1 /*- 2 * Copyright (c) 1982, 1986, 1989, 1991, 1993 3 * The Regents of the University of California. 4 * Copyright (c) 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/mbuf.h> 68 #include <sys/mount.h> 69 #include <sys/mutex.h> 70 #include <sys/namei.h> 71 #include <sys/proc.h> 72 #include <sys/protosw.h> 73 #include <sys/resourcevar.h> 74 #include <sys/socket.h> 75 #include <sys/socketvar.h> 76 #include <sys/signalvar.h> 77 #include <sys/stat.h> 78 #include <sys/sx.h> 79 #include <sys/sysctl.h> 80 #include <sys/systm.h> 81 #include <sys/taskqueue.h> 82 #include <sys/un.h> 83 #include <sys/unpcb.h> 84 #include <sys/vnode.h> 85 86 #include <security/mac/mac_framework.h> 87 88 #include <vm/uma.h> 89 90 static uma_zone_t unp_zone; 91 static unp_gen_t unp_gencnt; 92 static u_int unp_count; 93 94 static struct unp_head unp_shead, unp_dhead; 95 96 /* 97 * Unix communications domain. 98 * 99 * TODO: 100 * SEQPACKET, RDM 101 * rethink name space problems 102 * need a proper out-of-band 103 * lock pushdown 104 */ 105 static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL }; 106 static ino_t unp_ino; /* prototype for fake inode numbers */ 107 struct mbuf *unp_addsockcred(struct thread *, struct mbuf *); 108 109 /* 110 * Both send and receive buffers are allocated PIPSIZ bytes of buffering for 111 * stream sockets, although the total for sender and receiver is actually 112 * only PIPSIZ. 113 * 114 * Datagram sockets really use the sendspace as the maximum datagram size, 115 * and don't really want to reserve the sendspace. Their recvspace should be 116 * large enough for at least one max-size datagram plus address. 117 */ 118 #ifndef PIPSIZ 119 #define PIPSIZ 8192 120 #endif 121 static u_long unpst_sendspace = PIPSIZ; 122 static u_long unpst_recvspace = PIPSIZ; 123 static u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 124 static u_long unpdg_recvspace = 4*1024; 125 126 static int unp_rights; /* file descriptors in flight */ 127 128 SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW, 0, "Local domain"); 129 SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW, 0, "SOCK_STREAM"); 130 SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW, 0, "SOCK_DGRAM"); 131 132 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 133 &unpst_sendspace, 0, ""); 134 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 135 &unpst_recvspace, 0, ""); 136 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 137 &unpdg_sendspace, 0, ""); 138 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 139 &unpdg_recvspace, 0, ""); 140 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, ""); 141 142 /* 143 * Currently, UNIX domain sockets are protected by a single subsystem lock, 144 * which covers global data structures and variables, the contents of each 145 * per-socket unpcb structure, and the so_pcb field in sockets attached to 146 * the UNIX domain. This provides for a moderate degree of paralellism, as 147 * receive operations on UNIX domain sockets do not need to acquire the 148 * subsystem lock. Finer grained locking to permit send() without acquiring 149 * a global lock would be a logical next step. 150 * 151 * The UNIX domain socket lock preceds all socket layer locks, including the 152 * socket lock and socket buffer lock, permitting UNIX domain socket code to 153 * call into socket support routines without releasing its locks. 154 * 155 * Some caution is required in areas where the UNIX domain socket code enters 156 * VFS in order to create or find rendezvous points. This results in 157 * dropping of the UNIX domain socket subsystem lock, acquisition of the 158 * Giant lock, and potential sleeping. This increases the chances of races, 159 * and exposes weaknesses in the socket->protocol API by offering poor 160 * failure modes. 161 */ 162 static struct mtx unp_mtx; 163 #define UNP_LOCK_INIT() \ 164 mtx_init(&unp_mtx, "unp", NULL, MTX_DEF) 165 #define UNP_LOCK() mtx_lock(&unp_mtx) 166 #define UNP_UNLOCK() mtx_unlock(&unp_mtx) 167 #define UNP_LOCK_ASSERT() mtx_assert(&unp_mtx, MA_OWNED) 168 #define UNP_UNLOCK_ASSERT() mtx_assert(&unp_mtx, MA_NOTOWNED) 169 170 /* 171 * Garbage collection of cyclic file descriptor/socket references occurs 172 * asynchronously in a taskqueue context in order to avoid recursion and 173 * reentrance in the UNIX domain socket, file descriptor, and socket layer 174 * code. See unp_gc() for a full description. 175 */ 176 static struct task unp_gc_task; 177 178 static int unp_connect(struct socket *,struct sockaddr *, struct thread *); 179 static int unp_connect2(struct socket *so, struct socket *so2, int); 180 static void unp_disconnect(struct unpcb *); 181 static void unp_shutdown(struct unpcb *); 182 static void unp_drop(struct unpcb *, int); 183 static void unp_gc(__unused void *, int); 184 static void unp_scan(struct mbuf *, void (*)(struct file *)); 185 static void unp_mark(struct file *); 186 static void unp_discard(struct file *); 187 static void unp_freerights(struct file **, int); 188 static int unp_internalize(struct mbuf **, struct thread *); 189 static int unp_listen(struct socket *, struct unpcb *, int, 190 struct thread *); 191 192 /* 193 * Definitions of protocols supported in the LOCAL domain. 194 */ 195 static struct domain localdomain; 196 static struct protosw localsw[] = { 197 { 198 .pr_type = SOCK_STREAM, 199 .pr_domain = &localdomain, 200 .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS, 201 .pr_ctloutput = &uipc_ctloutput, 202 .pr_usrreqs = &uipc_usrreqs 203 }, 204 { 205 .pr_type = SOCK_DGRAM, 206 .pr_domain = &localdomain, 207 .pr_flags = PR_ATOMIC|PR_ADDR|PR_RIGHTS, 208 .pr_usrreqs = &uipc_usrreqs 209 }, 210 }; 211 212 static struct domain localdomain = { 213 .dom_family = AF_LOCAL, 214 .dom_name = "local", 215 .dom_init = unp_init, 216 .dom_externalize = unp_externalize, 217 .dom_dispose = unp_dispose, 218 .dom_protosw = localsw, 219 .dom_protoswNPROTOSW = &localsw[sizeof(localsw)/sizeof(localsw[0])] 220 }; 221 DOMAIN_SET(local); 222 223 static void 224 uipc_abort(struct socket *so) 225 { 226 struct unpcb *unp; 227 228 unp = sotounpcb(so); 229 KASSERT(unp != NULL, ("uipc_abort: unp == NULL")); 230 UNP_LOCK(); 231 unp_drop(unp, ECONNABORTED); 232 UNP_UNLOCK(); 233 } 234 235 static int 236 uipc_accept(struct socket *so, struct sockaddr **nam) 237 { 238 struct unpcb *unp; 239 const struct sockaddr *sa; 240 241 /* 242 * Pass back name of connected socket, if it was bound and we are 243 * still connected (our peer may have closed already!). 