uipc_usrreq.c revision 161040
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 161040 2006-08-07 12:02:43Z 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_NODE(_net, PF_LOCAL, local, CTLFLAG_RW, 0, "Local domain"); 128SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW, 0, "SOCK_STREAM"); 129SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW, 0, "SOCK_DGRAM"); 130 131SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 132 &unpst_sendspace, 0, ""); 133SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 134 &unpst_recvspace, 0, ""); 135SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 136 &unpdg_sendspace, 0, ""); 137SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 138 &unpdg_recvspace, 0, ""); 139SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, ""); 140 141/* 142 * Currently, UNIX domain sockets are protected by a single subsystem lock, 143 * which covers global data structures and variables, the contents of each 144 * per-socket unpcb structure, and the so_pcb field in sockets attached to 145 * the UNIX domain. This provides for a moderate degree of paralellism, as 146 * receive operations on UNIX domain sockets do not need to acquire the 147 * subsystem lock. Finer grained locking to permit send() without acquiring 148 * a global lock would be a logical next step. 149 * 150 * The UNIX domain socket lock preceds all socket layer locks, including the 151 * socket lock and socket buffer lock, permitting UNIX domain socket code to 152 * call into socket support routines without releasing its locks. 153 * 154 * Some caution is required in areas where the UNIX domain socket code enters 155 * VFS in order to create or find rendezvous points. This results in 156 * dropping of the UNIX domain socket subsystem lock, acquisition of the 157 * Giant lock, and potential sleeping. This increases the chances of races, 158 * and exposes weaknesses in the socket->protocol API by offering poor 159 * failure modes. 160 */ 161static struct mtx unp_mtx; 162#define UNP_LOCK_INIT() \ 163 mtx_init(&unp_mtx, "unp", NULL, MTX_DEF) 164#define UNP_LOCK() mtx_lock(&unp_mtx) 165#define UNP_UNLOCK() mtx_unlock(&unp_mtx) 166#define UNP_LOCK_ASSERT() mtx_assert(&unp_mtx, MA_OWNED) 167#define UNP_UNLOCK_ASSERT() mtx_assert(&unp_mtx, MA_NOTOWNED) 168 169/* 170 * Garbage collection of cyclic file descriptor/socket references occurs 171 * asynchronously in a taskqueue context in order to avoid recursion and 172 * reentrance in the UNIX domain socket, file descriptor, and socket layer 173 * code. See unp_gc() for a full description. 174 */ 175static struct task unp_gc_task; 176 177static int unp_connect(struct socket *,struct sockaddr *, struct thread *); 178static int unp_connect2(struct socket *so, struct socket *so2, int); 179static void unp_disconnect(struct unpcb *); 180static void unp_shutdown(struct unpcb *); 181static void unp_drop(struct unpcb *, int); 182static void unp_gc(__unused void *, int); 183static void unp_scan(struct mbuf *, void (*)(struct file *)); 184static void unp_mark(struct file *); 185static void unp_discard(struct file *); 186static void unp_freerights(struct file **, int); 187static int unp_internalize(struct mbuf **, struct thread *); 188static int unp_listen(struct socket *, struct unpcb *, int, 189 struct thread *); 190 191/* 192 * Definitions of protocols supported in the LOCAL domain. 193 */ 194static struct domain localdomain; 195static struct protosw localsw[] = { 196{ 197 .pr_type = SOCK_STREAM, 198 .pr_domain = &localdomain, 199 .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS, 200 .pr_ctloutput = &uipc_ctloutput, 201 .pr_usrreqs = &uipc_usrreqs 202}, 203{ 204 .pr_type = SOCK_DGRAM, 205 .pr_domain = &localdomain, 206 .pr_flags = PR_ATOMIC|PR_ADDR|PR_RIGHTS, 207 .pr_usrreqs = &uipc_usrreqs 208}, 209}; 210 211static struct domain localdomain = { 212 .dom_family = AF_LOCAL, 213 .dom_name = "local", 214 .dom_init = unp_init, 215 .dom_externalize = unp_externalize, 216 .dom_dispose = unp_dispose, 217 .dom_protosw = localsw, 218 .dom_protoswNPROTOSW = &localsw[sizeof(localsw)/sizeof(localsw[0])] 219}; 220DOMAIN_SET(local); 221 222static void 223uipc_abort(struct socket *so) 224{ 225 struct unpcb *unp; 226 227 unp = sotounpcb(so); 228 KASSERT(unp != NULL, ("uipc_abort: unp == NULL")); 229 UNP_LOCK(); 230 unp_drop(unp, ECONNABORTED); 231 UNP_UNLOCK(); 232} 233 234static int 235uipc_accept(struct socket *so, struct sockaddr **nam) 236{ 237 struct unpcb *unp; 238 const struct sockaddr *sa; 239 240 /* 241 * Pass back name of connected socket, if it was bound and we are 242 * still connected (our peer may have closed already!). 243 */ 244 unp = sotounpcb(so); 245 KASSERT(unp != NULL, ("uipc_accept: unp == NULL")); 246 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 247 UNP_LOCK(); 248 if (unp->unp_conn != NULL && unp->unp_conn->unp_addr != NULL) 249 sa = (struct sockaddr *) unp->unp_conn->unp_addr; 250 else 251 sa = &sun_noname; 252 bcopy(sa, *nam, sa->sa_len); 253 UNP_UNLOCK(); 254 return (0); 255} 256 257static int 258uipc_attach(struct socket *so, int proto, struct thread *td) 259{ 260 struct unpcb *unp; 261 int error; 262 263 KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL")); 264 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 265 switch (so->so_type) { 266 267 case SOCK_STREAM: 268 error = soreserve(so, unpst_sendspace, unpst_recvspace); 269 break; 270 271 case SOCK_DGRAM: 272 error = soreserve(so, unpdg_sendspace, unpdg_recvspace); 273 break; 274 275 default: 276 panic("unp_attach"); 277 } 278 if (error) 279 return (error); 280 } 281 unp = uma_zalloc(unp_zone, M_WAITOK | M_ZERO); 282 if (unp == NULL) 283 return (ENOBUFS); 284 LIST_INIT(&unp->unp_refs); 285 unp->unp_socket = so; 286 so->so_pcb = unp; 287 288 UNP_LOCK(); 289 unp->unp_gencnt = ++unp_gencnt; 290 unp_count++; 291 LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead 292 : &unp_shead, unp, unp_link); 293 UNP_UNLOCK(); 294 295 return (0); 296} 297 298static int 299uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 300{ 301 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 302 struct vattr vattr; 303 int error, namelen; 304 struct nameidata nd; 305 struct unpcb *unp; 306 struct vnode *vp; 307 struct mount *mp; 308 char *buf; 309 310 unp = sotounpcb(so); 311 KASSERT(unp != NULL, ("uipc_bind: unp == NULL")); 312 313 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 314 if (namelen <= 0) 315 return (EINVAL); 316 317 /* 318 * We don't allow simultaneous bind() calls on a single UNIX domain 319 * socket, so flag in-progress operations, and return an error if an 320 * operation is already in progress. 