1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22 /* 23 * Copyright 2006 Sun Microsystems, Inc. All rights reserved. 24 * Use is subject to license terms. 25 */ 26 27 #pragma ident "%Z%%M% %I% %E% SMI" 28 29 #include <sys/types.h> 30 #include <sys/t_lock.h> 31 #include <sys/param.h> 32 #include <sys/systm.h> 33 #include <sys/buf.h> 34 #include <sys/conf.h> 35 #include <sys/cred.h> 36 #include <sys/kmem.h> 37 #include <sys/sysmacros.h> 38 #include <sys/vfs.h> 39 #include <sys/vnode.h> 40 #include <sys/debug.h> 41 #include <sys/errno.h> 42 #include <sys/time.h> 43 #include <sys/file.h> 44 #include <sys/open.h> 45 #include <sys/user.h> 46 #include <sys/termios.h> 47 #include <sys/stream.h> 48 #include <sys/strsubr.h> 49 #include <sys/strsun.h> 50 #include <sys/esunddi.h> 51 #include <sys/flock.h> 52 #include <sys/modctl.h> 53 #include <sys/cmn_err.h> 54 #include <sys/mkdev.h> 55 #include <sys/pathname.h> 56 #include <sys/ddi.h> 57 #include <sys/stat.h> 58 #include <sys/fs/snode.h> 59 #include <sys/fs/dv_node.h> 60 #include <sys/zone.h> 61 62 #include <sys/socket.h> 63 #include <sys/socketvar.h> 64 #include <netinet/in.h> 65 #include <sys/un.h> 66 67 #include <sys/ucred.h> 68 69 #include <sys/tiuser.h> 70 #define _SUN_TPI_VERSION 2 71 #include <sys/tihdr.h> 72 73 #include <c2/audit.h> 74 75 #include <fs/sockfs/nl7c.h> 76 77 /* 78 * Macros that operate on struct cmsghdr. 79 * The CMSG_VALID macro does not assume that the last option buffer is padded. 80 */ 81 #define CMSG_CONTENT(cmsg) (&((cmsg)[1])) 82 #define CMSG_CONTENTLEN(cmsg) ((cmsg)->cmsg_len - sizeof (struct cmsghdr)) 83 #define CMSG_VALID(cmsg, start, end) \ 84 (ISALIGNED_cmsghdr(cmsg) && \ 85 ((uintptr_t)(cmsg) >= (uintptr_t)(start)) && \ 86 ((uintptr_t)(cmsg) < (uintptr_t)(end)) && \ 87 ((ssize_t)(cmsg)->cmsg_len >= sizeof (struct cmsghdr)) && \ 88 ((uintptr_t)(cmsg) + (cmsg)->cmsg_len <= (uintptr_t)(end))) 89 #define SO_LOCK_WAKEUP_TIME 3000 /* Wakeup time in milliseconds */ 90 91 static struct kmem_cache *socktpi_cache, *socktpi_unix_cache; 92 93 dev_t sockdev; /* For fsid in getattr */ 94 95 struct sockparams *sphead; 96 krwlock_t splist_lock; 97 98 struct socklist socklist; 99 100 static int sockfs_update(kstat_t *, int); 101 static int sockfs_snapshot(kstat_t *, void *, int); 102 103 extern void sendfile_init(); 104 105 extern void nl7c_init(void); 106 107 #define ADRSTRLEN (2 * sizeof (void *) + 1) 108 /* 109 * kernel structure for passing the sockinfo data back up to the user. 110 * the strings array allows us to convert AF_UNIX addresses into strings 111 * with a common method regardless of which n-bit kernel we're running. 112 */ 113 struct k_sockinfo { 114 struct sockinfo ks_si; 115 char ks_straddr[3][ADRSTRLEN]; 116 }; 117 118 /* 119 * Translate from a device pathname (e.g. "/dev/tcp") to a vnode. 120 * Returns with the vnode held. 121 */ 122 static int 123 sogetvp(char *devpath, vnode_t **vpp, int uioflag) 124 { 125 struct snode *csp; 126 vnode_t *vp, *dvp; 127 major_t maj; 128 int error; 129 130 ASSERT(uioflag == UIO_SYSSPACE || uioflag == UIO_USERSPACE); 131 /* 132 * Lookup the underlying filesystem vnode. 133 */ 134 error = lookupname(devpath, uioflag, FOLLOW, NULLVPP, &vp); 135 if (error) 136 return (error); 137 138 /* Check that it is the correct vnode */ 139 if (vp->v_type != VCHR) { 140 VN_RELE(vp); 141 return (ENOTSOCK); 142 } 143 144 /* 145 * If devpath went through devfs, the device should already 146 * be configured. If devpath is a mknod file, however, we 147 * need to make sure the device is properly configured. 148 * To do this, we do something similar to spec_open() 149 * except that we resolve to the minor/leaf level since 150 * we need to return a vnode. 151 */ 152 csp = VTOS(VTOS(vp)->s_commonvp); 153 if (!(csp->s_flag & SDIPSET)) { 154 char *pathname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 155 error = ddi_dev_pathname(vp->v_rdev, S_IFCHR, pathname); 156 if (error == 0) 157 error = devfs_lookupname(pathname, NULLVPP, &dvp); 158 VN_RELE(vp); 159 kmem_free(pathname, MAXPATHLEN); 160 if (error != 0) 161 return (ENXIO); 162 vp = dvp; /* use the devfs vp */ 163 } 164 165 /* device is configured at this point */ 166 maj = getmajor(vp->v_rdev); 167 if (!STREAMSTAB(maj)) { 168 VN_RELE(vp); 169 return (ENOSTR); 170 } 171 172 *vpp = vp; 173 return (0); 174 } 175 176 /* 177 * Add or delete (latter if devpath is NULL) an enter to the sockparams 178 * table. If devpathlen is zero the devpath with not be kmem_freed. Otherwise 179 * this routine assumes that the caller has kmem_alloced devpath/devpathlen 180 * for this routine to consume. 181 * The zero devpathlen could be used if the kernel wants to create entries 182 * itself by calling sockconfig(1,2,3, "/dev/tcp", 0); 183 */ 184 int 185 soconfig(int domain, int type, int protocol, 186 char *devpath, int devpathlen) 187 { 188 struct sockparams **spp; 189 struct sockparams *sp; 190 int error = 0; 191 192 dprint(0, ("soconfig(%d,%d,%d,%s,%d)\n", 193 domain, type, protocol, devpath, devpathlen)); 194 195 /* 196 * Look for an existing match. 197 */ 198 rw_enter(&splist_lock, RW_WRITER); 199 for (spp = &sphead; (sp = *spp) != NULL; spp = &sp->sp_next) { 200 if (sp->sp_domain == domain && 201 sp->sp_type == type && 202 sp->sp_protocol == protocol) { 203 break; 204 } 205 } 206 if (devpath == NULL) { 207 ASSERT(devpathlen == 0); 208 209 /* Delete existing entry */ 210 if (sp == NULL) { 211 error = ENXIO; 212 goto done; 213 } 214 /* Unlink and free existing entry */ 215 *spp = sp->sp_next; 216 ASSERT(sp->sp_vnode); 217 VN_RELE(sp->sp_vnode); 218 if (sp->sp_devpathlen != 0) 219 kmem_free(sp->sp_devpath, sp->sp_devpathlen); 220 kmem_free(sp, sizeof (*sp)); 221 } else { 222 vnode_t *vp; 223 224 /* Add new entry */ 225 if (sp != NULL) { 226 error = EEXIST; 227 goto done; 228 } 229 230 error = sogetvp(devpath, &vp, UIO_SYSSPACE); 231 if (error) { 232 dprint(0, ("soconfig: vp %s failed with %d\n", 233 devpath, error)); 234 goto done; 235 } 236 237 dprint(0, ("soconfig: %s => vp %p, dev 0x%lx\n", 238 devpath, vp, vp->v_rdev)); 239 240 sp = kmem_alloc(sizeof (*sp), KM_SLEEP); 241 sp->sp_domain = domain; 242 sp->sp_type = type; 243 sp->sp_protocol = protocol; 244 sp->sp_devpath = devpath; 245 sp->sp_devpathlen = devpathlen; 246 sp->sp_vnode = vp; 247 sp->sp_next = NULL; 248 *spp = sp; 249 } 250 done: 251 rw_exit(&splist_lock); 252 if (error) { 253 if (devpath != NULL) 254 kmem_free(devpath, devpathlen); 255 #ifdef SOCK_DEBUG 256 eprintline(error); 257 #endif /* SOCK_DEBUG */ 258 } 259 return (error); 260 } 261 262 /* 263 * Lookup an entry in the sockparams list based on the triple. 264 * If no entry is found and devpath is not NULL translate devpath to a 265 * vnode. Note that devpath is a pointer to a user address! 266 * Returns with the vnode held. 267 * 268 * When this routine uses devpath it does not create an entry in the sockparams 269 * list since this routine can run on behalf of any user and one user 270 * should not be able to effect the transport used by another user. 271 * 272 * In order to return the correct error this routine has to do wildcard scans 273 * of the list. The errors are (in decreasing precedence): 274 * EAFNOSUPPORT - address family not in list 275 * EPROTONOSUPPORT - address family supported but not protocol. 276 * EPROTOTYPE - address family and protocol supported but not socket type. 277 */ 278 vnode_t * 279 solookup(int domain, int type, int protocol, char *devpath, int *errorp) 280 { 281 struct sockparams *sp; 282 int error; 283 vnode_t *vp; 284 285 rw_enter(&splist_lock, RW_READER); 286 for (sp = sphead; sp != NULL; sp = sp->sp_next) { 287 if (sp->sp_domain == domain && 288 sp->sp_type == type && 289 sp->sp_protocol == protocol) { 290 break; 291 } 292 } 293 if (sp == NULL) { 294 dprint(0, ("solookup(%d,%d,%d) not found\n", 295 domain, type, protocol)); 296 if (devpath == NULL) { 297 /* Determine correct error code */ 298 int found = 0; 299 300 for (sp = sphead; sp != NULL; sp = sp->sp_next) { 301 if (sp->sp_domain == domain && found < 1) 302 found = 1; 303 if (sp->sp_domain == domain && 304 sp->sp_protocol == protocol && found < 2) 305 found = 2; 306 } 307 rw_exit(&splist_lock); 308 switch (found) { 309 case 0: 310 *errorp = EAFNOSUPPORT; 311 break; 312 case 1: 313 *errorp = EPROTONOSUPPORT; 314 break; 315 case 2: 316 *errorp = EPROTOTYPE; 317 break; 318 } 319 return (NULL); 320 } 321 rw_exit(&splist_lock); 322 323 /* 324 * Return vp based on devpath. 