uipc_usrreq.c revision 160868
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 160868 2006-07-31 23:00:05Z 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 } else { 603 if (unp->unp_conn == NULL) { 604 error = ENOTCONN; 605 break; 606 } 607 } 608 /* 609 * Because connect() and send() are non-atomic in a sendto() 610 * with a target address, it's possible that the socket will 611 * have disconnected before the send() can run. In that case 612 * return the slightly counter-intuitive but otherwise 613 * correct error that the socket is not connected. 614 */ 615 unp2 = unp->unp_conn; 616 if (unp2 == NULL) { 617 error = ENOTCONN; 618 break; 619 } 620 so2 = unp2->unp_socket; 621 if (unp->unp_addr != NULL) 622 from = (struct sockaddr *)unp->unp_addr; 623 else 624 from = &sun_noname; 625 if (unp2->unp_flags & UNP_WANTCRED) 626 control = unp_addsockcred(td, control); 627 SOCKBUF_LOCK(&so2->so_rcv); 628 if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) { 629 sorwakeup_locked(so2); 630 m = NULL; 631 control = NULL; 632 } else { 633 SOCKBUF_UNLOCK(&so2->so_rcv); 634 error = ENOBUFS; 635 } 636 if (nam != NULL) 637 unp_disconnect(unp); 638 break; 639 } 640 641 case SOCK_STREAM: 642 /* 643 * Connect if not connected yet. 644 * 645 * Note: A better implementation would complain if not equal 646 * to the peer's address. 647 */ 648 if ((so->so_state & SS_ISCONNECTED) == 0) { 649 if (nam != NULL) { 650 error = unp_connect(so, nam, td); 651 if (error) 652 break; /* XXX */ 653 } else { 654 error = ENOTCONN; 655 break; 656 } 657 } 658 659 /* Lockless read. */ 660 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 661 error = EPIPE; 662 break; 663 } 664 /* 665 * Because connect() and send() are non-atomic in a sendto() 666 * with a target address, it's possible that the socket will 667 * have disconnected before the send() can run. In that case 668 * return the slightly counter-intuitive but otherwise 669 * correct error that the socket is not connected. 670 */ 671 unp2 = unp->unp_conn; 672 if (unp2 == NULL) { 673 error = ENOTCONN; 674 break; 675 } 676 so2 = unp2->unp_socket; 677 SOCKBUF_LOCK(&so2->so_rcv); 678 if (unp2->unp_flags & UNP_WANTCRED) { 679 /* 680 * Credentials are passed only once on 681 * SOCK_STREAM. 682 */ 683 unp2->unp_flags &= ~UNP_WANTCRED; 684 control = unp_addsockcred(td, control); 685 } 686 /* 687 * Send to paired receive port, and then reduce send buffer 688 * hiwater marks to maintain backpressure. Wake up readers. 689 */ 690 if (control != NULL) { 691 if (sbappendcontrol_locked(&so2->so_rcv, m, control)) 692 control = NULL; 693 } else { 694 sbappend_locked(&so2->so_rcv, m); 695 } 696 mbcnt = so2->so_rcv.sb_mbcnt - unp2->unp_mbcnt; 697 unp2->unp_mbcnt = so2->so_rcv.sb_mbcnt; 698 sbcc = so2->so_rcv.sb_cc; 699 sorwakeup_locked(so2); 700 701 SOCKBUF_LOCK(&so->so_snd); 702 newhiwat = so->so_snd.sb_hiwat - (sbcc - unp2->unp_cc); 703 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat, 704 newhiwat, RLIM_INFINITY); 705 so->so_snd.sb_mbmax -= mbcnt; 706 SOCKBUF_UNLOCK(&so->so_snd); 707 708 unp2->unp_cc = sbcc; 709 m = NULL; 710 break; 711 712 default: 713 panic("uipc_send unknown socktype"); 714 } 715 716 /* 717 * SEND_EOF is equivalent to a SEND followed by 718 * a SHUTDOWN. 719 */ 720 if (flags & PRUS_EOF) { 721 socantsendmore(so); 722 unp_shutdown(unp); 723 } 724 UNP_UNLOCK(); 725 726 if (control != NULL && error != 0) 727 unp_dispose(control); 728 729release: 730 if (control != NULL) 731 m_freem(control); 732 if (m != NULL) 733 m_freem(m); 734 return (error); 735} 736 737static int 738uipc_sense(struct socket *so, struct stat *sb) 739{ 740 struct unpcb *unp; 741 struct socket *so2; 742 743 unp = sotounpcb(so); 744 KASSERT(unp != NULL, ("uipc_sense: unp == NULL")); 745 UNP_LOCK(); 746 sb->st_blksize = so->so_snd.sb_hiwat; 747 if (so->so_type == SOCK_STREAM && unp->unp_conn != NULL) { 748 so2 = unp->unp_conn->unp_socket; 749 sb->st_blksize += so2->so_rcv.sb_cc; 750 } 751 sb->st_dev = NODEV; 752 if (unp->unp_ino == 0) 753 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino; 754 sb->st_ino = unp->unp_ino; 755 UNP_UNLOCK(); 756 return (0); 757} 758 759static int 760uipc_shutdown(struct socket *so) 761{ 762 struct unpcb *unp; 763 764 unp = sotounpcb(so); 765 