244 */ 245 unp = sotounpcb(so); 246 KASSERT(unp != NULL, ("uipc_accept: unp == NULL")); 247 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 248 UNP_LOCK(); 249 if (unp->unp_conn != NULL && unp->unp_conn->unp_addr != NULL) 250 sa = (struct sockaddr *) unp->unp_conn->unp_addr; 251 else 252 sa = &sun_noname; 253 bcopy(sa, *nam, sa->sa_len); 254 UNP_UNLOCK(); 255 return (0); 256 } 257 258 static int 259 uipc_attach(struct socket *so, int proto, struct thread *td) 260 { 261 struct unpcb *unp; 262 int error; 263 264 KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL")); 265 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 266 switch (so->so_type) { 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->unp_refcount = 1; 289 UNP_LOCK(); 290 unp->unp_gencnt = ++unp_gencnt; 291 unp_count++; 292 LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead : &unp_shead, 293 unp, unp_link); 294 UNP_UNLOCK(); 295 296 return (0); 297 } 298 299 static int 300 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 301 { 302 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 303 struct vattr vattr; 304 int error, namelen; 305 struct nameidata nd; 306 struct unpcb *unp; 307 struct vnode *vp; 308 struct mount *mp; 309 char *buf; 310 311 unp = sotounpcb(so); 312 KASSERT(unp != NULL, ("uipc_bind: unp == NULL")); 313 314 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 315 if (namelen <= 0) 316 return (EINVAL); 317 318 /* 319 * We don't allow simultaneous bind() calls on a single UNIX domain 320 * socket, so flag in-progress operations, and return an error if an 321 * operation is already in progress. 322 * 323 * Historically, we have not allowed a socket to be rebound, so this 324 * also returns an error. Not allowing re-binding certainly 325 * simplifies the implementation and avoids a great many possible 326 * failure modes. 327 */ 328 UNP_LOCK(); 329 if (unp->unp_vnode != NULL) { 330 UNP_UNLOCK(); 331 return (EINVAL); 332 } 333 if (unp->unp_flags & UNP_BINDING) { 334 UNP_UNLOCK(); 335 return (EALREADY); 336 } 337 unp->unp_flags |= UNP_BINDING; 338 UNP_UNLOCK(); 339 340 buf = malloc(namelen + 1, M_TEMP, M_WAITOK); 341 strlcpy(buf, soun->sun_path, namelen + 1); 342 343 mtx_lock(&Giant); 344 restart: 345 mtx_assert(&Giant, MA_OWNED); 346 NDINIT(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME, UIO_SYSSPACE, 347 buf, td); 348 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 349 error = namei(&nd); 350 if (error) 351 goto error; 352 vp = nd.ni_vp; 353 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { 354 NDFREE(&nd, NDF_ONLY_PNBUF); 355 if (nd.ni_dvp == vp) 356 vrele(nd.ni_dvp); 357 else 358 vput(nd.ni_dvp); 359 if (vp != NULL) { 360 vrele(vp); 361 error = EADDRINUSE; 362 goto error; 363 } 364 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH); 365 if (error) 366 goto error; 367 goto restart; 368 } 369 VATTR_NULL(&vattr); 370 vattr.va_type = VSOCK; 371 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask); 372 #ifdef MAC 373 error = mac_check_vnode_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, 374 &vattr); 375 #endif 376 if (error == 0) { 377 VOP_LEASE(nd.ni_dvp, td, td->td_ucred, LEASE_WRITE); 378 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 379 } 380 NDFREE(&nd, NDF_ONLY_PNBUF); 381 vput(nd.ni_dvp); 382 if (error) { 383 vn_finished_write(mp); 384 goto error; 385 } 386 vp = nd.ni_vp; 387 ASSERT_VOP_LOCKED(vp, "uipc_bind"); 388 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK); 389 UNP_LOCK(); 390 vp->v_socket = unp->unp_socket; 391 unp->unp_vnode = vp; 392 unp->unp_addr = soun; 393 unp->unp_flags &= ~UNP_BINDING; 394 UNP_UNLOCK(); 395 VOP_UNLOCK(vp, 0, td); 396 vn_finished_write(mp); 397 mtx_unlock(&Giant); 398 free(buf, M_TEMP); 399 return (0); 400 error: 401 UNP_LOCK(); 402 unp->unp_flags &= ~UNP_BINDING; 403 UNP_UNLOCK(); 404 mtx_unlock(&Giant); 405 free(buf, M_TEMP); 406 return (error); 407 } 408 409 static int 410 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 411 { 412 int error; 413 414 KASSERT(td == curthread, ("uipc_connect: td != curthread")); 415 UNP_LOCK(); 416 error = unp_connect(so, nam, td); 417 UNP_UNLOCK(); 418 return (error); 419 } 420 421 /* 422 * XXXRW: Should also unbind? 423 */ 424 static void 425 uipc_close(struct socket *so) 426 { 427 struct unpcb *unp; 428 429 unp = sotounpcb(so); 430 KASSERT(unp != NULL, ("uipc_close: unp == NULL")); 431 UNP_LOCK(); 432 unp_disconnect(unp); 433 UNP_UNLOCK(); 434 } 435 436 int 437 uipc_connect2(struct socket *so1, struct socket *so2) 438 { 439 struct unpcb *unp; 440 int error; 441 442 unp = sotounpcb(so1); 443 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL")); 444 UNP_LOCK(); 445 error = unp_connect2(so1, so2, PRU_CONNECT2); 446 UNP_UNLOCK(); 447 return (error); 448 } 449 450 /* control is EOPNOTSUPP */ 451 452 static void 453 uipc_detach(struct socket *so) 454 { 455 struct sockaddr_un *saved_unp_addr; 456 struct unpcb *unp; 457 struct vnode *vp; 458 int freeunp, local_unp_rights; 459 460 unp = sotounpcb(so); 461 KASSERT(unp != NULL, ("uipc_detach: unp == NULL")); 462 UNP_LOCK(); 463 LIST_REMOVE(unp, unp_link); 464 unp->unp_gencnt = ++unp_gencnt; 465 --unp_count; 466 if ((vp = unp->unp_vnode) != NULL) { 467 /* 468 * XXXRW: should v_socket be frobbed only while holding 469 * Giant? 