uipc_usrreq.c revision 1.190
1 /* $NetBSD: uipc_usrreq.c,v 1.190 2019/02/04 10:11:34 mrg Exp $ */ 2 3 /*- 4 * Copyright (c) 1998, 2000, 2004, 2008, 2009 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility, 9 * NASA Ames Research Center, and by Andrew Doran. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 30 * POSSIBILITY OF SUCH DAMAGE. 31 */ 32 33 /* 34 * Copyright (c) 1982, 1986, 1989, 1991, 1993 35 * The Regents of the University of California. All rights reserved. 36 * 37 * Redistribution and use in source and binary forms, with or without 38 * modification, are permitted provided that the following conditions 39 * are met: 40 * 1. Redistributions of source code must retain the above copyright 41 * notice, this list of conditions and the following disclaimer. 42 * 2. Redistributions in binary form must reproduce the above copyright 43 * notice, this list of conditions and the following disclaimer in the 44 * documentation and/or other materials provided with the distribution. 45 * 3. Neither the name of the University nor the names of its contributors 46 * may be used to endorse or promote products derived from this software 47 * without specific prior written permission. 48 * 49 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 50 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 51 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 52 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 53 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 54 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 55 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 56 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 57 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 58 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 59 * SUCH DAMAGE. 60 * 61 * @(#)uipc_usrreq.c 8.9 (Berkeley) 5/14/95 62 */ 63 64 /* 65 * Copyright (c) 1997 Christopher G. Demetriou. All rights reserved. 66 * 67 * Redistribution and use in source and binary forms, with or without 68 * modification, are permitted provided that the following conditions 69 * are met: 70 * 1. Redistributions of source code must retain the above copyright 71 * notice, this list of conditions and the following disclaimer. 72 * 2. Redistributions in binary form must reproduce the above copyright 73 * notice, this list of conditions and the following disclaimer in the 74 * documentation and/or other materials provided with the distribution. 75 * 3. All advertising materials mentioning features or use of this software 76 * must display the following acknowledgement: 77 * This product includes software developed by the University of 78 * California, Berkeley and its contributors. 79 * 4. Neither the name of the University nor the names of its contributors 80 * may be used to endorse or promote products derived from this software 81 * without specific prior written permission. 82 * 83 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 84 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 85 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 86 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 87 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 88 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 89 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 90 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 91 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 92 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 93 * SUCH DAMAGE. 94 * 95 * @(#)uipc_usrreq.c 8.9 (Berkeley) 5/14/95 96 */ 97 98 #include <sys/cdefs.h> 99 __KERNEL_RCSID(0, "$NetBSD: uipc_usrreq.c,v 1.190 2019/02/04 10:11:34 mrg Exp $"); 100 101 #ifdef _KERNEL_OPT 102 #include "opt_compat_netbsd.h" 103 #endif 104 105 #include <sys/param.h> 106 #include <sys/systm.h> 107 #include <sys/proc.h> 108 #include <sys/filedesc.h> 109 #include <sys/domain.h> 110 #include <sys/protosw.h> 111 #include <sys/socket.h> 112 #include <sys/socketvar.h> 113 #include <sys/unpcb.h> 114 #include <sys/un.h> 115 #include <sys/namei.h> 116 #include <sys/vnode.h> 117 #include <sys/file.h> 118 #include <sys/stat.h> 119 #include <sys/mbuf.h> 120 #include <sys/kauth.h> 121 #include <sys/kmem.h> 122 #include <sys/atomic.h> 123 #include <sys/uidinfo.h> 124 #include <sys/kernel.h> 125 #include <sys/kthread.h> 126 #include <sys/compat_stub.h> 127 128 #include <compat/sys/socket.h> 129 130 /* 131 * Unix communications domain. 132 * 133 * TODO: 134 * RDM 135 * rethink name space problems 136 * need a proper out-of-band 137 * 138 * Notes on locking: 139 * 140 * The generic rules noted in uipc_socket2.c apply. In addition: 141 * 142 * o We have a global lock, uipc_lock. 143 * 144 * o All datagram sockets are locked by uipc_lock. 145 * 146 * o For stream socketpairs, the two endpoints are created sharing the same 147 * independent lock. Sockets presented to PRU_CONNECT2 must already have 148 * matching locks. 149 * 150 * o Stream sockets created via socket() start life with their own 151 * independent lock. 152 * 153 * o Stream connections to a named endpoint are slightly more complicated. 154 * Sockets that have called listen() have their lock pointer mutated to 155 * the global uipc_lock. When establishing a connection, the connecting 156 * socket also has its lock mutated to uipc_lock, which matches the head 157 * (listening socket). We create a new socket for accept() to return, and 158 * that also shares the head's lock. Until the connection is completely 159 * done on both ends, all three sockets are locked by uipc_lock. Once the 160 * connection is complete, the association with the head's lock is broken. 161 * The connecting socket and the socket returned from accept() have their 162 * lock pointers mutated away from uipc_lock, and back to the connecting 163 * socket's original, independent lock. The head continues to be locked 164 * by uipc_lock. 165 * 166 * o If uipc_lock is determined to be a significant source of contention, 167 * it could easily be hashed out. It is difficult to simply make it an 168 * independent lock because of visibility / garbage collection issues: 169 * if a socket has been associated with a lock at any point, that lock 170 * must remain valid until the socket is no longer visible in the system. 171 * The lock must not be freed or otherwise destroyed until any sockets 172 * that had referenced it have also been destroyed. 