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL")); 766 UNP_LOCK(); 767 socantsendmore(so); 768 unp_shutdown(unp); 769 UNP_UNLOCK(); 770 return (0); 771} 772 773static int 774uipc_sockaddr(struct socket *so, struct sockaddr **nam) 775{ 776 struct unpcb *unp; 777 const struct sockaddr *sa; 778 779 unp = sotounpcb(so); 780 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL")); 781 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 782 UNP_LOCK(); 783 if (unp->unp_addr != NULL) 784 sa = (struct sockaddr *) unp->unp_addr; 785 else 786 sa = &sun_noname; 787 bcopy(sa, *nam, sa->sa_len); 788 UNP_UNLOCK(); 789 return (0); 790} 791 792struct pr_usrreqs uipc_usrreqs = { 793 .pru_abort = uipc_abort, 794 .pru_accept = uipc_accept, 795 .pru_attach = uipc_attach, 796 .pru_bind = uipc_bind, 797 .pru_connect = uipc_connect, 798 .pru_connect2 = uipc_connect2, 799 .pru_detach = uipc_detach, 800 .pru_disconnect = uipc_disconnect, 801 .pru_listen = uipc_listen, 802 .pru_peeraddr = uipc_peeraddr, 803 .pru_rcvd = uipc_rcvd, 804 .pru_send = uipc_send, 805 .pru_sense = uipc_sense, 806 .pru_shutdown = uipc_shutdown, 807 .pru_sockaddr = uipc_sockaddr, 808 .pru_sosend = sosend_generic, 809 .pru_soreceive = soreceive_generic, 810 .pru_sopoll = sopoll_generic, 811 .pru_close = uipc_close, 812}; 813 814int 815uipc_ctloutput(struct socket *so, struct sockopt *sopt) 816{ 817 struct unpcb *unp; 818 struct xucred xu; 819 int error, optval; 820 821 if (sopt->sopt_level != 0) 822 return (EINVAL); 823 824 unp = sotounpcb(so); 825 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL")); 826 UNP_LOCK(); 827 error = 0; 828 switch (sopt->sopt_dir) { 829 case SOPT_GET: 830 switch (sopt->sopt_name) { 831 case LOCAL_PEERCRED: 832 if (unp->unp_flags & UNP_HAVEPC) 833 xu = unp->unp_peercred; 834 else { 835 if (so->so_type == SOCK_STREAM) 836 error = ENOTCONN; 837 else 838 error = EINVAL; 839 } 840 if (error == 0) 841 error = sooptcopyout(sopt, &xu, sizeof(xu)); 842 break; 843 case LOCAL_CREDS: 844 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0; 845 error = sooptcopyout(sopt, &optval, sizeof(optval)); 846 break; 847 case LOCAL_CONNWAIT: 848 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0; 849 error = sooptcopyout(sopt, &optval, sizeof(optval)); 850 break; 851 default: 852 error = EOPNOTSUPP; 853 break; 854 } 855 break; 856 case SOPT_SET: 857 switch (sopt->sopt_name) { 858 case LOCAL_CREDS: 859 case LOCAL_CONNWAIT: 860 error = sooptcopyin(sopt, &optval, sizeof(optval), 861 sizeof(optval)); 862 if (error) 863 break; 864 865#define OPTSET(bit) \ 866 if (optval) \ 867 unp->unp_flags |= bit; \ 868 else \ 869 unp->unp_flags &= ~bit; 870 871 switch (sopt->sopt_name) { 872 case LOCAL_CREDS: 873 OPTSET(UNP_WANTCRED); 874 break; 875 case LOCAL_CONNWAIT: 876 OPTSET(UNP_CONNWAIT); 877 break; 878 default: 879 break; 880 } 881 break; 882#undef OPTSET 883 default: 884 error = ENOPROTOOPT; 885 break; 886 } 887 break; 888 default: 889 error = EOPNOTSUPP; 890 break; 891 } 892 UNP_UNLOCK(); 893 return (error); 894} 895 896static int 897unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 898{ 899 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 900 struct vnode *vp; 901 struct socket *so2, *so3; 902 struct unpcb *unp, *unp2, *unp3; 903 int error, len; 904 struct nameidata nd; 905 char buf[SOCK_MAXADDRLEN]; 906 struct sockaddr *sa; 907 908 UNP_LOCK_ASSERT(); 909 910 unp = sotounpcb(so); 911 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 912 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 913 if (len <= 0) 914 return (EINVAL); 915 strlcpy(buf, soun->sun_path, len + 1); 916 if (unp->unp_flags & UNP_CONNECTING) { 917 UNP_UNLOCK(); 918 return (EALREADY); 919 } 920 UNP_UNLOCK(); 921 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 922 mtx_lock(&Giant); 923 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, td); 924 error = namei(&nd); 925 if (error) 926 vp = NULL; 927 else 928 vp = nd.ni_vp; 929 ASSERT_VOP_LOCKED(vp, "unp_connect"); 930 NDFREE(&nd, NDF_ONLY_PNBUF); 931 if (error) 932 goto bad; 933 934 if (vp->v_type != VSOCK) { 935 error = ENOTSOCK; 936 goto bad; 937 } 938 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); 939 if (error) 940 goto bad; 941 mtx_unlock(&Giant); 942 UNP_LOCK(); 943 unp = sotounpcb(so); 944 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 945 so2 = vp->v_socket; 946 if (so2 == NULL) { 947 error = ECONNREFUSED; 948 goto bad2; 949 } 950 if (so->so_type != so2->so_type) { 951 error = EPROTOTYPE; 952 goto bad2; 953 } 954 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 955 if (so2->so_options & SO_ACCEPTCONN) { 956 /* 957 * NB: drop locks here so unp_attach is entered w/o 958 * locks; this avoids a recursive lock of the head 959 * and holding sleep locks across a (potentially) 960 * blocking malloc. 