470 */ 471 unp->unp_vnode->v_socket = NULL; 472 unp->unp_vnode = NULL; 473 } 474 if (unp->unp_conn != NULL) 475 unp_disconnect(unp); 476 while (!LIST_EMPTY(&unp->unp_refs)) { 477 struct unpcb *ref = LIST_FIRST(&unp->unp_refs); 478 unp_drop(ref, ECONNRESET); 479 } 480 unp->unp_socket->so_pcb = NULL; 481 local_unp_rights = unp_rights; 482 saved_unp_addr = unp->unp_addr; 483 unp->unp_addr = NULL; 484 unp->unp_refcount--; 485 freeunp = (unp->unp_refcount == 0); 486 UNP_UNLOCK(); 487 if (saved_unp_addr != NULL) 488 FREE(saved_unp_addr, M_SONAME); 489 if (freeunp) 490 uma_zfree(unp_zone, unp); 491 if (vp) { 492 int vfslocked; 493 494 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 495 vrele(vp); 496 VFS_UNLOCK_GIANT(vfslocked); 497 } 498 if (local_unp_rights) 499 taskqueue_enqueue(taskqueue_thread, &unp_gc_task); 500 } 501 502 static int 503 uipc_disconnect(struct socket *so) 504 { 505 struct unpcb *unp; 506 507 unp = sotounpcb(so); 508 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL")); 509 UNP_LOCK(); 510 unp_disconnect(unp); 511 UNP_UNLOCK(); 512 return (0); 513 } 514 515 static int 516 uipc_listen(struct socket *so, int backlog, struct thread *td) 517 { 518 struct unpcb *unp; 519 int error; 520 521 unp = sotounpcb(so); 522 KASSERT(unp != NULL, ("uipc_listen: unp == NULL")); 523 UNP_LOCK(); 524 if (unp->unp_vnode == NULL) { 525 UNP_UNLOCK(); 526 return (EINVAL); 527 } 528 error = unp_listen(so, unp, backlog, td); 529 UNP_UNLOCK(); 530 return (error); 531 } 532 533 static int 534 uipc_peeraddr(struct socket *so, struct sockaddr **nam) 535 { 536 struct unpcb *unp; 537 const struct sockaddr *sa; 538 539 unp = sotounpcb(so); 540 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL")); 541 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 542 UNP_LOCK(); 543 if (unp->unp_conn != NULL && unp->unp_conn->unp_addr!= NULL) 544 sa = (struct sockaddr *) unp->unp_conn->unp_addr; 545 else { 546 /* 547 * XXX: It seems that this test always fails even when 548 * connection is established. So, this else clause is 549 * added as workaround to return PF_LOCAL sockaddr. 550 */ 551 sa = &sun_noname; 552 } 553 bcopy(sa, *nam, sa->sa_len); 554 UNP_UNLOCK(); 555 return (0); 556 } 557 558 static int 559 uipc_rcvd(struct socket *so, int flags) 560 { 561 struct unpcb *unp; 562 struct socket *so2; 563 u_int mbcnt, sbcc; 564 u_long newhiwat; 565 566 unp = sotounpcb(so); 567 KASSERT(unp != NULL, ("uipc_rcvd: unp == NULL")); 568 switch (so->so_type) { 569 case SOCK_DGRAM: 570 panic("uipc_rcvd DGRAM?"); 571 /*NOTREACHED*/ 572 573 case SOCK_STREAM: 574 /* 575 * Adjust backpressure on sender and wakeup any waiting to 576 * write. 577 */ 578 SOCKBUF_LOCK(&so->so_rcv); 579 mbcnt = so->so_rcv.sb_mbcnt; 580 sbcc = so->so_rcv.sb_cc; 581 SOCKBUF_UNLOCK(&so->so_rcv); 582 UNP_LOCK(); 583 if (unp->unp_conn == NULL) { 584 UNP_UNLOCK(); 585 break; 586 } 587 so2 = unp->unp_conn->unp_socket; 588 SOCKBUF_LOCK(&so2->so_snd); 589 so2->so_snd.sb_mbmax += unp->unp_mbcnt - mbcnt; 590 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - sbcc; 591 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat, 592 newhiwat, RLIM_INFINITY); 593 sowwakeup_locked(so2); 594 unp->unp_mbcnt = mbcnt; 595 unp->unp_cc = sbcc; 596 UNP_UNLOCK(); 597 break; 598 599 default: 600 panic("uipc_rcvd unknown socktype"); 601 } 602 return (0); 603 } 604 605 /* pru_rcvoob is EOPNOTSUPP */ 606 607 static int 608 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 609 struct mbuf *control, struct thread *td) 610 { 611 struct unpcb *unp, *unp2; 612 struct socket *so2; 613 u_int mbcnt, sbcc; 614 u_long newhiwat; 615 int error = 0; 616 617 unp = sotounpcb(so); 618 KASSERT(unp != NULL, ("uipc_send: unp == NULL")); 619 if (flags & PRUS_OOB) { 620 error = EOPNOTSUPP; 621 goto release; 622 } 623 624 if (control != NULL && (error = unp_internalize(&control, td))) 625 goto release; 626 627 UNP_LOCK(); 628 switch (so->so_type) { 629 case SOCK_DGRAM: 630 { 631 const struct sockaddr *from; 632 633 if (nam != NULL) { 634 if (unp->unp_conn != NULL) { 635 error = EISCONN; 636 break; 637 } 638 error = unp_connect(so, nam, td); 639 if (error) 640 break; 641 } 642 /* 643 * Because connect() and send() are non-atomic in a sendto() 644 * with a target address, it's possible that the socket will 645 * have disconnected before the send() can run. In that case 646 * return the slightly counter-intuitive but otherwise 647 * correct error that the socket is not connected. 648 */ 649 unp2 = unp->unp_conn; 650 if (unp2 == NULL) { 651 error = ENOTCONN; 652 break; 653 } 654 so2 = unp2->unp_socket; 655 if (unp->unp_addr != NULL) 656 from = (struct sockaddr *)unp->unp_addr; 657 else 658 from = &sun_noname; 659 if (unp2->unp_flags & UNP_WANTCRED) 660 control = unp_addsockcred(td, control); 661 SOCKBUF_LOCK(&so2->so_rcv); 662 if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) { 663 sorwakeup_locked(so2); 664 m = NULL; 665 control = NULL; 666 } else { 667 SOCKBUF_UNLOCK(&so2->so_rcv); 668 error = ENOBUFS; 669 } 670 if (nam != NULL) 671 unp_disconnect(unp); 672 break; 673 } 674 675 case SOCK_STREAM: 676 /* 677 * Connect if not connected yet. 678 * 679 * Note: A better implementation would complain if not equal 680 * to the peer's address. 681 */ 682 if ((so->so_state & SS_ISCONNECTED) == 0) { 683 if (nam != NULL) { 684 error = unp_connect(so, nam, td); 685 if (error) 686 break; /* XXX */ 687 } else { 688 error = ENOTCONN; 689 break; 690 } 691 } 692 693 /* Lockless read. */ 694 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 695 error = EPIPE; 696 break; 697 } 698 /* 699 * Because connect() and send() are non-atomic in a sendto() 700 * with a target address, it's possible that the socket will 701 * have disconnected before the send() can run. In that case 702 * return the slightly counter-intuitive but otherwise 703 * correct error that the socket is not connected. 704 */ 705 unp2 = unp->unp_conn; 706 if (unp2 == NULL) { 707 error = ENOTCONN; 708 break; 709 } 710 so2 = unp2->unp_socket; 711 SOCKBUF_LOCK(&so2->so_rcv); 712 if (unp2->unp_flags & UNP_WANTCRED) { 713 /* 714 * Credentials are passed only once on 715 * SOCK_STREAM. 716 */ 717 unp2->unp_flags &= ~UNP_WANTCRED; 718 control = unp_addsockcred(td, control); 719 } 720 /* 721 * Send to paired receive port, and then reduce send buffer 722 * hiwater marks to maintain backpressure. Wake up readers. 723 */ 724 if (control != NULL) { 725 if (sbappendcontrol_locked(&so2->so_rcv, m, control)) 726 control = NULL; 727 } else { 728 sbappend_locked(&so2->so_rcv, m); 729 } 730 mbcnt = so2->so_rcv.sb_mbcnt - unp2->unp_mbcnt; 731 unp2->unp_mbcnt = so2->so_rcv.sb_mbcnt; 732 sbcc = so2->so_rcv.sb_cc; 733 sorwakeup_locked(so2); 734 735 SOCKBUF_LOCK(&so->so_snd); 736 newhiwat = so->so_snd.sb_hiwat - (sbcc - unp2->unp_cc); 737 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat, 738 newhiwat, RLIM_INFINITY); 739 so->so_snd.sb_mbmax -= mbcnt; 740 SOCKBUF_UNLOCK(&so->so_snd); 741 742 unp2->unp_cc = sbcc; 743 m = NULL; 744 break; 745 746 default: 747 panic("uipc_send unknown socktype"); 748 } 749 750 /* 751 * SEND_EOF is equivalent to a SEND followed by 752 * a SHUTDOWN. 