321 * 322 * Historically, we have not allowed a socket to be rebound, so this 323 * also returns an error. Not allowing re-binding certainly 324 * simplifies the implementation and avoids a great many possible 325 * failure modes. 326 */ 327 UNP_LOCK(); 328 if (unp->unp_vnode != NULL) { 329 UNP_UNLOCK(); 330 return (EINVAL); 331 } 332 if (unp->unp_flags & UNP_BINDING) { 333 UNP_UNLOCK(); 334 return (EALREADY); 335 } 336 unp->unp_flags |= UNP_BINDING; 337 UNP_UNLOCK(); 338 339 buf = malloc(namelen + 1, M_TEMP, M_WAITOK); 340 strlcpy(buf, soun->sun_path, namelen + 1); 341 342 mtx_lock(&Giant); 343restart: 344 mtx_assert(&Giant, MA_OWNED); 345 NDINIT(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME, UIO_SYSSPACE, 346 buf, td); 347/* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 348 error = namei(&nd); 349 if (error) 350 goto error; 351 vp = nd.ni_vp; 352 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { 353 NDFREE(&nd, NDF_ONLY_PNBUF); 354 if (nd.ni_dvp == vp) 355 vrele(nd.ni_dvp); 356 else 357 vput(nd.ni_dvp); 358 if (vp != NULL) { 359 vrele(vp); 360 error = EADDRINUSE; 361 goto error; 362 } 363 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH); 364 if (error) 365 goto error; 366 goto restart; 367 } 368 VATTR_NULL(&vattr); 369 vattr.va_type = VSOCK; 370 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask); 371#ifdef MAC 372 error = mac_check_vnode_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, 373 &vattr); 374#endif 375 if (error == 0) { 376 VOP_LEASE(nd.ni_dvp, td, td->td_ucred, LEASE_WRITE); 377 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 378 } 379 NDFREE(&nd, NDF_ONLY_PNBUF); 380 vput(nd.ni_dvp); 381 if (error) { 382 vn_finished_write(mp); 383 goto error; 384 } 385 vp = nd.ni_vp; 386 ASSERT_VOP_LOCKED(vp, "uipc_bind"); 387 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK); 388 UNP_LOCK(); 389 vp->v_socket = unp->unp_socket; 390 unp->unp_vnode = vp; 391 unp->unp_addr = soun; 392 unp->unp_flags &= ~UNP_BINDING; 393 UNP_UNLOCK(); 394 VOP_UNLOCK(vp, 0, td); 395 vn_finished_write(mp); 396 mtx_unlock(&Giant); 397 free(buf, M_TEMP); 398 return (0); 399error: 400 UNP_LOCK(); 401 unp->unp_flags &= ~UNP_BINDING; 402 UNP_UNLOCK(); 403 mtx_unlock(&Giant); 404 free(buf, M_TEMP); 405 return (error); 406} 407 408static int 409uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 410{ 411 int error; 412 413 KASSERT(td == curthread, ("uipc_connect: td != curthread")); 414 UNP_LOCK(); 415 error = unp_connect(so, nam, td); 416 UNP_UNLOCK(); 417 return (error); 418} 419 420/* 421 * XXXRW: Should also unbind? 422 */ 423static void 424uipc_close(struct socket *so) 425{ 426 struct unpcb *unp; 427 428 unp = sotounpcb(so); 429 KASSERT(unp != NULL, ("uipc_close: unp == NULL")); 430 UNP_LOCK(); 431 unp_disconnect(unp); 432 UNP_UNLOCK(); 433} 434 435int 436uipc_connect2(struct socket *so1, struct socket *so2) 437{ 438 struct unpcb *unp; 439 int error; 440 441 unp = sotounpcb(so1); 442 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL")); 443 UNP_LOCK(); 444 error = unp_connect2(so1, so2, PRU_CONNECT2); 445 UNP_UNLOCK(); 446 return (error); 447} 448 449/* control is EOPNOTSUPP */ 450 451static void 452uipc_detach(struct socket *so) 453{ 454 int local_unp_rights; 455 struct unpcb *unp; 456 struct vnode *vp; 457 458 unp = sotounpcb(so); 459 KASSERT(unp != NULL, ("uipc_detach: unp == NULL")); 460 UNP_LOCK(); 461 LIST_REMOVE(unp, unp_link); 462 unp->unp_gencnt = ++unp_gencnt; 463 --unp_count; 464 if ((vp = unp->unp_vnode) != NULL) { 465 /* 466 * XXXRW: should v_socket be frobbed only while holding 467 * Giant? 468 */ 469 unp->unp_vnode->v_socket = NULL; 470 unp->unp_vnode = NULL; 471 } 472 if (unp->unp_conn != NULL) 473 unp_disconnect(unp); 474 while (!LIST_EMPTY(&unp->unp_refs)) { 475 struct unpcb *ref = LIST_FIRST(&unp->unp_refs); 476 unp_drop(ref, ECONNRESET); 477 } 478 unp->unp_socket->so_pcb = NULL; 479 local_unp_rights = unp_rights; 480 UNP_UNLOCK(); 481 if (unp->unp_addr != NULL) 482 FREE(unp->unp_addr, M_SONAME); 483 uma_zfree(unp_zone, unp); 484 if (vp) { 485 int vfslocked; 486 487 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 488 vrele(vp); 489 VFS_UNLOCK_GIANT(vfslocked); 490 } 491 if (local_unp_rights) 492 taskqueue_enqueue(taskqueue_thread, &unp_gc_task); 493} 494 495static int 496uipc_disconnect(struct socket *so) 497{ 498 struct unpcb *unp; 499 500 unp = sotounpcb(so); 501 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL")); 502 UNP_LOCK(); 503 unp_disconnect(unp); 504 UNP_UNLOCK(); 505 return (0); 506} 507 508static int 509uipc_listen(struct socket *so, int backlog, struct thread *td) 510{ 511 struct unpcb *unp; 512 int error; 513 514 unp = sotounpcb(so); 515 KASSERT(unp != NULL, ("uipc_listen: unp == NULL")); 516 UNP_LOCK(); 517 if (unp->unp_vnode == NULL) { 518 UNP_UNLOCK(); 519 return (EINVAL); 520 } 521 error = unp_listen(so, unp, backlog, td); 522 UNP_UNLOCK(); 523 return (error); 524} 525 526static int 527uipc_peeraddr(struct socket *so, struct sockaddr **nam) 528{ 529 struct unpcb *unp; 530 const struct sockaddr *sa; 531 532 unp = sotounpcb(so); 533 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL")); 534 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 535 UNP_LOCK(); 536 if (unp->unp_conn != NULL && unp->unp_conn->unp_addr!