961 */ 962 UNP_UNLOCK(); 963 so3 = sonewconn(so2, 0); 964 UNP_LOCK(); 965 } else 966 so3 = NULL; 967 if (so3 == NULL) { 968 error = ECONNREFUSED; 969 goto bad2; 970 } 971 unp = sotounpcb(so); 972 unp2 = sotounpcb(so2); 973 unp3 = sotounpcb(so3); 974 if (unp2->unp_addr != NULL) { 975 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); 976 unp3->unp_addr = (struct sockaddr_un *) sa; 977 sa = NULL; 978 } 979 /* 980 * unp_peercred management: 981 * 982 * The connecter's (client's) credentials are copied from its 983 * process structure at the time of connect() (which is now). 984 */ 985 cru2x(td->td_ucred, &unp3->unp_peercred); 986 unp3->unp_flags |= UNP_HAVEPC; 987 /* 988 * The receiver's (server's) credentials are copied from the 989 * unp_peercred member of socket on which the former called 990 * listen(); unp_listen() cached that process's credentials 991 * at that time so we can use them now. 992 */ 993 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED, 994 ("unp_connect: listener without cached peercred")); 995 memcpy(&unp->unp_peercred, &unp2->unp_peercred, 996 sizeof(unp->unp_peercred)); 997 unp->unp_flags |= UNP_HAVEPC; 998 if (unp2->unp_flags & UNP_WANTCRED) 999 unp3->unp_flags |= UNP_WANTCRED; 1000#ifdef MAC 1001 SOCK_LOCK(so); 1002 mac_set_socket_peer_from_socket(so, so3); 1003 mac_set_socket_peer_from_socket(so3, so); 1004 SOCK_UNLOCK(so); 1005#endif 1006 1007 so2 = so3; 1008 } 1009 error = unp_connect2(so, so2, PRU_CONNECT); 1010bad2: 1011 UNP_UNLOCK(); 1012 mtx_lock(&Giant); 1013bad: 1014 mtx_assert(&Giant, MA_OWNED); 1015 if (vp != NULL) 1016 vput(vp); 1017 mtx_unlock(&Giant); 1018 free(sa, M_SONAME); 1019 UNP_LOCK(); 1020 unp->unp_flags &= ~UNP_CONNECTING; 1021 return (error); 1022} 1023 1024static int 1025unp_connect2(struct socket *so, struct socket *so2, int req) 1026{ 1027 struct unpcb *unp = sotounpcb(so); 1028 struct unpcb *unp2; 1029 1030 UNP_LOCK_ASSERT(); 1031 1032 if (so2->so_type != so->so_type) 1033 return (EPROTOTYPE); 1034 unp2 = sotounpcb(so2); 1035 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL")); 1036 unp->unp_conn = unp2; 1037 switch (so->so_type) { 1038 1039 case SOCK_DGRAM: 1040 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 1041 soisconnected(so); 1042 break; 1043 1044 case SOCK_STREAM: 1045 unp2->unp_conn = unp; 1046 if (req == PRU_CONNECT && 1047 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 1048 soisconnecting(so); 1049 else 1050 soisconnected(so); 1051 soisconnected(so2); 1052 break; 1053 1054 default: 1055 panic("unp_connect2"); 1056 } 1057 return (0); 1058} 1059 1060static void 1061unp_disconnect(struct unpcb *unp) 1062{ 1063 struct unpcb *unp2 = unp->unp_conn; 1064 struct socket *so; 1065 1066 UNP_LOCK_ASSERT(); 1067 1068 if (unp2 == NULL) 1069 return; 1070 unp->unp_conn = NULL; 1071 switch (unp->unp_socket->so_type) { 1072 case SOCK_DGRAM: 1073 LIST_REMOVE(unp, unp_reflink); 1074 so = unp->unp_socket; 1075 SOCK_LOCK(so); 1076 so->so_state &= ~SS_ISCONNECTED; 1077 SOCK_UNLOCK(so); 1078 break; 1079 1080 case SOCK_STREAM: 1081 soisdisconnected(unp->unp_socket); 1082 unp2->unp_conn = NULL; 1083 soisdisconnected(unp2->unp_socket); 1084 break; 1085 } 1086} 1087 1088/* 1089 * unp_pcblist() assumes that UNIX domain socket memory is never reclaimed by 1090 * the zone (UMA_ZONE_NOFREE), and as such potentially stale pointers are 1091 * safe to reference. It first scans the list of struct unpcb's to generate 1092 * a pointer list, then it rescans its list one entry at a time to 1093 * externalize and copyout. It checks the generation number to see if a 1094 * struct unpcb has been reused, and will skip it if so. 1095 */ 1096static int 1097unp_pcblist(SYSCTL_HANDLER_ARGS) 1098{ 1099 int error, i, n; 1100 struct unpcb *unp, **unp_list; 1101 unp_gen_t gencnt; 1102 struct xunpgen *xug; 1103 struct unp_head *head; 1104 struct xunpcb *xu; 1105 1106 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead); 1107 1108 /* 1109 * The process of preparing the PCB list is too time-consuming and 1110 * resource-intensive to repeat twice on every request. 