173 */ 174 const struct sockaddr_un sun_noname = { 175 .sun_len = offsetof(struct sockaddr_un, sun_path), 176 .sun_family = AF_LOCAL, 177 }; 178 ino_t unp_ino; /* prototype for fake inode numbers */ 179 180 static struct mbuf * unp_addsockcred(struct lwp *, struct mbuf *); 181 static void unp_discard_later(file_t *); 182 static void unp_discard_now(file_t *); 183 static void unp_disconnect1(struct unpcb *); 184 static bool unp_drop(struct unpcb *, int); 185 static int unp_internalize(struct mbuf **); 186 static void unp_mark(file_t *); 187 static void unp_scan(struct mbuf *, void (*)(file_t *), int); 188 static void unp_shutdown1(struct unpcb *); 189 static void unp_thread(void *); 190 static void unp_thread_kick(void); 191 192 static kmutex_t *uipc_lock; 193 194 static kcondvar_t unp_thread_cv; 195 static lwp_t *unp_thread_lwp; 196 static SLIST_HEAD(,file) unp_thread_discard; 197 static int unp_defer; 198 199 /* Compat interface */ 200 201 struct mbuf * stub_compat_70_unp_addsockcred(lwp_t *, struct mbuf *); 202 203 struct mbuf * stub_compat_70_unp_addsockcred(struct lwp *lwp, 204 struct mbuf *control) 205 { 206 207 /* just copy our initial argument */ 208 return control; 209 } 210 211 bool *compat70_ocreds_valid = false; 212 213 /* 214 * Initialize Unix protocols. 215 */ 216 void 217 uipc_init(void) 218 { 219 int error; 220 221 uipc_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 222 cv_init(&unp_thread_cv, "unpgc"); 223 224 error = kthread_create(PRI_NONE, KTHREAD_MPSAFE, NULL, unp_thread, 225 NULL, &unp_thread_lwp, "unpgc"); 226 if (error != 0) 227 panic("uipc_init %d", error); 228 } 229 230 static void 231 unp_connid(struct lwp *l, struct unpcb *unp, int flags) 232 { 233 unp->unp_connid.unp_pid = l->l_proc->p_pid; 234 unp->unp_connid.unp_euid = kauth_cred_geteuid(l->l_cred); 235 unp->unp_connid.unp_egid = kauth_cred_getegid(l->l_cred); 236 unp->unp_flags |= flags; 237 } 238 239 /* 240 * A connection succeeded: disassociate both endpoints from the head's 241 * lock, and make them share their own lock. There is a race here: for 242 * a very brief time one endpoint will be locked by a different lock 243 * than the other end. However, since the current thread holds the old 244 * lock (the listening socket's lock, the head) access can still only be 245 * made to one side of the connection. 246 */ 247 static void 248 unp_setpeerlocks(struct socket *so, struct socket *so2) 249 { 250 struct unpcb *unp; 251 kmutex_t *lock; 252 253 KASSERT(solocked2(so, so2)); 254 255 /* 256 * Bail out if either end of the socket is not yet fully 257 * connected or accepted. We only break the lock association 258 * with the head when the pair of sockets stand completely 259 * on their own. 260 */ 261 KASSERT(so->so_head == NULL); 262 if (so2->so_head != NULL) 263 return; 264 265 /* 266 * Drop references to old lock. A third reference (from the 267 * queue head) must be held as we still hold its lock. Bonus: 268 * we don't need to worry about garbage collecting the lock. 269 */ 270 lock = so->so_lock; 271 KASSERT(lock == uipc_lock); 272 mutex_obj_free(lock); 273 mutex_obj_free(lock); 274 275 /* 276 * Grab stream lock from the initiator and share between the two 277 * endpoints. Issue memory barrier to ensure all modifications 278 * become globally visible before the lock change. so2 is 279 * assumed not to have a stream lock, because it was created 280 * purely for the server side to accept this connection and 281 * started out life using the domain-wide lock. 282 */ 283 unp = sotounpcb(so); 284 KASSERT(unp->unp_streamlock != NULL); 285 KASSERT(sotounpcb(so2)->unp_streamlock == NULL); 286 lock = unp->unp_streamlock; 287 unp->unp_streamlock = NULL; 288 mutex_obj_hold(lock); 289 membar_exit(); 290 /* 291 * possible race if lock is not held - see comment in 292 * uipc_usrreq(PRU_ACCEPT). 293 */ 294 KASSERT(mutex_owned(lock)); 295 solockreset(so, lock); 296 solockreset(so2, lock); 297 } 298 299 /* 300 * Reset a socket's lock back to the domain-wide lock. 301 */ 302 static void 303 unp_resetlock(struct socket *so) 304 { 305 kmutex_t *olock, *nlock; 306 struct unpcb *unp; 307 308 KASSERT(solocked(so)); 309 310 olock = so->so_lock; 311 nlock = uipc_lock; 312 if (olock == nlock) 313 return; 314 unp = sotounpcb(so); 315 KASSERT(unp->unp_streamlock == NULL); 316 unp->unp_streamlock = olock; 317 mutex_obj_hold(nlock); 318 mutex_enter(nlock); 319 solockreset(so, nlock); 320 mutex_exit(olock); 321 } 322 323 static void 324 unp_free(struct unpcb *unp) 325 { 326 if (unp->unp_addr) 327 free(unp->unp_addr, M_SONAME); 328 if (unp->unp_streamlock != NULL) 329 mutex_obj_free(unp->unp_streamlock); 330 kmem_free(unp, sizeof(*unp)); 331 } 332 333 static int 334 unp_output(struct mbuf *m, struct mbuf *control, struct unpcb *unp) 335 { 336 struct socket *so2; 337 const struct sockaddr_un *sun; 338 339 /* XXX: server side closed the socket */ 340 if (unp->unp_conn == NULL) 341 return ECONNREFUSED; 342 so2 = unp->unp_conn->unp_socket; 343 344 KASSERT(solocked(so2)); 345 346 if (unp->unp_addr) 347 sun = unp->unp_addr; 348 else 349 sun = &sun_noname; 350 if (unp->unp_conn->unp_flags & UNP_WANTCRED) 351 control = unp_addsockcred(curlwp, control); 352 if (unp->unp_conn->unp_flags & UNP_OWANTCRED) 353 MODULE_CALL_HOOK(uipc_unp_70_hook, (curlwp, control), 354 stub_compat_70_unp_addsockcred(curlwp, control), control); 355 if (sbappendaddr(&so2->so_rcv, (const struct sockaddr *)sun, m, 356 control) == 0) { 357 unp_dispose(control); 358 m_freem(control); 359 m_freem(m); 360 /* Don't call soroverflow because we're returning this 361 * error directly to the sender. */ 362 so2->so_rcv.sb_overflowed++; 363 return ENOBUFS; 364 } else { 365 sorwakeup(so2); 366 return 0; 367 } 368 } 369 370 static void 371 unp_setaddr(struct socket *so, struct sockaddr *nam, bool peeraddr) 372 { 373 const struct sockaddr_un *sun = NULL; 374 struct unpcb *unp; 375 376 KASSERT(solocked(so)); 377 unp = sotounpcb(so); 378 379 if (peeraddr) { 380 if (unp->unp_conn && unp->unp_conn->unp_addr) 381 sun = unp->unp_conn->unp_addr; 382 } else { 383 if (unp->unp_addr) 384 sun = unp->unp_addr; 385 } 386 if (sun == NULL) 387 sun = &sun_noname; 388 389 memcpy(nam, sun, sun->sun_len); 390 } 391 392 static int 393 unp_rcvd(struct socket *so, int flags, struct lwp *l) 394 { 395 struct unpcb *unp = sotounpcb(so); 396 struct socket *so2; 397 u_int newhiwat; 398 399 KASSERT(solocked(so)); 400 KASSERT(unp != NULL); 401 402 switch (so->so_type) { 403 404 case SOCK_DGRAM: 405 panic("uipc 1"); 406 /*NOTREACHED*/ 407 408 case SOCK_SEQPACKET: /* FALLTHROUGH */ 409 case SOCK_STREAM: 410 #define rcv (&so->so_rcv) 411 #define snd (&so2->so_snd) 412 if (unp->unp_conn == 0) 413 break; 414 so2 = unp->unp_conn->unp_socket; 415 KASSERT(solocked2(so, so2)); 416 /* 417 * Adjust backpressure on sender 418 * and wakeup any waiting to write. 