325 * Do not enter into table to avoid random users 326 * modifying the sockparams list. 327 */ 328 error = sogetvp(devpath, &vp, UIO_USERSPACE); 329 if (error) { 330 dprint(0, ("solookup: vp %p failed with %d\n", 331 devpath, error)); 332 *errorp = EPROTONOSUPPORT; 333 return (NULL); 334 } 335 dprint(0, ("solookup: %p => vp %p, dev 0x%lx\n", 336 devpath, vp, vp->v_rdev)); 337 338 return (vp); 339 } 340 dprint(0, ("solookup(%d,%d,%d) vp %p devpath %s\n", 341 domain, type, protocol, sp->sp_vnode, sp->sp_devpath)); 342 343 vp = sp->sp_vnode; 344 VN_HOLD(vp); 345 rw_exit(&splist_lock); 346 return (vp); 347 } 348 349 /* 350 * Return a socket vnode. 351 * 352 * Assumes that the caller is "passing" an VN_HOLD for accessvp i.e. 353 * when the socket is freed a VN_RELE will take place. 354 * 355 * Note that sockets assume that the driver will clone (either itself 356 * or by using the clone driver) i.e. a socket() call will always 357 * result in a new vnode being created. 358 */ 359 struct vnode * 360 makesockvp(struct vnode *accessvp, int domain, int type, int protocol) 361 { 362 kmem_cache_t *cp; 363 struct sonode *so; 364 struct vnode *vp; 365 time_t now; 366 dev_t dev; 367 368 cp = (domain == AF_UNIX) ? socktpi_unix_cache : socktpi_cache; 369 so = kmem_cache_alloc(cp, KM_SLEEP); 370 so->so_cache = cp; 371 so->so_obj = so; 372 vp = SOTOV(so); 373 now = gethrestime_sec(); 374 375 so->so_flag = 0; 376 ASSERT(so->so_accessvp == NULL); 377 so->so_accessvp = accessvp; 378 dev = accessvp->v_rdev; 379 380 /* 381 * Record in so_flag that it is a clone. 382 */ 383 if (getmajor(dev) == clone_major) { 384 so->so_flag |= SOCLONE; 385 } 386 so->so_dev = dev; 387 388 so->so_state = 0; 389 so->so_mode = 0; 390 391 so->so_fsid = sockdev; 392 so->so_atime = now; 393 so->so_mtime = now; 394 so->so_ctime = now; /* Never modified */ 395 so->so_count = 0; 396 397 so->so_family = (short)domain; 398 so->so_type = (short)type; 399 so->so_protocol = (short)protocol; 400 so->so_pushcnt = 0; 401 402 so->so_options = 0; 403 so->so_linger.l_onoff = 0; 404 so->so_linger.l_linger = 0; 405 so->so_sndbuf = 0; 406 so->so_rcvbuf = 0; 407 so->so_sndlowat = 0; 408 so->so_rcvlowat = 0; 409 #ifdef notyet 410 so->so_sndtimeo = 0; 411 so->so_rcvtimeo = 0; 412 #endif /* notyet */ 413 so->so_error = 0; 414 so->so_delayed_error = 0; 415 416 ASSERT(so->so_oobmsg == NULL); 417 so->so_oobcnt = 0; 418 so->so_oobsigcnt = 0; 419 so->so_pgrp = 0; 420 so->so_provinfo = NULL; 421 422 ASSERT(so->so_laddr_sa == NULL && so->so_faddr_sa == NULL); 423 so->so_laddr_len = so->so_faddr_len = 0; 424 so->so_laddr_maxlen = so->so_faddr_maxlen = 0; 425 so->so_eaddr_mp = NULL; 426 so->so_priv = NULL; 427 428 so->so_peercred = NULL; 429 430 ASSERT(so->so_ack_mp == NULL); 431 ASSERT(so->so_conn_ind_head == NULL); 432 ASSERT(so->so_conn_ind_tail == NULL); 433 ASSERT(so->so_ux_bound_vp == NULL); 434 ASSERT(so->so_unbind_mp == NULL); 435 436 vn_reinit(vp); 437 vp->v_vfsp = rootvfs; 438 vp->v_type = VSOCK; 439 vp->v_rdev = so->so_dev; 440 vn_exists(vp); 441 442 return (vp); 443 } 444 445 void 446 sockfree(struct sonode *so) 447 { 448 mblk_t *mp; 449 vnode_t *vp; 450 451 ASSERT(so->so_count == 0); 452 ASSERT(so->so_accessvp); 453 ASSERT(so->so_discon_ind_mp == NULL); 454 455 vp = so->so_accessvp; 456 VN_RELE(vp); 457 458 /* 459 * Protect so->so_[lf]addr_sa so that sockfs_snapshot() can safely 460 * indirect them. It also uses so_accessvp as a validity test. 461 */ 462 mutex_enter(&so->so_lock); 463 464 so->so_accessvp = NULL; 465 466 if (so->so_laddr_sa) { 467 ASSERT((caddr_t)so->so_faddr_sa == 468 (caddr_t)so->so_laddr_sa + so->so_laddr_maxlen); 469 ASSERT(so->so_faddr_maxlen == so->so_laddr_maxlen); 470 so->so_state &= ~(SS_LADDR_VALID | SS_FADDR_VALID); 471 kmem_free(so->so_laddr_sa, so->so_laddr_maxlen * 2); 472 so->so_laddr_sa = NULL; 473 so->so_laddr_len = so->so_laddr_maxlen = 0; 474 so->so_faddr_sa = NULL; 475 so->so_faddr_len = so->so_faddr_maxlen = 0; 476 } 477 478 mutex_exit(&so->so_lock); 479 480 if ((mp = so->so_eaddr_mp) != NULL) { 481 freemsg(mp); 482 so->so_eaddr_mp = NULL; 483 so->so_delayed_error = 0; 484 } 485 if ((mp = so->so_ack_mp) != NULL) { 486 freemsg(mp); 487 so->so_ack_mp = NULL; 488 } 489 if ((mp = so->so_conn_ind_head) != NULL) { 490 mblk_t *mp1; 491 492 while (mp) { 493 mp1 = mp->b_next; 494 mp->b_next = NULL; 495 freemsg(mp); 496 mp = mp1; 497 } 498 so->so_conn_ind_head = so->so_conn_ind_tail = NULL; 499 so->so_state &= ~SS_HASCONNIND; 500 } 501 #ifdef DEBUG 502 mutex_enter(&so->so_lock); 503 ASSERT(so_verify_oobstate(so)); 504 mutex_exit(&so->so_lock); 505 #endif /* DEBUG */ 506 if ((mp = so->so_oobmsg) != NULL) { 507 freemsg(mp); 508 so->so_oobmsg = NULL; 509 so->so_state &= ~(SS_OOBPEND|SS_HAVEOOBDATA|SS_HADOOBDATA); 510 } 511 512 if ((mp = so->so_nl7c_rcv_mp) != NULL) { 513 so->so_nl7c_rcv_mp = NULL; 514 freemsg(mp); 515 } 516 so->so_nl7c_rcv_rval = 0; 517 if (so->so_nl7c_uri != NULL) { 518 nl7c_urifree(so); 519 /* urifree() cleared nl7c_uri */ 520 } 521 if (so->so_nl7c_flags) { 522 so->so_nl7c_flags = 0; 523 } 524 525 ASSERT(so->so_ux_bound_vp == NULL); 526 if ((mp = so->so_unbind_mp) != NULL) { 527 freemsg(mp); 528 so->so_unbind_mp = NULL; 529 } 530 vn_invalid(SOTOV(so)); 531 532 if (so->so_peercred != NULL) 533 crfree(so->so_peercred); 534 535 kmem_cache_free(so->so_cache, so->so_obj); 536 } 537 538 /* 539 * Update the accessed, updated, or changed times in an sonode 540 * with the current time. 541 * 542 * Note that both SunOS 4.X and 4.4BSD sockets do not present reasonable 543 * attributes in a fstat call. (They return the current time and 0 for 544 * all timestamps, respectively.) We maintain the current timestamps 545 * here primarily so that should sockmod be popped the resulting 546 * file descriptor will behave like a stream w.r.t. the timestamps. 547 */ 548 void 549 so_update_attrs(struct sonode *so, int flag) 550 { 551 time_t now = gethrestime_sec(); 552 553 mutex_enter(&so->so_lock); 554 so->so_flag |= flag; 555 if (flag & SOACC) 556 so->so_atime = now; 557 if (flag & SOMOD) 558 so->so_mtime = now; 559 mutex_exit(&so->so_lock); 560 } 561 562 /*ARGSUSED*/ 563 static int 564 socktpi_constructor(void *buf, void *cdrarg, int kmflags) 565 { 566 struct sonode *so = buf; 567 struct vnode *vp; 568 569 so->so_nl7c_flags = 0; 570 so->so_nl7c_uri = NULL; 571 so->so_nl7c_rcv_mp = NULL; 572 573 so->so_oobmsg = NULL; 574 so->so_ack_mp = NULL; 575 so->so_conn_ind_head = NULL; 576 so->so_conn_ind_tail = NULL; 577 so->so_discon_ind_mp = NULL; 578 so->so_ux_bound_vp = NULL; 579 so->so_unbind_mp = NULL; 580 so->so_accessvp = NULL; 581 so->so_laddr_sa = NULL; 582 so->so_faddr_sa = NULL; 583 so->so_ops = &sotpi_sonodeops; 584 585 vp = vn_alloc(KM_SLEEP); 586 so->so_vnode = vp; 587 588 vn_setops(vp, socktpi_vnodeops); 589 vp->v_data = (caddr_t)so; 590 591 mutex_init(&so->so_lock, NULL, MUTEX_DEFAULT, NULL); 592 mutex_init(&so->so_plumb_lock, NULL, MUTEX_DEFAULT, NULL); 593 cv_init(&so->so_state_cv, NULL, CV_DEFAULT, NULL); 594 cv_init(&so->so_ack_cv, NULL, CV_DEFAULT, NULL); 595 cv_init(&so->so_connind_cv, NULL, CV_DEFAULT, NULL); 596 cv_init(&so->so_want_cv, NULL, CV_DEFAULT, NULL); 597 598 return (0); 599 } 600 601 /*ARGSUSED1*/ 602 static void 603 socktpi_destructor(void *buf, void *cdrarg) 604 { 605 struct sonode *so = buf; 606 struct vnode *vp = SOTOV(so); 607 608 ASSERT(so->so_nl7c_flags == 0); 609 ASSERT(so->so_nl7c_uri == NULL); 610 ASSERT(so->so_nl7c_rcv_mp == NULL); 611 612 ASSERT(so->so_oobmsg == NULL); 613 ASSERT(so->so_ack_mp == NULL); 614 ASSERT(so->so_conn_ind_head == NULL); 615 ASSERT(so->so_conn_ind_tail == NULL); 616 ASSERT(so->so_discon_ind_mp == NULL); 617 ASSERT(so->so_ux_bound_vp == NULL); 618 ASSERT(so->so_unbind_mp == NULL); 619 ASSERT(so->so_ops == &sotpi_sonodeops); 620 621 ASSERT(vn_matchops(vp, socktpi_vnodeops)); 622 ASSERT(vp->v_data == (caddr_t)so); 623 624 vn_free(vp); 625 626 mutex_destroy(&so->so_lock); 627 mutex_destroy(&so->so_plumb_lock); 628 cv_destroy(&so->so_state_cv); 629 cv_destroy(&so->so_ack_cv); 630 cv_destroy(&so->so_connind_cv); 631 cv_destroy(&so->so_want_cv); 632 } 633 634 static int 635 socktpi_unix_constructor(void *buf, void *cdrarg, int kmflags) 636 { 637 int retval; 638 639 if ((retval = socktpi_constructor(buf, cdrarg, kmflags)) == 0) { 640 struct sonode *so = (struct sonode *)buf; 641 642 mutex_enter(&socklist.sl_lock); 643 644 so->so_next = socklist.sl_list; 645 so->so_prev = NULL; 646 if (so->so_next != NULL) 647 so->so_next->so_prev = so; 648 socklist.sl_list = so; 649 650 mutex_exit(&socklist.sl_lock); 651 652 } 653 return (retval); 654 } 655 656 static void 657 socktpi_unix_destructor(void *buf, void *cdrarg) 658 { 659 struct sonode *so = (struct sonode *)buf; 660 661 mutex_enter(&socklist.sl_lock); 662 663 if (so->so_next != NULL) 664 so->so_next->so_prev = so->so_prev; 665 if (so->so_prev != NULL) 666 so->so_prev->so_next = so->so_next; 667 else 668 socklist.sl_list = so->so_next; 669 670 mutex_exit(&socklist.sl_lock); 671 672 socktpi_destructor(buf, cdrarg); 673 } 674 675 /* 676 * Init function called when sockfs is loaded. 677 */ 678 int 679 sockinit(int fstype, char *name) 680 { 681 static const fs_operation_def_t sock_vfsops_template[] = { 682 NULL, NULL 683 }; 684 int error; 685 major_t dev; 686 char *err_str; 687 688 error = vfs_setfsops(fstype, sock_vfsops_template, NULL); 689 if (error != 0) { 690 zcmn_err(GLOBAL_ZONEID, CE_WARN, 691 "sockinit: bad vfs ops template"); 692 return (error); 693 } 694 695 error = vn_make_ops(name, socktpi_vnodeops_template, &socktpi_vnodeops); 696 if (error != 0) { 697 err_str = "sockinit: bad sock vnode ops template"; 698 /* vn_make_ops() does not reset socktpi_vnodeops on failure. */ 699 socktpi_vnodeops = NULL; 700 goto failure; 701 } 702 703 error = sosctp_init(); 704 if (error != 0) { 705 err_str = NULL; 706 goto failure; 707 } 708 709 /* 710 * Create sonode caches. We create a special one for AF_UNIX so 711 * that we can track them for netstat(1m). 712 */ 713 socktpi_cache = kmem_cache_create("socktpi_cache", 714 sizeof (struct sonode), 0, socktpi_constructor, 715 socktpi_destructor, NULL, NULL, NULL, 0); 716 717 socktpi_unix_cache = kmem_cache_create("socktpi_unix_cache", 718 sizeof (struct sonode), 0, socktpi_unix_constructor, 719 socktpi_unix_destructor, NULL, NULL, NULL, 0); 720 721 /* 722 * Build initial list mapping socket parameters to vnode. 723 */ 724 rw_init(&splist_lock, NULL, RW_DEFAULT, NULL); 725 726 /* 727 * If sockets are needed before init runs /sbin/soconfig 728 * it is possible to preload the sockparams list here using 729 * calls like: 730 * sockconfig(1,2,3, "/dev/tcp", 0); 731 */ 732 733 /* 734 * Create a unique dev_t for use in so_fsid. 735 */ 736 737 if ((dev = getudev()) == (major_t)-1) 738 dev = 0; 739 sockdev = makedevice(dev, 0); 740 741 mutex_init(&socklist.sl_lock, NULL, MUTEX_DEFAULT, NULL); 742 sendfile_init(); 743 nl7c_init(); 744 745 return (0); 746 747 failure: 748 (void) vfs_freevfsops_by_type(fstype); 749 if (socktpi_vnodeops != NULL) 750 vn_freevnodeops(socktpi_vnodeops); 751 if (err_str != NULL) 752 zcmn_err(GLOBAL_ZONEID, CE_WARN, err_str); 753 return (error); 754 } 755 756 /* 757 * Caller must hold the mutex. Used to set SOLOCKED. 758 */ 759 void 760 so_lock_single(struct sonode *so) 761 { 762 ASSERT(MUTEX_HELD(&so->so_lock)); 763 764 while (so->so_flag & (SOLOCKED | SOASYNC_UNBIND)) { 765 so->so_flag |= SOWANT; 766 cv_wait_stop(&so->so_want_cv, &so->so_lock, 767 SO_LOCK_WAKEUP_TIME); 768 } 769 so->so_flag |= SOLOCKED; 770 } 771 772 /* 773 * Caller must hold the mutex and pass in SOLOCKED or SOASYNC_UNBIND. 774 * Used to clear SOLOCKED or SOASYNC_UNBIND. 775 */ 776 void 777 so_unlock_single(struct sonode *so, int flag) 778 { 779 ASSERT(MUTEX_HELD(&so->so_lock)); 780 ASSERT(flag & (SOLOCKED|SOASYNC_UNBIND)); 781 ASSERT((flag & ~(SOLOCKED|SOASYNC_UNBIND)) == 0); 782 ASSERT(so->so_flag & flag); 783 784 /* 785 * Process the T_DISCON_IND on so_discon_ind_mp. 786 * 787 * Call to so_drain_discon_ind will result in so_lock 788 * being dropped and re-acquired later. 789 */ 790 if (so->so_discon_ind_mp != NULL) 791 so_drain_discon_ind(so); 792 793 if (so->so_flag & SOWANT) 794 cv_broadcast(&so->so_want_cv); 795 so->so_flag &= ~(SOWANT|flag); 796 } 797 798 /* 799 * Caller must hold the mutex. Used to set SOREADLOCKED. 800 * If the caller wants nonblocking behavior it should set fmode. 801 */ 802 int 803 so_lock_read(struct sonode *so, int fmode) 804 { 805 ASSERT(MUTEX_HELD(&so->so_lock)); 806 807 while (so->so_flag & SOREADLOCKED) { 808 if (fmode & (FNDELAY|FNONBLOCK)) 809 return (EWOULDBLOCK); 810 so->so_flag |= SOWANT; 811 cv_wait_stop(&so->so_want_cv, &so->so_lock, 812 SO_LOCK_WAKEUP_TIME); 813 } 814 so->so_flag |= SOREADLOCKED; 815 return (0); 816 } 817 818 /* 819 * Like so_lock_read above but allows signals. 820 */ 821 int 822 so_lock_read_intr(struct sonode *so, int fmode) 823 { 824 ASSERT(MUTEX_HELD(&so->so_lock)); 825 826 while (so->so_flag & SOREADLOCKED) { 827 if (fmode & (FNDELAY|FNONBLOCK)) 828 return (EWOULDBLOCK); 829 so->so_flag |= SOWANT; 830 if (!cv_wait_sig(&so->so_want_cv, &so->so_lock)) 831 return (EINTR); 832 } 833 so->so_flag |= SOREADLOCKED; 834 return (0); 835 } 836 837 /* 838 * Caller must hold the mutex. Used to clear SOREADLOCKED, 839 * set in so_lock_read() or so_lock_read_intr(). 840 */ 841 void 842 so_unlock_read(struct sonode *so) 843 { 844 ASSERT(MUTEX_HELD(&so->so_lock)); 845 ASSERT(so->so_flag & SOREADLOCKED); 846 847 if (so->so_flag & SOWANT) 848 cv_broadcast(&so->so_want_cv); 849 so->so_flag &= ~(SOWANT|SOREADLOCKED); 850 } 851 852 /* 853 * Verify that the specified offset falls within the mblk and 854 * that the resulting pointer is aligned. 855 * Returns NULL if not. 856 */ 857 void * 858 sogetoff(mblk_t *mp, t_uscalar_t offset, 859 t_uscalar_t length, uint_t align_size) 860 { 861 uintptr_t ptr1, ptr2; 862 863 ASSERT(mp && mp->b_wptr >= mp->b_rptr); 864 ptr1 = (uintptr_t)mp->b_rptr + offset; 865 ptr2 = (uintptr_t)ptr1 + length; 866 if (ptr1 < (uintptr_t)mp->b_rptr || ptr2 > (uintptr_t)mp->b_wptr) { 867 eprintline(0); 868 return (NULL); 869 } 870 if ((ptr1 & (align_size - 1)) != 0) { 871 eprintline(0); 872 return (NULL); 873 } 874 return ((void *)ptr1); 875 } 876 877 /* 878 * Return the AF_UNIX underlying filesystem vnode matching a given name. 879 * Makes sure the sending and the destination sonodes are compatible. 880 * The vnode is returned held. 881 * 882 * The underlying filesystem VSOCK vnode has a v_stream pointer that 883 * references the actual stream head (hence indirectly the actual sonode). 884 */ 885 static int 886 so_ux_lookup(struct sonode *so, struct sockaddr_un *soun, int checkaccess, 887 vnode_t **vpp) 888 { 889 vnode_t *vp; /* Underlying filesystem vnode */ 890 vnode_t *svp; /* sockfs vnode */ 891 struct sonode *so2; 892 int error; 893 894 dprintso(so, 1, ("so_ux_lookup(%p) name <%s>\n", 895 so, soun->sun_path)); 896 897 error = lookupname(soun->sun_path, UIO_SYSSPACE, FOLLOW, NULLVPP, &vp); 898 if (error) { 899 eprintsoline(so, error); 900 return (error); 901 } 902 if (vp->v_type != VSOCK) { 903 error = ENOTSOCK; 904 eprintsoline(so, error); 905 goto done2; 906 } 907 908 if (checkaccess) { 909 /* 910 * Check that we have permissions to access the destination 911 * vnode. This check is not done in BSD but it is required 912 * by X/Open. 913 */ 914 if (error = VOP_ACCESS(vp, VREAD|VWRITE, 0, CRED())) { 915 eprintsoline(so, error); 916 goto done2; 917 } 918 } 919 920 /* 921 * Check if the remote socket has been closed. 