= NULL) 537 sa = (struct sockaddr *) unp->unp_conn->unp_addr; 538 else { 539 /* 540 * XXX: It seems that this test always fails even when 541 * connection is established. So, this else clause is 542 * added as workaround to return PF_LOCAL sockaddr. 543 */ 544 sa = &sun_noname; 545 } 546 bcopy(sa, *nam, sa->sa_len); 547 UNP_UNLOCK(); 548 return (0); 549} 550 551static int 552uipc_rcvd(struct socket *so, int flags) 553{ 554 struct unpcb *unp; 555 struct socket *so2; 556 u_int mbcnt, sbcc; 557 u_long newhiwat; 558 559 unp = sotounpcb(so); 560 KASSERT(unp != NULL, ("uipc_rcvd: unp == NULL")); 561 switch (so->so_type) { 562 case SOCK_DGRAM: 563 panic("uipc_rcvd DGRAM?"); 564 /*NOTREACHED*/ 565 566 case SOCK_STREAM: 567 /* 568 * Adjust backpressure on sender and wakeup any waiting to 569 * write. 570 */ 571 SOCKBUF_LOCK(&so->so_rcv); 572 mbcnt = so->so_rcv.sb_mbcnt; 573 sbcc = so->so_rcv.sb_cc; 574 SOCKBUF_UNLOCK(&so->so_rcv); 575 UNP_LOCK(); 576 if (unp->unp_conn == NULL) { 577 UNP_UNLOCK(); 578 break; 579 } 580 so2 = unp->unp_conn->unp_socket; 581 SOCKBUF_LOCK(&so2->so_snd); 582 so2->so_snd.sb_mbmax += unp->unp_mbcnt - mbcnt; 583 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - sbcc; 584 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat, 585 newhiwat, RLIM_INFINITY); 586 sowwakeup_locked(so2); 587 unp->unp_mbcnt = mbcnt; 588 unp->unp_cc = sbcc; 589 UNP_UNLOCK(); 590 break; 591 592 default: 593 panic("uipc_rcvd unknown socktype"); 594 } 595 return (0); 596} 597 598/* pru_rcvoob is EOPNOTSUPP */ 599 600static int 601uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 602 struct mbuf *control, struct thread *td) 603{ 604 struct unpcb *unp, *unp2; 605 struct socket *so2; 606 u_int mbcnt, sbcc; 607 u_long newhiwat; 608 int error = 0; 609 610 unp = sotounpcb(so); 611 KASSERT(unp != NULL, ("uipc_send: unp == NULL")); 612 if (flags & PRUS_OOB) { 613 error = EOPNOTSUPP; 614 goto release; 615 } 616 617 if (control != NULL && (error = unp_internalize(&control, td))) 618 goto release; 619 620 UNP_LOCK(); 621 switch (so->so_type) { 622 case SOCK_DGRAM: 623 { 624 const struct sockaddr *from; 625 626 if (nam != NULL) { 627 if (unp->unp_conn != NULL) { 628 error = EISCONN; 629 break; 630 } 631 error = unp_connect(so, nam, td); 632 if (error) 633 break; 634 } 635 /* 636 * Because connect() and send() are non-atomic in a sendto() 637 * with a target address, it's possible that the socket will 638 * have disconnected before the send() can run. In that case 639 * return the slightly counter-intuitive but otherwise 640 * correct error that the socket is not connected. 641 */ 642 unp2 = unp->unp_conn; 643 if (unp2 == NULL) { 644 error = ENOTCONN; 645 break; 646 } 647 so2 = unp2->unp_socket; 648 if (unp->unp_addr != NULL) 649 from = (struct sockaddr *)unp->unp_addr; 650 else 651 from = &sun_noname; 652 if (unp2->unp_flags & UNP_WANTCRED) 653 control = unp_addsockcred(td, control); 654 SOCKBUF_LOCK(&so2->so_rcv); 655 if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) { 656 sorwakeup_locked(so2); 657 m = NULL; 658 control = NULL; 659 } else { 660 SOCKBUF_UNLOCK(&so2->so_rcv); 661 error = ENOBUFS; 662 } 663 if (nam != NULL) 664 unp_disconnect(unp); 665 break; 666 } 667 668 case SOCK_STREAM: 669 /* 670 * Connect if not connected yet. 671 * 672 * Note: A better implementation would complain if not equal 673 * to the peer's address. 674 */ 675 if ((so->so_state & SS_ISCONNECTED) == 0) { 676 if (nam != NULL) { 677 error = unp_connect(so, nam, td); 678 if (error) 679 break; /* XXX */ 680 } else { 681 error = ENOTCONN; 682 break; 683 } 684 } 685 686 /* Lockless read. */ 687 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 688 error = EPIPE; 689 break; 690 } 691 /* 692 * Because connect() and send() are non-atomic in a sendto() 693 * with a target address, it's possible that the socket will 694 * have disconnected before the send() can run. In that case 695 * return the slightly counter-intuitive but otherwise 696 * correct error that the socket is not connected. 697 */ 698 unp2 = unp->unp_conn; 699 if (unp2 == NULL) { 700 error = ENOTCONN; 701 break; 702 } 703 so2 = unp2->unp_socket; 704 SOCKBUF_LOCK(&so2->so_rcv); 705 if (unp2->unp_flags & UNP_WANTCRED) { 706 /* 707 * Credentials are passed only once on 708 * SOCK_STREAM. 709 */ 710 unp2->unp_flags &= ~UNP_WANTCRED; 711 control = unp_addsockcred(td, control); 712 } 713 /* 714 * Send to paired receive port, and then reduce send buffer 715 * hiwater marks to maintain backpressure. Wake up readers. 716 */ 717 if (control != NULL) { 718 if (sbappendcontrol_locked(&so2->so_rcv, m, control)) 719 control = NULL; 720 } else { 721 sbappend_locked(&so2->so_rcv, m); 722 } 723 mbcnt = so2->so_rcv.sb_mbcnt - unp2->unp_mbcnt; 724 unp2->unp_mbcnt = so2->so_rcv.sb_mbcnt; 725 sbcc = so2->so_rcv.sb_cc; 726 sorwakeup_locked(so2); 727 728 SOCKBUF_LOCK(&so->so_snd); 729 newhiwat = so->so_snd.sb_hiwat - (sbcc - unp2->unp_cc); 730 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat, 731 newhiwat, RLIM_INFINITY); 732 so->so_snd.sb_mbmax -= mbcnt; 733 SOCKBUF_UNLOCK(&so->so_snd); 734 735 unp2->unp_cc = sbcc; 736 m = NULL; 737 break; 738 739 default: 740 panic("uipc_send unknown socktype"); 741 } 742 743 /* 744 * SEND_EOF is equivalent to a SEND followed by 745 * a SHUTDOWN. 