419 */ 420 snd->sb_mbmax += unp->unp_mbcnt - rcv->sb_mbcnt; 421 unp->unp_mbcnt = rcv->sb_mbcnt; 422 newhiwat = snd->sb_hiwat + unp->unp_cc - rcv->sb_cc; 423 (void)chgsbsize(so2->so_uidinfo, 424 &snd->sb_hiwat, newhiwat, RLIM_INFINITY); 425 unp->unp_cc = rcv->sb_cc; 426 sowwakeup(so2); 427 #undef snd 428 #undef rcv 429 break; 430 431 default: 432 panic("uipc 2"); 433 } 434 435 return 0; 436 } 437 438 static int 439 unp_recvoob(struct socket *so, struct mbuf *m, int flags) 440 { 441 KASSERT(solocked(so)); 442 443 return EOPNOTSUPP; 444 } 445 446 static int 447 unp_send(struct socket *so, struct mbuf *m, struct sockaddr *nam, 448 struct mbuf *control, struct lwp *l) 449 { 450 struct unpcb *unp = sotounpcb(so); 451 int error = 0; 452 u_int newhiwat; 453 struct socket *so2; 454 455 KASSERT(solocked(so)); 456 KASSERT(unp != NULL); 457 KASSERT(m != NULL); 458 459 /* 460 * Note: unp_internalize() rejects any control message 461 * other than SCM_RIGHTS, and only allows one. This 462 * has the side-effect of preventing a caller from 463 * forging SCM_CREDS. 464 */ 465 if (control) { 466 sounlock(so); 467 error = unp_internalize(&control); 468 solock(so); 469 if (error != 0) { 470 m_freem(control); 471 m_freem(m); 472 return error; 473 } 474 } 475 476 switch (so->so_type) { 477 478 case SOCK_DGRAM: { 479 KASSERT(so->so_lock == uipc_lock); 480 if (nam) { 481 if ((so->so_state & SS_ISCONNECTED) != 0) 482 error = EISCONN; 483 else { 484 /* 485 * Note: once connected, the 486 * socket's lock must not be 487 * dropped until we have sent 488 * the message and disconnected. 489 * This is necessary to prevent 490 * intervening control ops, like 491 * another connection. 492 */ 493 error = unp_connect(so, nam, l); 494 } 495 } else { 496 if ((so->so_state & SS_ISCONNECTED) == 0) 497 error = ENOTCONN; 498 } 499 if (error) { 500 unp_dispose(control); 501 m_freem(control); 502 m_freem(m); 503 return error; 504 } 505 error = unp_output(m, control, unp); 506 if (nam) 507 unp_disconnect1(unp); 508 break; 509 } 510 511 case SOCK_SEQPACKET: /* FALLTHROUGH */ 512 case SOCK_STREAM: 513 #define rcv (&so2->so_rcv) 514 #define snd (&so->so_snd) 515 if (unp->unp_conn == NULL) { 516 error = ENOTCONN; 517 break; 518 } 519 so2 = unp->unp_conn->unp_socket; 520 KASSERT(solocked2(so, so2)); 521 if (unp->unp_conn->unp_flags & UNP_WANTCRED) { 522 /* 523 * Credentials are passed only once on 524 * SOCK_STREAM and SOCK_SEQPACKET. 525 */ 526 unp->unp_conn->unp_flags &= ~UNP_WANTCRED; 527 control = unp_addsockcred(l, control); 528 } 529 if (unp->unp_conn->unp_flags & UNP_OWANTCRED) { 530 /* 531 * Credentials are passed only once on 532 * SOCK_STREAM and SOCK_SEQPACKET. 533 */ 534 unp->unp_conn->unp_flags &= ~UNP_OWANTCRED; 535 MODULE_CALL_HOOK(uipc_unp_70_hook, (curlwp, control), 536 stub_compat_70_unp_addsockcred(curlwp, control), 537 control); 538 } 539 /* 540 * Send to paired receive port, and then reduce 541 * send buffer hiwater marks to maintain backpressure. 542 * Wake up readers. 543 */ 544 if (control) { 545 if (sbappendcontrol(rcv, m, control) != 0) 546 control = NULL; 547 } else { 548 switch(so->so_type) { 549 case SOCK_SEQPACKET: 550 sbappendrecord(rcv, m); 551 break; 552 case SOCK_STREAM: 553 sbappend(rcv, m); 554 break; 555 default: 556 panic("uipc_usrreq"); 557 break; 558 } 559 } 560 snd->sb_mbmax -= 561 rcv->sb_mbcnt - unp->unp_conn->unp_mbcnt; 562 unp->unp_conn->unp_mbcnt = rcv->sb_mbcnt; 563 newhiwat = snd->sb_hiwat - 564 (rcv->sb_cc - unp->unp_conn->unp_cc); 565 (void)chgsbsize(so->so_uidinfo, 566 &snd->sb_hiwat, newhiwat, RLIM_INFINITY); 567 unp->unp_conn->unp_cc = rcv->sb_cc; 568 sorwakeup(so2); 569 #undef snd 570 #undef rcv 571 if (control != NULL) { 572 unp_dispose(control); 573 m_freem(control); 574 } 575 break; 576 577 default: 578 panic("uipc 4"); 579 } 580 581 return error; 582 } 583 584 static int 585 unp_sendoob(struct socket *so, struct mbuf *m, struct mbuf * control) 586 { 587 KASSERT(solocked(so)); 588 589 m_freem(m); 590 m_freem(control); 591 592 return EOPNOTSUPP; 593 } 594 595 /* 596 * Unix domain socket option processing. 597 */ 598 int 599 uipc_ctloutput(int op, struct socket *so, struct sockopt *sopt) 600 { 601 struct unpcb *unp = sotounpcb(so); 602 int optval = 0, error = 0; 603 604 KASSERT(solocked(so)); 605 606 if (sopt->sopt_level != 0) { 607 error = ENOPROTOOPT; 608 } else switch (op) { 609 610 case PRCO_SETOPT: 611 switch (sopt->sopt_name) { 612 case LOCAL_OCREDS: 613 if (!*compat70_ocreds_valid) { 614 error = ENOPROTOOPT; 615 break; 616 } 617 /* FALLTHROUGH */ 618 case LOCAL_CREDS: 619 case LOCAL_CONNWAIT: 620 error = sockopt_getint(sopt, &optval); 621 if (error) 622 break; 623 switch (sopt->sopt_name) { 624 #define OPTSET(bit) \ 625 if (optval) \ 626 unp->unp_flags |= (bit); \ 627 else \ 628 unp->unp_flags &= ~(bit); 629 630 case LOCAL_CREDS: 631 OPTSET(UNP_WANTCRED); 632 break; 633 case LOCAL_CONNWAIT: 634 OPTSET(UNP_CONNWAIT); 635 break; 636 case LOCAL_OCREDS: 637 OPTSET(UNP_OWANTCRED); 638 break; 639 } 640 break; 641 #undef OPTSET 642 643 default: 644 error = ENOPROTOOPT; 645 break; 646 } 647 break; 648 649 case PRCO_GETOPT: 650 sounlock(so); 651 switch (sopt->sopt_name) { 652 case LOCAL_PEEREID: 653 if (unp->unp_flags & UNP_EIDSVALID) { 654 error = sockopt_set(sopt, &unp->unp_connid, 655 sizeof(unp->unp_connid)); 656 } else { 657 error = EINVAL; 658 } 659 break; 660 case LOCAL_CREDS: 661 #define OPTBIT(bit) (unp->unp_flags & (bit) ? 1 : 0) 662 663 optval = OPTBIT(UNP_WANTCRED); 664 error = sockopt_setint(sopt, optval); 665 break; 666 case LOCAL_OCREDS: 667 if (*compat70_ocreds_valid) { 668 optval = OPTBIT(UNP_OWANTCRED); 669 error = sockopt_setint(sopt, optval); 670 break; 671 } 672 #undef OPTBIT 673 /* FALLTHROUGH */ 674 default: 675 error = ENOPROTOOPT; 676 break; 677 } 678 solock(so); 679 break; 680 } 681 return (error); 682 } 683 684 /* 685 * Both send and receive buffers are allocated PIPSIZ bytes of buffering 686 * for stream sockets, although the total for sender and receiver is 687 * actually only PIPSIZ. 688 * Datagram sockets really use the sendspace as the maximum datagram size, 689 * and don't really want to reserve the sendspace. Their recvspace should 690 * be large enough for at least one max-size datagram plus address. 691 */ 692 #ifndef PIPSIZ 693 #define PIPSIZ 8192 694 #endif 695 u_long unpst_sendspace = PIPSIZ; 696 u_long unpst_recvspace = PIPSIZ; 697 u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 698 u_long unpdg_recvspace = 16*1024; 699 700 u_int unp_rights; /* files in flight */ 701 u_int unp_rights_ratio = 2; /* limit, fraction of maxfiles */ 702 703 static int 704 unp_attach(struct socket *so, int proto) 705 { 706 struct unpcb *unp = sotounpcb(so); 707 u_long sndspc, rcvspc; 708 int error; 709 710 KASSERT(unp == NULL); 711 712 switch (so->so_type) { 713 case SOCK_SEQPACKET: 714 /* FALLTHROUGH */ 715 case SOCK_STREAM: 716 if (so->so_lock == NULL) { 717 so->so_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 718 solock(so); 719 } 720 sndspc = unpst_sendspace; 721 rcvspc = unpst_recvspace; 722 break; 723 724 case SOCK_DGRAM: 725 if (so->so_lock == NULL) { 726 mutex_obj_hold(uipc_lock); 727 so->so_lock = uipc_lock; 728 solock(so); 729 } 730 sndspc = unpdg_sendspace; 731 rcvspc = unpdg_recvspace; 732 break; 733 734 default: 735 panic("unp_attach"); 736 } 737 738 