753 */ 754 if (flags & PRUS_EOF) { 755 socantsendmore(so); 756 unp_shutdown(unp); 757 } 758 UNP_UNLOCK(); 759 760 if (control != NULL && error != 0) 761 unp_dispose(control); 762 763 release: 764 if (control != NULL) 765 m_freem(control); 766 if (m != NULL) 767 m_freem(m); 768 return (error); 769 } 770 771 static int 772 uipc_sense(struct socket *so, struct stat *sb) 773 { 774 struct unpcb *unp; 775 struct socket *so2; 776 777 unp = sotounpcb(so); 778 KASSERT(unp != NULL, ("uipc_sense: unp == NULL")); 779 UNP_LOCK(); 780 sb->st_blksize = so->so_snd.sb_hiwat; 781 if (so->so_type == SOCK_STREAM && unp->unp_conn != NULL) { 782 so2 = unp->unp_conn->unp_socket; 783 sb->st_blksize += so2->so_rcv.sb_cc; 784 } 785 sb->st_dev = NODEV; 786 if (unp->unp_ino == 0) 787 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino; 788 sb->st_ino = unp->unp_ino; 789 UNP_UNLOCK(); 790 return (0); 791 } 792 793 static int 794 uipc_shutdown(struct socket *so) 795 { 796 struct unpcb *unp; 797 798 unp = sotounpcb(so); 799 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL")); 800 UNP_LOCK(); 801 socantsendmore(so); 802 unp_shutdown(unp); 803 UNP_UNLOCK(); 804 return (0); 805 } 806 807 static int 808 uipc_sockaddr(struct socket *so, struct sockaddr **nam) 809 { 810 struct unpcb *unp; 811 const struct sockaddr *sa; 812 813 unp = sotounpcb(so); 814 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL")); 815 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 816 UNP_LOCK(); 817 if (unp->unp_addr != NULL) 818 sa = (struct sockaddr *) unp->unp_addr; 819 else 820 sa = &sun_noname; 821 bcopy(sa, *nam, sa->sa_len); 822 UNP_UNLOCK(); 823 return (0); 824 } 825 826 struct pr_usrreqs uipc_usrreqs = { 827 .pru_abort = uipc_abort, 828 .pru_accept = uipc_accept, 829 .pru_attach = uipc_attach, 830 .pru_bind = uipc_bind, 831 .pru_connect = uipc_connect, 832 .pru_connect2 = uipc_connect2, 833 .pru_detach = uipc_detach, 834 .pru_disconnect = uipc_disconnect, 835 .pru_listen = uipc_listen, 836 .pru_peeraddr = uipc_peeraddr, 837 .pru_rcvd = uipc_rcvd, 838 .pru_send = uipc_send, 839 .pru_sense = uipc_sense, 840 .pru_shutdown = uipc_shutdown, 841 .pru_sockaddr = uipc_sockaddr, 842 .pru_close = uipc_close, 843 }; 844 845 int 846 uipc_ctloutput(struct socket *so, struct sockopt *sopt) 847 { 848 struct unpcb *unp; 849 struct xucred xu; 850 int error, optval; 851 852 if (sopt->sopt_level != 0) 853 return (EINVAL); 854 855 unp = sotounpcb(so); 856 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL")); 857 error = 0; 858 switch (sopt->sopt_dir) { 859 case SOPT_GET: 860 switch (sopt->sopt_name) { 861 case LOCAL_PEERCRED: 862 UNP_LOCK(); 863 if (unp->unp_flags & UNP_HAVEPC) 864 xu = unp->unp_peercred; 865 else { 866 if (so->so_type == SOCK_STREAM) 867 error = ENOTCONN; 868 else 869 error = EINVAL; 870 } 871 UNP_UNLOCK(); 872 if (error == 0) 873 error = sooptcopyout(sopt, &xu, sizeof(xu)); 874 break; 875 case LOCAL_CREDS: 876 /* Unocked read. */ 877 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0; 878 error = sooptcopyout(sopt, &optval, sizeof(optval)); 879 break; 880 case LOCAL_CONNWAIT: 881 /* Unocked read. */ 882 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0; 883 error = sooptcopyout(sopt, &optval, sizeof(optval)); 884 break; 885 default: 886 error = EOPNOTSUPP; 887 break; 888 } 889 break; 890 case SOPT_SET: 891 switch (sopt->sopt_name) { 892 case LOCAL_CREDS: 893 case LOCAL_CONNWAIT: 894 error = sooptcopyin(sopt, &optval, sizeof(optval), 895 sizeof(optval)); 896 if (error) 897 break; 898 899 #define OPTSET(bit) \ 900 if (optval) \ 901 unp->unp_flags |= bit; \ 902 else \ 903 unp->unp_flags &= ~bit; 904 905 UNP_LOCK(); 906 switch (sopt->sopt_name) { 907 case LOCAL_CREDS: 908 OPTSET(UNP_WANTCRED); 909 break; 910 case LOCAL_CONNWAIT: 911 OPTSET(UNP_CONNWAIT); 912 break; 913 default: 914 break; 915 } 916 UNP_UNLOCK(); 917 break; 918 #undef OPTSET 919 default: 920 error = ENOPROTOOPT; 921 break; 922 } 923 break; 924 default: 925 error = EOPNOTSUPP; 926 break; 927 } 928 return (error); 929 } 930 931 static int 932 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 933 { 934 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 935 struct vnode *vp; 936 struct socket *so2, *so3; 937 struct unpcb *unp, *unp2, *unp3; 938 int error, len; 939 struct nameidata nd; 940 char buf[SOCK_MAXADDRLEN]; 941 struct sockaddr *sa; 942 943 UNP_LOCK_ASSERT(); 944 945 unp = sotounpcb(so); 946 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 947 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 948 if (len <= 0) 949 return (EINVAL); 950 strlcpy(buf, soun->sun_path, len + 1); 951 if (unp->unp_flags & UNP_CONNECTING) { 952 UNP_UNLOCK(); 953 return (EALREADY); 954 } 955 UNP_UNLOCK(); 956 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 957 mtx_lock(&Giant); 958 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, td); 959 error = namei(&nd); 960 if (error) 961 vp = NULL; 962 else 963 vp = nd.ni_vp; 964 ASSERT_VOP_LOCKED(vp, "unp_connect"); 965 NDFREE(&nd, NDF_ONLY_PNBUF); 966 if (error) 967 goto bad; 968 969 if (vp->v_type != VSOCK) { 970 error = ENOTSOCK; 971 goto bad; 972 } 973 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); 974 if (error) 975 goto bad; 976 mtx_unlock(&Giant); 977 UNP_LOCK(); 978 unp = sotounpcb(so); 979 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 980 so2 = vp->v_socket; 981 if (so2 == NULL) { 982 error = ECONNREFUSED; 983 goto bad2; 984 } 985 if (so->so_type != so2->so_type) { 986 error = EPROTOTYPE; 987 goto bad2; 988 } 989 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 990 if (so2->so_options & SO_ACCEPTCONN) { 991 /* 992 * NB: drop locks here so unp_attach is entered w/o 993 * locks; this avoids a recursive lock of the head 994 * and holding sleep locks across a (potentially) 995 * blocking malloc. 996 */ 997 UNP_UNLOCK(); 998 so3 = sonewconn(so2, 0); 999 UNP_LOCK(); 1000 } else 1001 so3 = NULL; 1002 if (so3 == NULL) { 1003 error = ECONNREFUSED; 1004 goto bad2; 1005 } 1006 unp = sotounpcb(so); 1007 unp2 = sotounpcb(so2); 1008 unp3 = sotounpcb(so3); 1009 if (unp2->unp_addr != NULL) { 1010 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); 1011 unp3->unp_addr = (struct sockaddr_un *) sa; 1012 sa = NULL; 1013 } 1014 /* 1015 * unp_peercred management: 1016 * 1017 * The connecter's (client's) credentials are copied from its 1018 * process structure at the time of connect() (which is now). 