922 * 923 * Synchronize with vn_rele_stream by holding v_lock while traversing 924 * v_stream->sd_vnode. 925 */ 926 mutex_enter(&vp->v_lock); 927 if (vp->v_stream == NULL) { 928 mutex_exit(&vp->v_lock); 929 if (so->so_type == SOCK_DGRAM) 930 error = EDESTADDRREQ; 931 else 932 error = ECONNREFUSED; 933 934 eprintsoline(so, error); 935 goto done2; 936 } 937 ASSERT(vp->v_stream->sd_vnode); 938 svp = vp->v_stream->sd_vnode; 939 /* 940 * holding v_lock on underlying filesystem vnode and acquiring 941 * it on sockfs vnode. Assumes that no code ever attempts to 942 * acquire these locks in the reverse order. 943 */ 944 VN_HOLD(svp); 945 mutex_exit(&vp->v_lock); 946 947 if (svp->v_type != VSOCK) { 948 error = ENOTSOCK; 949 eprintsoline(so, error); 950 goto done; 951 } 952 953 so2 = VTOSO(svp); 954 955 if (so->so_type != so2->so_type) { 956 error = EPROTOTYPE; 957 eprintsoline(so, error); 958 goto done; 959 } 960 961 VN_RELE(svp); 962 *vpp = vp; 963 return (0); 964 965 done: 966 VN_RELE(svp); 967 done2: 968 VN_RELE(vp); 969 return (error); 970 } 971 972 /* 973 * Verify peer address for connect and sendto/sendmsg. 974 * Since sendto/sendmsg would not get synchronous errors from the transport 975 * provider we have to do these ugly checks in the socket layer to 976 * preserve compatibility with SunOS 4.X. 977 */ 978 int 979 so_addr_verify(struct sonode *so, const struct sockaddr *name, 980 socklen_t namelen) 981 { 982 int family; 983 984 dprintso(so, 1, ("so_addr_verify(%p, %p, %d)\n", so, name, namelen)); 985 986 ASSERT(name != NULL); 987 988 family = so->so_family; 989 switch (family) { 990 case AF_INET: 991 if (name->sa_family != family) { 992 eprintsoline(so, EAFNOSUPPORT); 993 return (EAFNOSUPPORT); 994 } 995 if (namelen != (socklen_t)sizeof (struct sockaddr_in)) { 996 eprintsoline(so, EINVAL); 997 return (EINVAL); 998 } 999 break; 1000 case AF_INET6: { 1001 #ifdef DEBUG 1002 struct sockaddr_in6 *sin6; 1003 #endif /* DEBUG */ 1004 1005 if (name->sa_family != family) { 1006 eprintsoline(so, EAFNOSUPPORT); 1007 return (EAFNOSUPPORT); 1008 } 1009 if (namelen != (socklen_t)sizeof (struct sockaddr_in6)) { 1010 eprintsoline(so, EINVAL); 1011 return (EINVAL); 1012 } 1013 #ifdef DEBUG 1014 /* Verify that apps don't forget to clear sin6_scope_id etc */ 1015 sin6 = (struct sockaddr_in6 *)name; 1016 if (sin6->sin6_scope_id != 0 && 1017 !IN6_IS_ADDR_LINKSCOPE(&sin6->sin6_addr)) { 1018 zcmn_err(getzoneid(), CE_WARN, 1019 "connect/send* with uninitialized sin6_scope_id " 1020 "(%d) on socket. Pid = %d\n", 1021 (int)sin6->sin6_scope_id, (int)curproc->p_pid); 1022 } 1023 #endif /* DEBUG */ 1024 break; 1025 } 1026 case AF_UNIX: 1027 if (so->so_state & SS_FADDR_NOXLATE) { 1028 return (0); 1029 } 1030 if (namelen < (socklen_t)sizeof (short)) { 1031 eprintsoline(so, ENOENT); 1032 return (ENOENT); 1033 } 1034 if (name->sa_family != family) { 1035 eprintsoline(so, EAFNOSUPPORT); 1036 return (EAFNOSUPPORT); 1037 } 1038 /* MAXPATHLEN + soun_family + nul termination */ 1039 if (namelen > (socklen_t)(MAXPATHLEN + sizeof (short) + 1)) { 1040 eprintsoline(so, ENAMETOOLONG); 1041 return (ENAMETOOLONG); 1042 } 1043 1044 break; 1045 1046 default: 1047 /* 1048 * Default is don't do any length or sa_family check 1049 * to allow non-sockaddr style addresses. 1050 */ 1051 break; 1052 } 1053 1054 return (0); 1055 } 1056 1057 1058 /* 1059 * Translate an AF_UNIX sockaddr_un to the transport internal name. 1060 * Assumes caller has called so_addr_verify first. 1061 */ 1062 /*ARGSUSED*/ 1063 int 1064 so_ux_addr_xlate(struct sonode *so, struct sockaddr *name, 1065 socklen_t namelen, int checkaccess, 1066 void **addrp, socklen_t *addrlenp) 1067 { 1068 int error; 1069 struct sockaddr_un *soun; 1070 vnode_t *vp; 1071 void *addr; 1072 socklen_t addrlen; 1073 1074 dprintso(so, 1, ("so_ux_addr_xlate(%p, %p, %d, %d)\n", 1075 so, name, namelen, checkaccess)); 1076 1077 ASSERT(name != NULL); 1078 ASSERT(so->so_family == AF_UNIX); 1079 ASSERT(!(so->so_state & SS_FADDR_NOXLATE)); 1080 ASSERT(namelen >= (socklen_t)sizeof (short)); 1081 ASSERT(name->sa_family == AF_UNIX); 1082 soun = (struct sockaddr_un *)name; 1083 /* 1084 * Lookup vnode for the specified path name and verify that 1085 * it is a socket. 1086 */ 1087 error = so_ux_lookup(so, soun, checkaccess, &vp); 1088 if (error) { 1089 eprintsoline(so, error); 1090 return (error); 1091 } 1092 /* 1093 * Use the address of the peer vnode as the address to send 1094 * to. We release the peer vnode here. In case it has been 1095 * closed by the time the T_CONN_REQ or T_UNIDATA_REQ reaches the 1096 * transport the message will get an error or be dropped. 1097 */ 1098 so->so_ux_faddr.soua_vp = vp; 1099 so->so_ux_faddr.soua_magic = SOU_MAGIC_EXPLICIT; 1100 addr = &so->so_ux_faddr; 1101 addrlen = (socklen_t)sizeof (so->so_ux_faddr); 1102 dprintso(so, 1, ("ux_xlate UNIX: addrlen %d, vp %p\n", 1103 addrlen, vp)); 1104 VN_RELE(vp); 1105 *addrp = addr; 1106 *addrlenp = (socklen_t)addrlen; 1107 return (0); 1108 } 1109 1110 /* 1111 * Esballoc free function for messages that contain SO_FILEP option. 1112 * Decrement the reference count on the file pointers using closef. 1113 */ 1114 void 1115 fdbuf_free(struct fdbuf *fdbuf) 1116 { 1117 int i; 1118 struct file *fp; 1119 1120 dprint(1, ("fdbuf_free: %d fds\n", fdbuf->fd_numfd)); 1121 for (i = 0; i < fdbuf->fd_numfd; i++) { 1122 /* 1123 * We need pointer size alignment for fd_fds. On a LP64 1124 * kernel, the required alignment is 8 bytes while 1125 * the option headers and values are only 4 bytes 1126 * aligned. So its safer to do a bcopy compared to 1127 * assigning fdbuf->fd_fds[i] to fp. 1128 */ 1129 bcopy((char *)&fdbuf->fd_fds[i], (char *)&fp, sizeof (fp)); 1130 dprint(1, ("fdbuf_free: [%d] = %p\n", i, fp)); 1131 (void) closef(fp); 1132 } 1133 if (fdbuf->fd_ebuf != NULL) 1134 kmem_free(fdbuf->fd_ebuf, fdbuf->fd_ebuflen); 1135 kmem_free(fdbuf, fdbuf->fd_size); 1136 } 1137 1138 /* 1139 * Allocate an esballoc'ed message for AF_UNIX file descriptor passing. 1140 * Waits if memory is not available. 1141 */ 1142 mblk_t * 1143 fdbuf_allocmsg(int size, struct fdbuf *fdbuf) 1144 { 1145 uchar_t *buf; 1146 mblk_t *mp; 1147 1148 dprint(1, ("fdbuf_allocmsg: size %d, %d fds\n", size, fdbuf->fd_numfd)); 1149 buf = kmem_alloc(size, KM_SLEEP); 1150 fdbuf->fd_ebuf = (caddr_t)buf; 1151 fdbuf->fd_ebuflen = size; 1152 fdbuf->fd_frtn.free_func = fdbuf_free; 1153 fdbuf->fd_frtn.free_arg = (caddr_t)fdbuf; 1154 1155 mp = esballoc_wait(buf, size, BPRI_MED, &fdbuf->fd_frtn); 1156 mp->b_datap->db_type = M_PROTO; 1157 return (mp); 1158 } 1159 1160 /* 1161 * Extract file descriptors from a fdbuf. 1162 * Return list in rights/rightslen. 1163 */ 1164 /*ARGSUSED*/ 1165 static int 1166 fdbuf_extract(struct fdbuf *fdbuf, void *rights, int rightslen) 1167 { 1168 int i, fd; 1169 int *rp; 1170 struct file *fp; 1171 int numfd; 1172 1173 dprint(1, ("fdbuf_extract: %d fds, len %d\n", 1174 fdbuf->fd_numfd, rightslen)); 1175 1176 numfd = fdbuf->fd_numfd; 1177 ASSERT(rightslen == numfd * (int)sizeof (int)); 1178 1179 /* 1180 * Allocate a file descriptor and increment the f_count. 1181 * The latter is needed since we always call fdbuf_free 1182 * which performs a closef. 