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