1 /* $NetBSD: vfs_mount.c,v 1.25 2013/11/27 17:25:46 christos Exp $ */ 2 3 /*- 4 * Copyright (c) 1997-2011 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, by Charles M. Hannum, 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) 1989, 1993 35 * The Regents of the University of California. All rights reserved. 36 * (c) UNIX System Laboratories, Inc. 37 * All or some portions of this file are derived from material licensed 38 * to the University of California by American Telephone and Telegraph 39 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 40 * the permission of UNIX System Laboratories, Inc. 41 * 42 * Redistribution and use in source and binary forms, with or without 43 * modification, are permitted provided that the following conditions 44 * are met: 45 * 1. Redistributions of source code must retain the above copyright 46 * notice, this list of conditions and the following disclaimer. 47 * 2. Redistributions in binary form must reproduce the above copyright 48 * notice, this list of conditions and the following disclaimer in the 49 * documentation and/or other materials provided with the distribution. 50 * 3. Neither the name of the University nor the names of its contributors 51 * may be used to endorse or promote products derived from this software 52 * without specific prior written permission. 53 * 54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 64 * SUCH DAMAGE. 65 * 66 * @(#)vfs_subr.c 8.13 (Berkeley) 4/18/94 67 */ 68 69 #include <sys/cdefs.h> 70 __KERNEL_RCSID(0, "$NetBSD: vfs_mount.c,v 1.25 2013/11/27 17:25:46 christos Exp $"); 71 72 #include <sys/param.h> 73 #include <sys/kernel.h> 74 75 #include <sys/atomic.h> 76 #include <sys/buf.h> 77 #include <sys/conf.h> 78 #include <sys/fcntl.h> 79 #include <sys/filedesc.h> 80 #include <sys/device.h> 81 #include <sys/kauth.h> 82 #include <sys/kmem.h> 83 #include <sys/module.h> 84 #include <sys/mount.h> 85 #include <sys/namei.h> 86 #include <sys/extattr.h> 87 #include <sys/syscallargs.h> 88 #include <sys/sysctl.h> 89 #include <sys/systm.h> 90 #include <sys/vfs_syscalls.h> 91 #include <sys/vnode.h> 92 93 #include <miscfs/genfs/genfs.h> 94 #include <miscfs/syncfs/syncfs.h> 95 #include <miscfs/specfs/specdev.h> 96 97 /* Root filesystem and device. */ 98 vnode_t * rootvnode; 99 device_t root_device; 100 101 /* Mounted filesystem list. */ 102 struct mntlist mountlist; 103 kmutex_t mountlist_lock; 104 105 kmutex_t mntvnode_lock; 106 kmutex_t vfs_list_lock; 107 108 static specificdata_domain_t mount_specificdata_domain; 109 static kmutex_t mntid_lock; 110 111 static kmutex_t mountgen_lock; 112 static uint64_t mountgen; 113 114 void 115 vfs_mount_sysinit(void) 116 { 117 118 TAILQ_INIT(&mountlist); 119 mutex_init(&mountlist_lock, MUTEX_DEFAULT, IPL_NONE); 120 mutex_init(&mntvnode_lock, MUTEX_DEFAULT, IPL_NONE); 121 mutex_init(&vfs_list_lock, MUTEX_DEFAULT, IPL_NONE); 122 123 mount_specificdata_domain = specificdata_domain_create(); 124 mutex_init(&mntid_lock, MUTEX_DEFAULT, IPL_NONE); 125 mutex_init(&mountgen_lock, MUTEX_DEFAULT, IPL_NONE); 126 