_UNLOCK(fp); 1740 continue; 1741 } 1742 /* 1743 * If all refs are from msgs, and it's not marked accessible 1744 * then it must be referenced from some unreachable cycle of 1745 * (shut-down) FDs, so include it in our list of FDs to 1746 * remove. 1747 */ 1748 if (fp->f_count == fp->f_msgcount && !(fp->f_gcflag & FMARK)) { 1749 *fpp++ = fp; 1750 nunref++; 1751 fp->f_count++; 1752 } 1753 FILE_UNLOCK(fp); 1754 } 1755 sx_sunlock(&filelist_lock); 1756 /* 1757 * For each FD on our hit list, do the following two things: 1758 */ 1759 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1760 struct file *tfp = *fpp; 1761 FILE_LOCK(tfp); 1762 if (tfp->f_type == DTYPE_SOCKET && 1763 tfp->f_data != NULL) { 1764 FILE_UNLOCK(tfp); 1765 sorflush(tfp->f_data); 1766 } else { 1767 FILE_UNLOCK(tfp); 1768 } 1769 } 1770 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1771 closef(*fpp, (struct thread *) NULL); 1772 unp_recycled++; 1773 } 1774 free(extra_ref, M_TEMP); 1775} 1776 1777void 1778unp_dispose(struct mbuf *m) 1779{ 1780 1781 if (m) 1782 unp_scan(m, unp_discard); 1783} 1784 1785static int 1786unp_listen(struct socket *so, struct unpcb *unp, int backlog, 1787 struct thread *td) 1788{ 1789 int error; 1790 1791 UNP_LOCK_ASSERT(); 1792 1793 SOCK_LOCK(so); 1794 error = solisten_proto_check(so); 1795 if (error == 0) { 1796 cru2x(td->td_ucred, &unp->unp_peercred); 1797 unp->unp_flags |= UNP_HAVEPCCACHED; 1798 solisten_proto(so, backlog); 1799 } 1800 SOCK_UNLOCK(so); 1801 return (error); 1802} 1803 1804static void 1805unp_scan(struct mbuf *m0, void (*op)(struct file *)) 1806{ 1807 struct mbuf *m; 1808 struct file **rp; 1809 struct cmsghdr *cm; 1810 void *data; 1811 int i; 1812 socklen_t clen, datalen; 1813 int qfds; 1814 1815 while (m0 != NULL) { 1816 for (m = m0; m; m = m->m_next) { 1817 if (m->m_type != MT_CONTROL) 1818 continue; 1819 1820 cm = mtod(m, struct cmsghdr *); 1821 clen = m->m_len; 1822 1823 while (cm != NULL) { 1824 if (sizeof(*cm) > clen || cm->cmsg_len > clen) 1825 break; 1826 1827 data = CMSG_DATA(cm); 1828 datalen = (caddr_t)cm + cm->cmsg_len 1829 - (caddr_t)data; 1830 1831 if (cm->cmsg_level == SOL_SOCKET && 1832 cm->cmsg_type == SCM_RIGHTS) { 1833 qfds = datalen / sizeof (struct file *); 1834 rp = data; 1835 for (i = 0; i < qfds; i++) 1836 (*op)(*rp++); 1837 } 1838 1839 if (CMSG_SPACE(datalen) < clen) { 1840 clen -= CMSG_SPACE(datalen); 1841 cm = (struct cmsghdr *) 1842 ((caddr_t)cm + CMSG_SPACE(datalen)); 1843 } else { 1844 clen = 0; 1845 cm = NULL; 1846 } 1847 } 1848 } 1849 m0 = m0->m_act; 1850 } 1851} 1852 1853static void 1854unp_mark(struct file *fp) 1855{ 1856 if (fp->f_gcflag & FMARK) 1857 return; 1858 unp_defer++; 1859 fp->f_gcflag |= (FMARK|FDEFER); 1860} 1861 1862static void 1863unp_discard(struct file *fp) 1864{ 1865 UNP_LOCK(); 1866 FILE_LOCK(fp); 1867 fp->f_msgcount--; 1868 unp_rights--; 1869 FILE_UNLOCK(fp); 1870 UNP_UNLOCK(); 1871 (void) closef(fp, (struct thread *)NULL); 1872} 1873