1019 */ 1020 cru2x(td->td_ucred, &unp3->unp_peercred); 1021 unp3->unp_flags |= UNP_HAVEPC; 1022 /* 1023 * The receiver's (server's) credentials are copied from the 1024 * unp_peercred member of socket on which the former called 1025 * listen(); unp_listen() cached that process's credentials 1026 * at that time so we can use them now. 1027 */ 1028 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED, 1029 ("unp_connect: listener without cached peercred")); 1030 memcpy(&unp->unp_peercred, &unp2->unp_peercred, 1031 sizeof(unp->unp_peercred)); 1032 unp->unp_flags |= UNP_HAVEPC; 1033 if (unp2->unp_flags & UNP_WANTCRED) 1034 unp3->unp_flags |= UNP_WANTCRED; 1035 #ifdef MAC 1036 SOCK_LOCK(so); 1037 mac_set_socket_peer_from_socket(so, so3); 1038 mac_set_socket_peer_from_socket(so3, so); 1039 SOCK_UNLOCK(so); 1040 #endif 1041 1042 so2 = so3; 1043 } 1044 error = unp_connect2(so, so2, PRU_CONNECT); 1045 bad2: 1046 UNP_UNLOCK(); 1047 mtx_lock(&Giant); 1048 bad: 1049 mtx_assert(&Giant, MA_OWNED); 1050 if (vp != NULL) 1051 vput(vp); 1052 mtx_unlock(&Giant); 1053 free(sa, M_SONAME); 1054 UNP_LOCK(); 1055 unp->unp_flags &= ~UNP_CONNECTING; 1056 return (error); 1057 } 1058 1059 static int 1060 unp_connect2(struct socket *so, struct socket *so2, int req) 1061 { 1062 struct unpcb *unp = sotounpcb(so); 1063 struct unpcb *unp2; 1064 1065 UNP_LOCK_ASSERT(); 1066 1067 if (so2->so_type != so->so_type) 1068 return (EPROTOTYPE); 1069 unp2 = sotounpcb(so2); 1070 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL")); 1071 unp->unp_conn = unp2; 1072 switch (so->so_type) { 1073 case SOCK_DGRAM: 1074 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 1075 soisconnected(so); 1076 break; 1077 1078 case SOCK_STREAM: 1079 unp2->unp_conn = unp; 1080 if (req == PRU_CONNECT && 1081 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 1082 soisconnecting(so); 1083 else 1084 soisconnected(so); 1085 soisconnected(so2); 1086 break; 1087 1088 default: 1089 panic("unp_connect2"); 1090 } 1091 return (0); 1092 } 1093 1094 static void 1095 unp_disconnect(struct unpcb *unp) 1096 { 1097 struct unpcb *unp2 = unp->unp_conn; 1098 struct socket *so; 1099 1100 UNP_LOCK_ASSERT(); 1101 1102 if (unp2 == NULL) 1103 return; 1104 unp->unp_conn = NULL; 1105 switch (unp->unp_socket->so_type) { 1106 case SOCK_DGRAM: 1107 LIST_REMOVE(unp, unp_reflink); 1108 so = unp->unp_socket; 1109 SOCK_LOCK(so); 1110 so->so_state &= ~SS_ISCONNECTED; 1111 SOCK_UNLOCK(so); 1112 break; 1113 1114 case SOCK_STREAM: 1115 soisdisconnected(unp->unp_socket); 1116 unp2->unp_conn = NULL; 1117 soisdisconnected(unp2->unp_socket); 1118 break; 1119 } 1120 } 1121 1122 /* 1123 * unp_pcblist() assumes that UNIX domain socket memory is never reclaimed by 1124 * the zone (UMA_ZONE_NOFREE), and as such potentially stale pointers are 1125 * safe to reference. It first scans the list of struct unpcb's to generate 1126 * a pointer list, then it rescans its list one entry at a time to 1127 * externalize and copyout. It checks the generation number to see if a 1128 * struct unpcb has been reused, and will skip it if so. 1129 */ 1130 static int 1131 unp_pcblist(SYSCTL_HANDLER_ARGS) 1132 { 1133 int error, i, n; 1134 int freeunp; 1135 struct unpcb *unp, **unp_list; 1136 unp_gen_t gencnt; 1137 struct xunpgen *xug; 1138 struct unp_head *head; 1139 struct xunpcb *xu; 1140 1141 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead); 1142 1143 /* 1144 * The process of preparing the PCB list is too time-consuming and 1145 * resource-intensive to repeat twice on every request. 1146 */ 1147 if (req->oldptr == NULL) { 1148 n = unp_count; 1149 req->oldidx = 2 * (sizeof *xug) 1150 + (n + n/8) * sizeof(struct xunpcb); 1151 return (0); 1152 } 1153 1154 if (req->newptr != NULL) 1155 return (EPERM); 1156 1157 /* 1158 * OK, now we're committed to doing something. 1159 */ 1160 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK); 1161 UNP_LOCK(); 1162 gencnt = unp_gencnt; 1163 n = unp_count; 1164 UNP_UNLOCK(); 1165 1166 xug->xug_len = sizeof *xug; 1167 xug->xug_count = n; 1168 xug->xug_gen = gencnt; 1169 xug->xug_sogen = so_gencnt; 1170 error = SYSCTL_OUT(req, xug, sizeof *xug); 1171 if (error) { 1172 free(xug, M_TEMP); 1173 return (error); 1174 } 1175 1176 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 1177 1178 UNP_LOCK(); 1179 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 1180 unp = LIST_NEXT(unp, unp_link)) { 1181 if (unp->unp_gencnt <= gencnt) { 1182 if (cr_cansee(req->td->td_ucred, 1183 unp->unp_socket->so_cred)) 1184 continue; 1185 unp_list[i++] = unp; 1186 unp->unp_refcount++; 1187 } 1188 } 1189 UNP_UNLOCK(); 1190 n = i; /* In case we lost some during malloc. */ 1191 1192 error = 0; 1193 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO); 1194 for (i = 0; i < n; i++) { 1195 unp = unp_list[i]; 1196 UNP_LOCK(); 1197 unp->unp_refcount--; 1198 if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) { 1199 xu->xu_len = sizeof *xu; 1200 xu->xu_unpp = unp; 1201 /* 1202 * XXX - need more locking here to protect against 1203 * connect/disconnect races for SMP. 1204 */ 1205 if (unp->unp_addr != NULL) 1206 bcopy(unp->unp_addr, &xu->xu_addr, 1207 unp->unp_addr->sun_len); 1208 if (unp->unp_conn != NULL && 1209 unp->unp_conn->unp_addr != NULL) 1210 bcopy(unp->unp_conn->unp_addr, 1211 &xu->xu_caddr, 1212 unp->unp_conn->unp_addr->sun_len); 1213 bcopy(unp, &xu->xu_unp, sizeof *unp); 1214 sotoxsocket(unp->unp_socket, &xu->xu_socket); 1215 UNP_UNLOCK(); 1216 error = SYSCTL_OUT(req, xu, sizeof *xu); 1217 } else { 1218 freeunp = (unp->unp_refcount == 0); 1219 UNP_UNLOCK(); 1220 if (freeunp) 1221 uma_zfree(unp_zone, unp); 1222 } 1223 } 1224 free(xu, M_TEMP); 1225 if (!error) { 1226 /* 1227 * Give the user an updated idea of our state. If the 1228 * generation differs from what we told her before, she knows 1229 * that something happened while we were processing this 1230 * request, and it might be necessary to retry. 