1111 */ 1112 if (req->oldptr == NULL) { 1113 n = unp_count; 1114 req->oldidx = 2 * (sizeof *xug) 1115 + (n + n/8) * sizeof(struct xunpcb); 1116 return (0); 1117 } 1118 1119 if (req->newptr != NULL) 1120 return (EPERM); 1121 1122 /* 1123 * OK, now we're committed to doing something. 1124 */ 1125 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK); 1126 UNP_LOCK(); 1127 gencnt = unp_gencnt; 1128 n = unp_count; 1129 UNP_UNLOCK(); 1130 1131 xug->xug_len = sizeof *xug; 1132 xug->xug_count = n; 1133 xug->xug_gen = gencnt; 1134 xug->xug_sogen = so_gencnt; 1135 error = SYSCTL_OUT(req, xug, sizeof *xug); 1136 if (error) { 1137 free(xug, M_TEMP); 1138 return (error); 1139 } 1140 1141 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 1142 1143 UNP_LOCK(); 1144 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 1145 unp = LIST_NEXT(unp, unp_link)) { 1146 if (unp->unp_gencnt <= gencnt) { 1147 if (cr_cansee(req->td->td_ucred, 1148 unp->unp_socket->so_cred)) 1149 continue; 1150 unp_list[i++] = unp; 1151 } 1152 } 1153 UNP_UNLOCK(); 1154 n = i; /* In case we lost some during malloc. */ 1155 1156 error = 0; 1157 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO); 1158 for (i = 0; i < n; i++) { 1159 unp = unp_list[i]; 1160 if (unp->unp_gencnt <= gencnt) { 1161 xu->xu_len = sizeof *xu; 1162 xu->xu_unpp = unp; 1163 /* 1164 * XXX - need more locking here to protect against 1165 * connect/disconnect races for SMP. 1166 */ 1167 if (unp->unp_addr != NULL) 1168 bcopy(unp->unp_addr, &xu->xu_addr, 1169 unp->unp_addr->sun_len); 1170 if (unp->unp_conn != NULL && 1171 unp->unp_conn->unp_addr != NULL) 1172 bcopy(unp->unp_conn->unp_addr, 1173 &xu->xu_caddr, 1174 unp->unp_conn->unp_addr->sun_len); 1175 bcopy(unp, &xu->xu_unp, sizeof *unp); 1176 sotoxsocket(unp->unp_socket, &xu->xu_socket); 1177 error = SYSCTL_OUT(req, xu, sizeof *xu); 1178 } 1179 } 1180 free(xu, M_TEMP); 1181 if (!error) { 1182 /* 1183 * Give the user an updated idea of our state. If the 1184 * generation differs from what we told her before, she knows 1185 * that something happened while we were processing this 1186 * request, and it might be necessary to retry. 1187 */ 1188 xug->xug_gen = unp_gencnt; 1189 xug->xug_sogen = so_gencnt; 1190 xug->xug_count = unp_count; 1191 error = SYSCTL_OUT(req, xug, sizeof *xug); 1192 } 1193 free(unp_list, M_TEMP); 1194 free(xug, M_TEMP); 1195 return (error); 1196} 1197 1198SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD, 1199 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 1200 "List of active local datagram sockets"); 1201SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD, 1202 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 1203 "List of active local stream sockets"); 1204 1205static void 1206unp_shutdown(struct unpcb *unp) 1207{ 1208 struct socket *so; 1209 1210 UNP_LOCK_ASSERT(); 1211 1212 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn && 1213 (so = unp->unp_conn->unp_socket)) 1214 socantrcvmore(so); 1215} 1216 1217static void 1218unp_drop(struct unpcb *unp, int errno) 1219{ 1220 struct socket *so = unp->unp_socket; 1221 1222 UNP_LOCK_ASSERT(); 1223 1224 so->so_error = errno; 1225 unp_disconnect(unp); 1226} 1227 1228static void 1229unp_freerights(struct file **rp, int fdcount) 1230{ 1231 int i; 1232 struct file *fp; 1233 1234 for (i = 0; i < fdcount; i++) { 1235 fp = *rp; 1236 /* 1237 * Zero the pointer before calling unp_discard since it may 1238 * end up in unp_gc().. 