746 */ 747 if (flags & PRUS_EOF) { 748 socantsendmore(so); 749 unp_shutdown(unp); 750 } 751 UNP_UNLOCK(); 752 753 if (control != NULL && error != 0) 754 unp_dispose(control); 755 756release: 757 if (control != NULL) 758 m_freem(control); 759 if (m != NULL) 760 m_freem(m); 761 return (error); 762} 763 764static int 765uipc_sense(struct socket *so, struct stat *sb) 766{ 767 struct unpcb *unp; 768 struct socket *so2; 769 770 unp = sotounpcb(so); 771 KASSERT(unp != NULL, ("uipc_sense: unp == NULL")); 772 UNP_LOCK(); 773 sb->st_blksize = so->so_snd.sb_hiwat; 774 if (so->so_type == SOCK_STREAM && unp->unp_conn != NULL) { 775 so2 = unp->unp_conn->unp_socket; 776 sb->st_blksize += so2->so_rcv.sb_cc; 777 } 778 sb->st_dev = NODEV; 779 if (unp->unp_ino == 0) 780 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino; 781 sb->st_ino = unp->unp_ino; 782 UNP_UNLOCK(); 783 return (0); 784} 785 786static int 787uipc_shutdown(struct socket *so) 788{ 789 struct unpcb *unp; 790 791 unp = sotounpcb(so); 792 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL")); 793 UNP_LOCK(); 794 socantsendmore(so); 795 unp_shutdown(unp); 796 UNP_UNLOCK(); 797 return (0); 798} 799 800static int 801uipc_sockaddr(struct socket *so, struct sockaddr **nam) 802{ 803 struct unpcb *unp; 804 const struct sockaddr *sa; 805 806 unp = sotounpcb(so); 807 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL")); 808 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 809 UNP_LOCK(); 810 if (unp->unp_addr != NULL) 811 sa = (struct sockaddr *) unp->unp_addr; 812 else 813 sa = &sun_noname; 814 bcopy(sa, *nam, sa->sa_len); 815 UNP_UNLOCK(); 816 return (0); 817} 818 819struct pr_usrreqs uipc_usrreqs = { 820 .pru_abort = uipc_abort, 821 .pru_accept = uipc_accept, 822 .pru_attach = uipc_attach, 823 .pru_bind = uipc_bind, 824 .pru_connect = uipc_connect, 825 .pru_connect2 = uipc_connect2, 826 .pru_detach = uipc_detach, 827 .pru_disconnect = uipc_disconnect, 828 .pru_listen = uipc_listen, 829 .pru_peeraddr = uipc_peeraddr, 830 .pru_rcvd = uipc_rcvd, 831 .pru_send = uipc_send, 832 .pru_sense = uipc_sense, 833 .pru_shutdown = uipc_shutdown, 834 .pru_sockaddr = uipc_sockaddr, 835 .pru_close = uipc_close, 836}; 837 838int 839uipc_ctloutput(struct socket *so, struct sockopt *sopt) 840{ 841 struct unpcb *unp; 842 struct xucred xu; 843 int error, optval; 844 845 if (sopt->sopt_level != 0) 846 return (EINVAL); 847 848 unp = sotounpcb(so); 849 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL")); 850 UNP_LOCK(); 851 error = 0; 852 switch (sopt->sopt_dir) { 853 case SOPT_GET: 854 switch (sopt->sopt_name) { 855 case LOCAL_PEERCRED: 856 if (unp->unp_flags & UNP_HAVEPC) 857 xu = unp->unp_peercred; 858 else { 859 if (so->so_type == SOCK_STREAM) 860 error = ENOTCONN; 861 else 862 error = EINVAL; 863 } 864 if (error == 0) 865 error = sooptcopyout(sopt, &xu, sizeof(xu)); 866 break; 867 case LOCAL_CREDS: 868 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0; 869 error = sooptcopyout(sopt, &optval, sizeof(optval)); 870 break; 871 case LOCAL_CONNWAIT: 872 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0; 873 error = sooptcopyout(sopt, &optval, sizeof(optval)); 874 break; 875 default: 876 error = EOPNOTSUPP; 877 break; 878 } 879 break; 880 case SOPT_SET: 881 switch (sopt->sopt_name) { 882 case LOCAL_CREDS: 883 case LOCAL_CONNWAIT: 884 error = sooptcopyin(sopt, &optval, sizeof(optval), 885 sizeof(optval)); 886 if (error) 887 break; 888 889#define OPTSET(bit) \ 890 if (optval) \ 891 unp->unp_flags |= bit; \ 892 else \ 893 unp->unp_flags &= ~bit; 894 895 switch (sopt->sopt_name) { 896 case LOCAL_CREDS: 897 OPTSET(UNP_WANTCRED); 898 break; 899 case LOCAL_CONNWAIT: 900 OPTSET(UNP_CONNWAIT); 901 break; 902 default: 903 break; 904 } 905 break; 906#undef OPTSET 907 default: 908 error = ENOPROTOOPT; 909 break; 910 } 911 break; 912 default: 913 error = EOPNOTSUPP; 914 break; 915 } 916 UNP_UNLOCK(); 917 return (error); 918} 919 920static int 921unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 922{ 923 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 924 struct vnode *vp; 925 struct socket *so2, *so3; 926 struct unpcb *unp, *unp2, *unp3; 927 int error, len; 928 struct nameidata nd; 929 char buf[SOCK_MAXADDRLEN]; 930 struct sockaddr *sa; 931 932 UNP_LOCK_ASSERT(); 933 934 unp = sotounpcb(so); 935 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 936 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 937 if (len <= 0) 938 return (EINVAL); 939 strlcpy(buf, soun->sun_path, len + 1); 940 if (unp->unp_flags & UNP_CONNECTING) { 941 UNP_UNLOCK(); 942 return (EALREADY); 943 } 944 UNP_UNLOCK(); 945 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 946 mtx_lock(&Giant); 947 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, td); 948 error = namei(&nd); 949 if (error) 950 vp = NULL; 951 else 952 vp = nd.ni_vp; 953 ASSERT_VOP_LOCKED(vp, "unp_connect"); 954 NDFREE(&nd, NDF_ONLY_PNBUF); 955 if (error) 956 goto bad; 957 958 if (vp->v_type != VSOCK) { 959 error = ENOTSOCK; 960 goto bad; 961 } 962 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); 963 if (error) 964 goto bad; 965 mtx_unlock(&Giant); 966 UNP_LOCK(); 967 unp = sotounpcb(so); 968 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 969 so2 = vp->v_socket; 970 if (so2 == NULL) { 971 error = ECONNREFUSED; 972 goto bad2; 973 } 974 if (so->so_type != so2->so_type) { 975 error = EPROTOTYPE; 976 goto bad2; 977 } 978 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 979 if (so2->so_options & SO_ACCEPTCONN) { 980 /* 981 * NB: drop locks here so unp_attach is entered w/o 982 * locks; this avoids a recursive lock of the head 983 * and holding sleep locks across a (potentially) 984 * blocking malloc. 985 */ 986 UNP_UNLOCK(); 987 so3 = sonewconn(so2, 0); 988 UNP_LOCK(); 989 } else 990 so3 = NULL; 991 if (so3 == NULL) { 992 error = ECONNREFUSED; 993 goto bad2; 994 } 995 unp = sotounpcb(so); 996 unp2 = sotounpcb(so2); 997 unp3 = sotounpcb(so3); 998 if (unp2->unp_addr != NULL) { 999 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); 1000 unp3->unp_addr = (struct sockaddr_un *) sa; 1001 sa = NULL; 1002 } 1003 /* 1004 * unp_peercred management: 1005 * 1006 * The connecter's (client's) credentials are copied from its 1007 * process structure at the time of connect() (which is now). 