1231 */ 1232 xug->xug_gen = unp_gencnt; 1233 xug->xug_sogen = so_gencnt; 1234 xug->xug_count = unp_count; 1235 error = SYSCTL_OUT(req, xug, sizeof *xug); 1236 } 1237 free(unp_list, M_TEMP); 1238 free(xug, M_TEMP); 1239 return (error); 1240 } 1241 1242 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD, 1243 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 1244 "List of active local datagram sockets"); 1245 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD, 1246 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 1247 "List of active local stream sockets"); 1248 1249 static void 1250 unp_shutdown(struct unpcb *unp) 1251 { 1252 struct socket *so; 1253 1254 UNP_LOCK_ASSERT(); 1255 1256 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn && 1257 (so = unp->unp_conn->unp_socket)) 1258 socantrcvmore(so); 1259 } 1260 1261 static void 1262 unp_drop(struct unpcb *unp, int errno) 1263 { 1264 struct socket *so = unp->unp_socket; 1265 1266 UNP_LOCK_ASSERT(); 1267 1268 so->so_error = errno; 1269 unp_disconnect(unp); 1270 } 1271 1272 static void 1273 unp_freerights(struct file **rp, int fdcount) 1274 { 1275 int i; 1276 struct file *fp; 1277 1278 for (i = 0; i < fdcount; i++) { 1279 fp = *rp; 1280 /* 1281 * Zero the pointer before calling unp_discard since it may 1282 * end up in unp_gc().. 1283 * 1284 * XXXRW: This is less true than it used to be. 1285 */ 1286 *rp++ = 0; 1287 unp_discard(fp); 1288 } 1289 } 1290 1291 int 1292 unp_externalize(struct mbuf *control, struct mbuf **controlp) 1293 { 1294 struct thread *td = curthread; /* XXX */ 1295 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1296 int i; 1297 int *fdp; 1298 struct file **rp; 1299 struct file *fp; 1300 void *data; 1301 socklen_t clen = control->m_len, datalen; 1302 int error, newfds; 1303 int f; 1304 u_int newlen; 1305 1306 UNP_UNLOCK_ASSERT(); 1307 1308 error = 0; 1309 if (controlp != NULL) /* controlp == NULL => free control messages */ 1310 *controlp = NULL; 1311 1312 while (cm != NULL) { 1313 if (sizeof(*cm) > clen || cm->cmsg_len > clen) { 1314 error = EINVAL; 1315 break; 1316 } 1317 1318 data = CMSG_DATA(cm); 1319 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1320 1321 if (cm->cmsg_level == SOL_SOCKET 1322 && cm->cmsg_type == SCM_RIGHTS) { 1323 newfds = datalen / sizeof(struct file *); 1324 rp = data; 1325 1326 /* If we're not outputting the descriptors free them. */ 1327 if (error || controlp == NULL) { 1328 unp_freerights(rp, newfds); 1329 goto next; 1330 } 1331 FILEDESC_LOCK(td->td_proc->p_fd); 1332 /* if the new FD's will not fit free them. */ 1333 if (!fdavail(td, newfds)) { 1334 FILEDESC_UNLOCK(td->td_proc->p_fd); 1335 error = EMSGSIZE; 1336 unp_freerights(rp, newfds); 1337 goto next; 1338 } 1339 /* 1340 * Now change each pointer to an fd in the global 1341 * table to an integer that is the index to the local 1342 * fd table entry that we set up to point to the 1343 * global one we are transferring. 1344 */ 1345 newlen = newfds * sizeof(int); 1346 *controlp = sbcreatecontrol(NULL, newlen, 1347 SCM_RIGHTS, SOL_SOCKET); 1348 if (*controlp == NULL) { 1349 FILEDESC_UNLOCK(td->td_proc->p_fd); 1350 error = E2BIG; 1351 unp_freerights(rp, newfds); 1352 goto next; 1353 } 1354 1355 fdp = (int *) 1356 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1357 for (i = 0; i < newfds; i++) { 1358 if (fdalloc(td, 0, &f)) 1359 panic("unp_externalize fdalloc failed"); 1360 fp = *rp++; 1361 td->td_proc->p_fd->fd_ofiles[f] = fp; 1362 FILE_LOCK(fp); 1363 fp->f_msgcount--; 1364 FILE_UNLOCK(fp); 1365 unp_rights--; 1366 *fdp++ = f; 1367 } 1368 FILEDESC_UNLOCK(td->td_proc->p_fd); 1369 } else { 1370 /* We can just copy anything else across. */ 1371 if (error || controlp == NULL) 1372 goto next; 1373 *controlp = sbcreatecontrol(NULL, datalen, 1374 cm->cmsg_type, cm->cmsg_level); 1375 if (*controlp == NULL) { 1376 error = ENOBUFS; 1377 goto next; 1378 } 1379 bcopy(data, 1380 CMSG_DATA(mtod(*controlp, struct cmsghdr *)), 1381 datalen); 1382 } 1383 1384 controlp = &(*controlp)->m_next; 1385 1386 next: 1387 if (CMSG_SPACE(datalen) < clen) { 1388 clen -= CMSG_SPACE(datalen); 1389 cm = (struct cmsghdr *) 1390 ((caddr_t)cm + CMSG_SPACE(datalen)); 1391 } else { 1392 clen = 0; 1393 cm = NULL; 1394 } 1395 } 1396 1397 m_freem(control); 1398 1399 return (error); 1400 } 1401 1402 static void 1403 unp_zone_change(void *tag) 1404 { 1405 1406 uma_zone_set_max(unp_zone, maxsockets); 1407 } 1408 1409 void 1410 unp_init(void) 1411 { 1412 1413 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL, 1414 NULL, NULL, UMA_ALIGN_PTR, 0); 1415 if (unp_zone == NULL) 1416 panic("unp_init"); 1417 uma_zone_set_max(unp_zone, maxsockets); 1418 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change, 1419 NULL, EVENTHANDLER_PRI_ANY); 1420 LIST_INIT(&unp_dhead); 1421 LIST_INIT(&unp_shead); 1422 TASK_INIT(&unp_gc_task, 0, unp_gc, NULL); 1423 UNP_LOCK_INIT(); 1424 } 1425 1426 static int 1427 unp_internalize(struct mbuf **controlp, struct thread *td) 1428 { 1429 struct mbuf *control = *controlp; 1430 struct proc *p = td->td_proc; 1431 struct filedesc *fdescp = p->p_fd; 1432 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1433 struct cmsgcred *cmcred; 1434 struct file **rp; 1435 struct file *fp; 1436 struct timeval *tv; 1437 int i, fd, *fdp; 1438 void *data; 1439 socklen_t clen = control->m_len, datalen; 1440 int error, oldfds; 1441 u_int newlen; 1442 1443 UNP_UNLOCK_ASSERT(); 1444 1445 error = 0; 1446 *controlp = NULL; 1447 1448 while (cm != NULL) { 1449 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET 1450 || cm->cmsg_len > clen) { 1451 error = EINVAL; 1452 goto out; 1453 } 1454 1455 data = CMSG_DATA(cm); 1456 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1457 1458 switch (cm->cmsg_type) { 1459 /* 1460 * Fill in credential information. 1461 */ 1462 case SCM_CREDS: 1463 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred), 1464 SCM_CREDS, SOL_SOCKET); 1465 if (*controlp == NULL) { 1466 error = ENOBUFS; 1467 goto out; 1468 } 1469 1470 cmcred = (struct cmsgcred *) 1471 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1472 cmcred->cmcred_pid = p->p_pid; 1473 cmcred->cmcred_uid = td->td_ucred->cr_ruid; 1474 cmcred->cmcred_gid = td->td_ucred->cr_rgid; 1475 cmcred->cmcred_euid = td->td_ucred->cr_uid; 1476 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups, 1477 CMGROUP_MAX); 1478 for (i = 0; i < cmcred->cmcred_ngroups; i++) 1479 cmcred->cmcred_groups[i] = 1480 td->td_ucred->cr_groups[i]; 1481 break; 1482 1483 case SCM_RIGHTS: 1484 oldfds = datalen / sizeof (int); 1485 /* 1486 * Check that all the FDs passed in refer to legal 1487 * files. If not, reject the entire operation. 