1239 * 1240 * XXXRW: This is less true than it used to be. 1241 */ 1242 *rp++ = 0; 1243 unp_discard(fp); 1244 } 1245} 1246 1247int 1248unp_externalize(struct mbuf *control, struct mbuf **controlp) 1249{ 1250 struct thread *td = curthread; /* XXX */ 1251 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1252 int i; 1253 int *fdp; 1254 struct file **rp; 1255 struct file *fp; 1256 void *data; 1257 socklen_t clen = control->m_len, datalen; 1258 int error, newfds; 1259 int f; 1260 u_int newlen; 1261 1262 UNP_UNLOCK_ASSERT(); 1263 1264 error = 0; 1265 if (controlp != NULL) /* controlp == NULL => free control messages */ 1266 *controlp = NULL; 1267 1268 while (cm != NULL) { 1269 if (sizeof(*cm) > clen || cm->cmsg_len > clen) { 1270 error = EINVAL; 1271 break; 1272 } 1273 1274 data = CMSG_DATA(cm); 1275 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1276 1277 if (cm->cmsg_level == SOL_SOCKET 1278 && cm->cmsg_type == SCM_RIGHTS) { 1279 newfds = datalen / sizeof(struct file *); 1280 rp = data; 1281 1282 /* If we're not outputting the descriptors free them. */ 1283 if (error || controlp == NULL) { 1284 unp_freerights(rp, newfds); 1285 goto next; 1286 } 1287 FILEDESC_LOCK(td->td_proc->p_fd); 1288 /* if the new FD's will not fit free them. */ 1289 if (!fdavail(td, newfds)) { 1290 FILEDESC_UNLOCK(td->td_proc->p_fd); 1291 error = EMSGSIZE; 1292 unp_freerights(rp, newfds); 1293 goto next; 1294 } 1295 /* 1296 * Now change each pointer to an fd in the global 1297 * table to an integer that is the index to the local 1298 * fd table entry that we set up to point to the 1299 * global one we are transferring. 1300 */ 1301 newlen = newfds * sizeof(int); 1302 *controlp = sbcreatecontrol(NULL, newlen, 1303 SCM_RIGHTS, SOL_SOCKET); 1304 if (*controlp == NULL) { 1305 FILEDESC_UNLOCK(td->td_proc->p_fd); 1306 error = E2BIG; 1307 unp_freerights(rp, newfds); 1308 goto next; 1309 } 1310 1311 fdp = (int *) 1312 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1313 for (i = 0; i < newfds; i++) { 1314 if (fdalloc(td, 0, &f)) 1315 panic("unp_externalize fdalloc failed"); 1316 fp = *rp++; 1317 td->td_proc->p_fd->fd_ofiles[f] = fp; 1318 FILE_LOCK(fp); 1319 fp->f_msgcount--; 1320 FILE_UNLOCK(fp); 1321 unp_rights--; 1322 *fdp++ = f; 1323 } 1324 FILEDESC_UNLOCK(td->td_proc->p_fd); 1325 } else { 1326 /* We can just copy anything else across. */ 1327 if (error || controlp == NULL) 1328 goto next; 1329 *controlp = sbcreatecontrol(NULL, datalen, 1330 cm->cmsg_type, cm->cmsg_level); 1331 if (*controlp == NULL) { 1332 error = ENOBUFS; 1333 goto next; 1334 } 1335 bcopy(data, 1336 CMSG_DATA(mtod(*controlp, struct cmsghdr *)), 1337 datalen); 1338 } 1339 1340 controlp = &(*controlp)->m_next; 1341 1342next: 1343 if (CMSG_SPACE(datalen) < clen) { 1344 clen -= CMSG_SPACE(datalen); 1345 cm = (struct cmsghdr *) 1346 ((caddr_t)cm + CMSG_SPACE(datalen)); 1347 } else { 1348 clen = 0; 1349 cm = NULL; 1350 } 1351 } 1352 1353 m_freem(control); 1354 1355 return (error); 1356} 1357 1358static void 1359unp_zone_change(void *tag) 1360{ 1361 1362 uma_zone_set_max(unp_zone, maxsockets); 1363} 1364 1365void 1366unp_init(void) 1367{ 1368 1369 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL, 1370 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 1371 if (unp_zone == NULL) 1372 panic("unp_init"); 1373 uma_zone_set_max(unp_zone, maxsockets); 1374 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change, 1375 NULL, EVENTHANDLER_PRI_ANY); 1376 LIST_INIT(&unp_dhead); 1377 LIST_INIT(&unp_shead); 1378 TASK_INIT(&unp_gc_task, 0, unp_gc, NULL); 1379 UNP_LOCK_INIT(); 1380} 1381 1382static int 1383unp_internalize(struct mbuf **controlp, struct thread *td) 1384{ 1385 struct mbuf *control = *controlp; 1386 struct proc *p = td->td_proc; 1387 struct filedesc *fdescp = p->p_fd; 1388 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1389 struct cmsgcred *cmcred; 1390 struct file **rp; 1391 struct file *fp; 1392 struct timeval *tv; 1393 int i, fd, *fdp; 1394 void *data; 1395 socklen_t clen = control->m_len, datalen; 1396 int error, oldfds; 1397 u_int newlen; 1398 1399 UNP_UNLOCK_ASSERT(); 1400 1401 error = 0; 1402 *controlp = NULL; 1403 1404 while (cm != NULL) { 1405 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET 1406 || cm->cmsg_len > clen) { 1407 error = EINVAL; 1408 goto out; 1409 } 1410 1411 data = CMSG_DATA(cm); 1412 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1413 1414 switch (cm->cmsg_type) { 1415 /* 1416 * Fill in credential information. 