1183 */ 1184 rp = (int *)rights; 1185 for (i = 0; i < numfd; i++) { 1186 if ((fd = ufalloc(0)) == -1) 1187 goto cleanup; 1188 /* 1189 * We need pointer size alignment for fd_fds. On a LP64 1190 * kernel, the required alignment is 8 bytes while 1191 * the option headers and values are only 4 bytes 1192 * aligned. So its safer to do a bcopy compared to 1193 * assigning fdbuf->fd_fds[i] to fp. 1194 */ 1195 bcopy((char *)&fdbuf->fd_fds[i], (char *)&fp, sizeof (fp)); 1196 mutex_enter(&fp->f_tlock); 1197 fp->f_count++; 1198 mutex_exit(&fp->f_tlock); 1199 setf(fd, fp); 1200 *rp++ = fd; 1201 #ifdef C2_AUDIT 1202 if (audit_active) 1203 audit_fdrecv(fd, fp); 1204 #endif 1205 dprint(1, ("fdbuf_extract: [%d] = %d, %p refcnt %d\n", 1206 i, fd, fp, fp->f_count)); 1207 } 1208 return (0); 1209 1210 cleanup: 1211 /* 1212 * Undo whatever partial work the loop above has done. 1213 */ 1214 { 1215 int j; 1216 1217 rp = (int *)rights; 1218 for (j = 0; j < i; j++) { 1219 dprint(0, 1220 ("fdbuf_extract: cleanup[%d] = %d\n", j, *rp)); 1221 (void) closeandsetf(*rp++, NULL); 1222 } 1223 } 1224 1225 return (EMFILE); 1226 } 1227 1228 /* 1229 * Insert file descriptors into an fdbuf. 1230 * Returns a kmem_alloc'ed fdbuf. The fdbuf should be freed 1231 * by calling fdbuf_free(). 1232 */ 1233 int 1234 fdbuf_create(void *rights, int rightslen, struct fdbuf **fdbufp) 1235 { 1236 int numfd, i; 1237 int *fds; 1238 struct file *fp; 1239 struct fdbuf *fdbuf; 1240 int fdbufsize; 1241 1242 dprint(1, ("fdbuf_create: len %d\n", rightslen)); 1243 1244 numfd = rightslen / (int)sizeof (int); 1245 1246 fdbufsize = (int)FDBUF_HDRSIZE + (numfd * (int)sizeof (struct file *)); 1247 fdbuf = kmem_alloc(fdbufsize, KM_SLEEP); 1248 fdbuf->fd_size = fdbufsize; 1249 fdbuf->fd_numfd = 0; 1250 fdbuf->fd_ebuf = NULL; 1251 fdbuf->fd_ebuflen = 0; 1252 fds = (int *)rights; 1253 for (i = 0; i < numfd; i++) { 1254 if ((fp = getf(fds[i])) == NULL) { 1255 fdbuf_free(fdbuf); 1256 return (EBADF); 1257 } 1258 dprint(1, ("fdbuf_create: [%d] = %d, %p refcnt %d\n", 1259 i, fds[i], fp, fp->f_count)); 1260 mutex_enter(&fp->f_tlock); 1261 fp->f_count++; 1262 mutex_exit(&fp->f_tlock); 1263 /* 1264 * The maximum alignment for fdbuf (or any option header 1265 * and its value) it 4 bytes. On a LP64 kernel, the alignment 1266 * is not sufficient for pointers (fd_fds in this case). Since 1267 * we just did a kmem_alloc (we get a double word alignment), 1268 * we don't need to do anything on the send side (we loose 1269 * the double word alignment because fdbuf goes after an 1270 * option header (eg T_unitdata_req) which is only 4 byte 1271 * aligned). We take care of this when we extract the file 1272 * descriptor in fdbuf_extract or fdbuf_free. 1273 */ 1274 fdbuf->fd_fds[i] = fp; 1275 fdbuf->fd_numfd++; 1276 releasef(fds[i]); 1277 #ifdef C2_AUDIT 1278 if (audit_active) 1279 audit_fdsend(fds[i], fp, 0); 1280 #endif 1281 } 1282 *fdbufp = fdbuf; 1283 return (0); 1284 } 1285 1286 static int 1287 fdbuf_optlen(int rightslen) 1288 { 1289 int numfd; 1290 1291 numfd = rightslen / (int)sizeof (int); 1292 1293 return ((int)FDBUF_HDRSIZE + (numfd * (int)sizeof (struct file *))); 1294 } 1295 1296 static t_uscalar_t 1297 fdbuf_cmsglen(int fdbuflen) 1298 { 1299 return (t_uscalar_t)((fdbuflen - FDBUF_HDRSIZE) / 1300 (int)sizeof (struct file *) * (int)sizeof (int)); 1301 } 1302 1303 1304 /* 1305 * Return non-zero if the mblk and fdbuf are consistent. 1306 */ 1307 static int 1308 fdbuf_verify(mblk_t *mp, struct fdbuf *fdbuf, int fdbuflen) 1309 { 1310 if (fdbuflen >= FDBUF_HDRSIZE && 1311 fdbuflen == fdbuf->fd_size) { 1312 frtn_t *frp = mp->b_datap->db_frtnp; 1313 /* 1314 * Check that the SO_FILEP portion of the 1315 * message has not been modified by 1316 * the loopback transport. The sending sockfs generates 1317 * a message that is esballoc'ed with the free function 1318 * being fdbuf_free() and where free_arg contains the 1319 * identical information as the SO_FILEP content. 1320 * 1321 * If any of these constraints are not satisfied we 1322 * silently ignore the option. 1323 */ 1324 ASSERT(mp); 1325 if (frp != NULL && 1326 frp->free_func == fdbuf_free && 1327 frp->free_arg != NULL && 1328 bcmp(frp->free_arg, fdbuf, fdbuflen) == 0) { 1329 dprint(1, ("fdbuf_verify: fdbuf %p len %d\n", 1330 fdbuf, fdbuflen)); 1331 return (1); 1332 } else { 1333 zcmn_err(getzoneid(), CE_WARN, 1334 "sockfs: mismatched fdbuf content (%p)", 1335 (void *)mp); 1336 return (0); 1337 } 1338 } else { 1339 zcmn_err(getzoneid(), CE_WARN, 1340 "sockfs: mismatched fdbuf len %d, %d\n", 1341 fdbuflen, fdbuf->fd_size); 1342 return (0); 1343 } 1344 } 1345 1346 /* 1347 * When the file descriptors returned by sorecvmsg can not be passed 1348 * to the application this routine will cleanup the references on 1349 * the files. Start at startoff bytes into the buffer. 1350 */ 1351 static void 1352 close_fds(void *fdbuf, int fdbuflen, int startoff) 1353 { 1354 int *fds = (int *)fdbuf; 1355 int numfd = fdbuflen / (int)sizeof (int); 1356 int i; 1357 1358 dprint(1, ("close_fds(%p, %d, %d)\n", fdbuf, fdbuflen, startoff)); 1359 1360 for (i = 0; i < numfd; i++) { 1361 if (startoff < 0) 1362 startoff = 0; 1363 if (startoff < (int)sizeof (int)) { 1364 /* 1365 * This file descriptor is partially or fully after 1366 * the offset 1367 */ 1368 dprint(0, 1369 ("close_fds: cleanup[%d] = %d\n", i, fds[i])); 1370 (void) closeandsetf(fds[i], NULL); 1371 } 1372 startoff -= (int)sizeof (int); 1373 } 1374 } 1375 1376 /* 1377 * Close all file descriptors contained in the control part starting at 1378 * the startoffset. 1379 */ 1380 void 1381 so_closefds(void *control, t_uscalar_t controllen, int oldflg, 1382 int startoff) 1383 { 1384 struct cmsghdr *cmsg; 1385 1386 if (control == NULL) 1387 return; 1388 1389 if (oldflg) { 1390 close_fds(control, controllen, startoff); 1391 return; 1392 } 1393 /* Scan control part for file descriptors. */ 1394 for (cmsg = (struct cmsghdr *)control; 1395 CMSG_VALID(cmsg, control, (uintptr_t)control + controllen); 1396 cmsg = CMSG_NEXT(cmsg)) { 1397 if (cmsg->cmsg_level == SOL_SOCKET && 1398 cmsg->cmsg_type == SCM_RIGHTS) { 1399 close_fds(CMSG_CONTENT(cmsg), 1400 (int)CMSG_CONTENTLEN(cmsg), 1401 startoff - (int)sizeof (struct cmsghdr)); 1402 } 1403 startoff -= cmsg->cmsg_len; 1404 } 1405 } 1406 1407 /* 1408 * Returns a pointer/length for the file descriptors contained 1409 * in the control buffer. Returns with *fdlenp == -1 if there are no 1410 * file descriptor options present. This is different than there being 1411 * a zero-length file descriptor option. 1412 * Fail if there are multiple SCM_RIGHT cmsgs. 1413 */ 1414 int 1415 so_getfdopt(void *control, t_uscalar_t controllen, int oldflg, 1416 void **fdsp, int *fdlenp) 1417 { 1418 struct cmsghdr *cmsg; 1419 void *fds; 1420 int fdlen; 1421 1422 if (control == NULL) { 1423 *fdsp = NULL; 1424 *fdlenp = -1; 1425 return (0); 1426 } 1427 1428 if (oldflg) { 1429 *fdsp = control; 1430 if (controllen == 0) 1431 *fdlenp = -1; 1432 else 1433 *fdlenp = controllen; 1434 dprint(1, ("so_getfdopt: old %d\n", *fdlenp)); 1435 return (0); 1436 } 1437 1438 fds = NULL; 1439 fdlen = 0; 1440 1441 for (cmsg = (struct cmsghdr *)control; 1442 CMSG_VALID(cmsg, control, (uintptr_t)control + controllen); 1443 cmsg = CMSG_NEXT(cmsg)) { 1444 if (cmsg->cmsg_level == SOL_SOCKET && 1445 cmsg->cmsg_type == SCM_RIGHTS) { 1446 if (fds != NULL) 1447 return (EINVAL); 1448 fds = CMSG_CONTENT(cmsg); 1449 fdlen = (int)CMSG_CONTENTLEN(cmsg); 1450 dprint(1, ("so_getfdopt: new %lu\n", 1451 (size_t)CMSG_CONTENTLEN(cmsg))); 1452 } 1453 } 1454 if (fds == NULL) { 1455 dprint(1, ("so_getfdopt: NONE\n")); 1456 *fdlenp = -1; 1457 } else 1458 *fdlenp = fdlen; 1459 *fdsp = fds; 1460 return (0); 1461 } 1462 1463 /* 1464 * Return the length of the options including any file descriptor options. 1465 */ 1466 t_uscalar_t 1467 so_optlen(void *control, t_uscalar_t controllen, int oldflg) 1468 { 1469 struct cmsghdr *cmsg; 1470 t_uscalar_t optlen = 0; 1471 t_uscalar_t len; 1472 1473 if (control == NULL) 1474 return (0); 1475 1476 if (oldflg) 1477 return ((t_uscalar_t)(sizeof (struct T_opthdr) + 1478 fdbuf_optlen(controllen))); 1479 1480 for (cmsg = (struct cmsghdr *)control; 1481 CMSG_VALID(cmsg, control, (uintptr_t)control + controllen); 1482 cmsg = CMSG_NEXT(cmsg)) { 1483 if (cmsg->cmsg_level == SOL_SOCKET && 1484 cmsg->cmsg_type == SCM_RIGHTS) { 1485 len = fdbuf_optlen((int)CMSG_CONTENTLEN(cmsg)); 1486 } else { 1487 len = (t_uscalar_t)CMSG_CONTENTLEN(cmsg); 1488 } 1489 optlen += (t_uscalar_t)(_TPI_ALIGN_TOPT(len) + 1490 sizeof (struct T_opthdr)); 1491 } 1492 dprint(1, ("so_optlen: controllen %d, flg %d -> optlen %d\n", 1493 controllen, oldflg, optlen)); 1494 return (optlen); 1495 } 1496 1497 /* 1498 * Copy options from control to the mblk. Skip any file descriptor options. 