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("$FreeBSD$"); 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 89 static uma_zone_t unp_zone; 90 static unp_gen_t unp_gencnt; 91 static u_int unp_count; 92 93 static 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 */ 104 static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL }; 105 static ino_t unp_ino; /* prototype for fake inode numbers */ 106 struct 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 120 static u_long unpst_sendspace = PIPSIZ; 121 static u_long unpst_recvspace = PIPSIZ; 122 static u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 123 static u_long unpdg_recvspace = 4*1024; 124 125 static int unp_rights; /* file descriptors in flight */ 126 127 SYSCTL_DECL(_net_local_stream); 128 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 129 &unpst_sendspace, 0, ""); 130 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 131 &unpst_recvspace, 0, ""); 132 SYSCTL_DECL(_net_local_dgram); 133 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 134 &unpdg_sendspace, 0, ""); 135 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 136 &unpdg_recvspace, 0, ""); 137 SYSCTL_DECL(_net_local); 138 SYSCTL_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 */ 160 static 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 */ 174 static struct task unp_gc_task; 175 176 static int unp_connect(struct socket *,struct sockaddr *, struct thread *); 177 static int unp_connect2(struct socket *so, struct socket *so2, int); 178 static void unp_disconnect(struct unpcb *); 179 static void unp_shutdown(struct unpcb *); 180 static void unp_drop(struct unpcb *, int); 181 static void unp_gc(__unused void *, int); 182 static void unp_scan(struct mbuf *, void (*)(struct file *)); 183 static void unp_mark(struct file *); 184 static void unp_discard(struct file *); 185 static void unp_freerights(struct file **, int); 186 static int unp_internalize(struct mbuf **, struct thread *); 187 static int unp_listen(struct socket *, struct unpcb *, int, 188 struct thread *); 189 190 static void 191 uipc_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 } 201 202 static int 203 uipc_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 } 224 225 static int 226 uipc_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 } 265 266 static int 267 uipc_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); 311 restart: 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); 367 error: 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 } 375 376 static int 377 uipc_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 */ 391 static void 392 uipc_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 } 402 403 int 404 uipc_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 */ 418 419 static void 420 uipc_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 } 462 463 static int 464 uipc_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 } 475 476 static int 477 uipc_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 } 493 494 static int 495 uipc_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 } 518 519 static int 520 uipc_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 */ 567 568 static int 569 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 570 struct mbuf *control, struct thread *td) 571 { 572 struct unpcb *unp, *unp2; 573 struct socket *so2; 574 u_int mbcnt, sbcc; 575 u_long newhiwat; 576 int error = 0; 577 578 unp = sotounpcb(so); 579 KASSERT(unp != NULL, ("uipc_send: unp == NULL")); 580 if (flags & PRUS_OOB) { 581 error = EOPNOTSUPP; 582 goto release; 583 } 584 585 if (control != NULL && (error = unp_internalize(&control, td))) 586 goto release; 587 588 UNP_LOCK(); 589 switch (so->so_type) { 590 case SOCK_DGRAM: 591 { 592 const struct sockaddr *from; 593 594 if (nam != NULL) { 595 if (unp->unp_conn != NULL) { 596 error = EISCONN; 597 break; 598 } 599 error = unp_connect(so, nam, td); 600 if (error) 601 break; 602 } 603 /* 604 * Because connect() and send() are non-atomic in a sendto() 605 * with a target address, it's possible that the socket will 606 * have disconnected before the send() can run. In that case 607 * return the slightly counter-intuitive but otherwise 608 * correct error that the socket is not connected. 609 */ 610 unp2 = unp->unp_conn; 611 if (unp2 == NULL) { 612 error = ENOTCONN; 613 break; 614 } 615 so2 = unp2->unp_socket; 616 if (unp->unp_addr != NULL) 617 from = (struct sockaddr *)unp->unp_addr; 618 else 619 from = &sun_noname; 620 if (unp2->unp_flags & UNP_WANTCRED) 621 control = unp_addsockcred(td, control); 622 SOCKBUF_LOCK(&so2->so_rcv); 623 if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) { 624 sorwakeup_locked(so2); 625 m = NULL; 626 control = NULL; 627 } else { 628 SOCKBUF_UNLOCK(&so2->so_rcv); 629 error = ENOBUFS; 630 } 631 if (nam != NULL) 632 unp_disconnect(unp); 633 break; 634 } 635 636 case SOCK_STREAM: 637 /* 638 * Connect if not connected yet. 639 * 640 * Note: A better implementation would complain if not equal 641 * to the peer's address. 