1008 */ 1009 cru2x(td->td_ucred, &unp3->unp_peercred); 1010 unp3->unp_flags |= UNP_HAVEPC; 1011 /* 1012 * The receiver's (server's) credentials are copied from the 1013 * unp_peercred member of socket on which the former called 1014 * listen(); unp_listen() cached that process's credentials 1015 * at that time so we can use them now. 1016 */ 1017 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED, 1018 ("unp_connect: listener without cached peercred")); 1019 memcpy(&unp->unp_peercred, &unp2->unp_peercred, 1020 sizeof(unp->unp_peercred)); 1021 unp->unp_flags |= UNP_HAVEPC; 1022 if (unp2->unp_flags & UNP_WANTCRED) 1023 unp3->unp_flags |= UNP_WANTCRED; 1024#ifdef MAC 1025 SOCK_LOCK(so); 1026 mac_set_socket_peer_from_socket(so, so3); 1027 mac_set_socket_peer_from_socket(so3, so); 1028 SOCK_UNLOCK(so); 1029#endif 1030 1031 so2 = so3; 1032 } 1033 error = unp_connect2(so, so2, PRU_CONNECT); 1034bad2: 1035 UNP_UNLOCK(); 1036 mtx_lock(&Giant); 1037bad: 1038 mtx_assert(&Giant, MA_OWNED); 1039 if (vp != NULL) 1040 vput(vp); 1041 mtx_unlock(&Giant); 1042 free(sa, M_SONAME); 1043 UNP_LOCK(); 1044 unp->unp_flags &= ~UNP_CONNECTING; 1045 return (error); 1046} 1047 1048static int 1049unp_connect2(struct socket *so, struct socket *so2, int req) 1050{ 1051 struct unpcb *unp = sotounpcb(so); 1052 struct unpcb *unp2; 1053 1054 UNP_LOCK_ASSERT(); 1055 1056 if (so2->so_type != so->so_type) 1057 return (EPROTOTYPE); 1058 unp2 = sotounpcb(so2); 1059 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL")); 1060 unp->unp_conn = unp2; 1061 switch (so->so_type) { 1062 1063 case SOCK_DGRAM: 1064 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 1065 soisconnected(so); 1066 break; 1067 1068 case SOCK_STREAM: 1069 unp2->unp_conn = unp; 1070 if (req == PRU_CONNECT && 1071 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 1072 soisconnecting(so); 1073 else 1074 soisconnected(so); 1075 soisconnected(so2); 1076 break; 1077 1078 default: 1079 panic("unp_connect2"); 1080 } 1081 return (0); 1082} 1083 1084static void 1085unp_disconnect(struct unpcb *unp) 1086{ 1087 struct unpcb *unp2 = unp->unp_conn; 1088 struct socket *so; 1089 1090 UNP_LOCK_ASSERT(); 1091 1092 if (unp2 == NULL) 1093 return; 1094 unp->unp_conn = NULL; 1095 switch (unp->unp_socket->so_type) { 1096 case SOCK_DGRAM: 1097 LIST_REMOVE(unp, unp_reflink); 1098 so = unp->unp_socket; 1099 SOCK_LOCK(so); 1100 so->so_state &= ~SS_ISCONNECTED; 1101 SOCK_UNLOCK(so); 1102 break; 1103 1104 case SOCK_STREAM: 1105 soisdisconnected(unp->unp_socket); 1106 unp2->unp_conn = NULL; 1107 soisdisconnected(unp2->unp_socket); 1108 break; 1109 } 1110} 1111 1112/* 1113 * unp_pcblist() assumes that UNIX domain socket memory is never reclaimed by 1114 * the zone (UMA_ZONE_NOFREE), and as such potentially stale pointers are 1115 * safe to reference. It first scans the list of struct unpcb's to generate 1116 * a pointer list, then it rescans its list one entry at a time to 1117 * externalize and copyout. It checks the generation number to see if a 1118 * struct unpcb has been reused, and will skip it if so. 1119 */ 1120static int 1121unp_pcblist(SYSCTL_HANDLER_ARGS) 1122{ 1123 int error, i, n; 1124 struct unpcb *unp, **unp_list; 1125 unp_gen_t gencnt; 1126 struct xunpgen *xug; 1127 struct unp_head *head; 1128 struct xunpcb *xu; 1129 1130 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead); 1131 1132 /* 1133 * The process of preparing the PCB list is too time-consuming and 1134 * resource-intensive to repeat twice on every request. 1135 */ 1136 if (req->oldptr == NULL) { 1137 n = unp_count; 1138 req->oldidx = 2 * (sizeof *xug) 1139 + (n + n/8) * sizeof(struct xunpcb); 1140 return (0); 1141 } 1142 1143 if (req->newptr != NULL) 1144 return (EPERM); 1145 1146 /* 1147 * OK, now we're committed to doing something. 1148 */ 1149 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK); 1150 UNP_LOCK(); 1151 gencnt = unp_gencnt; 1152 n = unp_count; 1153 UNP_UNLOCK(); 1154 1155 xug->xug_len = sizeof *xug; 1156 xug->xug_count = n; 1157 xug->xug_gen = gencnt; 1158 xug->xug_sogen = so_gencnt; 1159 error = SYSCTL_OUT(req, xug, sizeof *xug); 1160 if (error) { 1161 free(xug, M_TEMP); 1162 return (error); 1163 } 1164 1165 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 1166 1167 UNP_LOCK(); 1168 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 1169 unp = LIST_NEXT(unp, unp_link)) { 1170 if (unp->unp_gencnt <= gencnt) { 1171 if (cr_cansee(req->td->td_ucred, 1172 unp->unp_socket->so_cred)) 1173 continue; 1174 unp_list[i++] = unp; 1175 } 1176 } 1177 UNP_UNLOCK(); 1178 n = i; /* In case we lost some during malloc. */ 1179 1180 error = 0; 1181 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO); 1182 for (i = 0; i < n; i++) { 1183 unp = unp_list[i]; 1184 if (unp->unp_gencnt <= gencnt) { 1185 xu->xu_len = sizeof *xu; 1186 xu->xu_unpp = unp; 1187 /* 1188 * XXX - need more locking here to protect against 1189 * connect/disconnect races for SMP. 1190 */ 1191 if (unp->unp_addr != NULL) 1192 bcopy(unp->unp_addr, &xu->xu_addr, 1193 unp->unp_addr->sun_len); 1194 if (unp->unp_conn != NULL && 1195 unp->unp_conn->unp_addr != NULL) 1196 bcopy(unp->unp_conn->unp_addr, 1197 &xu->xu_caddr, 1198 unp->unp_conn->unp_addr->sun_len); 1199 bcopy(unp, &xu->xu_unp, sizeof *unp); 1200 sotoxsocket(unp->unp_socket, &xu->xu_socket); 1201 error = SYSCTL_OUT(req, xu, sizeof *xu); 1202 } 1203 } 1204 free(xu, M_TEMP); 1205 if (!error) { 1206 /* 1207 * Give the user an updated idea of our state. If the 1208 * generation differs from what we told her before, she knows 1209 * that something happened while we were processing this 1210 * request, and it might be necessary to retry. 1211 */ 1212 xug->xug_gen = unp_gencnt; 1213 xug->xug_sogen = so_gencnt; 1214 xug->xug_count = unp_count; 1215 error = SYSCTL_OUT(req, xug, sizeof *xug); 1216 } 1217 free(unp_list, M_TEMP); 1218 free(xug, M_TEMP); 1219 return (error); 1220} 1221 1222SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD, 1223 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 1224 "List of active local datagram sockets"); 1225SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD, 1226 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 1227 "List of active local stream sockets"); 1228 1229static void 1230unp_shutdown(struct unpcb *unp) 1231{ 1232 struct socket *so; 1233 1234 UNP_LOCK_ASSERT(); 1235 1236 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn && 1237 (so = unp->unp_conn->unp_socket)) 1238 socantrcvmore(so); 1239} 1240 1241static void 1242unp_drop(struct unpcb *unp, int errno) 1243{ 1244 struct socket *so = unp->unp_socket; 1245 1246 UNP_LOCK_ASSERT(); 1247 1248 so->so_error = errno; 1249 unp_disconnect(unp); 1250} 1251 1252static void 1253unp_freerights(struct file **rp, int fdcount) 1254{ 1255 int i; 1256 struct file *fp; 1257 1258 for (i = 0; i < fdcount; i++) { 1259 fp = *rp; 1260 /* 1261 * Zero the pointer before calling unp_discard since it may 1262 * end up in unp_gc().. 