1499 */ 1500 void 1501 so_cmsg2opt(void *control, t_uscalar_t controllen, int oldflg, mblk_t *mp) 1502 { 1503 struct T_opthdr toh; 1504 struct cmsghdr *cmsg; 1505 1506 if (control == NULL) 1507 return; 1508 1509 if (oldflg) { 1510 /* No real options - caller has handled file descriptors */ 1511 return; 1512 } 1513 for (cmsg = (struct cmsghdr *)control; 1514 CMSG_VALID(cmsg, control, (uintptr_t)control + controllen); 1515 cmsg = CMSG_NEXT(cmsg)) { 1516 /* 1517 * Note: The caller handles file descriptors prior 1518 * to calling this function. 1519 */ 1520 t_uscalar_t len; 1521 1522 if (cmsg->cmsg_level == SOL_SOCKET && 1523 cmsg->cmsg_type == SCM_RIGHTS) 1524 continue; 1525 1526 len = (t_uscalar_t)CMSG_CONTENTLEN(cmsg); 1527 toh.level = cmsg->cmsg_level; 1528 toh.name = cmsg->cmsg_type; 1529 toh.len = len + (t_uscalar_t)sizeof (struct T_opthdr); 1530 toh.status = 0; 1531 1532 soappendmsg(mp, &toh, sizeof (toh)); 1533 soappendmsg(mp, CMSG_CONTENT(cmsg), len); 1534 mp->b_wptr += _TPI_ALIGN_TOPT(len) - len; 1535 ASSERT(mp->b_wptr <= mp->b_datap->db_lim); 1536 } 1537 } 1538 1539 /* 1540 * Return the length of the control message derived from the options. 1541 * Exclude SO_SRCADDR and SO_UNIX_CLOSE options. Include SO_FILEP. 1542 * When oldflg is set only include SO_FILEP. 1543 */ 1544 t_uscalar_t 1545 so_cmsglen(mblk_t *mp, void *opt, t_uscalar_t optlen, int oldflg) 1546 { 1547 t_uscalar_t cmsglen = 0; 1548 struct T_opthdr *tohp; 1549 t_uscalar_t len; 1550 t_uscalar_t last_roundup = 0; 1551 1552 ASSERT(__TPI_TOPT_ISALIGNED(opt)); 1553 1554 for (tohp = (struct T_opthdr *)opt; 1555 tohp && _TPI_TOPT_VALID(tohp, opt, (uintptr_t)opt + optlen); 1556 tohp = _TPI_TOPT_NEXTHDR(opt, optlen, tohp)) { 1557 dprint(1, ("so_cmsglen: level 0x%x, name %d, len %d\n", 1558 tohp->level, tohp->name, tohp->len)); 1559 if (tohp->level == SOL_SOCKET && 1560 (tohp->name == SO_SRCADDR || 1561 tohp->name == SO_UNIX_CLOSE)) { 1562 continue; 1563 } 1564 if (tohp->level == SOL_SOCKET && tohp->name == SO_FILEP) { 1565 struct fdbuf *fdbuf; 1566 int fdbuflen; 1567 1568 fdbuf = (struct fdbuf *)_TPI_TOPT_DATA(tohp); 1569 fdbuflen = (int)_TPI_TOPT_DATALEN(tohp); 1570 1571 if (!fdbuf_verify(mp, fdbuf, fdbuflen)) 1572 continue; 1573 if (oldflg) { 1574 cmsglen += fdbuf_cmsglen(fdbuflen); 1575 continue; 1576 } 1577 len = fdbuf_cmsglen(fdbuflen); 1578 } else { 1579 if (oldflg) 1580 continue; 1581 len = (t_uscalar_t)_TPI_TOPT_DATALEN(tohp); 1582 } 1583 /* 1584 * Exlucde roundup for last option to not set 1585 * MSG_CTRUNC when the cmsg fits but the padding doesn't fit. 1586 */ 1587 last_roundup = (t_uscalar_t) 1588 (ROUNDUP_cmsglen(len + (int)sizeof (struct cmsghdr)) - 1589 (len + (int)sizeof (struct cmsghdr))); 1590 cmsglen += (t_uscalar_t)(len + (int)sizeof (struct cmsghdr)) + 1591 last_roundup; 1592 } 1593 cmsglen -= last_roundup; 1594 dprint(1, ("so_cmsglen: optlen %d, flg %d -> cmsglen %d\n", 1595 optlen, oldflg, cmsglen)); 1596 return (cmsglen); 1597 } 1598 1599 /* 1600 * Copy options from options to the control. Convert SO_FILEP to 1601 * file descriptors. 1602 * Returns errno or zero. 1603 */ 1604 int 1605 so_opt2cmsg(mblk_t *mp, void *opt, t_uscalar_t optlen, int oldflg, 1606 void *control, t_uscalar_t controllen) 1607 { 1608 struct T_opthdr *tohp; 1609 struct cmsghdr *cmsg; 1610 struct fdbuf *fdbuf; 1611 int fdbuflen; 1612 int error; 1613 1614 cmsg = (struct cmsghdr *)control; 1615 1616 ASSERT(__TPI_TOPT_ISALIGNED(opt)); 1617 1618 for (tohp = (struct T_opthdr *)opt; 1619 tohp && _TPI_TOPT_VALID(tohp, opt, (uintptr_t)opt + optlen); 1620 tohp = _TPI_TOPT_NEXTHDR(opt, optlen, tohp)) { 1621 dprint(1, ("so_opt2cmsg: level 0x%x, name %d, len %d\n", 1622 tohp->level, tohp->name, tohp->len)); 1623 1624 if (tohp->level == SOL_SOCKET && 1625 (tohp->name == SO_SRCADDR || 1626 tohp->name == SO_UNIX_CLOSE)) { 1627 continue; 1628 } 1629 ASSERT((uintptr_t)cmsg <= (uintptr_t)control + controllen); 1630 if (tohp->level == SOL_SOCKET && tohp->name == SO_FILEP) { 1631 fdbuf = (struct fdbuf *)_TPI_TOPT_DATA(tohp); 1632 fdbuflen = (int)_TPI_TOPT_DATALEN(tohp); 1633 1634 if (!fdbuf_verify(mp, fdbuf, fdbuflen)) 1635 return (EPROTO); 1636 if (oldflg) { 1637 error = fdbuf_extract(fdbuf, control, 1638 (int)controllen); 1639 if (error != 0) 1640 return (error); 1641 continue; 1642 } else { 1643 int fdlen; 1644 1645 fdlen = (int)fdbuf_cmsglen( 1646 (int)_TPI_TOPT_DATALEN(tohp)); 1647 1648 cmsg->cmsg_level = tohp->level; 1649 cmsg->cmsg_type = SCM_RIGHTS; 1650 cmsg->cmsg_len = (socklen_t)(fdlen + 1651 sizeof (struct cmsghdr)); 1652 1653 error = fdbuf_extract(fdbuf, 1654 CMSG_CONTENT(cmsg), fdlen); 1655 if (error != 0) 1656 return (error); 1657 } 1658 } else if (tohp->level == SOL_SOCKET && 1659 tohp->name == SCM_TIMESTAMP) { 1660 timestruc_t *timestamp; 1661 1662 if (oldflg) 1663 continue; 1664 1665 cmsg->cmsg_level = tohp->level; 1666 cmsg->cmsg_type = tohp->name; 1667 1668 timestamp = 1669 (timestruc_t *)P2ROUNDUP((intptr_t)&tohp[1], 1670 sizeof (intptr_t)); 1671 1672 if (get_udatamodel() == DATAMODEL_NATIVE) { 1673 struct timeval *time_native; 1674 1675 cmsg->cmsg_len = sizeof (struct timeval) + 1676 sizeof (struct cmsghdr); 1677 time_native = 1678 (struct timeval *)CMSG_CONTENT(cmsg); 1679 time_native->tv_sec = timestamp->tv_sec; 1680 time_native->tv_usec = 1681 timestamp->tv_nsec / (NANOSEC / MICROSEC); 1682 } else { 1683 struct timeval32 *time32; 1684 1685 cmsg->cmsg_len = sizeof (struct timeval32) + 1686 sizeof (struct cmsghdr); 1687 time32 = (struct timeval32 *)CMSG_CONTENT(cmsg); 1688 time32->tv_sec = (time32_t)timestamp->tv_sec; 1689 time32->tv_usec = 1690 (int32_t)(timestamp->tv_nsec / 1691 (NANOSEC / MICROSEC)); 1692 } 1693 1694 } else { 1695 if (oldflg) 1696 continue; 1697 1698 cmsg->cmsg_level = tohp->level; 1699 cmsg->cmsg_type = tohp->name; 1700 cmsg->cmsg_len = (socklen_t)(_TPI_TOPT_DATALEN(tohp) + 1701 sizeof (struct cmsghdr)); 1702 1703 /* copy content to control data part */ 1704 bcopy(&tohp[1], CMSG_CONTENT(cmsg), 1705 CMSG_CONTENTLEN(cmsg)); 1706 } 1707 /* move to next CMSG structure! */ 1708 cmsg = CMSG_NEXT(cmsg); 1709 } 1710 return (0); 1711 } 1712 1713 /* 1714 * Extract the SO_SRCADDR option value if present. 1715 */ 1716 void 1717 so_getopt_srcaddr(void *opt, t_uscalar_t optlen, void **srcp, 1718 t_uscalar_t *srclenp) 1719 { 1720 struct T_opthdr *tohp; 1721 1722 ASSERT(__TPI_TOPT_ISALIGNED(opt)); 1723 1724 ASSERT(srcp != NULL && srclenp != NULL); 1725 *srcp = NULL; 1726 *srclenp = 0; 1727 1728 for (tohp = (struct T_opthdr *)opt; 1729 tohp && _TPI_TOPT_VALID(tohp, opt, (uintptr_t)opt + optlen); 1730 tohp = _TPI_TOPT_NEXTHDR(opt, optlen, tohp)) { 1731 dprint(1, ("so_getopt_srcaddr: level 0x%x, name %d, len %d\n", 1732 tohp->level, tohp->name, tohp->len)); 1733 if (tohp->level == SOL_SOCKET && 1734 tohp->name == SO_SRCADDR) { 1735 *srcp = _TPI_TOPT_DATA(tohp); 1736 *srclenp = (t_uscalar_t)_TPI_TOPT_DATALEN(tohp); 1737 } 1738 } 1739 } 1740 1741 /* 1742 * Verify if the SO_UNIX_CLOSE option is present. 1743 */ 1744 int 1745 so_getopt_unix_close(void *opt, t_uscalar_t optlen) 1746 { 1747 struct T_opthdr *tohp; 1748 1749 ASSERT(__TPI_TOPT_ISALIGNED(opt)); 1750 1751 for (tohp = (struct T_opthdr *)opt; 1752 tohp && _TPI_TOPT_VALID(tohp, opt, (uintptr_t)opt + optlen); 1753 tohp = _TPI_TOPT_NEXTHDR(opt, optlen, tohp)) { 1754 dprint(1, 1755 ("so_getopt_unix_close: level 0x%x, name %d, len %d\n", 1756 tohp->level, tohp->name, tohp->len)); 1757 if (tohp->level == SOL_SOCKET && 1758 tohp->name == SO_UNIX_CLOSE) 1759 return (1); 1760 } 1761 return (0); 1762 } 1763 1764 /* 1765 * Allocate an M_PROTO message. 