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 739 error = soreserve(so, sndspc, rcvspc); 740 if (error) { 741 return error; 742 } 743 } 744 745 unp = kmem_zalloc(sizeof(*unp), KM_SLEEP); 746 nanotime(&unp->unp_ctime); 747 unp->unp_socket = so; 748 so->so_pcb = unp; 749 750 KASSERT(solocked(so)); 751 return 0; 752 } 753 754 static void 755 unp_detach(struct socket *so) 756 { 757 struct unpcb *unp; 758 vnode_t *vp; 759 760 unp = sotounpcb(so); 761 KASSERT(unp != NULL); 762 KASSERT(solocked(so)); 763 retry: 764 if ((vp = unp->unp_vnode) != NULL) { 765 sounlock(so); 766 /* Acquire v_interlock to protect against unp_connect(). */ 767 /* XXXAD racy */ 768 mutex_enter(vp->v_interlock); 769 vp->v_socket = NULL; 770 mutex_exit(vp->v_interlock); 771 vrele(vp); 772 solock(so); 773 unp->unp_vnode = NULL; 774 } 775 if (unp->unp_conn) 776 unp_disconnect1(unp); 777 while (unp->unp_refs) { 778 KASSERT(solocked2(so, unp->unp_refs->unp_socket)); 779 if (unp_drop(unp->unp_refs, ECONNRESET)) { 780 solock(so); 781 goto retry; 782 } 783 } 784 soisdisconnected(so); 785 so->so_pcb = NULL; 786 if (unp_rights) { 787 /* 788 * Normally the receive buffer is flushed later, in sofree, 789 * but if our receive buffer holds references to files that 790 * are now garbage, we will enqueue those file references to 791 * the garbage collector and kick it into action. 792 */ 793 sorflush(so); 794 unp_free(unp); 795 unp_thread_kick(); 796 } else 797 unp_free(unp); 798 } 799 800 static int 801 unp_accept(struct socket *so, struct sockaddr *nam) 802 { 803 struct unpcb *unp = sotounpcb(so); 804 struct socket *so2; 805 806 KASSERT(solocked(so)); 807 KASSERT(nam != NULL); 808 809 /* XXX code review required to determine if unp can ever be NULL */ 810 if (unp == NULL) 811 return EINVAL; 812 813 KASSERT(so->so_lock == uipc_lock); 814 /* 815 * Mark the initiating STREAM socket as connected *ONLY* 816 * after it's been accepted. This prevents a client from 817 * overrunning a server and receiving ECONNREFUSED. 818 */ 819 if (unp->unp_conn == NULL) { 820 /* 821 * This will use the empty socket and will not 822 * allocate. 823 */ 824 unp_setaddr(so, nam, true); 825 return 0; 826 } 827 so2 = unp->unp_conn->unp_socket; 828 if (so2->so_state & SS_ISCONNECTING) { 829 KASSERT(solocked2(so, so->so_head)); 830 KASSERT(solocked2(so2, so->so_head)); 831 soisconnected(so2); 832 } 833 /* 834 * If the connection is fully established, break the 835 * association with uipc_lock and give the connected 836 * pair a separate lock to share. 837 * There is a race here: sotounpcb(so2)->unp_streamlock 838 * is not locked, so when changing so2->so_lock 839 * another thread can grab it while so->so_lock is still 840 * pointing to the (locked) uipc_lock. 841 * this should be harmless, except that this makes 842 * solocked2() and solocked() unreliable. 843 * Another problem is that unp_setaddr() expects the 844 * the socket locked. Grabing sotounpcb(so2)->unp_streamlock 845 * fixes both issues. 846 */ 847 mutex_enter(sotounpcb(so2)->unp_streamlock); 848 unp_setpeerlocks(so2, so); 849 /* 850 * Only now return peer's address, as we may need to 851 * block in order to allocate memory. 852 * 853 * XXX Minor race: connection can be broken while 854 * lock is dropped in unp_setaddr(). We will return 855 * error == 0 and sun_noname as the peer address. 856 */ 857 unp_setaddr(so, nam, true); 858 /* so_lock now points to unp_streamlock */ 859 mutex_exit(so2->so_lock); 860 return 0; 861 } 862 863 static int 864 unp_ioctl(struct socket *so, u_long cmd, void *nam, struct ifnet *ifp) 865 { 866 return EOPNOTSUPP; 867 } 868 869 static int 870 unp_stat(struct socket *so, struct stat *ub) 871 { 872 struct unpcb *unp; 873 struct socket *so2; 874 875 KASSERT(solocked(so)); 876 877 unp = sotounpcb(so); 878 if (unp == NULL) 879 return EINVAL; 880 881 ub->st_blksize = so->so_snd.sb_hiwat; 882 switch (so->so_type) { 883 case SOCK_SEQPACKET: /* FALLTHROUGH */ 884 case SOCK_STREAM: 885 if (unp->unp_conn == 0) 886 break; 887 888 so2 = unp->unp_conn->unp_socket; 889 KASSERT(solocked2(so, so2)); 890 ub->st_blksize += so2->so_rcv.sb_cc; 891 break; 892 default: 893 break; 894 } 895 ub->st_dev = NODEV; 896 if (unp->unp_ino == 0) 897 unp->unp_ino = unp_ino++; 898 ub->st_atimespec = ub->st_mtimespec = ub->st_ctimespec = unp->unp_ctime; 899 ub->st_ino = unp->unp_ino; 900 return (0); 901 } 902 903 static int 904 unp_peeraddr(struct socket *so, struct sockaddr *nam) 905 { 906 KASSERT(solocked(so)); 907 KASSERT(sotounpcb(so) != NULL); 908 KASSERT(nam != NULL); 909 910 unp_setaddr(so, nam, true); 911 return 0; 912 } 913 914 static int 915 unp_sockaddr(struct socket *so, struct sockaddr *nam) 916 { 917 KASSERT(solocked(so)); 918 KASSERT(sotounpcb(so) != NULL); 919 KASSERT(nam != NULL); 920 921 unp_setaddr(so, nam, false); 922 return 0; 923 } 924 925 /* 926 * we only need to perform this allocation until syscalls other than 927 * bind are adjusted to use sockaddr_big. 928 */ 929 static struct sockaddr_un * 930 makeun_sb(struct sockaddr *nam, size_t *addrlen) 931 { 932 struct sockaddr_un *sun; 933 934 *addrlen = nam->sa_len + 1; 935 sun = malloc(*addrlen, M_SONAME, M_WAITOK); 936 memcpy(sun, nam, nam->sa_len); 937 *(((char *)sun) + nam->sa_len) = '\0'; 938 return sun; 939 } 940 941 static int 942 unp_bind(struct socket *so, struct sockaddr *nam, struct lwp *l) 943 { 944 struct sockaddr_un *sun; 945 struct unpcb *unp; 946 vnode_t *vp; 947 struct vattr vattr; 948 size_t addrlen; 949 int error; 950 struct pathbuf *pb; 951 struct nameidata nd; 952 proc_t *p; 953 954 unp = sotounpcb(so); 955 956 KASSERT(solocked(so)); 957 KASSERT(unp != NULL); 958 KASSERT(nam != NULL); 959 960 if (unp->unp_vnode != NULL) 961 return (EINVAL); 962 if ((unp->unp_flags & UNP_BUSY) != 0) { 963 /* 964 * EALREADY may not be strictly accurate, but since this 965 * is a major application error it's hardly a big deal. 966 */ 967 return (EALREADY); 968 } 969 unp->unp_flags |= UNP_BUSY; 970 sounlock(so); 971 972 p = l->l_proc; 973 sun = makeun_sb(nam, &addrlen); 974 975 pb = pathbuf_create(sun->sun_path); 976 if (pb == NULL) { 977 error = ENOMEM; 978 goto bad; 979 } 980 NDINIT(&nd, CREATE, FOLLOW | LOCKPARENT | TRYEMULROOT, pb); 981 982 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 983 if ((error = namei(&nd)) != 0) { 984 pathbuf_destroy(pb); 985 goto bad; 986 } 987 vp = nd.ni_vp; 988 if (vp != NULL) { 989 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 990 if (nd.ni_dvp == vp) 991 vrele(nd.ni_dvp); 992 else 993 vput(nd.ni_dvp); 994 vrele(vp); 995 pathbuf_destroy(pb); 996 error = EADDRINUSE; 997 goto bad; 998 } 999 vattr_null(&vattr); 1000 vattr.va_type = VSOCK; 1001 vattr.va_mode = ACCESSPERMS & ~(p->p_cwdi->cwdi_cmask); 1002 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 1003 if (error) { 1004 vput(nd.ni_dvp); 1005 pathbuf_destroy(pb); 1006 goto bad; 1007 } 1008 vp = nd.ni_vp; 1009 