1488 */ 1489 fdp = data; 1490 FILEDESC_LOCK(fdescp); 1491 for (i = 0; i < oldfds; i++) { 1492 fd = *fdp++; 1493 if ((unsigned)fd >= fdescp->fd_nfiles || 1494 fdescp->fd_ofiles[fd] == NULL) { 1495 FILEDESC_UNLOCK(fdescp); 1496 error = EBADF; 1497 goto out; 1498 } 1499 fp = fdescp->fd_ofiles[fd]; 1500 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { 1501 FILEDESC_UNLOCK(fdescp); 1502 error = EOPNOTSUPP; 1503 goto out; 1504 } 1505 1506 } 1507 /* 1508 * Now replace the integer FDs with pointers to the 1509 * associated global file table entry.. 1510 */ 1511 newlen = oldfds * sizeof(struct file *); 1512 *controlp = sbcreatecontrol(NULL, newlen, 1513 SCM_RIGHTS, SOL_SOCKET); 1514 if (*controlp == NULL) { 1515 FILEDESC_UNLOCK(fdescp); 1516 error = E2BIG; 1517 goto out; 1518 } 1519 1520 fdp = data; 1521 rp = (struct file **) 1522 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1523 for (i = 0; i < oldfds; i++) { 1524 fp = fdescp->fd_ofiles[*fdp++]; 1525 *rp++ = fp; 1526 FILE_LOCK(fp); 1527 fp->f_count++; 1528 fp->f_msgcount++; 1529 FILE_UNLOCK(fp); 1530 unp_rights++; 1531 } 1532 FILEDESC_UNLOCK(fdescp); 1533 break; 1534 1535 case SCM_TIMESTAMP: 1536 *controlp = sbcreatecontrol(NULL, sizeof(*tv), 1537 SCM_TIMESTAMP, SOL_SOCKET); 1538 if (*controlp == NULL) { 1539 error = ENOBUFS; 1540 goto out; 1541 } 1542 tv = (struct timeval *) 1543 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1544 microtime(tv); 1545 break; 1546 1547 default: 1548 error = EINVAL; 1549 goto out; 1550 } 1551 1552 controlp = &(*controlp)->m_next; 1553 1554 if (CMSG_SPACE(datalen) < clen) { 1555 clen -= CMSG_SPACE(datalen); 1556 cm = (struct cmsghdr *) 1557 ((caddr_t)cm + CMSG_SPACE(datalen)); 1558 } else { 1559 clen = 0; 1560 cm = NULL; 1561 } 1562 } 1563 1564 out: 1565 m_freem(control); 1566 1567 return (error); 1568 } 1569 1570 struct mbuf * 1571 unp_addsockcred(struct thread *td, struct mbuf *control) 1572 { 1573 struct mbuf *m, *n, *n_prev; 1574 struct sockcred *sc; 1575 const struct cmsghdr *cm; 1576 int ngroups; 1577 int i; 1578 1579 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); 1580 1581 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET); 1582 if (m == NULL) 1583 return (control); 1584 1585 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *)); 1586 sc->sc_uid = td->td_ucred->cr_ruid; 1587 sc->sc_euid = td->td_ucred->cr_uid; 1588 sc->sc_gid = td->td_ucred->cr_rgid; 1589 sc->sc_egid = td->td_ucred->cr_gid; 1590 sc->sc_ngroups = ngroups; 1591 for (i = 0; i < sc->sc_ngroups; i++) 1592 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 1593 1594 /* 1595 * Unlink SCM_CREDS control messages (struct cmsgcred), since just 1596 * created SCM_CREDS control message (struct sockcred) has another 1597 * format. 1598 */ 1599 if (control != NULL) 1600 for (n = control, n_prev = NULL; n != NULL;) { 1601 cm = mtod(n, struct cmsghdr *); 1602 if (cm->cmsg_level == SOL_SOCKET && 1603 cm->cmsg_type == SCM_CREDS) { 1604 if (n_prev == NULL) 1605 control = n->m_next; 1606 else 1607 n_prev->m_next = n->m_next; 1608 n = m_free(n); 1609 } else { 1610 n_prev = n; 1611 n = n->m_next; 1612 } 1613 } 1614 1615 /* Prepend it to the head. */ 1616 m->m_next = control; 1617 1618 return (m); 1619 } 1620 1621 /* 1622 * unp_defer indicates whether additional work has been defered for a future 1623 * pass through unp_gc(). It is thread local and does not require explicit 1624 * synchronization. 1625 */ 1626 static int unp_defer; 1627 1628 static int unp_taskcount; 1629 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, ""); 1630 1631 static int unp_recycled; 1632 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, ""); 1633 1634 static void 1635 unp_gc(__unused void *arg, int pending) 1636 { 1637 struct file *fp, *nextfp; 1638 struct socket *so; 1639 struct file **extra_ref, **fpp; 1640 int nunref, i; 1641 int nfiles_snap; 1642 int nfiles_slack = 20; 1643 1644 unp_taskcount++; 1645 unp_defer = 0; 1646 /* 1647 * Before going through all this, set all FDs to be NOT deferred and 1648 * NOT externally accessible. 1649 */ 1650 sx_slock(&filelist_lock); 1651 LIST_FOREACH(fp, &filehead, f_list) 1652 fp->f_gcflag &= ~(FMARK|FDEFER); 1653 do { 1654 KASSERT(unp_defer >= 0, ("unp_gc: unp_defer %d", unp_defer)); 1655 LIST_FOREACH(fp, &filehead, f_list) { 1656 FILE_LOCK(fp); 1657 /* 1658 * If the file is not open, skip it -- could be a 1659 * file in the process of being opened, or in the 1660 * process of being closed. If the file is 1661 * "closing", it may have been marked for deferred 1662 * consideration. Clear the flag now if so. 1663 */ 1664 if (fp->f_count == 0) { 1665 if (fp->f_gcflag & FDEFER) 1666 unp_defer--; 1667 fp->f_gcflag &= ~(FMARK|FDEFER); 1668 FILE_UNLOCK(fp); 1669 continue; 1670 } 1671 /* 1672 * If we already marked it as 'defer' in a 1673 * previous pass, then try to process it this 1674 * time and un-mark it. 1675 */ 1676 if (fp->f_gcflag & FDEFER) { 1677 fp->f_gcflag &= ~FDEFER; 1678 unp_defer--; 1679 } else { 1680 /* 1681 * if it's not deferred, then check if it's 1682 * already marked.. if so skip it 1683 */ 1684 if (fp->f_gcflag & FMARK) { 1685 FILE_UNLOCK(fp); 1686 continue; 1687 } 1688 /* 1689 * If all references are from messages in 1690 * transit, then skip it. it's not externally 1691 * accessible. 1692 */ 1693 if (fp->f_count == fp->f_msgcount) { 1694 FILE_UNLOCK(fp); 1695 continue; 1696 } 1697 /* 1698 * If it got this far then it must be 1699 * externally accessible. 1700 */ 1701 fp->f_gcflag |= FMARK; 1702 } 1703 /* 1704 * Either it was deferred, or it is externally 1705 * accessible and not already marked so. Now check 1706 * if it is possibly one of OUR sockets. 1707 */ 1708 if (fp->f_type != DTYPE_SOCKET || 1709 (so = fp->f_data) == NULL) { 1710 FILE_UNLOCK(fp); 1711 continue; 1712 } 1713 if (so->so_proto->pr_domain != &localdomain || 1714 (so->so_proto->pr_flags & PR_RIGHTS) == 0) { 1715 FILE_UNLOCK(fp); 1716 continue; 1717 } 1718 1719 /* 1720 * Tell any other threads that do a subsequent 1721 * fdrop() that we are scanning the message 1722 * buffers. 