642 */ 643 if ((so->so_state & SS_ISCONNECTED) == 0) { 644 if (nam != NULL) { 645 error = unp_connect(so, nam, td); 646 if (error) 647 break; /* XXX */ 648 } else { 649 error = ENOTCONN; 650 break; 651 } 652 } 653 654 /* Lockless read. */ 655 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 656 error = EPIPE; 657 break; 658 } 659 /* 660 * Because connect() and send() are non-atomic in a sendto() 661 * with a target address, it's possible that the socket will 662 * have disconnected before the send() can run. In that case 663 * return the slightly counter-intuitive but otherwise 664 * correct error that the socket is not connected. 665 */ 666 unp2 = unp->unp_conn; 667 if (unp2 == NULL) { 668 error = ENOTCONN; 669 break; 670 } 671 so2 = unp2->unp_socket; 672 SOCKBUF_LOCK(&so2->so_rcv); 673 if (unp2->unp_flags & UNP_WANTCRED) { 674 /* 675 * Credentials are passed only once on 676 * SOCK_STREAM. 677 */ 678 unp2->unp_flags &= ~UNP_WANTCRED; 679 control = unp_addsockcred(td, control); 680 } 681 /* 682 * Send to paired receive port, and then reduce send buffer 683 * hiwater marks to maintain backpressure. Wake up readers. 684 */ 685 if (control != NULL) { 686 if (sbappendcontrol_locked(&so2->so_rcv, m, control)) 687 control = NULL; 688 } else { 689 sbappend_locked(&so2->so_rcv, m); 690 } 691 mbcnt = so2->so_rcv.sb_mbcnt - unp2->unp_mbcnt; 692 unp2->unp_mbcnt = so2->so_rcv.sb_mbcnt; 693 sbcc = so2->so_rcv.sb_cc; 694 sorwakeup_locked(so2); 695 696 SOCKBUF_LOCK(&so->so_snd); 697 newhiwat = so->so_snd.sb_hiwat - (sbcc - unp2->unp_cc); 698 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat, 699 newhiwat, RLIM_INFINITY); 700 so->so_snd.sb_mbmax -= mbcnt; 701 SOCKBUF_UNLOCK(&so->so_snd); 702 703 unp2->unp_cc = sbcc; 704 m = NULL; 705 break; 706 707 default: 708 panic("uipc_send unknown socktype"); 709 } 710 711 /* 712 * SEND_EOF is equivalent to a SEND followed by 713 * a SHUTDOWN. 714 */ 715 if (flags & PRUS_EOF) { 716 socantsendmore(so); 717 unp_shutdown(unp); 718 } 719 UNP_UNLOCK(); 720 721 if (control != NULL && error != 0) 722 unp_dispose(control); 723 724 release: 725 if (control != NULL) 726 m_freem(control); 727 if (m != NULL) 728 m_freem(m); 729 return (error); 730 } 731 732 static int 733 uipc_sense(struct socket *so, struct stat *sb) 734 { 735 struct unpcb *unp; 736 struct socket *so2; 737 738 unp = sotounpcb(so); 739 KASSERT(unp != NULL, ("uipc_sense: unp == NULL")); 740 UNP_LOCK(); 741 sb->st_blksize = so->so_snd.sb_hiwat; 742 if (so->so_type == SOCK_STREAM && unp->unp_conn != NULL) { 743 so2 = unp->unp_conn->unp_socket; 744 sb->st_blksize += so2->so_rcv.sb_cc; 745 } 746 sb->st_dev = NODEV; 747 if (unp->unp_ino == 0) 748 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino; 749 sb->st_ino = unp->unp_ino; 750 UNP_UNLOCK(); 751 return (0); 752 } 753 754 static int 755 uipc_shutdown(struct socket *so) 756 { 757 struct unpcb *unp; 758 759 unp = sotounpcb(so); 760 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL")); 761 UNP_LOCK(); 762 socantsendmore(so); 763 unp_shutdown(unp); 764 UNP_UNLOCK(); 765 return (0); 766 } 767 768 static int 769 uipc_sockaddr(struct socket *so, struct sockaddr **nam) 770 { 771 struct unpcb *unp; 772 const struct sockaddr *sa; 773 774 unp = sotounpcb(so); 775 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL")); 776 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 777 UNP_LOCK(); 778 if (unp->unp_addr != NULL) 779 sa = (struct sockaddr *) unp->unp_addr; 780 else 781 sa = &sun_noname; 782 bcopy(sa, *nam, sa->sa_len); 783 UNP_UNLOCK(); 784 return (0); 785 } 786 787 struct pr_usrreqs uipc_usrreqs = { 788 .pru_abort = uipc_abort, 789 .pru_accept = uipc_accept, 790 .pru_attach = uipc_attach, 791 .pru_bind = uipc_bind, 792 .pru_connect = uipc_connect, 793 .pru_connect2 = uipc_connect2, 794 .pru_detach = uipc_detach, 795 .pru_disconnect = uipc_disconnect, 796 .pru_listen = uipc_listen, 797 .pru_peeraddr = uipc_peeraddr, 798 .pru_rcvd = uipc_rcvd, 799 .pru_send = uipc_send, 800 .pru_sense = uipc_sense, 801 .pru_shutdown = uipc_shutdown, 802 .pru_sockaddr = uipc_sockaddr, 803 .pru_sosend = sosend_generic, 804 .pru_soreceive = soreceive_generic, 805 .pru_sopoll = sopoll_generic, 806 .pru_close = uipc_close, 807 }; 808 809 int 810 uipc_ctloutput(struct socket *so, struct sockopt *sopt) 811 { 812 struct unpcb *unp; 813 struct xucred xu; 814 int error, optval; 815 816 if (sopt->sopt_level != 0) 