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1010 solock(so); 1011 vp->v_socket = unp->unp_socket; 1012 unp->unp_vnode = vp; 1013 unp->unp_addrlen = addrlen; 1014 unp->unp_addr = sun; 1015 VOP_UNLOCK(vp); 1016 vput(nd.ni_dvp); 1017 unp->unp_flags &= ~UNP_BUSY; 1018 pathbuf_destroy(pb); 1019 return (0); 1020 1021 bad: 1022 free(sun, M_SONAME); 1023 solock(so); 1024 unp->unp_flags &= ~UNP_BUSY; 1025 return (error); 1026 } 1027 1028 static int 1029 unp_listen(struct socket *so, struct lwp *l) 1030 { 1031 struct unpcb *unp = sotounpcb(so); 1032 1033 KASSERT(solocked(so)); 1034 KASSERT(unp != NULL); 1035 1036 /* 1037 * If the socket can accept a connection, it must be 1038 * locked by uipc_lock. 1039 */ 1040 unp_resetlock(so); 1041 if (unp->unp_vnode == NULL) 1042 return EINVAL; 1043 1044 unp_connid(l, unp, UNP_EIDSBIND); 1045 return 0; 1046 } 1047 1048 static int 1049 unp_disconnect(struct socket *so) 1050 { 1051 KASSERT(solocked(so)); 1052 KASSERT(sotounpcb(so) != NULL); 1053 1054 unp_disconnect1(sotounpcb(so)); 1055 return 0; 1056 } 1057 1058 static int 1059 unp_shutdown(struct socket *so) 1060 { 1061 KASSERT(solocked(so)); 1062 KASSERT(sotounpcb(so) != NULL); 1063 1064 socantsendmore(so); 1065 unp_shutdown1(sotounpcb(so)); 1066 return 0; 1067 } 1068 1069 static int 1070 unp_abort(struct socket *so) 1071 { 1072 KASSERT(solocked(so)); 1073 KASSERT(sotounpcb(so) != NULL); 1074 1075 (void)unp_drop(sotounpcb(so), ECONNABORTED); 1076 KASSERT(so->so_head == NULL); 1077 KASSERT(so->so_pcb != NULL); 1078 unp_detach(so); 1079 return 0; 1080 } 1081 1082 static int 1083 unp_connect1(struct socket *so, struct socket *so2, struct lwp *l) 1084 { 1085 struct unpcb *unp = sotounpcb(so); 1086 struct unpcb *unp2; 1087 1088 if (so2->so_type != so->so_type) 1089 return EPROTOTYPE; 1090 1091 /* 1092 * All three sockets involved must be locked by same lock: 1093 * 1094 * local endpoint (so) 1095 * remote endpoint (so2) 1096 * queue head (so2->so_head, only if PR_CONNREQUIRED) 1097 */ 1098 KASSERT(solocked2(so, so2)); 1099 KASSERT(so->so_head == NULL); 1100 if (so2->so_head != NULL) { 1101 KASSERT(so2->so_lock == uipc_lock); 1102 KASSERT(solocked2(so2, so2->so_head)); 1103 } 1104 1105 unp2 = sotounpcb(so2); 1106 unp->unp_conn = unp2; 1107 1108 switch (so->so_type) { 1109 1110 case SOCK_DGRAM: 1111 unp->unp_nextref = unp2->unp_refs; 1112 unp2->unp_refs = unp; 1113 soisconnected(so); 1114 break; 1115 1116 case SOCK_SEQPACKET: /* FALLTHROUGH */ 1117 case SOCK_STREAM: 1118 1119 /* 1120 * SOCK_SEQPACKET and SOCK_STREAM cases are handled by callers 1121 * which are unp_connect() or unp_connect2(). 1122 */ 1123 1124 break; 1125 1126 default: 1127 panic("unp_connect1"); 1128 } 1129 1130 return 0; 1131 } 1132 1133 int 1134 unp_connect(struct socket *so, struct sockaddr *nam, struct lwp *l) 1135 { 1136 struct sockaddr_un *sun; 1137 vnode_t *vp; 1138 struct socket *so2, *so3; 1139 struct unpcb *unp, *unp2, *unp3; 1140 size_t addrlen; 1141 int error; 1142 struct pathbuf *pb; 1143 struct nameidata nd; 1144 1145 unp = sotounpcb(so); 1146 if ((unp->unp_flags & UNP_BUSY) != 0) { 1147 /* 1148 * EALREADY may not be strictly accurate, but since this 1149 * is a major application error it's hardly a big deal. 1150 */ 1151 return (EALREADY); 1152 } 1153 unp->unp_flags |= UNP_BUSY; 1154 sounlock(so); 1155 1156 sun = makeun_sb(nam, &addrlen); 1157 pb = pathbuf_create(sun->sun_path); 1158 if (pb == NULL) { 1159 error = ENOMEM; 1160 goto bad2; 1161 } 1162 1163 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb); 1164 1165 if ((error = namei(&nd)) != 0) { 1166 pathbuf_destroy(pb); 1167 goto bad2; 1168 } 1169 vp = nd.ni_vp; 1170 pathbuf_destroy(pb); 1171 if (vp->v_type != VSOCK) { 1172 error = ENOTSOCK; 1173 goto bad; 1174 } 1175 if ((error = VOP_ACCESS(vp, VWRITE, l->l_cred)) != 0) 1176 goto bad; 1177 /* Acquire v_interlock to protect against unp_detach(). */ 1178 mutex_enter(vp->v_interlock); 1179 so2 = vp->v_socket; 1180 if (so2 == NULL) { 1181 mutex_exit(vp->v_interlock); 1182 error = ECONNREFUSED; 1183 goto bad; 1184 } 1185 if (so->so_type != so2->so_type) { 1186 mutex_exit(vp->v_interlock); 1187 error = EPROTOTYPE; 1188 goto bad; 1189 } 1190 solock(so); 1191 unp_resetlock(so); 1192 mutex_exit(vp->v_interlock); 1193 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) { 1194 /* 1195 * This may seem somewhat fragile but is OK: if we can 1196 * see SO_ACCEPTCONN set on the endpoint, then it must 1197 * be locked by the domain-wide uipc_lock. 1198 */ 1199 KASSERT((so2->so_options & SO_ACCEPTCONN) == 0 || 1200 so2->so_lock == uipc_lock); 1201 if ((so2->so_options & SO_ACCEPTCONN) == 0 || 1202 (so3 = sonewconn(so2, false)) == NULL) { 1203 error = ECONNREFUSED; 1204 sounlock(so); 1205 goto bad; 1206 } 1207 unp2 = sotounpcb(so2); 1208 unp3 = sotounpcb(so3); 1209 if (unp2->unp_addr) { 1210 unp3->unp_addr = malloc(unp2->unp_addrlen, 1211 M_SONAME, M_WAITOK); 1212 memcpy(unp3->unp_addr, unp2->unp_addr, 1213 unp2->unp_addrlen); 1214 unp3->unp_addrlen = unp2->unp_addrlen; 1215 } 1216 unp3->unp_flags = unp2->unp_flags; 1217 so2 = so3; 1218 /* 1219 * The connector's (client's) credentials are copied from its 1220 * process structure at the time of connect() (which is now). 1221 */ 1222 unp_connid(l, unp3, UNP_EIDSVALID); 1223 /* 1224 * The receiver's (server's) credentials are copied from the 1225 * unp_peercred member of socket on which the former called 1226 * listen(); unp_listen() cached that process's credentials 1227 * at that time so we can use them now. 1228 */ 1229 if (unp2->unp_flags & UNP_EIDSBIND) { 1230 memcpy(&unp->unp_connid, &unp2->unp_connid, 1231 sizeof(unp->unp_connid)); 1232 unp->unp_flags |= UNP_EIDSVALID; 1233 } 1234 } 1235 error = unp_connect1(so, so2, l); 1236 if (error) { 1237 sounlock(so); 1238 goto bad; 1239 } 1240 unp2 = sotounpcb(so2); 1241 switch (so->so_type) { 1242 1243 /* 1244 * SOCK_DGRAM and default cases are handled in prior call to 1245 * unp_connect1(), do not add a default case without fixing 1246 * unp_connect1(). 1247 */ 1248 1249 case SOCK_SEQPACKET: /* FALLTHROUGH */ 1250 case SOCK_STREAM: 1251 unp2->unp_conn = unp; 1252 if ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT) 1253 soisconnecting(so); 1254 else 1255 soisconnected(so); 1256 soisconnected(so2); 1257 /* 1258 * If the connection is fully established, break the 1259 * association with uipc_lock and give the connected 1260 * pair a seperate lock to share. 1261 */ 1262 KASSERT(so2->so_head != NULL); 1263 unp_setpeerlocks(so, so2); 1264 break; 1265 1266 } 1267 sounlock(so); 1268 bad: 1269 vput(vp); 1270 bad2: 1271 free(sun, M_SONAME); 1272 solock(so); 1273 unp->unp_flags &= ~UNP_BUSY; 1274 return (error); 1275 } 1276 1277 int 1278 unp_connect2(struct socket *so, struct socket *so2) 1279 { 1280 struct unpcb *unp = sotounpcb(so); 1281 struct unpcb *unp2; 1282 int error = 0; 1283 1284 KASSERT(solocked2(so, so2)); 1285 1286 error = unp_connect1(so, so2, curlwp); 1287 if (error) 1288 return error; 1289 1290 unp2 = sotounpcb(so2); 1291 switch (so->so_type) { 1292 1293 /* 1294 * SOCK_DGRAM and default cases are handled in prior call to 1295 * unp_connect1(), do not add a default case without fixing 1296 * unp_connect1(). 