mountgen = 0; 127 } 128 129 struct mount * 130 vfs_mountalloc(struct vfsops *vfsops, vnode_t *vp) 131 { 132 struct mount *mp; 133 int error __diagused; 134 135 mp = kmem_zalloc(sizeof(*mp), KM_SLEEP); 136 if (mp == NULL) 137 return NULL; 138 139 mp->mnt_op = vfsops; 140 mp->mnt_refcnt = 1; 141 TAILQ_INIT(&mp->mnt_vnodelist); 142 mutex_init(&mp->mnt_unmounting, MUTEX_DEFAULT, IPL_NONE); 143 mutex_init(&mp->mnt_renamelock, MUTEX_DEFAULT, IPL_NONE); 144 mutex_init(&mp->mnt_updating, MUTEX_DEFAULT, IPL_NONE); 145 error = vfs_busy(mp, NULL); 146 KASSERT(error == 0); 147 mp->mnt_vnodecovered = vp; 148 mount_initspecific(mp); 149 150 mutex_enter(&mountgen_lock); 151 mp->mnt_gen = mountgen++; 152 mutex_exit(&mountgen_lock); 153 154 return mp; 155 } 156 157 /* 158 * vfs_rootmountalloc: lookup a filesystem type, and if found allocate and 159 * initialize a mount structure for it. 160 * 161 * Devname is usually updated by mount(8) after booting. 162 */ 163 int 164 vfs_rootmountalloc(const char *fstypename, const char *devname, 165 struct mount **mpp) 166 { 167 struct vfsops *vfsp = NULL; 168 struct mount *mp; 169 170 mutex_enter(&vfs_list_lock); 171 LIST_FOREACH(vfsp, &vfs_list, vfs_list) 172 if (!strncmp(vfsp->vfs_name, fstypename, 173 sizeof(mp->mnt_stat.f_fstypename))) 174 break; 175 if (vfsp == NULL) { 176 mutex_exit(&vfs_list_lock); 177 return (ENODEV); 178 } 179 vfsp->vfs_refcount++; 180 mutex_exit(&vfs_list_lock); 181 182 if ((mp = vfs_mountalloc(vfsp, NULL)) == NULL) 183 return ENOMEM; 184 mp->mnt_flag = MNT_RDONLY; 185 (void)strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfs_name, 186 sizeof(mp->mnt_stat.f_fstypename)); 187 mp->mnt_stat.f_mntonname[0] = '/'; 188 mp->mnt_stat.f_mntonname[1] = '\0'; 189 mp->mnt_stat.f_mntfromname[sizeof(mp->mnt_stat.f_mntfromname) - 1] = 190 '\0'; 191 (void)copystr(devname, mp->mnt_stat.f_mntfromname, 192 sizeof(mp->mnt_stat.f_mntfromname) - 1, 0); 193 *mpp = mp; 194 return 0; 195 } 196 197 /* 198 * vfs_getnewfsid: get a new unique fsid. 199 */ 200 void 201 vfs_getnewfsid(struct mount *mp) 202 { 203 static u_short xxxfs_mntid; 204 fsid_t tfsid; 205 int mtype; 206 207 mutex_enter(&mntid_lock); 208 mtype = makefstype(mp->mnt_op->vfs_name); 209 mp->mnt_stat.f_fsidx.__fsid_val[0] = makedev(mtype, 0); 210 mp->mnt_stat.f_fsidx.__fsid_val[1] = mtype; 211 mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0]; 212 if (xxxfs_mntid == 0) 213 ++xxxfs_mntid; 214 tfsid.__fsid_val[0] = makedev(mtype & 0xff, xxxfs_mntid); 215 tfsid.__fsid_val[1] = mtype; 216 if (!TAILQ_EMPTY(&mountlist)) { 217 while (vfs_getvfs(&tfsid)) { 218 tfsid.__fsid_val[0]++; 219 xxxfs_mntid++; 220 } 221 } 222 mp->mnt_stat.f_fsidx.__fsid_val[0] = tfsid.__fsid_val[0]; 223 mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0]; 224 mutex_exit(&mntid_lock); 225 } 226 227 /* 228 * Lookup a mount point by filesystem identifier. 229 * 230 * XXX Needs to add a reference to the mount point. 231 */ 232 struct mount * 233 vfs_getvfs(fsid_t *fsid) 234 { 235 struct mount *mp; 236 237 mutex_enter(&mountlist_lock); 238 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 239 if (mp->mnt_stat.f_fsidx.