817 return (EINVAL); 818 819 unp = sotounpcb(so); 820 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL")); 821 UNP_LOCK(); 822 error = 0; 823 switch (sopt->sopt_dir) { 824 case SOPT_GET: 825 switch (sopt->sopt_name) { 826 case LOCAL_PEERCRED: 827 if (unp->unp_flags & UNP_HAVEPC) 828 xu = unp->unp_peercred; 829 else { 830 if (so->so_type == SOCK_STREAM) 831 error = ENOTCONN; 832 else 833 error = EINVAL; 834 } 835 if (error == 0) 836 error = sooptcopyout(sopt, &xu, sizeof(xu)); 837 break; 838 case LOCAL_CREDS: 839 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0; 840 error = sooptcopyout(sopt, &optval, sizeof(optval)); 841 break; 842 case LOCAL_CONNWAIT: 843 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0; 844 error = sooptcopyout(sopt, &optval, sizeof(optval)); 845 break; 846 default: 847 error = EOPNOTSUPP; 848 break; 849 } 850 break; 851 case SOPT_SET: 852 switch (sopt->sopt_name) { 853 case LOCAL_CREDS: 854 case LOCAL_CONNWAIT: 855 error = sooptcopyin(sopt, &optval, sizeof(optval), 856 sizeof(optval)); 857 if (error) 858 break; 859 860 #define OPTSET(bit) \ 861 if (optval) \ 862 unp->unp_flags |= bit; \ 863 else \ 864 unp->unp_flags &= ~bit; 865 866 switch (sopt->sopt_name) { 867 case LOCAL_CREDS: 868 OPTSET(UNP_WANTCRED); 869 break; 870 case LOCAL_CONNWAIT: 871 OPTSET(UNP_CONNWAIT); 872 break; 873 default: 874 break; 875 } 876 break; 877 #undef OPTSET 878 default: 879 error = ENOPROTOOPT; 880 break; 881 } 882 break; 883 default: 884 error = EOPNOTSUPP; 885 break; 886 } 887 UNP_UNLOCK(); 888 return (error); 889 } 890 891 static int 892 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 893 { 894 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 895 struct vnode *vp; 896 struct socket *so2, *so3; 897 struct unpcb *unp, *unp2, *unp3; 898 int error, len; 899 struct nameidata nd; 900 char buf[SOCK_MAXADDRLEN]; 901 struct sockaddr *sa; 902 903 UNP_LOCK_ASSERT(); 904 905 unp = sotounpcb(so); 906 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 907 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 908 if (len <= 0) 909 return (EINVAL); 910 strlcpy(buf, soun->sun_path, len + 1); 911 if (unp->unp_flags & UNP_CONNECTING) { 912 UNP_UNLOCK(); 913 return (EALREADY); 914 } 915 UNP_UNLOCK(); 916 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 917 mtx_lock(&Giant); 918 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, td); 919 error = namei(&nd); 920 if (error) 921 vp = NULL; 922 else 923 vp = nd.ni_vp; 924 ASSERT_VOP_LOCKED(vp, "unp_connect"); 925 NDFREE(&nd, NDF_ONLY_PNBUF); 926 if (error) 927 goto bad; 928 929 if (vp->v_type != VSOCK) { 930 error = ENOTSOCK; 931 goto bad; 932 } 933 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); 934 if (error) 935 goto bad; 936 mtx_unlock(&Giant); 937 UNP_LOCK(); 938 unp = sotounpcb(so); 939 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 940 so2 = vp->v_socket; 941 if (so2 == NULL) { 942 error = ECONNREFUSED; 943 goto bad2; 944 } 945 if (so->so_type != so2->so_type) { 946 error = EPROTOTYPE; 947 goto bad2; 948 } 949 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 950 if (so2->so_options & SO_ACCEPTCONN) { 951 /* 952 * NB: drop locks here so unp_attach is entered w/o 953 * locks; this avoids a recursive lock of the head 954 * and holding sleep locks across a (potentially) 955 * blocking malloc. 956 */ 957 UNP_UNLOCK(); 958 so3 = sonewconn(so2, 0); 959 UNP_LOCK(); 960 } else 961 so3 = NULL; 962 if (so3 == NULL) { 963 error = ECONNREFUSED; 964 goto bad2; 965 } 966 unp = sotounpcb(so); 967 unp2 = sotounpcb(so2); 968 unp3 = sotounpcb(so3); 969 if (unp2->unp_addr != NULL) { 970 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); 971 unp3->unp_addr = (struct sockaddr_un *) sa; 972 sa = NULL; 973 } 974 /* 975 * unp_peercred management: 976 * 977 * The connecter's (client's) credentials are copied from its 978 * process structure at the time of connect() (which is now). 979 */ 980 cru2x(td->td_ucred, &unp3->unp_peercred); 981 unp3->unp_flags |= UNP_HAVEPC; 982 /* 983 * The receiver's (server's) credentials are copied from the 984 * unp_peercred member of socket on which the former called 985 * listen(); unp_listen() cached that process's credentials 986 * at that time so we can use them now. 987 */ 988 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED, 989 ("unp_connect: listener without cached peercred")); 990 memcpy(&unp->unp_peercred, &unp2->unp_peercred, 991 sizeof(unp->unp_peercred)); 992 unp->unp_flags |= UNP_HAVEPC; 993 if (unp2->unp_flags & UNP_WANTCRED) 994 unp3->unp_flags |= UNP_WANTCRED; 995 #ifdef MAC 996 SOCK_LOCK(so); 997 mac_set_socket_peer_from_socket(so, so3); 998 mac_set_socket_peer_from_socket(so3, so); 999 SOCK_UNLOCK(so); 1000 #endif 1001 