1297 */ 1298 1299 case SOCK_SEQPACKET: /* FALLTHROUGH */ 1300 case SOCK_STREAM: 1301 unp2->unp_conn = unp; 1302 soisconnected(so); 1303 soisconnected(so2); 1304 break; 1305 1306 } 1307 return error; 1308 } 1309 1310 static void 1311 unp_disconnect1(struct unpcb *unp) 1312 { 1313 struct unpcb *unp2 = unp->unp_conn; 1314 struct socket *so; 1315 1316 if (unp2 == 0) 1317 return; 1318 unp->unp_conn = 0; 1319 so = unp->unp_socket; 1320 switch (so->so_type) { 1321 case SOCK_DGRAM: 1322 if (unp2->unp_refs == unp) 1323 unp2->unp_refs = unp->unp_nextref; 1324 else { 1325 unp2 = unp2->unp_refs; 1326 for (;;) { 1327 KASSERT(solocked2(so, unp2->unp_socket)); 1328 if (unp2 == 0) 1329 panic("unp_disconnect1"); 1330 if (unp2->unp_nextref == unp) 1331 break; 1332 unp2 = unp2->unp_nextref; 1333 } 1334 unp2->unp_nextref = unp->unp_nextref; 1335 } 1336 unp->unp_nextref = 0; 1337 so->so_state &= ~SS_ISCONNECTED; 1338 break; 1339 1340 case SOCK_SEQPACKET: /* FALLTHROUGH */ 1341 case SOCK_STREAM: 1342 KASSERT(solocked2(so, unp2->unp_socket)); 1343 soisdisconnected(so); 1344 unp2->unp_conn = 0; 1345 soisdisconnected(unp2->unp_socket); 1346 break; 1347 } 1348 } 1349 1350 static void 1351 unp_shutdown1(struct unpcb *unp) 1352 { 1353 struct socket *so; 1354 1355 switch(unp->unp_socket->so_type) { 1356 case SOCK_SEQPACKET: /* FALLTHROUGH */ 1357 case SOCK_STREAM: 1358 if (unp->unp_conn && (so = unp->unp_conn->unp_socket)) 1359 socantrcvmore(so); 1360 break; 1361 default: 1362 break; 1363 } 1364 } 1365 1366 static bool 1367 unp_drop(struct unpcb *unp, int errno) 1368 { 1369 struct socket *so = unp->unp_socket; 1370 1371 KASSERT(solocked(so)); 1372 1373 so->so_error = errno; 1374 unp_disconnect1(unp); 1375 if (so->so_head) { 1376 so->so_pcb = NULL; 1377 /* sofree() drops the socket lock */ 1378 sofree(so); 1379 unp_free(unp); 1380 return true; 1381 } 1382 return false; 1383 } 1384 1385 #ifdef notdef 1386 unp_drain(void) 1387 { 1388 1389 } 1390 #endif 1391 1392 int 1393 unp_externalize(struct mbuf *rights, struct lwp *l, int flags) 1394 { 1395 struct cmsghdr * const cm = mtod(rights, struct cmsghdr *); 1396 struct proc * const p = l->l_proc; 1397 file_t **rp; 1398 int error = 0; 1399 1400 const size_t nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) / 1401 sizeof(file_t *); 1402 if (nfds == 0) 1403 goto noop; 1404 1405 int * const fdp = kmem_alloc(nfds * sizeof(int), KM_SLEEP); 1406 rw_enter(&p->p_cwdi->cwdi_lock, RW_READER); 1407 1408 /* Make sure the recipient should be able to see the files.. */ 1409 rp = (file_t **)CMSG_DATA(cm); 1410 for (size_t i = 0; i < nfds; i++) { 1411 file_t * const fp = *rp++; 1412 if (fp == NULL) { 1413 error = EINVAL; 1414 goto out; 1415 } 1416 /* 1417 * If we are in a chroot'ed directory, and 1418 * someone wants to pass us a directory, make 1419 * sure it's inside the subtree we're allowed 1420 * to access. 1421 */ 1422 if (p->p_cwdi->cwdi_rdir != NULL && fp->f_type == DTYPE_VNODE) { 1423 vnode_t *vp = fp->f_vnode; 1424 if ((vp->v_type == VDIR) && 1425 !vn_isunder(vp, p->p_cwdi->cwdi_rdir, l)) { 1426 error = EPERM; 1427 goto out; 1428 } 1429 } 1430 } 1431 1432 restart: 1433 /* 1434 * First loop -- allocate file descriptor table slots for the 1435 * new files. 1436 */ 1437 for (size_t i = 0; i < nfds; i++) { 1438 if ((error = fd_alloc(p, 0, &fdp[i])) != 0) { 1439 /* 1440 * Back out what we've done so far. 1441 */ 1442 while (i-- > 0) { 1443 fd_abort(p, NULL, fdp[i]); 1444 } 1445 if (error == ENOSPC) { 1446 fd_tryexpand(p); 1447 error = 0; 1448 goto restart; 1449 } 1450 /* 1451 * This is the error that has historically 1452 * been returned, and some callers may 1453 * expect it. 1454 */ 1455 error = EMSGSIZE; 1456 goto out; 1457 } 1458 } 1459 1460 /* 1461 * Now that adding them has succeeded, update all of the 1462 * file passing state and affix the descriptors. 1463 */ 1464 rp = (file_t **)CMSG_DATA(cm); 1465 int *ofdp = (int *)CMSG_DATA(cm); 1466 for (size_t i = 0; i < nfds; i++) { 1467 file_t * const fp = *rp++; 1468 const int fd = fdp[i]; 1469 atomic_dec_uint(&unp_rights); 1470 fd_set_exclose(l, fd, (flags & O_CLOEXEC) != 0); 1471 fd_affix(p, fp, fd); 1472 /* 1473 * Done with this file pointer, replace it with a fd; 1474 */ 1475 *ofdp++ = fd; 1476 mutex_enter(&fp->f_lock); 1477 fp->f_msgcount--; 1478 mutex_exit(&fp->f_lock); 1479 /* 1480 * Note that fd_affix() adds a reference to the file. 1481 * The file may already have been closed by another 1482 * LWP in the process, so we must drop the reference 1483 * added by unp_internalize() with closef(). 1484 */ 1485 closef(fp); 1486 } 1487 1488 /* 1489 * Adjust length, in case of transition from large file_t 1490 * pointers to ints. 1491 */ 1492 if (sizeof(file_t *) != sizeof(int)) { 1493 cm->cmsg_len = CMSG_LEN(nfds * sizeof(int)); 1494 rights->m_len = CMSG_SPACE(nfds * sizeof(int)); 1495 } 1496 out: 1497 if (__predict_false(error != 0)) { 1498 file_t **const fpp = (file_t **)CMSG_DATA(cm); 1499 for (size_t i = 0; i < nfds; i++) 1500 unp_discard_now(fpp[i]); 1501 /* 1502 * Truncate the array so that nobody will try to interpret 1503 * what is now garbage in it. 1504 */ 1505 cm->cmsg_len = CMSG_LEN(0); 1506 rights->m_len = CMSG_SPACE(0); 1507 } 1508 rw_exit(&p->p_cwdi->cwdi_lock); 1509 kmem_free(fdp, nfds * sizeof(int)); 1510 1511 noop: 1512 /* 1513 * Don't disclose kernel memory in the alignment space. 1514 */ 1515 KASSERT(cm->cmsg_len <= rights->m_len); 1516 memset(&mtod(rights, char *)[cm->cmsg_len], 0, rights->m_len - 1517 cm->cmsg_len); 1518 return error; 1519 } 1520 1521 static int 1522 unp_internalize(struct mbuf **controlp) 1523 { 1524 filedesc_t *fdescp = curlwp->l_fd; 1525 struct mbuf *control = *controlp; 1526 struct cmsghdr *newcm, *cm = mtod(control, struct cmsghdr *); 1527 file_t **rp, **files; 1528 file_t *fp; 1529 int i, fd, *fdp; 1530 int nfds, error; 1531 u_int maxmsg; 1532 1533 error = 0; 1534 newcm = NULL; 1535 1536 /* Sanity check the control message header. */ 1537 if (cm->cmsg_type != SCM_RIGHTS || cm->cmsg_level != SOL_SOCKET || 1538 cm->cmsg_len > control->m_len || 1539 cm->cmsg_len < CMSG_ALIGN(sizeof(*cm))) 1540 return (EINVAL); 1541 1542 /* 1543 * Verify that the file descriptors are valid, and acquire 1544 * a reference to each. 1545 */ 1546 nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) / sizeof(int); 1547 fdp = (int *)CMSG_DATA(cm); 1548 maxmsg = maxfiles / unp_rights_ratio; 1549 for (i = 0; i < nfds; i++) { 1550 fd = *fdp++; 1551 if (atomic_inc_uint_nv(&unp_rights) > maxmsg) { 1552 atomic_dec_uint(&unp_rights); 1553 nfds = i; 1554 error = EAGAIN; 1555 goto out; 1556 } 1557 if ((fp = fd_getfile(fd)) == NULL 1558 || fp->f_type == DTYPE_KQUEUE) { 1559 if (fp) 1560 fd_putfile(fd); 1561 atomic_dec_uint(&unp_rights); 1562 nfds = i; 1563 error = EBADF; 1564 goto out; 1565 } 1566 } 1567 1568 /* Allocate new space and copy header into it. */ 1569 newcm = malloc(CMSG_SPACE(nfds * sizeof(file_t *)), M_MBUF, M_WAITOK); 1570 if (newcm == NULL) { 1571 error = E2BIG; 1572 goto out; 1573 } 1574 memcpy(newcm, cm, sizeof(struct cmsghdr)); 1575 files = (file_t **)CMSG_DATA(newcm); 1576 1577 /* 1578 * Transform the file descriptors into file_t pointers, in 1579 * reverse order so that if pointers are bigger than ints, the 1580 * int won't get until we're done. No need to lock, as we have 1581 * already validated the descriptors with fd_getfile(). 