1417 */ 1418 case SCM_CREDS: 1419 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred), 1420 SCM_CREDS, SOL_SOCKET); 1421 if (*controlp == NULL) { 1422 error = ENOBUFS; 1423 goto out; 1424 } 1425 1426 cmcred = (struct cmsgcred *) 1427 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1428 cmcred->cmcred_pid = p->p_pid; 1429 cmcred->cmcred_uid = td->td_ucred->cr_ruid; 1430 cmcred->cmcred_gid = td->td_ucred->cr_rgid; 1431 cmcred->cmcred_euid = td->td_ucred->cr_uid; 1432 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups, 1433 CMGROUP_MAX); 1434 for (i = 0; i < cmcred->cmcred_ngroups; i++) 1435 cmcred->cmcred_groups[i] = 1436 td->td_ucred->cr_groups[i]; 1437 break; 1438 1439 case SCM_RIGHTS: 1440 oldfds = datalen / sizeof (int); 1441 /* 1442 * Check that all the FDs passed in refer to legal 1443 * files. If not, reject the entire operation. 1444 */ 1445 fdp = data; 1446 FILEDESC_LOCK(fdescp); 1447 for (i = 0; i < oldfds; i++) { 1448 fd = *fdp++; 1449 if ((unsigned)fd >= fdescp->fd_nfiles || 1450 fdescp->fd_ofiles[fd] == NULL) { 1451 FILEDESC_UNLOCK(fdescp); 1452 error = EBADF; 1453 goto out; 1454 } 1455 fp = fdescp->fd_ofiles[fd]; 1456 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { 1457 FILEDESC_UNLOCK(fdescp); 1458 error = EOPNOTSUPP; 1459 goto out; 1460 } 1461 1462 } 1463 /* 1464 * Now replace the integer FDs with pointers to the 1465 * associated global file table entry.. 1466 */ 1467 newlen = oldfds * sizeof(struct file *); 1468 *controlp = sbcreatecontrol(NULL, newlen, 1469 SCM_RIGHTS, SOL_SOCKET); 1470 if (*controlp == NULL) { 1471 FILEDESC_UNLOCK(fdescp); 1472 error = E2BIG; 1473 goto out; 1474 } 1475 1476 fdp = data; 1477 rp = (struct file **) 1478 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1479 for (i = 0; i < oldfds; i++) { 1480 fp = fdescp->fd_ofiles[*fdp++]; 1481 *rp++ = fp; 1482 FILE_LOCK(fp); 1483 fp->f_count++; 1484 fp->f_msgcount++; 1485 FILE_UNLOCK(fp); 1486 unp_rights++; 1487 } 1488 FILEDESC_UNLOCK(fdescp); 1489 break; 1490 1491 case SCM_TIMESTAMP: 1492 *controlp = sbcreatecontrol(NULL, sizeof(*tv), 1493 SCM_TIMESTAMP, SOL_SOCKET); 1494 if (*controlp == NULL) { 1495 error = ENOBUFS; 1496 goto out; 1497 } 1498 tv = (struct timeval *) 1499 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1500 microtime(tv); 1501 break; 1502 1503 default: 1504 error = EINVAL; 1505 goto out; 1506 } 1507 1508 controlp = &(*controlp)->m_next; 1509 1510 if (CMSG_SPACE(datalen) < clen) { 1511 clen -= CMSG_SPACE(datalen); 1512 cm = (struct cmsghdr *) 1513 ((caddr_t)cm + CMSG_SPACE(datalen)); 1514 } else { 1515 clen = 0; 1516 cm = NULL; 1517 } 1518 } 1519 1520out: 1521 m_freem(control); 1522 1523 return (error); 1524} 1525 1526struct mbuf * 1527unp_addsockcred(struct thread *td, struct mbuf *control) 1528{ 1529 struct mbuf *m, *n, *n_prev; 1530 struct sockcred *sc; 1531 const struct cmsghdr *cm; 1532 int ngroups; 1533 int i; 1534 1535 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); 1536 1537 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET); 1538 if (m == NULL) 1539 return (control); 1540 1541 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *)); 1542 sc->sc_uid = td->td_ucred->cr_ruid; 1543 sc->sc_euid = td->td_ucred->cr_uid; 1544 sc->sc_gid = td->td_ucred->cr_rgid; 1545 sc->sc_egid = td->td_ucred->cr_gid; 1546 sc->sc_ngroups = ngroups; 1547 for (i = 0; i < sc->sc_ngroups; i++) 1548 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 1549 1550 /* 1551 * Unlink SCM_CREDS control messages (struct cmsgcred), since just 1552 * created SCM_CREDS control message (struct sockcred) has another 1553 * format. 1554 */ 1555 if (control != NULL) 1556 for (n = control, n_prev = NULL; n != NULL;) { 1557 cm = mtod(n, struct cmsghdr *); 1558 if (cm->cmsg_level == SOL_SOCKET && 1559 cm->cmsg_type == SCM_CREDS) { 1560 if (n_prev == NULL) 1561 control = n->m_next; 1562 else 1563 n_prev->m_next = n->m_next; 1564 n = m_free(n); 1565 } else { 1566 n_prev = n; 1567 n = n->m_next; 1568 } 1569 } 1570 1571 /* Prepend it to the head. */ 1572 m->m_next = control; 1573 1574 return (m); 1575} 1576 1577/* 1578 * unp_defer indicates whether additional work has been defered for a future 1579 * pass through unp_gc(). It is thread local and does not require explicit 1580 * synchronization. 