1766 * 1767 * If allocation fails the behavior depends on sleepflg: 1768 * _ALLOC_NOSLEEP fail immediately 1769 * _ALLOC_INTR sleep for memory until a signal is caught 1770 * _ALLOC_SLEEP sleep forever. Don't return NULL. 1771 */ 1772 mblk_t * 1773 soallocproto(size_t size, int sleepflg) 1774 { 1775 mblk_t *mp; 1776 1777 /* Round up size for reuse */ 1778 size = MAX(size, 64); 1779 mp = allocb(size, BPRI_MED); 1780 if (mp == NULL) { 1781 int error; /* Dummy - error not returned to caller */ 1782 1783 switch (sleepflg) { 1784 case _ALLOC_SLEEP: 1785 mp = allocb_wait(size, BPRI_MED, STR_NOSIG, &error); 1786 ASSERT(mp); 1787 break; 1788 case _ALLOC_INTR: 1789 mp = allocb_wait(size, BPRI_MED, 0, &error); 1790 if (mp == NULL) { 1791 /* Caught signal while sleeping for memory */ 1792 eprintline(ENOBUFS); 1793 return (NULL); 1794 } 1795 break; 1796 case _ALLOC_NOSLEEP: 1797 default: 1798 eprintline(ENOBUFS); 1799 return (NULL); 1800 } 1801 } 1802 DB_TYPE(mp) = M_PROTO; 1803 return (mp); 1804 } 1805 1806 /* 1807 * Allocate an M_PROTO message with a single component. 1808 * len is the length of buf. size is the amount to allocate. 1809 * 1810 * buf can be NULL with a non-zero len. 1811 * This results in a bzero'ed chunk being placed the message. 1812 */ 1813 mblk_t * 1814 soallocproto1(const void *buf, ssize_t len, ssize_t size, int sleepflg) 1815 { 1816 mblk_t *mp; 1817 1818 if (size == 0) 1819 size = len; 1820 1821 ASSERT(size >= len); 1822 /* Round up size for reuse */ 1823 size = MAX(size, 64); 1824 mp = soallocproto(size, sleepflg); 1825 if (mp == NULL) 1826 return (NULL); 1827 mp->b_datap->db_type = M_PROTO; 1828 if (len != 0) { 1829 if (buf != NULL) 1830 bcopy(buf, mp->b_wptr, len); 1831 else 1832 bzero(mp->b_wptr, len); 1833 mp->b_wptr += len; 1834 } 1835 return (mp); 1836 } 1837 1838 /* 1839 * Append buf/len to mp. 1840 * The caller has to ensure that there is enough room in the mblk. 1841 * 1842 * buf can be NULL with a non-zero len. 1843 * This results in a bzero'ed chunk being placed the message. 1844 */ 1845 void 1846 soappendmsg(mblk_t *mp, const void *buf, ssize_t len) 1847 { 1848 ASSERT(mp); 1849 1850 if (len != 0) { 1851 /* Assert for room left */ 1852 ASSERT(mp->b_datap->db_lim - mp->b_wptr >= len); 1853 if (buf != NULL) 1854 bcopy(buf, mp->b_wptr, len); 1855 else 1856 bzero(mp->b_wptr, len); 1857 } 1858 mp->b_wptr += len; 1859 } 1860 1861 /* 1862 * Create a message using two kernel buffers. 1863 * If size is set that will determine the allocation size (e.g. for future 1864 * soappendmsg calls). If size is zero it is derived from the buffer 1865 * lengths. 1866 */ 1867 mblk_t * 1868 soallocproto2(const void *buf1, ssize_t len1, const void *buf2, ssize_t len2, 1869 ssize_t size, int sleepflg) 1870 { 1871 mblk_t *mp; 1872 1873 if (size == 0) 1874 size = len1 + len2; 1875 ASSERT(size >= len1 + len2); 1876 1877 mp = soallocproto1(buf1, len1, size, sleepflg); 1878 if (mp) 1879 soappendmsg(mp, buf2, len2); 1880 return (mp); 1881 } 1882 1883 /* 1884 * Create a message using three kernel buffers. 1885 * If size is set that will determine the allocation size (for future 1886 * soappendmsg calls). If size is zero it is derived from the buffer 1887 * lengths. 1888 */ 1889 mblk_t * 1890 soallocproto3(const void *buf1, ssize_t len1, const void *buf2, ssize_t len2, 1891 const void *buf3, ssize_t len3, ssize_t size, int sleepflg) 1892 { 1893 mblk_t *mp; 1894 1895 if (size == 0) 1896 size = len1 + len2 +len3; 1897 ASSERT(size >= len1 + len2 + len3); 1898 1899 mp = soallocproto1(buf1, len1, size, sleepflg); 1900 if (mp != NULL) { 1901 soappendmsg(mp, buf2, len2); 1902 soappendmsg(mp, buf3, len3); 1903 } 1904 return (mp); 1905 } 1906 1907 #ifdef DEBUG 1908 char * 1909 pr_state(uint_t state, uint_t mode) 1910 { 1911 static char buf[1024]; 1912 1913 buf[0] = 0; 1914 if (state & SS_ISCONNECTED) 1915 strcat(buf, "ISCONNECTED "); 1916 if (state & SS_ISCONNECTING) 1917 strcat(buf, "ISCONNECTING "); 1918 if (state & SS_ISDISCONNECTING) 1919 strcat(buf, "ISDISCONNECTING "); 1920 if (state & SS_CANTSENDMORE) 1921 strcat(buf, "CANTSENDMORE "); 1922 1923 if (state & SS_CANTRCVMORE) 1924 strcat(buf, "CANTRCVMORE "); 1925 if (state & SS_ISBOUND) 1926 strcat(buf, "ISBOUND "); 1927 if (state & SS_NDELAY) 1928 strcat(buf, "NDELAY "); 1929 if (state & SS_NONBLOCK) 1930 strcat(buf, "NONBLOCK "); 1931 1932 if (state & SS_ASYNC) 1933 strcat(buf, "ASYNC "); 1934 if (state & SS_ACCEPTCONN) 1935 strcat(buf, "ACCEPTCONN "); 1936 if (state & SS_HASCONNIND) 1937 strcat(buf, "HASCONNIND "); 1938 if (state & SS_SAVEDEOR) 1939 strcat(buf, "SAVEDEOR "); 1940 1941 if (state & SS_RCVATMARK) 1942 strcat(buf, "RCVATMARK "); 1943 if (state & SS_OOBPEND) 1944 strcat(buf, "OOBPEND "); 1945 if (state & SS_HAVEOOBDATA) 1946 strcat(buf, "HAVEOOBDATA "); 1947 if (state & SS_HADOOBDATA) 1948 strcat(buf, "HADOOBDATA "); 1949 1950 if (state & SS_FADDR_NOXLATE) 1951 strcat(buf, "FADDR_NOXLATE "); 1952 1953 if (mode & SM_PRIV) 1954 strcat(buf, "PRIV "); 1955 if (mode & SM_ATOMIC) 1956 strcat(buf, "ATOMIC "); 1957 if (mode & SM_ADDR) 1958 strcat(buf, "ADDR "); 1959 if (mode & SM_CONNREQUIRED) 1960 strcat(buf, "CONNREQUIRED "); 1961 1962 if (mode & SM_FDPASSING) 1963 strcat(buf, "FDPASSING "); 1964 if (mode & SM_EXDATA) 1965 strcat(buf, "EXDATA "); 1966 if (mode & SM_OPTDATA) 1967 strcat(buf, "OPTDATA "); 1968 if (mode & SM_BYTESTREAM) 1969 strcat(buf, "BYTESTREAM "); 1970 return (buf); 1971 } 1972 1973 char * 1974 pr_addr(int family, struct sockaddr *addr, t_uscalar_t addrlen) 1975 { 1976 static char buf[1024]; 1977 1978 if (addr == NULL || addrlen == 0) { 1979 sprintf(buf, "(len %d) %p", addrlen, addr); 1980 return (buf); 1981 } 1982 switch (family) { 1983 case AF_INET: { 1984 struct sockaddr_in sin; 1985 1986 bcopy(addr, &sin, sizeof (sin)); 1987 1988 (void) sprintf(buf, "(len %d) %x/%d", 1989 addrlen, ntohl(sin.sin_addr.s_addr), 1990 ntohs(sin.sin_port)); 1991 break; 1992 } 1993 case AF_INET6: { 1994 struct sockaddr_in6 sin6; 1995 uint16_t *piece = (uint16_t *)&sin6.sin6_addr; 1996 1997 bcopy((char *)addr, (char *)&sin6, sizeof (sin6)); 1998 sprintf(buf, "(len %d) %x:%x:%x:%x:%x:%x:%x:%x/%d", 1999 addrlen, 2000 ntohs(piece[0]), ntohs(piece[1]), 2001 ntohs(piece[2]), ntohs(piece[3]), 2002 ntohs(piece[4]), ntohs(piece[5]), 2003 ntohs(piece[6]), ntohs(piece[7]), 2004 ntohs(sin6.sin6_port)); 2005 break; 2006 } 2007 case AF_UNIX: { 2008 struct sockaddr_un *soun = (struct sockaddr_un *)addr; 2009 2010 (void) sprintf(buf, "(len %d) %s", 2011 addrlen, 2012 (soun == NULL) ? "(none)" : soun->sun_path); 2013 break; 2014 } 2015 default: 2016 (void) sprintf(buf, "(unknown af %d)", family); 2017 break; 2018 } 2019 return (buf); 2020 } 2021 2022 /* The logical equivalence operator (a if-and-only-if b) */ 2023 #define EQUIV(a, b) (((a) && (b)) || (!(a) && (!(b)))) 2024 2025 /* 2026 * Verify limitations and invariants on oob state. 2027 * Return 1 if OK, otherwise 0 so that it can be used as 2028 * ASSERT(verify_oobstate(so)); 2029 */ 2030 int 2031 so_verify_oobstate(struct sonode *so) 2032 { 2033 ASSERT(MUTEX_HELD(&so->so_lock)); 2034 2035 /* 2036 * The possible state combinations are: 2037 * 0 2038 * SS_OOBPEND 2039 * SS_OOBPEND|SS_HAVEOOBDATA 2040 * SS_OOBPEND|SS_HADOOBDATA 2041 * SS_HADOOBDATA 2042 */ 2043 switch (so->so_state & (SS_OOBPEND|SS_HAVEOOBDATA|SS_HADOOBDATA)) { 2044 case 0: 2045 case SS_OOBPEND: 2046 case SS_OOBPEND|SS_HAVEOOBDATA: 2047 case SS_OOBPEND|SS_HADOOBDATA: 2048 case SS_HADOOBDATA: 2049 break; 2050 default: 2051 printf("Bad oob state 1 (%p): counts %d/%d state %s\n", 2052 so, so->so_oobsigcnt, 2053 so->so_oobcnt, pr_state(so->so_state, so->so_mode)); 2054 return (0); 2055 } 2056 2057 /* SS_RCVATMARK should only be set when SS_OOBPEND is set */ 2058 if ((so->so_state & (SS_RCVATMARK|SS_OOBPEND)) == SS_RCVATMARK) { 2059 printf("Bad oob state 2 (%p): counts %d/%d state %s\n", 2060 so, so->so_oobsigcnt, 2061 so->so_oobcnt, pr_state(so->so_state, so->so_mode)); 2062 return (0); 2063 } 2064 2065 /* 2066 * (so_oobsigcnt != 0 or SS_RCVATMARK) iff SS_OOBPEND 2067 */ 2068 if (!EQUIV((so->so_oobsigcnt != 0) || (so->so_state & SS_RCVATMARK), 2069 so->so_state & SS_OOBPEND)) { 2070 printf("Bad oob state 3 (%p): counts %d/%d state %s\n", 2071 so, so->so_oobsigcnt, 2072 so->so_oobcnt, pr_state(so->so_state, so->so_mode)); 2073 return (0); 2074 } 2075 2076 /* 2077 * Unless SO_OOBINLINE we have so_oobmsg != NULL iff SS_HAVEOOBDATA 2078 */ 2079 if (!