1582 */ 1583 fdp = (int *)CMSG_DATA(cm) + nfds; 1584 rp = files + nfds; 1585 for (i = 0; i < nfds; i++) { 1586 fp = fdescp->fd_dt->dt_ff[*--fdp]->ff_file; 1587 KASSERT(fp != NULL); 1588 mutex_enter(&fp->f_lock); 1589 *--rp = fp; 1590 fp->f_count++; 1591 fp->f_msgcount++; 1592 mutex_exit(&fp->f_lock); 1593 } 1594 1595 out: 1596 /* Release descriptor references. */ 1597 fdp = (int *)CMSG_DATA(cm); 1598 for (i = 0; i < nfds; i++) { 1599 fd_putfile(*fdp++); 1600 if (error != 0) { 1601 atomic_dec_uint(&unp_rights); 1602 } 1603 } 1604 1605 if (error == 0) { 1606 if (control->m_flags & M_EXT) { 1607 m_freem(control); 1608 *controlp = control = m_get(M_WAIT, MT_CONTROL); 1609 } 1610 MEXTADD(control, newcm, CMSG_SPACE(nfds * sizeof(file_t *)), 1611 M_MBUF, NULL, NULL); 1612 cm = newcm; 1613 /* 1614 * Adjust message & mbuf to note amount of space 1615 * actually used. 1616 */ 1617 cm->cmsg_len = CMSG_LEN(nfds * sizeof(file_t *)); 1618 control->m_len = CMSG_SPACE(nfds * sizeof(file_t *)); 1619 } 1620 1621 return error; 1622 } 1623 1624 struct mbuf * 1625 unp_addsockcred(struct lwp *l, struct mbuf *control) 1626 { 1627 struct sockcred *sc; 1628 struct mbuf *m; 1629 void *p; 1630 1631 m = sbcreatecontrol1(&p, SOCKCREDSIZE(kauth_cred_ngroups(l->l_cred)), 1632 SCM_CREDS, SOL_SOCKET, M_WAITOK); 1633 if (m == NULL) 1634 return control; 1635 1636 sc = p; 1637 sc->sc_pid = l->l_proc->p_pid; 1638 sc->sc_uid = kauth_cred_getuid(l->l_cred); 1639 sc->sc_euid = kauth_cred_geteuid(l->l_cred); 1640 sc->sc_gid = kauth_cred_getgid(l->l_cred); 1641 sc->sc_egid = kauth_cred_getegid(l->l_cred); 1642 sc->sc_ngroups = kauth_cred_ngroups(l->l_cred); 1643 1644 for (int i = 0; i < sc->sc_ngroups; i++) 1645 sc->sc_groups[i] = kauth_cred_group(l->l_cred, i); 1646 1647 return m_add(control, m); 1648 } 1649 1650 /* 1651 * Do a mark-sweep GC of files in the system, to free up any which are 1652 * caught in flight to an about-to-be-closed socket. Additionally, 1653 * process deferred file closures. 1654 */ 1655 static void 1656 unp_gc(file_t *dp) 1657 { 1658 extern struct domain unixdomain; 1659 file_t *fp, *np; 1660 struct socket *so, *so1; 1661 u_int i, oflags, rflags; 1662 bool didwork; 1663 1664 KASSERT(curlwp == unp_thread_lwp); 1665 KASSERT(mutex_owned(&filelist_lock)); 1666 1667 /* 1668 * First, process deferred file closures. 1669 */ 1670 while (!SLIST_EMPTY(&unp_thread_discard)) { 1671 fp = SLIST_FIRST(&unp_thread_discard); 1672 KASSERT(fp->f_unpcount > 0); 1673 KASSERT(fp->f_count > 0); 1674 KASSERT(fp->f_msgcount > 0); 1675 KASSERT(fp->f_count >= fp->f_unpcount); 1676 KASSERT(fp->f_count >= fp->f_msgcount); 1677 KASSERT(fp->f_msgcount >= fp->f_unpcount); 1678 SLIST_REMOVE_HEAD(&unp_thread_discard, f_unplist); 1679 i = fp->f_unpcount; 1680 fp->f_unpcount = 0; 1681 mutex_exit(&filelist_lock); 1682 for (; i != 0; i--) { 1683 unp_discard_now(fp); 1684 } 1685 mutex_enter(&filelist_lock); 1686 } 1687 1688 /* 1689 * Clear mark bits. Ensure that we don't consider new files 1690 * entering the file table during this loop (they will not have 1691 * FSCAN set). 1692 */ 1693 unp_defer = 0; 1694 LIST_FOREACH(fp, &filehead, f_list) { 1695 for (oflags = fp->f_flag;; oflags = rflags) { 1696 rflags = atomic_cas_uint(&fp->f_flag, oflags, 1697 (oflags | FSCAN) & ~(FMARK|FDEFER)); 1698 if (__predict_true(oflags == rflags)) { 1699 break; 1700 } 1701 } 1702 } 1703 1704 /* 1705 * Iterate over the set of sockets, marking ones believed (based on 1706 * refcount) to be referenced from a process, and marking for rescan 1707 * sockets which are queued on a socket. Recan continues descending 1708 * and searching for sockets referenced by sockets (FDEFER), until 1709 * there are no more socket->socket references to be discovered. 1710 */ 1711 do { 1712 didwork = false; 1713 for (fp = LIST_FIRST(&filehead); fp != NULL; fp = np) { 1714 KASSERT(mutex_owned(&filelist_lock)); 1715 np = LIST_NEXT(fp, f_list); 1716 mutex_enter(&fp->f_lock); 1717 if ((fp->f_flag & FDEFER) != 0) { 1718 atomic_and_uint(&fp->f_flag, ~FDEFER); 1719 unp_defer--; 1720 if (fp->f_count == 0) { 1721 /* 1722 * XXX: closef() doesn't pay attention 1723 * to FDEFER 1724 */ 1725 mutex_exit(&fp->f_lock); 1726 continue; 1727 } 1728 } else { 1729 if (fp->f_count == 0 || 1730 (fp->f_flag & FMARK) != 0 || 1731 fp->f_count == fp->f_msgcount || 1732 fp->f_unpcount != 0) { 1733 mutex_exit(&fp->f_lock); 1734 continue; 1735 } 1736 } 1737 atomic_or_uint(&fp->f_flag, FMARK); 1738 1739 if (fp->f_type != DTYPE_SOCKET || 1740 (so = fp->f_socket) == NULL || 1741 so->so_proto->pr_domain != &unixdomain || 1742 (so->so_proto->pr_flags & PR_RIGHTS) == 0) { 1743 mutex_exit(&fp->f_lock); 1744 continue; 1745 } 1746 1747 /* Gain file ref, mark our position, and unlock. */ 1748 didwork = true; 1749 LIST_INSERT_AFTER(fp, dp, f_list); 1750 fp->f_count++; 1751 mutex_exit(&fp->f_lock); 1752 mutex_exit(&filelist_lock); 1753 1754 /* 1755 * Mark files referenced from sockets queued on the 1756 * accept queue as well. 1757 */ 1758 solock(so); 1759 unp_scan(so->so_rcv.sb_mb, unp_mark, 0); 1760 if ((so->so_options & SO_ACCEPTCONN) != 0) { 1761 TAILQ_FOREACH(so1, &so->so_q0, so_qe) { 1762 unp_scan(so1->so_rcv.sb_mb, unp_mark, 0); 1763 } 1764 TAILQ_FOREACH(so1, &so->so_q, so_qe) { 1765 unp_scan(so1->so_rcv.sb_mb, unp_mark, 0); 1766 } 1767 } 1768 sounlock(so); 1769 1770 /* Re-lock and restart from where we left off. */ 1771 closef(fp); 1772 mutex_enter(&filelist_lock); 1773 np = LIST_NEXT(dp, f_list); 1774 LIST_REMOVE(dp, f_list); 1775 } 1776 /* 1777 * Bail early if we did nothing in the loop above. Could 1778 * happen because of concurrent activity causing unp_defer 1779 * to get out of sync. 1780 */ 1781 } while (unp_defer != 0 && didwork); 1782 1783 /* 1784 * Sweep pass. 1785 * 1786 * We grab an extra reference to each of the files that are 1787 * not otherwise accessible and then free the rights that are 1788 * stored in messages on them. 1789 */ 1790 for (fp = LIST_FIRST(&filehead); fp != NULL; fp = np) { 1791 KASSERT(mutex_owned(&filelist_lock)); 1792 np = LIST_NEXT(fp, f_list); 1793 mutex_enter(&fp->f_lock); 1794 1795 /* 1796 * Ignore non-sockets. 