1002 so2 = so3; 1003 } 1004 error = unp_connect2(so, so2, PRU_CONNECT); 1005 bad2: 1006 UNP_UNLOCK(); 1007 mtx_lock(&Giant); 1008 bad: 1009 mtx_assert(&Giant, MA_OWNED); 1010 if (vp != NULL) 1011 vput(vp); 1012 mtx_unlock(&Giant); 1013 free(sa, M_SONAME); 1014 UNP_LOCK(); 1015 unp->unp_flags &= ~UNP_CONNECTING; 1016 return (error); 1017 } 1018 1019 static int 1020 unp_connect2(struct socket *so, struct socket *so2, int req) 1021 { 1022 struct unpcb *unp = sotounpcb(so); 1023 struct unpcb *unp2; 1024 1025 UNP_LOCK_ASSERT(); 1026 1027 if (so2->so_type != so->so_type) 1028 return (EPROTOTYPE); 1029 unp2 = sotounpcb(so2); 1030 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL")); 1031 unp->unp_conn = unp2; 1032 switch (so->so_type) { 1033 1034 case SOCK_DGRAM: 1035 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 1036 soisconnected(so); 1037 break; 1038 1039 case SOCK_STREAM: 1040 unp2->unp_conn = unp; 1041 if (req == PRU_CONNECT && 1042 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 1043 soisconnecting(so); 1044 else 1045 soisconnected(so); 1046 soisconnected(so2); 1047 break; 1048 1049 default: 1050 panic("unp_connect2"); 1051 } 1052 return (0); 1053 } 1054 1055 static void 1056 unp_disconnect(struct unpcb *unp) 1057 { 1058 struct unpcb *unp2 = unp->unp_conn; 1059 struct socket *so; 1060 1061 UNP_LOCK_ASSERT(); 1062 1063 if (unp2 == NULL) 1064 return; 1065 unp->unp_conn = NULL; 1066 switch (unp->unp_socket->so_type) { 1067 case SOCK_DGRAM: 1068 LIST_REMOVE(unp, unp_reflink); 1069 so = unp->unp_socket; 1070 SOCK_LOCK(so); 1071 so->so_state &= ~SS_ISCONNECTED; 1072 SOCK_UNLOCK(so); 1073 break; 1074 1075 case SOCK_STREAM: 1076 soisdisconnected(unp->unp_socket); 1077 unp2->unp_conn = NULL; 1078 soisdisconnected(unp2->unp_socket); 1079 break; 1080 } 1081 } 1082 1083 /* 1084 * unp_pcblist() assumes that UNIX domain socket memory is never reclaimed by 1085 * the zone (UMA_ZONE_NOFREE), and as such potentially stale pointers are 1086 * safe to reference. It first scans the list of struct unpcb's to generate 1087 * a pointer list, then it rescans its list one entry at a time to 1088 * externalize and copyout. It checks the generation number to see if a 1089 * struct unpcb has been reused, and will skip it if so. 1090 */ 1091 static int 1092 unp_pcblist(SYSCTL_HANDLER_ARGS) 1093 { 1094 int error, i, n; 1095 struct unpcb *unp, **unp_list; 1096 unp_gen_t gencnt; 1097 struct xunpgen *xug; 1098 struct unp_head *head; 1099 struct xunpcb *xu; 1100 1101 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead); 1102 1103 /* 1104 * The process of preparing the PCB list is too time-consuming and 1105 * resource-intensive to repeat twice on every request. 1106 */ 1107 if (req->oldptr == NULL) { 1108 n = unp_count; 1109 req->oldidx = 2 * (sizeof *xug) 1110 + (n + n/8) * sizeof(struct xunpcb); 1111 return (0); 1112 } 1113 1114 if (req->newptr != NULL) 1115 return (EPERM); 1116 1117 /* 1118 * OK, now we're committed to doing something. 1119 */ 1120 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK); 1121 UNP_LOCK(); 1122 gencnt = unp_gencnt; 1123 n = unp_count; 1124 UNP_UNLOCK(); 1125 1126 xug->xug_len = sizeof *xug; 1127 xug->xug_count = n; 1128 xug->xug_gen = gencnt; 1129 xug->xug_sogen = so_gencnt; 1130 error = SYSCTL_OUT(req, xug, sizeof *xug); 1131 if (error) { 1132 free(xug, M_TEMP); 1133 return (error); 1134 } 1135 1136 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 1137 1138 UNP_LOCK(); 1139 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 1140 unp = LIST_NEXT(unp, unp_link)) { 1141 if (unp->unp_gencnt <= gencnt) { 1142 if (cr_cansee(req->td->td_ucred, 1143 unp->unp_socket->so_cred)) 1144 continue; 1145 unp_list[i++] = unp; 1146 } 1147 } 1148 UNP_UNLOCK(); 1149 n = i; /* In case we lost some during malloc. */ 1150 1151 error = 0; 1152 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO); 1153 for (i = 0; i < n; i++) { 1154 unp = unp_list[i]; 1155 if (unp->unp_gencnt <= gencnt) { 1156 xu->xu_len = sizeof *xu; 1157 xu->xu_unpp = unp; 1158 /* 1159 * XXX - need more locking here to protect against 1160 * connect/disconnect races for SMP. 1161 */ 1162 if (unp->unp_addr != NULL) 1163 bcopy(unp->unp_addr, &xu->xu_addr, 1164 unp->unp_addr->sun_len); 1165 if (unp->unp_conn != NULL && 1166 unp->unp_conn->unp_addr != NULL) 1167 bcopy(unp->unp_conn->unp_addr, 1168 &xu->xu_caddr, 1169 unp->unp_conn->unp_addr->sun_len); 1170 bcopy(unp, &xu->xu_unp, sizeof *unp); 1171 sotoxsocket(unp->unp_socket, &xu->xu_socket); 1172 error = SYSCTL_OUT(req, xu, sizeof *xu); 1173 } 1174 } 1175 free(xu, M_TEMP); 1176 if (!error) { 1177 /* 1178 * Give the user an updated idea of our state. If the 1179 * generation differs from what we told her before, she knows 1180 * that something happened while we were processing this 1181 * request, and it might be necessary to retry. 