1263 * 1264 * XXXRW: This is less true than it used to be. 1265 */ 1266 *rp++ = 0; 1267 unp_discard(fp); 1268 } 1269} 1270 1271int 1272unp_externalize(struct mbuf *control, struct mbuf **controlp) 1273{ 1274 struct thread *td = curthread; /* XXX */ 1275 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1276 int i; 1277 int *fdp; 1278 struct file **rp; 1279 struct file *fp; 1280 void *data; 1281 socklen_t clen = control->m_len, datalen; 1282 int error, newfds; 1283 int f; 1284 u_int newlen; 1285 1286 UNP_UNLOCK_ASSERT(); 1287 1288 error = 0; 1289 if (controlp != NULL) /* controlp == NULL => free control messages */ 1290 *controlp = NULL; 1291 1292 while (cm != NULL) { 1293 if (sizeof(*cm) > clen || cm->cmsg_len > clen) { 1294 error = EINVAL; 1295 break; 1296 } 1297 1298 data = CMSG_DATA(cm); 1299 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1300 1301 if (cm->cmsg_level == SOL_SOCKET 1302 && cm->cmsg_type == SCM_RIGHTS) { 1303 newfds = datalen / sizeof(struct file *); 1304 rp = data; 1305 1306 /* If we're not outputting the descriptors free them. */ 1307 if (error || controlp == NULL) { 1308 unp_freerights(rp, newfds); 1309 goto next; 1310 } 1311 FILEDESC_LOCK(td->td_proc->p_fd); 1312 /* if the new FD's will not fit free them. */ 1313 if (!fdavail(td, newfds)) { 1314 FILEDESC_UNLOCK(td->td_proc->p_fd); 1315 error = EMSGSIZE; 1316 unp_freerights(rp, newfds); 1317 goto next; 1318 } 1319 /* 1320 * Now change each pointer to an fd in the global 1321 * table to an integer that is the index to the local 1322 * fd table entry that we set up to point to the 1323 * global one we are transferring. 1324 */ 1325 newlen = newfds * sizeof(int); 1326 *controlp = sbcreatecontrol(NULL, newlen, 1327 SCM_RIGHTS, SOL_SOCKET); 1328 if (*controlp == NULL) { 1329 FILEDESC_UNLOCK(td->td_proc->p_fd); 1330 error = E2BIG; 1331 unp_freerights(rp, newfds); 1332 goto next; 1333 } 1334 1335 fdp = (int *) 1336 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1337 for (i = 0; i < newfds; i++) { 1338 if (fdalloc(td, 0, &f)) 1339 panic("unp_externalize fdalloc failed"); 1340 fp = *rp++; 1341 td->td_proc->p_fd->fd_ofiles[f] = fp; 1342 FILE_LOCK(fp); 1343 fp->f_msgcount--; 1344 FILE_UNLOCK(fp); 1345 unp_rights--; 1346 *fdp++ = f; 1347 } 1348 FILEDESC_UNLOCK(td->td_proc->p_fd); 1349 } else { 1350 /* We can just copy anything else across. */ 1351 if (error || controlp == NULL) 1352 goto next; 1353 *controlp = sbcreatecontrol(NULL, datalen, 1354 cm->cmsg_type, cm->cmsg_level); 1355 if (*controlp == NULL) { 1356 error = ENOBUFS; 1357 goto next; 1358 } 1359 bcopy(data, 1360 CMSG_DATA(mtod(*controlp, struct cmsghdr *)), 1361 datalen); 1362 } 1363 1364 controlp = &(*controlp)->m_next; 1365 1366next: 1367 if (CMSG_SPACE(datalen) < clen) { 1368 clen -= CMSG_SPACE(datalen); 1369 cm = (struct cmsghdr *) 1370 ((caddr_t)cm + CMSG_SPACE(datalen)); 1371 } else { 1372 clen = 0; 1373 cm = NULL; 1374 } 1375 } 1376 1377 m_freem(control); 1378 1379 return (error); 1380} 1381 1382static void 1383unp_zone_change(void *tag) 1384{ 1385 1386 uma_zone_set_max(unp_zone, maxsockets); 1387} 1388 1389void 1390unp_init(void) 1391{ 1392 1393 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL, 1394 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 1395 if (unp_zone == NULL) 1396 panic("unp_init"); 1397 uma_zone_set_max(unp_zone, maxsockets); 1398 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change, 1399 NULL, EVENTHANDLER_PRI_ANY); 1400 LIST_INIT(&unp_dhead); 1401 LIST_INIT(&unp_shead); 1402 TASK_INIT(&unp_gc_task, 0, unp_gc, NULL); 1403 UNP_LOCK_INIT(); 1404} 1405 1406static int 1407unp_internalize(struct mbuf **controlp, struct thread *td) 1408{ 1409 struct mbuf *control = *controlp; 1410 struct proc *p = td->td_proc; 1411 struct filedesc *fdescp = p->p_fd; 1412 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1413 struct cmsgcred *cmcred; 1414 struct file **rp; 1415 struct file *fp; 1416 struct timeval *tv; 1417 int i, fd, *fdp; 1418 void *data; 1419 socklen_t clen = control->m_len, datalen; 1420 int error, oldfds; 1421 u_int newlen; 1422 1423 UNP_UNLOCK_ASSERT(); 1424 1425 error = 0; 1426 *controlp = NULL; 1427 1428 while (cm != NULL) { 1429 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET 1430 || cm->cmsg_len > clen) { 1431 error = EINVAL; 1432 goto out; 1433 } 1434 1435 data = CMSG_DATA(cm); 1436 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1437 1438 switch (cm->cmsg_type) { 1439 /* 1440 * Fill in credential information. 1441 */ 1442 case SCM_CREDS: 1443 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred), 1444 SCM_CREDS, SOL_SOCKET); 1445 if (*controlp == NULL) { 1446 error = ENOBUFS; 1447 goto out; 1448 } 1449 1450 cmcred = (struct cmsgcred *) 1451 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1452 cmcred->cmcred_pid = p->p_pid; 1453 cmcred->cmcred_uid = td->td_ucred->cr_ruid; 1454 cmcred->cmcred_gid = td->td_ucred->cr_rgid; 1455 cmcred->cmcred_euid = td->td_ucred->cr_uid; 1456 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups, 1457 CMGROUP_MAX); 1458 for (i = 0; i < cmcred->cmcred_ngroups; i++) 1459 cmcred->cmcred_groups[i] = 1460 td->td_ucred->cr_groups[i]; 1461 break; 1462 1463 case SCM_RIGHTS: 1464 oldfds = datalen / sizeof (int); 1465 /* 1466 * Check that all the FDs passed in refer to legal 1467 * files. If not, reject the entire operation. 