1797 * Ignore dead sockets, or sockets with pending close. 1798 * Ignore sockets obviously referenced elsewhere. 1799 * Ignore sockets marked as referenced by our scan. 1800 * Ignore new sockets that did not exist during the scan. 1801 */ 1802 if (fp->f_type != DTYPE_SOCKET || 1803 fp->f_count == 0 || fp->f_unpcount != 0 || 1804 fp->f_count != fp->f_msgcount || 1805 (fp->f_flag & (FMARK | FSCAN)) != FSCAN) { 1806 mutex_exit(&fp->f_lock); 1807 continue; 1808 } 1809 1810 /* Gain file ref, mark our position, and unlock. */ 1811 LIST_INSERT_AFTER(fp, dp, f_list); 1812 fp->f_count++; 1813 mutex_exit(&fp->f_lock); 1814 mutex_exit(&filelist_lock); 1815 1816 /* 1817 * Flush all data from the socket's receive buffer. 1818 * This will cause files referenced only by the 1819 * socket to be queued for close. 1820 */ 1821 so = fp->f_socket; 1822 solock(so); 1823 sorflush(so); 1824 sounlock(so); 1825 1826 /* Re-lock and restart from where we left off. */ 1827 closef(fp); 1828 mutex_enter(&filelist_lock); 1829 np = LIST_NEXT(dp, f_list); 1830 LIST_REMOVE(dp, f_list); 1831 } 1832 } 1833 1834 /* 1835 * Garbage collector thread. While SCM_RIGHTS messages are in transit, 1836 * wake once per second to garbage collect. Run continually while we 1837 * have deferred closes to process. 1838 */ 1839 static void 1840 unp_thread(void *cookie) 1841 { 1842 file_t *dp; 1843 1844 /* Allocate a dummy file for our scans. */ 1845 if ((dp = fgetdummy()) == NULL) { 1846 panic("unp_thread"); 1847 } 1848 1849 mutex_enter(&filelist_lock); 1850 for (;;) { 1851 KASSERT(mutex_owned(&filelist_lock)); 1852 if (SLIST_EMPTY(&unp_thread_discard)) { 1853 if (unp_rights != 0) { 1854 (void)cv_timedwait(&unp_thread_cv, 1855 &filelist_lock, hz); 1856 } else { 1857 cv_wait(&unp_thread_cv, &filelist_lock); 1858 } 1859 } 1860 unp_gc(dp); 1861 } 1862 /* NOTREACHED */ 1863 } 1864 1865 /* 1866 * Kick the garbage collector into action if there is something for 1867 * it to process. 1868 */ 1869 static void 1870 unp_thread_kick(void) 1871 { 1872 1873 if (!SLIST_EMPTY(&unp_thread_discard) || unp_rights != 0) { 1874 mutex_enter(&filelist_lock); 1875 cv_signal(&unp_thread_cv); 1876 mutex_exit(&filelist_lock); 1877 } 1878 } 1879 1880 void 1881 unp_dispose(struct mbuf *m) 1882 { 1883 1884 if (m) 1885 unp_scan(m, unp_discard_later, 1); 1886 } 1887 1888 void 1889 unp_scan(struct mbuf *m0, void (*op)(file_t *), int discard) 1890 { 1891 struct mbuf *m; 1892 file_t **rp, *fp; 1893 struct cmsghdr *cm; 1894 int i, qfds; 1895 1896 while (m0) { 1897 for (m = m0; m; m = m->m_next) { 1898 if (m->m_type != MT_CONTROL || 1899 m->m_len < sizeof(*cm)) { 1900 continue; 1901 } 1902 cm = mtod(m, struct cmsghdr *); 1903 if (cm->cmsg_level != SOL_SOCKET || 1904 cm->cmsg_type != SCM_RIGHTS) 1905 continue; 1906 qfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) 1907 / sizeof(file_t *); 1908 rp = (file_t **)CMSG_DATA(cm); 1909 for (i = 0; i < qfds; i++) { 1910 fp = *rp; 1911 if (discard) { 1912 *rp = 0; 1913 } 1914 (*op)(fp); 1915 rp++; 1916 } 1917 } 1918 m0 = m0->m_nextpkt; 1919 } 1920 } 1921 1922 void 1923 unp_mark(file_t *fp) 1924 { 1925 1926 if (fp == NULL) 1927 return; 1928 1929 /* If we're already deferred, don't screw up the defer count */ 1930 mutex_enter(&fp->f_lock); 1931 if (fp->f_flag & (FMARK | FDEFER)) { 1932 mutex_exit(&fp->f_lock); 1933 return; 1934 } 1935 1936 /* 1937 * Minimize the number of deferrals... Sockets are the only type of 1938 * file which can hold references to another file, so just mark 1939 * other files, and defer unmarked sockets for the next pass. 1940 */ 1941 if (fp->f_type == DTYPE_SOCKET) { 1942 unp_defer++; 1943 KASSERT(fp->f_count != 0); 1944 atomic_or_uint(&fp->f_flag, FDEFER); 1945 } else { 1946 atomic_or_uint(&fp->f_flag, FMARK); 1947 } 1948 mutex_exit(&fp->f_lock); 1949 } 1950 1951 static void 1952 unp_discard_now(file_t *fp) 1953 { 1954 1955 if (fp == NULL) 1956 return; 1957 1958 KASSERT(fp->f_count > 0); 1959 KASSERT(fp->f_msgcount > 0); 1960 1961 mutex_enter(&fp->f_lock); 1962 fp->f_msgcount--; 1963 mutex_exit(&fp->f_lock); 1964 atomic_dec_uint(&unp_rights); 1965 (void)closef(fp); 1966 } 1967 1968 static void 1969 unp_discard_later(file_t *fp) 1970 { 1971 1972 if (fp == NULL) 1973 return; 1974 1975 KASSERT(fp->f_count > 0); 1976 KASSERT(fp->f_msgcount > 0); 1977 1978 mutex_enter(&filelist_lock); 1979 if (fp->f_unpcount++ == 0) { 1980 SLIST_INSERT_HEAD(&unp_thread_discard, fp, f_unplist); 1981 } 1982 mutex_exit(&filelist_lock); 1983 } 1984 1985 void 1986 unp_sysctl_create(struct sysctllog **clog) 1987 { 1988 sysctl_createv(clog, 0, NULL, NULL, 1989 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1990 CTLTYPE_LONG, "sendspace", 1991 SYSCTL_DESCR("Default stream send space"), 1992 NULL, 0, &unpst_sendspace, 0, 1993 CTL_NET, PF_LOCAL, SOCK_STREAM, CTL_CREATE, CTL_EOL); 1994 sysctl_createv(clog, 0, NULL, NULL, 1995 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1996 CTLTYPE_LONG, "recvspace", 1997 SYSCTL_DESCR("Default stream recv space"), 1998 NULL, 0, &unpst_recvspace, 0, 1999 CTL_NET, PF_LOCAL, SOCK_STREAM, CTL_CREATE, CTL_EOL); 2000 sysctl_createv(clog, 0, NULL, NULL, 2001 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2002 CTLTYPE_LONG, "sendspace", 2003 SYSCTL_DESCR("Default datagram send space"), 2004 NULL, 0, &unpdg_sendspace, 0, 2005 CTL_NET, PF_LOCAL, SOCK_DGRAM, CTL_CREATE, CTL_EOL); 2006 sysctl_createv(clog, 0, NULL, NULL, 2007 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2008 CTLTYPE_LONG, "recvspace", 2009 SYSCTL_DESCR("Default datagram recv space"), 2010 NULL, 0, &unpdg_recvspace, 0, 2011 CTL_NET, PF_LOCAL, SOCK_DGRAM, CTL_CREATE, CTL_EOL); 2012 sysctl_createv(clog, 0, NULL, NULL, 2013 CTLFLAG_PERMANENT|CTLFLAG_READONLY, 2014 CTLTYPE_INT, "inflight", 2015 SYSCTL_DESCR("File descriptors in flight"), 2016 NULL, 0, &unp_rights, 0, 2017 CTL_NET, PF_LOCAL, CTL_CREATE, CTL_EOL); 2018 sysctl_createv(clog, 0, NULL, NULL, 2019 CTLFLAG_PERMANENT|CTLFLAG_READONLY, 2020 CTLTYPE_INT, "deferred", 2021 SYSCTL_DESCR("File descriptors deferred for close"), 2022 NULL, 0, &unp_defer, 0, 2023 CTL_NET, PF_LOCAL, CTL_CREATE, CTL_EOL); 2024 } 2025 2026 const struct pr_usrreqs unp_usrreqs = { 2027 .pr_attach = unp_attach, 2028 .pr_detach = unp_detach, 2029 .pr_accept = unp_accept, 2030 .pr_bind = unp_bind, 2031 .pr_listen = unp_listen, 2032 .pr_connect = unp_connect, 2033 .pr_connect2 = unp_connect2, 2034 .pr_disconnect = unp_disconnect, 2035 .pr_shutdown = unp_shutdown, 2036 .pr_abort = unp_abort, 2037 .pr_ioctl = unp_ioctl, 2038 .pr_stat = unp_stat, 2039 .pr_peeraddr = unp_peeraddr, 2040 .pr_sockaddr = unp_sockaddr, 2041 .pr_rcvd = unp_rcvd, 2042 .pr_recvoob = unp_recvoob, 2043 .pr_send = unp_send, 2044 .pr_sendoob = unp_sendoob, 2045 }; 2046
Indexes created Sat Sep 20 22:09:52 GMT 2025