1182 */ 1183 xug->xug_gen = unp_gencnt; 1184 xug->xug_sogen = so_gencnt; 1185 xug->xug_count = unp_count; 1186 error = SYSCTL_OUT(req, xug, sizeof *xug); 1187 } 1188 free(unp_list, M_TEMP); 1189 free(xug, M_TEMP); 1190 return (error); 1191 } 1192 1193 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD, 1194 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 1195 "List of active local datagram sockets"); 1196 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD, 1197 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 1198 "List of active local stream sockets"); 1199 1200 static void 1201 unp_shutdown(struct unpcb *unp) 1202 { 1203 struct socket *so; 1204 1205 UNP_LOCK_ASSERT(); 1206 1207 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn && 1208 (so = unp->unp_conn->unp_socket)) 1209 socantrcvmore(so); 1210 } 1211 1212 static void 1213 unp_drop(struct unpcb *unp, int errno) 1214 { 1215 struct socket *so = unp->unp_socket; 1216 1217 UNP_LOCK_ASSERT(); 1218 1219 so->so_error = errno; 1220 unp_disconnect(unp); 1221 } 1222 1223 static void 1224 unp_freerights(struct file **rp, int fdcount) 1225 { 1226 int i; 1227 struct file *fp; 1228 1229 for (i = 0; i < fdcount; i++) { 1230 fp = *rp; 1231 /* 1232 * Zero the pointer before calling unp_discard since it may 1233 * end up in unp_gc().. 1234 * 1235 * XXXRW: This is less true than it used to be. 1236 */ 1237 *rp++ = 0; 1238 unp_discard(fp); 1239 } 1240 } 1241 1242 int 1243 unp_externalize(struct mbuf *control, struct mbuf **controlp) 1244 { 1245 struct thread *td = curthread; /* XXX */ 1246 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1247 int i; 1248 int *fdp; 1249 struct file **rp; 1250 struct file *fp; 1251 void *data; 1252 socklen_t clen = control->m_len, datalen; 1253 int error, newfds; 1254 int f; 1255 u_int newlen; 1256 1257 UNP_UNLOCK_ASSERT(); 1258 1259 error = 0; 1260 if (controlp != NULL) /* controlp == NULL => free control messages */ 1261 *controlp = NULL; 1262 1263 while (cm != NULL) { 1264 if (sizeof(*cm) > clen || cm->cmsg_len > clen) { 1265 error = EINVAL; 1266 break; 1267 } 1268 1269 data = CMSG_DATA(cm); 1270 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1271 1272 if (cm->cmsg_level == SOL_SOCKET 1273 && cm->cmsg_type == SCM_RIGHTS) { 1274 newfds = datalen / sizeof(struct file *); 1275 rp = data; 1276 1277 /* If we're not outputting the descriptors free them. */ 1278 if (error || controlp == NULL) { 1279 unp_freerights(rp, newfds); 1280 goto next; 1281 } 1282 FILEDESC_LOCK(td->td_proc->p_fd); 1283 /* if the new FD's will not fit free them. */ 1284 if (!fdavail(td, newfds)) { 1285 FILEDESC_UNLOCK(td->td_proc->p_fd); 1286 error = EMSGSIZE; 1287 unp_freerights(rp, newfds); 1288 goto next; 1289 } 1290 /* 1291 * Now change each pointer to an fd in the global 1292 * table to an integer that is the index to the local 1293 * fd table entry that we set up to point to the 1294 * global one we are transferring. 1295 */ 1296 newlen = newfds * sizeof(int); 1297 *controlp = sbcreatecontrol(NULL, newlen, 1298 SCM_RIGHTS, SOL_SOCKET); 1299 if (*controlp == NULL) { 1300 FILEDESC_UNLOCK(td->td_proc->p_fd); 1301 error = E2BIG; 1302 unp_freerights(rp, newfds); 1303 goto next; 1304 } 1305 1306 fdp = (int *) 1307 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1308 for (i = 0; i < newfds; i++) { 1309 if (fdalloc(td, 0, &f)) 1310 panic("unp_externalize fdalloc failed"); 1311 fp = *rp++; 1312 td->td_proc->p_fd->fd_ofiles[f] = fp; 1313 FILE_LOCK(fp); 1314 fp->f_msgcount--; 1315 FILE_UNLOCK(fp); 1316 unp_rights--; 1317 *fdp++ = f; 1318 } 1319 FILEDESC_UNLOCK(td->td_proc->p_fd); 1320 } else { 1321 /* We can just copy anything else across. */ 1322 if (error || controlp == NULL) 1323 goto next; 1324 *controlp = sbcreatecontrol(NULL, datalen, 1325 cm->cmsg_type, cm->cmsg_level); 1326 if (*controlp == NULL) { 1327 error = ENOBUFS; 1328 goto next; 1329 } 1330 bcopy(data, 1331 CMSG_DATA(mtod(*controlp, struct cmsghdr *)), 1332 datalen); 1333 } 1334 1335 controlp = &(*controlp)->m_next; 1336 1337 next: 1338 if (CMSG_SPACE(datalen) < clen) { 1339 clen -= CMSG_SPACE(datalen); 1340 cm = (struct cmsghdr *) 1341 ((caddr_t)cm + CMSG_SPACE(datalen)); 1342 } else { 1343 clen = 0; 1344 cm = NULL; 1345 } 1346 } 1347 1348 m_freem(control); 1349 1350 return (error); 1351 } 1352 1353 static void 1354 unp_zone_change(void *tag) 1355 { 1356 1357 uma_zone_set_max(unp_zone, maxsockets); 1358 } 1359 1360 void 1361 unp_init(void) 1362 { 1363 1364 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL, 1365 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 1366 if (unp_zone == NULL) 1367 panic("unp_init"); 