1468 */ 1469 fdp = data; 1470 FILEDESC_LOCK(fdescp); 1471 for (i = 0; i < oldfds; i++) { 1472 fd = *fdp++; 1473 if ((unsigned)fd >= fdescp->fd_nfiles || 1474 fdescp->fd_ofiles[fd] == NULL) { 1475 FILEDESC_UNLOCK(fdescp); 1476 error = EBADF; 1477 goto out; 1478 } 1479 fp = fdescp->fd_ofiles[fd]; 1480 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { 1481 FILEDESC_UNLOCK(fdescp); 1482 error = EOPNOTSUPP; 1483 goto out; 1484 } 1485 1486 } 1487 /* 1488 * Now replace the integer FDs with pointers to the 1489 * associated global file table entry.. 1490 */ 1491 newlen = oldfds * sizeof(struct file *); 1492 *controlp = sbcreatecontrol(NULL, newlen, 1493 SCM_RIGHTS, SOL_SOCKET); 1494 if (*controlp == NULL) { 1495 FILEDESC_UNLOCK(fdescp); 1496 error = E2BIG; 1497 goto out; 1498 } 1499 1500 fdp = data; 1501 rp = (struct file **) 1502 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1503 for (i = 0; i < oldfds; i++) { 1504 fp = fdescp->fd_ofiles[*fdp++]; 1505 *rp++ = fp; 1506 FILE_LOCK(fp); 1507 fp->f_count++; 1508 fp->f_msgcount++; 1509 FILE_UNLOCK(fp); 1510 unp_rights++; 1511 } 1512 FILEDESC_UNLOCK(fdescp); 1513 break; 1514 1515 case SCM_TIMESTAMP: 1516 *controlp = sbcreatecontrol(NULL, sizeof(*tv), 1517 SCM_TIMESTAMP, SOL_SOCKET); 1518 if (*controlp == NULL) { 1519 error = ENOBUFS; 1520 goto out; 1521 } 1522 tv = (struct timeval *) 1523 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1524 microtime(tv); 1525 break; 1526 1527 default: 1528 error = EINVAL; 1529 goto out; 1530 } 1531 1532 controlp = &(*controlp)->m_next; 1533 1534 if (CMSG_SPACE(datalen) < clen) { 1535 clen -= CMSG_SPACE(datalen); 1536 cm = (struct cmsghdr *) 1537 ((caddr_t)cm + CMSG_SPACE(datalen)); 1538 } else { 1539 clen = 0; 1540 cm = NULL; 1541 } 1542 } 1543 1544out: 1545 m_freem(control); 1546 1547 return (error); 1548} 1549 1550struct mbuf * 1551unp_addsockcred(struct thread *td, struct mbuf *control) 1552{ 1553 struct mbuf *m, *n, *n_prev; 1554 struct sockcred *sc; 1555 const struct cmsghdr *cm; 1556 int ngroups; 1557 int i; 1558 1559 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); 1560 1561 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET); 1562 if (m == NULL) 1563 return (control); 1564 1565 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *)); 1566 sc->sc_uid = td->td_ucred->cr_ruid; 1567 sc->sc_euid = td->td_ucred->cr_uid; 1568 sc->sc_gid = td->td_ucred->cr_rgid; 1569 sc->sc_egid = td->td_ucred->cr_gid; 1570 sc->sc_ngroups = ngroups; 1571 for (i = 0; i < sc->sc_ngroups; i++) 1572 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 1573 1574 /* 1575 * Unlink SCM_CREDS control messages (struct cmsgcred), since just 1576 * created SCM_CREDS control message (struct sockcred) has another 1577 * format. 1578 */ 1579 if (control != NULL) 1580 for (n = control, n_prev = NULL; n != NULL;) { 1581 cm = mtod(n, struct cmsghdr *); 1582 if (cm->cmsg_level == SOL_SOCKET && 1583 cm->cmsg_type == SCM_CREDS) { 1584 if (n_prev == NULL) 1585 control = n->m_next; 1586 else 1587 n_prev->m_next = n->m_next; 1588 n = m_free(n); 1589 } else { 1590 n_prev = n; 1591 n = n->m_next; 1592 } 1593 } 1594 1595 /* Prepend it to the head. */ 1596 m->m_next = control; 1597 1598 return (m); 1599} 1600 1601/* 1602 * unp_defer indicates whether additional work has been defered for a future 1603 * pass through unp_gc(). It is thread local and does not require explicit 1604 * synchronization. 1605 */ 1606static int unp_defer; 1607 1608static int unp_taskcount; 1609SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, ""); 1610 1611static int unp_recycled; 1612SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, ""); 1613 1614static void 1615unp_gc(__unused void *arg, int pending) 1616{ 1617 struct file *fp, *nextfp; 1618 struct socket *so; 1619 struct file **extra_ref, **fpp; 1620 int nunref, i; 1621 int nfiles_snap; 1622 int nfiles_slack = 20; 1623 1624 unp_taskcount++; 1625 unp_defer = 0; 1626 /* 1627 * Before going through all this, set all FDs to be NOT defered and 1628 * NOT externally accessible. 1629 */ 1630 sx_slock(&filelist_lock); 1631 LIST_FOREACH(fp, &filehead, f_list) 1632 fp->f_gcflag &= ~(FMARK|FDEFER); 1633 do { 1634 KASSERT(unp_defer >= 0, ("unp_gc: unp_defer %d", unp_defer)); 1635 LIST_FOREACH(fp, &filehead, f_list) { 1636 FILE_LOCK(fp); 1637 /* 1638 * If the file is not open, skip it -- could be a 1639 * file in the process of being opened, or in the 1640 * process of being closed. If the file is 1641 * "closing", it may have been marked for deferred 1642 * consideration. Clear the flag now if so. 1643 */ 1644 if (fp->f_count == 0) { 1645 if (fp->f_gcflag & FDEFER) 1646 unp_defer--; 1647 fp->f_gcflag &= ~(FMARK|FDEFER); 1648 FILE_UNLOCK(fp); 1649 continue; 1650 } 1651 /* 1652 * If we already marked it as 'defer' in a previous 1653 * pass, then try process it this time and un-mark 1654 * it. 1655 */ 1656 if (fp->f_gcflag & FDEFER) { 1657 fp->f_gcflag &= ~FDEFER; 1658 unp_defer--; 1659 } else { 1660 /* 1661 * if it's not defered, then check if it's 1662 * already marked.. if so skip it 1663 */ 1664 if (fp->f_gcflag & FMARK) { 1665 FILE_UNLOCK(fp); 1666 continue; 1667 } 1668 /* 1669 * If all references are from messages in 1670 * transit, then skip it. it's not externally 1671 * accessible. 1672 */ 1673 if (fp->f_count == fp->f_msgcount) { 1674 FILE_UNLOCK(fp); 1675 continue; 1676 } 1677 /* 1678 * If it got this far then it must be 1679 * externally accessible. 1680 */ 1681 fp->f_gcflag |= FMARK; 1682 } 1683 /* 1684 * Either it was defered, or it is externally 1685 * accessible and not already marked so. Now check 1686 * if it is possibly one of OUR sockets. 1687 */ 1688 if (fp->f_type != DTYPE_SOCKET || 1689 (so = fp->f_data) == NULL) { 1690 FILE_UNLOCK(fp); 1691 continue; 1692 } 1693 FILE_UNLOCK(fp); 1694 if (so->so_proto->pr_domain != &localdomain || 1695 (so->so_proto->pr_flags&PR_RIGHTS) == 0) 1696 continue; 1697 /* 1698 * So, Ok, it's one of our sockets and it IS 1699 * externally accessible (or was defered). Now we 1700 * look to see if we hold any file descriptors in its 1701 * message buffers. Follow those links and mark them 1702 * as accessible too. 1703 */ 1704 SOCKBUF_LOCK(&so->so_rcv); 1705 unp_scan(so->so_rcv.sb_mb, unp_mark); 1706 SOCKBUF_UNLOCK(&so->so_rcv); 1707 } 1708 } while (unp_defer); 1709 sx_sunlock(&filelist_lock); 1710 /* 1711 * XXXRW: The following comments need updating for a post-SMPng and 1712 * deferred unp_gc() world, but are still generally accurate. 