__fsid_val[0] == fsid->__fsid_val[0] && 240 mp->mnt_stat.f_fsidx.__fsid_val[1] == fsid->__fsid_val[1]) { 241 mutex_exit(&mountlist_lock); 242 return (mp); 243 } 244 } 245 mutex_exit(&mountlist_lock); 246 return NULL; 247 } 248 249 /* 250 * Drop a reference to a mount structure, freeing if the last reference. 251 */ 252 void 253 vfs_destroy(struct mount *mp) 254 { 255 256 if (__predict_true((int)atomic_dec_uint_nv(&mp->mnt_refcnt) > 0)) { 257 return; 258 } 259 260 /* 261 * Nothing else has visibility of the mount: we can now 262 * free the data structures. 263 */ 264 KASSERT(mp->mnt_refcnt == 0); 265 specificdata_fini(mount_specificdata_domain, &mp->mnt_specdataref); 266 mutex_destroy(&mp->mnt_unmounting); 267 mutex_destroy(&mp->mnt_updating); 268 mutex_destroy(&mp->mnt_renamelock); 269 if (mp->mnt_op != NULL) { 270 vfs_delref(mp->mnt_op); 271 } 272 kmem_free(mp, sizeof(*mp)); 273 } 274 275 /* 276 * Mark a mount point as busy, and gain a new reference to it. Used to 277 * prevent the file system from being unmounted during critical sections. 278 * 279 * vfs_busy can be called multiple times and by multiple threads 280 * and must be accompanied by the same number of vfs_unbusy calls. 281 * 282 * => The caller must hold a pre-existing reference to the mount. 283 * => Will fail if the file system is being unmounted, or is unmounted. 284 */ 285 int 286 vfs_busy(struct mount *mp, struct mount **nextp) 287 { 288 289 KASSERT(mp->mnt_refcnt > 0); 290 291 mutex_enter(&mp->mnt_unmounting); 292 if (__predict_false((mp->mnt_iflag & IMNT_GONE) != 0)) { 293 mutex_exit(&mp->mnt_unmounting); 294 if (nextp != NULL) { 295 KASSERT(mutex_owned(&mountlist_lock)); 296 *nextp = TAILQ_NEXT(mp, mnt_list); 297 } 298 return ENOENT; 299 } 300 ++mp->mnt_busynest; 301 KASSERT(mp->mnt_busynest != 0); 302 mutex_exit(&mp->mnt_unmounting); 303 if (nextp != NULL) { 304 mutex_exit(&mountlist_lock); 305 } 306 atomic_inc_uint(&mp->mnt_refcnt); 307 return 0; 308 } 309 310 /* 311 * Unbusy a busy filesystem. 312 * 313 * Every successful vfs_busy() call must be undone by a vfs_unbusy() call. 314 * 315 * => If keepref is true, preserve reference added by vfs_busy(). 316 * => If nextp != NULL, acquire mountlist_lock. 317 */ 318 void 319 vfs_unbusy(struct mount *mp, bool keepref, struct mount **nextp) 320 { 321 322 KASSERT(mp->mnt_refcnt > 0); 323 324 if (nextp != NULL) { 325 mutex_enter(&mountlist_lock); 326 } 327 mutex_enter(&mp->mnt_unmounting); 328 KASSERT(mp->mnt_busynest != 0); 329 mp->mnt_busynest--; 330 mutex_exit(&mp->mnt_unmounting); 331 if (!keepref) { 332 vfs_destroy(mp); 333 } 334 if (nextp != NULL) { 335 KASSERT(mutex_owned(&mountlist_lock)); 336 *nextp = TAILQ_NEXT(mp, mnt_list); 337 } 338 } 339 340 /* 341 * Insert a marker vnode into a mount's vnode list, after the 342 * specified vnode. mntvnode_lock must be held. 343 */ 344 void 345 vmark(vnode_t *mvp, vnode_t *vp) 346 { 347 struct mount *mp = mvp->v_mount; 348 349 KASSERT(mutex_owned(&mntvnode_lock)); 350 KASSERT((mvp->v_iflag & VI_MARKER) != 0); 351 KASSERT(vp->v_mount == mp); 352 353 TAILQ_INSERT_AFTER(&mp->mnt_vnodelist, vp, mvp, v_mntvnodes); 354 } 355 356 /* 357 * Remove a marker vnode from a mount's vnode list, and return 358 * a pointer to the next vnode in the list. mntvnode_lock must 359 * be held. 360 */ 361 vnode_t * 362 vunmark(vnode_t *mvp) 363 { 364 struct mount *mp = mvp->v_mount; 365 vnode_t *vp; 366 367 KASSERT(mutex_owned(&mntvnode_lock)); 368 KASSERT((mvp->v_iflag & VI_MARKER) != 0); 369 370 vp = TAILQ_NEXT(mvp, v_mntvnodes); 371 TAILQ_REMOVE(&mp->mnt_vnodelist, mvp, v_mntvnodes); 372 373 KASSERT(vp == NULL || vp->v_mount == mp); 374 375 return vp; 376 } 377 378 /* 379 * Move a vnode from one mount queue to another. 380 */ 381 void 382 vfs_insmntque(vnode_t *vp, struct mount *mp) 383 { 384 struct mount *omp; 385 386 KASSERT(mp == NULL || (mp->mnt_iflag & IMNT_UNMOUNT) == 0 || 387 vp->v_tag == VT_VFS); 388 389 mutex_enter(&mntvnode_lock); 390 /* 391 * Delete from old mount point vnode list, if on one. 392 */ 393 if ((omp = vp->v_mount) != NULL) 394 TAILQ_REMOVE(&vp->v_mount->mnt_vnodelist, vp, v_mntvnodes); 395 /* 396 * Insert into list of vnodes for the new mount point, if 397 * available. The caller must take a reference on the mount 398 * structure and donate to the vnode. 399 */ 400 if ((vp->v_mount = mp) != NULL) 401 TAILQ_INSERT_TAIL(&mp->mnt_vnodelist, vp, v_mntvnodes); 402 mutex_exit(&mntvnode_lock); 403 404 if (omp != NULL) { 405 /* Release reference to old mount. */ 406 vfs_destroy(omp); 407 } 408 } 409 410 /* 411 * Remove any vnodes in the vnode table belonging to mount point mp. 412 * 413 * If FORCECLOSE is not specified, there should not be any active ones, 414 * return error if any are found (nb: this is a user error, not a 415 * system error). If FORCECLOSE is specified, detach any active vnodes 416 * that are found. 417 * 418 * If WRITECLOSE is set, only flush out regular file vnodes open for 419 * writing. 420 * 421 * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped. 422 */ 423 #ifdef DEBUG 424 int busyprt = 0; /* print out busy vnodes */ 425 struct ctldebug debug1 = { "busyprt", &busyprt }; 426 #endif 427 428 static vnode_t * 429 vflushnext(vnode_t *mvp, int *when) 430 { 431 432 if (hardclock_ticks > *when) { 433 mutex_exit(&mntvnode_lock); 434 yield(); 435 mutex_enter(&mntvnode_lock); 436 *when = hardclock_ticks + hz / 10; 437 } 438 return vunmark(mvp); 439 } 440 441 int 442 vflush(struct mount *mp, vnode_t *skipvp, int flags) 443 { 444 vnode_t *vp, *mvp; 445 int busy = 0, when = 0; 446 447 /* First, flush out any vnode references from vrele_list. */ 448 vrele_flush(); 449 450 /* Allocate a marker vnode. */ 451 mvp = vnalloc(mp); 452 453 /* 454 * NOTE: not using the TAILQ_FOREACH here since in this loop vgone() 455 * and vclean() are called. 456 */ 457 mutex_enter(&mntvnode_lock); 458 for (vp = TAILQ_FIRST(&mp->mnt_vnodelist); 459 vp != NULL; 460 vp = vflushnext(mvp, &when)) { 461 vmark(mvp, vp); 462 if (vp->v_mount != mp || vismarker(vp)) 463 continue; 464 /* 465 * Skip over a selected vnode. 