(so->so_options & SO_OOBINLINE) && 2080 !EQUIV(so->so_oobmsg != NULL, so->so_state & SS_HAVEOOBDATA)) { 2081 printf("Bad oob state 4 (%p): counts %d/%d state %s\n", 2082 so, so->so_oobsigcnt, 2083 so->so_oobcnt, pr_state(so->so_state, so->so_mode)); 2084 return (0); 2085 } 2086 if (so->so_oobsigcnt < so->so_oobcnt) { 2087 printf("Bad oob state 5 (%p): counts %d/%d state %s\n", 2088 so, so->so_oobsigcnt, 2089 so->so_oobcnt, pr_state(so->so_state, so->so_mode)); 2090 return (0); 2091 } 2092 return (1); 2093 } 2094 #undef EQUIV 2095 2096 #endif /* DEBUG */ 2097 2098 /* initialize sockfs zone specific kstat related items */ 2099 void * 2100 sock_kstat_init(zoneid_t zoneid) 2101 { 2102 kstat_t *ksp; 2103 2104 ksp = kstat_create_zone("sockfs", 0, "sock_unix_list", "misc", 2105 KSTAT_TYPE_RAW, 0, KSTAT_FLAG_VAR_SIZE|KSTAT_FLAG_VIRTUAL, zoneid); 2106 2107 if (ksp != NULL) { 2108 ksp->ks_update = sockfs_update; 2109 ksp->ks_snapshot = sockfs_snapshot; 2110 ksp->ks_lock = &socklist.sl_lock; 2111 ksp->ks_private = (void *)(uintptr_t)zoneid; 2112 kstat_install(ksp); 2113 } 2114 2115 return (ksp); 2116 } 2117 2118 /* tear down sockfs zone specific kstat related items */ 2119 /*ARGSUSED*/ 2120 void 2121 sock_kstat_fini(zoneid_t zoneid, void *arg) 2122 { 2123 kstat_t *ksp = (kstat_t *)arg; 2124 2125 if (ksp != NULL) { 2126 ASSERT(zoneid == (zoneid_t)(uintptr_t)ksp->ks_private); 2127 kstat_delete(ksp); 2128 } 2129 } 2130 2131 /* 2132 * Zones: 2133 * Note that nactive is going to be different for each zone. 2134 * This means we require kstat to call sockfs_update and then sockfs_snapshot 2135 * for the same zone, or sockfs_snapshot will be taken into the wrong size 2136 * buffer. This is safe, but if the buffer is too small, user will not be 2137 * given details of all sockets. However, as this kstat has a ks_lock, kstat 2138 * driver will keep it locked between the update and the snapshot, so no 2139 * other process (zone) can currently get inbetween resulting in a wrong size 2140 * buffer allocation. 2141 */ 2142 static int 2143 sockfs_update(kstat_t *ksp, int rw) 2144 { 2145 uint_t nactive = 0; /* # of active AF_UNIX sockets */ 2146 struct sonode *so; /* current sonode on socklist */ 2147 zoneid_t myzoneid = (zoneid_t)(uintptr_t)ksp->ks_private; 2148 2149 ASSERT((zoneid_t)(uintptr_t)ksp->ks_private == getzoneid()); 2150 2151 if (rw == KSTAT_WRITE) { /* bounce all writes */ 2152 return (EACCES); 2153 } 2154 2155 for (so = socklist.sl_list; so != NULL; so = so->so_next) { 2156 if (so->so_accessvp != NULL && so->so_zoneid == myzoneid) { 2157 nactive++; 2158 } 2159 } 2160 ksp->ks_ndata = nactive; 2161 ksp->ks_data_size = nactive * sizeof (struct k_sockinfo); 2162 2163 return (0); 2164 } 2165 2166 static int 2167 sockfs_snapshot(kstat_t *ksp, void *buf, int rw) 2168 { 2169 int ns; /* # of sonodes we've copied */ 2170 struct sonode *so; /* current sonode on socklist */ 2171 struct k_sockinfo *pksi; /* where we put sockinfo data */ 2172 t_uscalar_t sn_len; /* soa_len */ 2173 zoneid_t myzoneid = (zoneid_t)(uintptr_t)ksp->ks_private; 2174 2175 ASSERT((zoneid_t)(uintptr_t)ksp->ks_private == getzoneid()); 2176 2177 ksp->ks_snaptime = gethrtime(); 2178 2179 if (rw == KSTAT_WRITE) { /* bounce all writes */ 2180 return (EACCES); 2181 } 2182 2183 /* 2184 * for each sonode on the socklist, we massage the important 2185 * info into buf, in k_sockinfo format. 2186 */ 2187 pksi = (struct k_sockinfo *)buf; 2188 for (ns = 0, so = socklist.sl_list; so != NULL; so = so->so_next) { 2189 /* only stuff active sonodes and the same zone: */ 2190 if (so->so_accessvp == NULL || so->so_zoneid != myzoneid) { 2191 continue; 2192 } 2193 2194 /* 2195 * If the sonode was activated between the update and the 2196 * snapshot, we're done - as this is only a snapshot. 2197 */ 2198 if ((caddr_t)(pksi) >= (caddr_t)buf + ksp->ks_data_size) { 2199 break; 2200 } 2201 2202 /* copy important info into buf: */ 2203 pksi->ks_si.si_size = sizeof (struct k_sockinfo); 2204 pksi->ks_si.si_family = so->so_family; 2205 pksi->ks_si.si_type = so->so_type; 2206 pksi->ks_si.si_flag = so->so_flag; 2207 pksi->ks_si.si_state = so->so_state; 2208 pksi->ks_si.si_serv_type = so->so_serv_type; 2209 pksi->ks_si.si_ux_laddr_sou_magic = so->so_ux_laddr.soua_magic; 2210 pksi->ks_si.si_ux_faddr_sou_magic = so->so_ux_faddr.soua_magic; 2211 pksi->ks_si.si_laddr_soa_len = so->so_laddr.soa_len; 2212 pksi->ks_si.si_faddr_soa_len = so->so_faddr.soa_len; 2213 pksi->ks_si.si_szoneid = so->so_zoneid; 2214 2215 mutex_enter(&so->so_lock); 2216 2217 if (so->so_laddr_sa != NULL) { 2218 ASSERT(so->so_laddr_sa->sa_data != NULL); 2219 sn_len = so->so_laddr_len; 2220 ASSERT(sn_len <= sizeof (short) + 2221 sizeof (pksi->ks_si.si_laddr_sun_path)); 2222 2223 pksi->ks_si.si_laddr_family = 2224 so->so_laddr_sa->sa_family; 2225 if (sn_len != 0) { 2226 /* AF_UNIX socket names are NULL terminated */ 2227 (void) strncpy(pksi->ks_si.si_laddr_sun_path, 2228 so->so_laddr_sa->sa_data, 2229 sizeof (pksi->ks_si.si_laddr_sun_path)); 2230 sn_len = strlen(pksi->ks_si.si_laddr_sun_path); 2231 } 2232 pksi->ks_si.si_laddr_sun_path[sn_len] = 0; 2233 } 2234 2235 if (so->so_faddr_sa != NULL) { 2236 ASSERT(so->so_faddr_sa->sa_data != NULL); 2237 sn_len = so->so_faddr_len; 2238 ASSERT(sn_len <= sizeof (short) + 2239 sizeof (pksi->ks_si.si_faddr_sun_path)); 2240 2241 pksi->ks_si.si_faddr_family = 2242 so->so_faddr_sa->sa_family; 2243 if (sn_len != 0) { 2244 (void) strncpy(pksi->ks_si.si_faddr_sun_path, 2245 so->so_faddr_sa->sa_data, 2246 sizeof (pksi->ks_si.si_faddr_sun_path)); 2247 sn_len = strlen(pksi->ks_si.si_faddr_sun_path); 2248 } 2249 pksi->ks_si.si_faddr_sun_path[sn_len] = 0; 2250 } 2251 2252 mutex_exit(&so->so_lock); 2253 2254 (void) sprintf(pksi->ks_straddr[0], "%p", (void *)so); 2255 (void) sprintf(pksi->ks_straddr[1], "%p", 2256 (void *)so->so_ux_laddr.soua_vp); 2257 (void) sprintf(pksi->ks_straddr[2], "%p", 2258 (void *)so->so_ux_faddr.soua_vp); 2259 2260 ns++; 2261 pksi++; 2262 } 2263 2264 ksp->ks_ndata = ns; 2265 return (0); 2266 } 2267 2268 ssize_t 2269 soreadfile(file_t *fp, uchar_t *buf, u_offset_t fileoff, int *err, size_t size) 2270 { 2271 struct uio auio; 2272 struct iovec aiov[MSG_MAXIOVLEN]; 2273 register vnode_t *vp; 2274 int ioflag, rwflag; 2275 ssize_t cnt; 2276 int error = 0; 2277 int iovcnt = 0; 2278 short fflag; 2279 2280 vp = fp->f_vnode; 2281 fflag = fp->f_flag; 2282 2283 rwflag = 0; 2284 aiov[0].iov_base = (caddr_t)buf; 2285 aiov[0].iov_len = size; 2286 iovcnt = 1; 2287 cnt = (ssize_t)size; 2288 (void) VOP_RWLOCK(vp, rwflag, NULL); 2289 2290 auio.uio_loffset = fileoff; 2291 auio.uio_iov = aiov; 2292 auio.uio_iovcnt = iovcnt; 2293 auio.uio_resid = cnt; 2294 auio.uio_segflg = UIO_SYSSPACE; 2295 auio.uio_llimit = MAXOFFSET_T; 2296 auio.uio_fmode = fflag; 2297 auio.uio_extflg = UIO_COPY_CACHED; 2298 2299 ioflag = auio.uio_fmode & (FAPPEND|FSYNC|FDSYNC|FRSYNC); 2300 2301 /* If read sync is not asked for, filter sync flags */ 2302 if ((ioflag & FRSYNC) == 0) 2303 ioflag &= ~(FSYNC|FDSYNC); 2304 error = VOP_READ(vp, &auio, ioflag, fp->f_cred, NULL); 2305 cnt -= auio.uio_resid; 2306 2307 VOP_RWUNLOCK(vp, rwflag, NULL); 2308 2309 if (error == EINTR && cnt != 0) 2310 error = 0; 2311 out: 2312 if (error != 0) { 2313 *err = error; 2314 return (0); 2315 } else { 2316 *err = 0; 2317 return (cnt); 2318 } 2319 } 2320