1713 * 1714 * We grab an extra reference to each of the file table entries that 1715 * are not otherwise accessible and then free the rights that are 1716 * stored in messages on them. 1717 * 1718 * The bug in the orginal code is a little tricky, so I'll describe 1719 * what's wrong with it here. 1720 * 1721 * It is incorrect to simply unp_discard each entry for f_msgcount 1722 * times -- consider the case of sockets A and B that contain 1723 * references to each other. On a last close of some other socket, 1724 * we trigger a gc since the number of outstanding rights (unp_rights) 1725 * is non-zero. If during the sweep phase the gc code unp_discards, 1726 * we end up doing a (full) closef on the descriptor. A closef on A 1727 * results in the following chain. Closef calls soo_close, which 1728 * calls soclose. Soclose calls first (through the switch 1729 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply 1730 * returns because the previous instance had set unp_gcing, and we 1731 * return all the way back to soclose, which marks the socket with 1732 * SS_NOFDREF, and then calls sofree. Sofree calls sorflush to free 1733 * up the rights that are queued in messages on the socket A, i.e., 1734 * the reference on B. The sorflush calls via the dom_dispose switch 1735 * unp_dispose, which unp_scans with unp_discard. This second 1736 * instance of unp_discard just calls closef on B. 1737 * 1738 * Well, a similar chain occurs on B, resulting in a sorflush on B, 1739 * which results in another closef on A. Unfortunately, A is already 1740 * being closed, and the descriptor has already been marked with 1741 * SS_NOFDREF, and soclose panics at this point. 1742 * 1743 * Here, we first take an extra reference to each inaccessible 1744 * descriptor. Then, we call sorflush ourself, since we know it is a 1745 * Unix domain socket anyhow. After we destroy all the rights 1746 * carried in messages, we do a last closef to get rid of our extra 1747 * reference. This is the last close, and the unp_detach etc will 1748 * shut down the socket. 1749 * 1750 * 91/09/19, bsy@cs.cmu.edu 1751 */ 1752again: 1753 nfiles_snap = openfiles + nfiles_slack; /* some slack */ 1754 extra_ref = malloc(nfiles_snap * sizeof(struct file *), M_TEMP, 1755 M_WAITOK); 1756 sx_slock(&filelist_lock); 1757 if (nfiles_snap < openfiles) { 1758 sx_sunlock(&filelist_lock); 1759 free(extra_ref, M_TEMP); 1760 nfiles_slack += 20; 1761 goto again; 1762 } 1763 for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref; 1764 fp != NULL; fp = nextfp) { 1765 nextfp = LIST_NEXT(fp, f_list); 1766 FILE_LOCK(fp); 1767 /* 1768 * If it's not open, skip it 1769 */ 1770 if (fp->f_count == 0) { 1771 FILE_UNLOCK(fp); 1772 continue; 1773 } 1774 /* 1775 * If all refs are from msgs, and it's not marked accessible 1776 * then it must be referenced from some unreachable cycle of 1777 * (shut-down) FDs, so include it in our list of FDs to 1778 * remove. 1779 */ 1780 if (fp->f_count == fp->f_msgcount && !(fp->f_gcflag & FMARK)) { 1781 *fpp++ = fp; 1782 nunref++; 1783 fp->f_count++; 1784 } 1785 FILE_UNLOCK(fp); 1786 } 1787 sx_sunlock(&filelist_lock); 1788 /* 1789 * For each FD on our hit list, do the following two things: 1790 */ 1791 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1792 struct file *tfp = *fpp; 1793 FILE_LOCK(tfp); 1794 if (tfp->f_type == DTYPE_SOCKET && 1795 tfp->f_data != NULL) { 1796 FILE_UNLOCK(tfp); 1797 sorflush(tfp->f_data); 1798 } else { 1799 FILE_UNLOCK(tfp); 1800 } 1801 } 1802 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1803 closef(*fpp, (struct thread *) NULL); 1804 unp_recycled++; 1805 } 1806 free(extra_ref, M_TEMP); 1807} 1808 1809void 1810unp_dispose(struct mbuf *m) 1811{ 1812 1813 if (m) 1814 unp_scan(m, unp_discard); 1815} 1816 1817static int 1818unp_listen(struct socket *so, struct unpcb *unp, int backlog, 1819 struct thread *td) 1820{ 1821 int error; 1822 1823 UNP_LOCK_ASSERT(); 1824 1825 SOCK_LOCK(so); 1826 error = solisten_proto_check(so); 1827 if (error == 0) { 1828 cru2x(td->td_ucred, &unp->unp_peercred); 1829 unp->unp_flags |= UNP_HAVEPCCACHED; 1830 solisten_proto(so, backlog); 1831 } 1832 SOCK_UNLOCK(so); 1833 return (error); 1834} 1835 1836static void 1837unp_scan(struct mbuf *m0, void (*op)(struct file *)) 1838{ 1839 struct mbuf *m; 1840 struct file **rp; 1841 struct cmsghdr *cm; 1842 void *data; 1843 int i; 1844 socklen_t clen, datalen; 1845 int qfds; 1846 1847 while (m0 != NULL) { 1848 for (m = m0; m; m = m->m_next) { 1849 if (m->m_type != MT_CONTROL) 1850 continue; 1851 1852 cm = mtod(m, struct cmsghdr *); 1853 clen = m->m_len; 1854 1855 while (cm != NULL) { 1856 if (sizeof(*cm) > clen || cm->cmsg_len > clen) 1857 break; 1858 1859 data = CMSG_DATA(cm); 1860 datalen = (caddr_t)cm + cm->cmsg_len 1861 - (caddr_t)data; 1862 1863 if (cm->cmsg_level == SOL_SOCKET && 1864 cm->cmsg_type == SCM_RIGHTS) { 1865 qfds = datalen / sizeof (struct file *); 1866 rp = data; 1867 for (i = 0; i < qfds; i++) 1868 (*op)(*rp++); 1869 } 1870 1871 if (CMSG_SPACE(datalen) < clen) { 1872 clen -= CMSG_SPACE(datalen); 1873 cm = (struct cmsghdr *) 1874 ((caddr_t)cm + CMSG_SPACE(datalen)); 1875 } else { 1876 clen = 0; 1877 cm = NULL; 1878 } 1879 } 1880 } 1881 m0 = m0->m_act; 1882 } 1883} 1884 1885static void 1886unp_mark(struct file *fp) 1887{ 1888 if (fp->f_gcflag & FMARK) 1889 return; 1890 unp_defer++; 1891 fp->f_gcflag |= (FMARK|FDEFER); 1892} 1893 1894static void 1895unp_discard(struct file *fp) 1896{ 1897 UNP_LOCK(); 1898 FILE_LOCK(fp); 1899 fp->f_msgcount--; 1900 unp_rights--; 1901 FILE_UNLOCK(fp); 1902 UNP_UNLOCK(); 1903 (void) closef(fp, (struct thread *)NULL); 1904} 1905