1368 uma_zone_set_max(unp_zone, maxsockets); 1369 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change, 1370 NULL, EVENTHANDLER_PRI_ANY); 1371 LIST_INIT(&unp_dhead); 1372 LIST_INIT(&unp_shead); 1373 TASK_INIT(&unp_gc_task, 0, unp_gc, NULL); 1374 UNP_LOCK_INIT(); 1375 } 1376 1377 static int 1378 unp_internalize(struct mbuf **controlp, struct thread *td) 1379 { 1380 struct mbuf *control = *controlp; 1381 struct proc *p = td->td_proc; 1382 struct filedesc *fdescp = p->p_fd; 1383 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1384 struct cmsgcred *cmcred; 1385 struct file **rp; 1386 struct file *fp; 1387 struct timeval *tv; 1388 int i, fd, *fdp; 1389 void *data; 1390 socklen_t clen = control->m_len, datalen; 1391 int error, oldfds; 1392 u_int newlen; 1393 1394 UNP_UNLOCK_ASSERT(); 1395 1396 error = 0; 1397 *controlp = NULL; 1398 1399 while (cm != NULL) { 1400 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET 1401 || cm->cmsg_len > clen) { 1402 error = EINVAL; 1403 goto out; 1404 } 1405 1406 data = CMSG_DATA(cm); 1407 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1408 1409 switch (cm->cmsg_type) { 1410 /* 1411 * Fill in credential information. 1412 */ 1413 case SCM_CREDS: 1414 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred), 1415 SCM_CREDS, SOL_SOCKET); 1416 if (*controlp == NULL) { 1417 error = ENOBUFS; 1418 goto out; 1419 } 1420 1421 cmcred = (struct cmsgcred *) 1422 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1423 cmcred->cmcred_pid = p->p_pid; 1424 cmcred->cmcred_uid = td->td_ucred->cr_ruid; 1425 cmcred->cmcred_gid = td->td_ucred->cr_rgid; 1426 cmcred->cmcred_euid = td->td_ucred->cr_uid; 1427 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups, 1428 CMGROUP_MAX); 1429 for (i = 0; i < cmcred->cmcred_ngroups; i++) 1430 cmcred->cmcred_groups[i] = 1431 td->td_ucred->cr_groups[i]; 1432 break; 1433 1434 case SCM_RIGHTS: 1435 oldfds = datalen / sizeof (int); 1436 /* 1437 * Check that all the FDs passed in refer to legal 1438 * files. If not, reject the entire operation. 1439 */ 1440 fdp = data; 1441 FILEDESC_LOCK(fdescp); 1442 for (i = 0; i < oldfds; i++) { 1443 fd = *fdp++; 1444 if ((unsigned)fd >= fdescp->fd_nfiles || 1445 fdescp->fd_ofiles[fd] == NULL) { 1446 FILEDESC_UNLOCK(fdescp); 1447 error = EBADF; 1448 goto out; 1449 } 1450 fp = fdescp->fd_ofiles[fd]; 1451 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { 1452 FILEDESC_UNLOCK(fdescp); 1453 error = EOPNOTSUPP; 1454 goto out; 1455 } 1456 1457 } 1458 /* 1459 * Now replace the integer FDs with pointers to the 1460 * associated global file table entry.. 1461 */ 1462 newlen = oldfds * sizeof(struct file *); 1463 *controlp = sbcreatecontrol(NULL, newlen, 1464 SCM_RIGHTS, SOL_SOCKET); 1465 if (*controlp == NULL) { 1466 FILEDESC_UNLOCK(fdescp); 1467 error = E2BIG; 1468 goto out; 1469 } 1470 1471 fdp = data; 1472 rp = (struct file **) 1473 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1474 for (i = 0; i < oldfds; i++) { 1475 fp = fdescp->fd_ofiles[*fdp++]; 1476 *rp++ = fp; 1477 FILE_LOCK(fp); 1478 fp->f_count++; 1479 fp->f_msgcount++; 1480 FILE_UNLOCK(fp); 1481 unp_rights++; 1482 } 1483 FILEDESC_UNLOCK(fdescp); 1484 break; 1485 1486 case SCM_TIMESTAMP: 1487 *controlp = sbcreatecontrol(NULL, sizeof(*tv), 1488 SCM_TIMESTAMP, SOL_SOCKET); 1489 if (*controlp == NULL) { 1490 error = ENOBUFS; 1491 goto out; 1492 } 1493 tv = (struct timeval *) 1494 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1495 microtime(tv); 1496 break; 1497 1498 default: 1499 error = EINVAL; 1500 goto out; 1501 } 1502 1503 controlp = &(*controlp)->m_next; 1504 1505 if (CMSG_SPACE(datalen) < clen) { 1506 clen -= CMSG_SPACE(datalen); 1507 cm = (struct cmsghdr *) 1508 ((caddr_t)cm + CMSG_SPACE(datalen)); 1509 } else { 1510 clen = 0; 1511 cm = NULL; 1512 } 1513 } 1514 1515 out: 1516 m_freem(control); 1517 1518 return (error); 1519 } 1520 1521 struct mbuf * 1522 unp_addsockcred(struct thread *td, struct mbuf *control) 1523 { 1524 struct mbuf *m, *n, *n_prev; 1525 struct sockcred *sc; 1526 const struct cmsghdr *cm; 1527 int ngroups; 1528 int i; 1529 1530 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); 1531 1532 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET); 1533 if (m == NULL) 1534 return (control); 1535 1536 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *)); 1537 sc->sc_uid = td->td_ucred->cr_ruid; 1538 sc->sc_euid = td->td_ucred->cr_uid; 1539 sc->sc_gid = td->td_ucred->cr_rgid; 1540 sc->sc_egid = td->td_ucred->cr_gid; 1541 sc->sc_ngroups = ngroups; 1542 for (i = 0; i < sc->sc_ngroups; i++) 1543 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 1544 1545 /* 1546 * Unlink SCM_CREDS control messages (struct cmsgcred), since just 1547 * created SCM_CREDS control message (struct sockcred) has another 1548 * format. 