1581 */ 1582static int unp_defer; 1583 1584static int unp_taskcount; 1585SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, ""); 1586 1587static int unp_recycled; 1588SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, ""); 1589 1590static void 1591unp_gc(__unused void *arg, int pending) 1592{ 1593 struct file *fp, *nextfp; 1594 struct socket *so; 1595 struct file **extra_ref, **fpp; 1596 int nunref, i; 1597 int nfiles_snap; 1598 int nfiles_slack = 20; 1599 1600 unp_taskcount++; 1601 unp_defer = 0; 1602 /* 1603 * Before going through all this, set all FDs to be NOT defered and 1604 * NOT externally accessible. 1605 */ 1606 sx_slock(&filelist_lock); 1607 LIST_FOREACH(fp, &filehead, f_list) 1608 fp->f_gcflag &= ~(FMARK|FDEFER); 1609 do { 1610 KASSERT(unp_defer >= 0, ("unp_gc: unp_defer %d", unp_defer)); 1611 LIST_FOREACH(fp, &filehead, f_list) { 1612 FILE_LOCK(fp); 1613 /* 1614 * If the file is not open, skip it -- could be a 1615 * file in the process of being opened, or in the 1616 * process of being closed. If the file is 1617 * "closing", it may have been marked for deferred 1618 * consideration. Clear the flag now if so. 1619 */ 1620 if (fp->f_count == 0) { 1621 if (fp->f_gcflag & FDEFER) 1622 unp_defer--; 1623 fp->f_gcflag &= ~(FMARK|FDEFER); 1624 FILE_UNLOCK(fp); 1625 continue; 1626 } 1627 /* 1628 * If we already marked it as 'defer' in a previous 1629 * pass, then try process it this time and un-mark 1630 * it. 1631 */ 1632 if (fp->f_gcflag & FDEFER) { 1633 fp->f_gcflag &= ~FDEFER; 1634 unp_defer--; 1635 } else { 1636 /* 1637 * if it's not defered, then check if it's 1638 * already marked.. if so skip it 1639 */ 1640 if (fp->f_gcflag & FMARK) { 1641 FILE_UNLOCK(fp); 1642 continue; 1643 } 1644 /* 1645 * If all references are from messages in 1646 * transit, then skip it. it's not externally 1647 * accessible. 1648 */ 1649 if (fp->f_count == fp->f_msgcount) { 1650 FILE_UNLOCK(fp); 1651 continue; 1652 } 1653 /* 1654 * If it got this far then it must be 1655 * externally accessible. 1656 */ 1657 fp->f_gcflag |= FMARK; 1658 } 1659 /* 1660 * Either it was defered, or it is externally 1661 * accessible and not already marked so. Now check 1662 * if it is possibly one of OUR sockets. 1663 */ 1664 if (fp->f_type != DTYPE_SOCKET || 1665 (so = fp->f_data) == NULL) { 1666 FILE_UNLOCK(fp); 1667 continue; 1668 } 1669 FILE_UNLOCK(fp); 1670 if (so->so_proto->pr_domain != &localdomain || 1671 (so->so_proto->pr_flags&PR_RIGHTS) == 0) 1672 continue; 1673 /* 1674 * So, Ok, it's one of our sockets and it IS 1675 * externally accessible (or was defered). Now we 1676 * look to see if we hold any file descriptors in its 1677 * message buffers. Follow those links and mark them 1678 * as accessible too. 1679 */ 1680 SOCKBUF_LOCK(&so->so_rcv); 1681 unp_scan(so->so_rcv.sb_mb, unp_mark); 1682 SOCKBUF_UNLOCK(&so->so_rcv); 1683 } 1684 } while (unp_defer); 1685 sx_sunlock(&filelist_lock); 1686 /* 1687 * XXXRW: The following comments need updating for a post-SMPng and 1688 * deferred unp_gc() world, but are still generally accurate. 1689 * 1690 * We grab an extra reference to each of the file table entries that 1691 * are not otherwise accessible and then free the rights that are 1692 * stored in messages on them. 1693 * 1694 * The bug in the orginal code is a little tricky, so I'll describe 1695 * what's wrong with it here. 1696 * 1697 * It is incorrect to simply unp_discard each entry for f_msgcount 1698 * times -- consider the case of sockets A and B that contain 1699 * references to each other. On a last close of some other socket, 1700 * we trigger a gc since the number of outstanding rights (unp_rights) 1701 * is non-zero. If during the sweep phase the gc code unp_discards, 1702 * we end up doing a (full) closef on the descriptor. A closef on A 1703 * results in the following chain. Closef calls soo_close, which 1704 * calls soclose. Soclose calls first (through the switch 1705 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply 1706 * returns because the previous instance had set unp_gcing, and we 1707 * return all the way back to soclose, which marks the socket with 1708 * SS_NOFDREF, and then calls sofree. Sofree calls sorflush to free 1709 * up the rights that are queued in messages on the socket A, i.e., 1710 * the reference on B. The sorflush calls via the dom_dispose switch 1711 * unp_dispose, which unp_scans with unp_discard. This second 1712 * instance of unp_discard just calls closef on B. 1713 * 1714 * Well, a similar chain occurs on B, resulting in a sorflush on B, 1715 * which results in another closef on A. Unfortunately, A is already 1716 * being closed, and the descriptor has already been marked with 1717 * SS_NOFDREF, and soclose panics at this point. 