1723 */ 1724 fp->f_gcflag |= FWAIT; 1725 FILE_UNLOCK(fp); 1726 1727 /* 1728 * So, Ok, it's one of our sockets and it IS 1729 * externally accessible (or was deferred). Now we 1730 * look to see if we hold any file descriptors in its 1731 * message buffers. Follow those links and mark them 1732 * as accessible too. 1733 */ 1734 SOCKBUF_LOCK(&so->so_rcv); 1735 unp_scan(so->so_rcv.sb_mb, unp_mark); 1736 SOCKBUF_UNLOCK(&so->so_rcv); 1737 1738 /* 1739 * Wake up any threads waiting in fdrop(). 1740 */ 1741 FILE_LOCK(fp); 1742 fp->f_gcflag &= ~FWAIT; 1743 wakeup(&fp->f_gcflag); 1744 FILE_UNLOCK(fp); 1745 } 1746 } while (unp_defer); 1747 sx_sunlock(&filelist_lock); 1748 /* 1749 * XXXRW: The following comments need updating for a post-SMPng and 1750 * deferred unp_gc() world, but are still generally accurate. 1751 * 1752 * We grab an extra reference to each of the file table entries that 1753 * are not otherwise accessible and then free the rights that are 1754 * stored in messages on them. 1755 * 1756 * The bug in the orginal code is a little tricky, so I'll describe 1757 * what's wrong with it here. 1758 * 1759 * It is incorrect to simply unp_discard each entry for f_msgcount 1760 * times -- consider the case of sockets A and B that contain 1761 * references to each other. On a last close of some other socket, 1762 * we trigger a gc since the number of outstanding rights (unp_rights) 1763 * is non-zero. If during the sweep phase the gc code unp_discards, 1764 * we end up doing a (full) closef on the descriptor. A closef on A 1765 * results in the following chain. Closef calls soo_close, which 1766 * calls soclose. Soclose calls first (through the switch 1767 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply 1768 * returns because the previous instance had set unp_gcing, and we 1769 * return all the way back to soclose, which marks the socket with 1770 * SS_NOFDREF, and then calls sofree. Sofree calls sorflush to free 1771 * up the rights that are queued in messages on the socket A, i.e., 1772 * the reference on B. The sorflush calls via the dom_dispose switch 1773 * unp_dispose, which unp_scans with unp_discard. This second 1774 * instance of unp_discard just calls closef on B. 1775 * 1776 * Well, a similar chain occurs on B, resulting in a sorflush on B, 1777 * which results in another closef on A. Unfortunately, A is already 1778 * being closed, and the descriptor has already been marked with 1779 * SS_NOFDREF, and soclose panics at this point. 1780 * 1781 * Here, we first take an extra reference to each inaccessible 1782 * descriptor. Then, we call sorflush ourself, since we know it is a 1783 * Unix domain socket anyhow. After we destroy all the rights 1784 * carried in messages, we do a last closef to get rid of our extra 1785 * reference. This is the last close, and the unp_detach etc will 1786 * shut down the socket. 1787 * 1788 * 91/09/19, bsy@cs.cmu.edu 1789 */ 1790 again: 1791 nfiles_snap = openfiles + nfiles_slack; /* some slack */ 1792 extra_ref = malloc(nfiles_snap * sizeof(struct file *), M_TEMP, 1793 M_WAITOK); 1794 sx_slock(&filelist_lock); 1795 if (nfiles_snap < openfiles) { 1796 sx_sunlock(&filelist_lock); 1797 free(extra_ref, M_TEMP); 1798 nfiles_slack += 20; 1799 goto again; 1800 } 1801 for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref; 1802 fp != NULL; fp = nextfp) { 1803 nextfp = LIST_NEXT(fp, f_list); 1804 FILE_LOCK(fp); 1805 /* 1806 * If it's not open, skip it 1807 */ 1808 if (fp->f_count == 0) { 1809 FILE_UNLOCK(fp); 1810 continue; 1811 } 1812 /* 1813 * If all refs are from msgs, and it's not marked accessible 1814 * then it must be referenced from some unreachable cycle of 1815 * (shut-down) FDs, so include it in our list of FDs to 1816 * remove. 1817 */ 1818 if (fp->f_count == fp->f_msgcount && !(fp->f_gcflag & FMARK)) { 1819 *fpp++ = fp; 1820 nunref++; 1821 fp->f_count++; 1822 } 1823 FILE_UNLOCK(fp); 1824 } 1825 sx_sunlock(&filelist_lock); 1826 /* 1827 * For each FD on our hit list, do the following two things: 1828 */ 1829 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1830 struct file *tfp = *fpp; 1831 FILE_LOCK(tfp); 1832 if (tfp->f_type == DTYPE_SOCKET && 1833 tfp->f_data != NULL) { 1834 FILE_UNLOCK(tfp); 1835 sorflush(tfp->f_data); 1836 } else { 1837 FILE_UNLOCK(tfp); 1838 } 1839 } 1840 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1841 closef(*fpp, (struct thread *) NULL); 1842 unp_recycled++; 1843 } 1844 free(extra_ref, M_TEMP); 1845 } 1846 1847 void 1848 unp_dispose(struct mbuf *m) 1849 { 1850 1851 if (m) 1852 unp_scan(m, unp_discard); 1853 } 1854 1855 static int 1856 unp_listen(struct socket *so, struct unpcb *unp, int backlog, 1857 struct thread *td) 1858 { 1859 int error; 1860 1861 UNP_LOCK_ASSERT(); 1862 1863 SOCK_LOCK(so); 1864 error = solisten_proto_check(so); 1865 if (error == 0) { 1866 cru2x(td->td_ucred, &unp->unp_peercred); 1867 unp->unp_flags |= UNP_HAVEPCCACHED; 1868 solisten_proto(so, backlog); 1869 } 1870 SOCK_UNLOCK(so); 1871 return (error); 1872 } 1873 1874 static void 1875 unp_scan(struct mbuf *m0, void (*op)(struct file *)) 1876 { 1877 struct mbuf *m; 1878 struct file **rp; 1879 struct cmsghdr *cm; 1880 void *data; 1881 int i; 1882 socklen_t clen, datalen; 1883 int qfds; 1884 1885 while (m0 != NULL) { 1886 for (m = m0; m; m = m->m_next) { 1887 if (m->m_type != MT_CONTROL) 1888 continue; 1889 1890 cm = mtod(m, struct cmsghdr *); 1891 clen = m->m_len; 1892 1893 while (cm != NULL) { 1894 if (sizeof(*cm) > clen || cm->cmsg_len > clen) 1895 break; 1896 1897 data = CMSG_DATA(cm); 1898 datalen = (caddr_t)cm + cm->cmsg_len 1899 - (caddr_t)data; 1900 1901 if (cm->cmsg_level == SOL_SOCKET && 1902 cm->cmsg_type == SCM_RIGHTS) { 1903 qfds = datalen / sizeof (struct file *); 1904 rp = data; 1905 for (i = 0; i < qfds; i++) 1906 (*op)(*rp++); 1907 } 1908 1909 if (CMSG_SPACE(datalen) < clen) { 1910 clen -= CMSG_SPACE(datalen); 1911 cm = (struct cmsghdr *) 1912 ((caddr_t)cm + CMSG_SPACE(datalen)); 1913 } else { 1914 clen = 0; 1915 cm = NULL; 1916 } 1917 } 1918 } 1919 m0 = m0->m_act; 1920 } 1921 } 1922 1923 static void 1924 unp_mark(struct file *fp) 1925 { 1926 if (fp->f_gcflag & FMARK) 1927 return; 1928 unp_defer++; 1929 fp->f_gcflag |= (FMARK|FDEFER); 1930 } 1931 1932 static void 1933 unp_discard(struct file *fp) 1934 { 1935 UNP_LOCK(); 1936 FILE_LOCK(fp); 1937 fp->f_msgcount--; 1938 unp_rights--; 1939 FILE_UNLOCK(fp); 1940 UNP_UNLOCK(); 1941 (void) closef(fp, (struct thread *)NULL); 1942 } 1943