1549 */ 1550 if (control != NULL) 1551 for (n = control, n_prev = NULL; n != NULL;) { 1552 cm = mtod(n, struct cmsghdr *); 1553 if (cm->cmsg_level == SOL_SOCKET && 1554 cm->cmsg_type == SCM_CREDS) { 1555 if (n_prev == NULL) 1556 control = n->m_next; 1557 else 1558 n_prev->m_next = n->m_next; 1559 n = m_free(n); 1560 } else { 1561 n_prev = n; 1562 n = n->m_next; 1563 } 1564 } 1565 1566 /* Prepend it to the head. */ 1567 m->m_next = control; 1568 1569 return (m); 1570 } 1571 1572 /* 1573 * unp_defer indicates whether additional work has been defered for a future 1574 * pass through unp_gc(). It is thread local and does not require explicit 1575 * synchronization. 1576 */ 1577 static int unp_defer; 1578 1579 static int unp_taskcount; 1580 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, ""); 1581 1582 static int unp_recycled; 1583 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, ""); 1584 1585 static void 1586 unp_gc(__unused void *arg, int pending) 1587 { 1588 struct file *fp, *nextfp; 1589 struct socket *so; 1590 struct file **extra_ref, **fpp; 1591 int nunref, i; 1592 int nfiles_snap; 1593 int nfiles_slack = 20; 1594 1595 unp_taskcount++; 1596 unp_defer = 0; 1597 /* 1598 * Before going through all this, set all FDs to be NOT defered and 1599 * NOT externally accessible. 1600 */ 1601 sx_slock(&filelist_lock); 1602 LIST_FOREACH(fp, &filehead, f_list) 1603 fp->f_gcflag &= ~(FMARK|FDEFER); 1604 do { 1605 KASSERT(unp_defer >= 0, ("unp_gc: unp_defer %d", unp_defer)); 1606 LIST_FOREACH(fp, &filehead, f_list) { 1607 FILE_LOCK(fp); 1608 /* 1609 * If the file is not open, skip it -- could be a 1610 * file in the process of being opened, or in the 1611 * process of being closed. If the file is 1612 * "closing", it may have been marked for deferred 1613 * consideration. Clear the flag now if so. 1614 */ 1615 if (fp->f_count == 0) { 1616 if (fp->f_gcflag & FDEFER) 1617 unp_defer--; 1618 fp->f_gcflag &= ~(FMARK|FDEFER); 1619 FILE_UNLOCK(fp); 1620 continue; 1621 } 1622 /* 1623 * If we already marked it as 'defer' in a previous 1624 * pass, then try process it this time and un-mark 1625 * it. 1626 */ 1627 if (fp->f_gcflag & FDEFER) { 1628 fp->f_gcflag &= ~FDEFER; 1629 unp_defer--; 1630 } else { 1631 /* 1632 * if it's not defered, then check if it's 1633 * already marked.. if so skip it 1634 */ 1635 if (fp->f_gcflag & FMARK) { 1636 FILE_UNLOCK(fp); 1637 continue; 1638 } 1639 /* 1640 * If all references are from messages in 1641 * transit, then skip it. it's not externally 1642 * accessible. 1643 */ 1644 if (fp->f_count == fp->f_msgcount) { 1645 FILE_UNLOCK(fp); 1646 continue; 1647 } 1648 /* 1649 * If it got this far then it must be 1650 * externally accessible. 1651 */ 1652 fp->f_gcflag |= FMARK; 1653 } 1654 /* 1655 * Either it was defered, or it is externally 1656 * accessible and not already marked so. Now check 1657 * if it is possibly one of OUR sockets. 1658 */ 1659 if (fp->f_type != DTYPE_SOCKET || 1660 (so = fp->f_data) == NULL) { 1661 FILE_UNLOCK(fp); 1662 continue; 1663 } 1664 FILE_UNLOCK(fp); 1665 if (so->so_proto->pr_domain != &localdomain || 1666 (so->so_proto->pr_flags&PR_RIGHTS) == 0) 1667 continue; 1668 /* 1669 * So, Ok, it's one of our sockets and it IS 1670 * externally accessible (or was defered). Now we 1671 * look to see if we hold any file descriptors in its 1672 * message buffers. Follow those links and mark them 1673 * as accessible too. 1674 */ 1675 SOCKBUF_LOCK(&so->so_rcv); 1676 unp_scan(so->so_rcv.sb_mb, unp_mark); 1677 SOCKBUF_UNLOCK(&so->so_rcv); 1678 } 1679 } while (unp_defer); 1680 sx_sunlock(&filelist_lock); 1681 /* 1682 * XXXRW: The following comments need updating for a post-SMPng and 1683 * deferred unp_gc() world, but are still generally accurate. 1684 * 1685 * We grab an extra reference to each of the file table entries that 1686 * are not otherwise accessible and then free the rights that are 1687 * stored in messages on them. 1688 * 1689 * The bug in the orginal code is a little tricky, so I'll describe 1690 * what's wrong with it here. 1691 * 1692 * It is incorrect to simply unp_discard each entry for f_msgcount 1693 * times -- consider the case of sockets A and B that contain 1694 * references to each other. On a last close of some other socket, 1695 * we trigger a gc since the number of outstanding rights (unp_rights) 1696 * is non-zero. If during the sweep phase the gc code unp_discards, 1697 * we end up doing a (full) closef on the descriptor. A closef on A 1698 * results in the following chain. Closef calls soo_close, which 1699 * calls soclose. Soclose calls first (through the switch 1700 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply 1701 * returns because the previous instance had set unp_gcing, and we 1702 * return all the way back to soclose, which marks the socket with 1703 * SS_NOFDREF, and then calls sofree. Sofree calls sorflush to free 1704 * up the rights that are queued in messages on the socket A, i.e., 1705 * the reference on B. The sorflush calls via the dom_dispose switch 1706 * unp_dispose, which unp_scans with unp_discard. This second 1707 * instance of unp_discard just calls closef on B. 1708 * 1709 * Well, a similar chain occurs on B, resulting in a sorflush on B, 1710 * which results in another closef on A. Unfortunately, A is already 1711 * being closed, and the descriptor has already been marked with 1712 * SS_NOFDREF, and soclose panics at this point. 1713 * 1714 * Here, we first take an extra reference to each inaccessible 1715 * descriptor. Then, we call sorflush ourself, since we know it is a 1716 * Unix domain socket anyhow. After we destroy all the rights 1717 * carried in messages, we do a last closef to get rid of our extra 1718 * reference. This is the last close, and the unp_detach etc will 1719 * shut down the socket. 1720 * 1721 * 91/09/19, bsy@cs.cmu.edu 1722 */ 1723 again: 1724 nfiles_snap = openfiles + nfiles_slack; /* some slack */ 1725 extra_ref = malloc(nfiles_snap * sizeof(struct file *), M_TEMP, 1726 M_WAITOK); 1727 sx_slock(&filelist_lock); 1728 if (nfiles_snap < openfiles) { 1729 sx_sunlock(&filelist_lock); 1730 free(extra_ref, M_TEMP); 1731 nfiles_slack += 20; 1732 goto again; 1733 } 1734 for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref; 1735 fp != NULL; fp = nextfp) { 1736 nextfp = LIST_NEXT(fp, f_list); 1737 FILE_LOCK(fp); 1738 /* 1739 * If it's not open, skip it 1740 */ 1741 if (fp->f_count == 0) { 1742 FILE_UNLOCK(fp); 1743 continue; 1744 } 1745 /* 1746 * If all refs are from msgs, and it's not marked accessible 1747 * then it must be referenced from some unreachable cycle of 1748 * (shut-down) FDs, so include it in our list of FDs to 1749 * remove. 1750 */ 1751 if (fp->f_count == fp->f_msgcount && !(fp->f_gcflag & FMARK)) { 1752 *fpp++ = fp; 1753 nunref++; 1754 fp->f_count++; 1755 } 1756 FILE_UNLOCK(fp); 1757 } 1758 sx_sunlock(&filelist_lock); 1759 /* 1760 * For each FD on our hit list, do the following two things: 1761 */ 1762 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1763 struct file *tfp = *fpp; 1764 FILE_LOCK(tfp); 1765 if (tfp->f_type == DTYPE_SOCKET && 1766 tfp->f_data != NULL) { 1767 FILE_UNLOCK(tfp); 1768 sorflush(tfp->f_data); 1769 } else { 1770 FILE_UNLOCK(tfp); 1771 } 1772 } 1773 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1774 closef(*fpp, (struct thread *) NULL); 1775 unp_recycled++; 1776 } 1777 free(extra_ref, M_TEMP); 1778 } 1779 1780 void 1781 unp_dispose(struct mbuf *m) 1782 { 1783 1784 if (m) 1785 unp_scan(m, unp_discard); 1786 } 1787 1788 static int 1789 unp_listen(struct socket *so, struct unpcb *unp, int backlog, 1790 struct thread *td) 1791 { 1792 int error; 1793 1794 UNP_LOCK_ASSERT(); 1795 1796 SOCK_LOCK(so); 1797 error = solisten_proto_check(so); 1798 if (error == 0) { 1799 cru2x(td->td_ucred, &unp->unp_peercred); 1800 unp->unp_flags |= UNP_HAVEPCCACHED; 1801 solisten_proto(so, backlog); 1802 } 1803 SOCK_UNLOCK(so); 1804 return (error); 1805 } 1806 1807 static void 1808 unp_scan(struct mbuf *m0, void (*op)(struct file *)) 1809 { 1810 struct mbuf *m; 1811 struct file **rp; 1812 struct cmsghdr *cm; 1813 void *data; 1814 int i; 1815 socklen_t clen, datalen; 1816 int qfds; 1817 1818 while (m0 != NULL) { 1819 for (m = m0; m; m = m->m_next) { 1820 if (m->m_type != MT_CONTROL) 1821 continue; 1822 1823 cm = mtod(m, struct cmsghdr *); 1824 clen = m->m_len; 1825 1826 while (cm != NULL) { 1827 if (sizeof(*cm) > clen || cm->cmsg_len > clen) 1828 break; 1829 1830 data = CMSG_DATA(cm); 1831 datalen = (caddr_t)cm + cm->cmsg_len 1832 - (caddr_t)data; 1833 1834 if (cm->cmsg_level == SOL_SOCKET && 1835 cm->cmsg_type == SCM_RIGHTS) { 1836 qfds = datalen / sizeof (struct file *); 1837 rp = data; 1838 for (i = 0; i < qfds; i++) 1839 (*op)(*rp++); 1840 } 1841 1842 if (CMSG_SPACE(datalen) < clen) { 1843 clen -= CMSG_SPACE(datalen); 1844 cm = (struct cmsghdr *) 1845 ((caddr_t)cm + CMSG_SPACE(datalen)); 1846 } else { 1847 clen = 0; 1848 cm = NULL; 1849 } 1850 } 1851 } 1852 m0 = m0->m_act; 1853 } 1854 } 1855 1856 static void 1857 unp_mark(struct file *fp) 1858 { 1859 if (fp->f_gcflag & FMARK) 1860 return; 1861 unp_defer++; 1862 fp->f_gcflag |= (FMARK|FDEFER); 1863 } 1864 1865 static void 1866 unp_discard(struct file *fp) 1867 { 1868 UNP_LOCK(); 1869 FILE_LOCK(fp); 1870 fp->f_msgcount--; 1871 unp_rights--; 1872 FILE_UNLOCK(fp); 1873 UNP_UNLOCK(); 1874 (void) closef(fp, (struct thread *)NULL); 1875 } 1876