1718 * 1719 * Here, we first take an extra reference to each inaccessible 1720 * descriptor. Then, we call sorflush ourself, since we know it is a 1721 * Unix domain socket anyhow. After we destroy all the rights 1722 * carried in messages, we do a last closef to get rid of our extra 1723 * reference. This is the last close, and the unp_detach etc will 1724 * shut down the socket. 1725 * 1726 * 91/09/19, bsy@cs.cmu.edu 1727 */ 1728again: 1729 nfiles_snap = openfiles + nfiles_slack; /* some slack */ 1730 extra_ref = malloc(nfiles_snap * sizeof(struct file *), M_TEMP, 1731 M_WAITOK); 1732 sx_slock(&filelist_lock); 1733 if (nfiles_snap < openfiles) { 1734 sx_sunlock(&filelist_lock); 1735 free(extra_ref, M_TEMP); 1736 nfiles_slack += 20; 1737 goto again; 1738 } 1739 for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref; 1740 fp != NULL; fp = nextfp) { 1741 nextfp = LIST_NEXT(fp, f_list); 1742 FILE_LOCK(fp); 1743 /* 1744 * If it's not open, skip it 1745 */ 1746 if (fp->f_count == 0) { 1747 FILE_UNLOCK(fp); 1748 continue; 1749 } 1750 /* 1751 * If all refs are from msgs, and it's not marked accessible 1752 * then it must be referenced from some unreachable cycle of 1753 * (shut-down) FDs, so include it in our list of FDs to 1754 * remove. 1755 */ 1756 if (fp->f_count == fp->f_msgcount && !(fp->f_gcflag & FMARK)) { 1757 *fpp++ = fp; 1758 nunref++; 1759 fp->f_count++; 1760 } 1761 FILE_UNLOCK(fp); 1762 } 1763 sx_sunlock(&filelist_lock); 1764 /* 1765 * For each FD on our hit list, do the following two things: 1766 */ 1767 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1768 struct file *tfp = *fpp; 1769 FILE_LOCK(tfp); 1770 if (tfp->f_type == DTYPE_SOCKET && 1771 tfp->f_data != NULL) { 1772 FILE_UNLOCK(tfp); 1773 sorflush(tfp->f_data); 1774 } else { 1775 FILE_UNLOCK(tfp); 1776 } 1777 } 1778 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1779 closef(*fpp, (struct thread *) NULL); 1780 unp_recycled++; 1781 } 1782 free(extra_ref, M_TEMP); 1783} 1784 1785void 1786unp_dispose(struct mbuf *m) 1787{ 1788 1789 if (m) 1790 unp_scan(m, unp_discard); 1791} 1792 1793static int 1794unp_listen(struct socket *so, struct unpcb *unp, int backlog, 1795 struct thread *td) 1796{ 1797 int error; 1798 1799 UNP_LOCK_ASSERT(); 1800 1801 SOCK_LOCK(so); 1802 error = solisten_proto_check(so); 1803 if (error == 0) { 1804 cru2x(td->td_ucred, &unp->unp_peercred); 1805 unp->unp_flags |= UNP_HAVEPCCACHED; 1806 solisten_proto(so, backlog); 1807 } 1808 SOCK_UNLOCK(so); 1809 return (error); 1810} 1811 1812static void 1813unp_scan(struct mbuf *m0, void (*op)(struct file *)) 1814{ 1815 struct mbuf *m; 1816 struct file **rp; 1817 struct cmsghdr *cm; 1818 void *data; 1819 int i; 1820 socklen_t clen, datalen; 1821 int qfds; 1822 1823 while (m0 != NULL) { 1824 for (m = m0; m; m = m->m_next) { 1825 if (m->m_type != MT_CONTROL) 1826 continue; 1827 1828 cm = mtod(m, struct cmsghdr *); 1829 clen = m->m_len; 1830 1831 while (cm != NULL) { 1832 if (sizeof(*cm) > clen || cm->cmsg_len > clen) 1833 break; 1834 1835 data = CMSG_DATA(cm); 1836 datalen = (caddr_t)cm + cm->cmsg_len 1837 - (caddr_t)data; 1838 1839 if (cm->cmsg_level == SOL_SOCKET && 1840 cm->cmsg_type == SCM_RIGHTS) { 1841 qfds = datalen / sizeof (struct file *); 1842 rp = data; 1843 for (i = 0; i < qfds; i++) 1844 (*op)(*rp++); 1845 } 1846 1847 if (CMSG_SPACE(datalen) < clen) { 1848 clen -= CMSG_SPACE(datalen); 1849 cm = (struct cmsghdr *) 1850 ((caddr_t)cm + CMSG_SPACE(datalen)); 1851 } else { 1852 clen = 0; 1853 cm = NULL; 1854 } 1855 } 1856 } 1857 m0 = m0->m_act; 1858 } 1859} 1860 1861static void 1862unp_mark(struct file *fp) 1863{ 1864 if (fp->f_gcflag & FMARK) 1865 return; 1866 unp_defer++; 1867 fp->f_gcflag |= (FMARK|FDEFER); 1868} 1869 1870static void 1871unp_discard(struct file *fp) 1872{ 1873 UNP_LOCK(); 1874 FILE_LOCK(fp); 1875 fp->f_msgcount--; 1876 unp_rights--; 1877 FILE_UNLOCK(fp); 1878 UNP_UNLOCK(); 1879 (void) closef(fp, (struct thread *)NULL); 1880} 1881