466 */ 467 if (vp == skipvp) 468 continue; 469 /* 470 * First try to recycle the vnode. 471 */ 472 if (vrecycle(vp, &mntvnode_lock)) { 473 mutex_enter(&mntvnode_lock); 474 continue; 475 } 476 mutex_enter(vp->v_interlock); 477 /* 478 * Ignore clean but still referenced vnodes. 479 */ 480 if ((vp->v_iflag & VI_CLEAN) != 0) { 481 mutex_exit(vp->v_interlock); 482 continue; 483 } 484 /* 485 * Skip over a vnodes marked VSYSTEM. 486 */ 487 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) { 488 mutex_exit(vp->v_interlock); 489 continue; 490 } 491 /* 492 * If WRITECLOSE is set, only flush out regular file 493 * vnodes open for writing. 494 */ 495 if ((flags & WRITECLOSE) && 496 (vp->v_writecount == 0 || vp->v_type != VREG)) { 497 mutex_exit(vp->v_interlock); 498 continue; 499 } 500 /* 501 * If FORCECLOSE is set, forcibly close the vnode. 502 * For block or character devices, revert to an 503 * anonymous device. For all other files, just 504 * kill them. 505 */ 506 if (flags & FORCECLOSE) { 507 mutex_exit(&mntvnode_lock); 508 if (vget(vp, 0) == 0) 509 vgone(vp); 510 mutex_enter(&mntvnode_lock); 511 continue; 512 } 513 #ifdef DEBUG 514 if (busyprt) 515 vprint("vflush: busy vnode", vp); 516 #endif 517 mutex_exit(vp->v_interlock); 518 busy++; 519 } 520 mutex_exit(&mntvnode_lock); 521 vnfree(mvp); 522 if (busy) 523 return (EBUSY); 524 return (0); 525 } 526 527 /* 528 * Remove clean vnodes from a mountpoint's vnode list. 529 */ 530 void 531 vfs_scrubvnlist(struct mount *mp) 532 { 533 vnode_t *vp, *nvp; 534 535 retry: 536 mutex_enter(&mntvnode_lock); 537 TAILQ_FOREACH_SAFE(vp, &mp->mnt_vnodelist, v_mntvnodes, nvp) { 538 mutex_enter(vp->v_interlock); 539 if ((vp->v_iflag & VI_CLEAN) != 0) { 540 TAILQ_REMOVE(&mp->mnt_vnodelist, vp, v_mntvnodes); 541 vp->v_mount = NULL; 542 mutex_exit(&mntvnode_lock); 543 mutex_exit(vp->v_interlock); 544 vfs_destroy(mp); 545 goto retry; 546 } 547 mutex_exit(vp->v_interlock); 548 } 549 mutex_exit(&mntvnode_lock); 550 } 551 552 /* 553 * Mount a file system. 554 */ 555 556 /* 557 * Scan all active processes to see if any of them have a current or root 558 * directory onto which the new filesystem has just been mounted. If so, 559 * replace them with the new mount point. 560 */ 561 static void 562 mount_checkdirs(vnode_t *olddp) 563 { 564 vnode_t *newdp, *rele1, *rele2; 565 struct cwdinfo *cwdi; 566 struct proc *p; 567 bool retry; 568 569 if (olddp->v_usecount == 1) { 570 return; 571 } 572 if (VFS_ROOT(olddp->v_mountedhere, &newdp)) 573 panic("mount: lost mount"); 574 575 do { 576 retry = false; 577 mutex_enter(proc_lock); 578 PROCLIST_FOREACH(p, &allproc) { 579 if ((cwdi = p->p_cwdi) == NULL) 580 continue; 581 /* 582 * Cannot change to the old directory any more, 583 * so even if we see a stale value it is not a 584 * problem. 585 */ 586 if (cwdi->cwdi_cdir != olddp && 587 cwdi->cwdi_rdir != olddp) 588 continue; 589 retry = true; 590 rele1 = NULL; 591 rele2 = NULL; 592 atomic_inc_uint(&cwdi->cwdi_refcnt); 593 mutex_exit(proc_lock); 594 rw_enter(&cwdi->cwdi_lock, RW_WRITER); 595 if (cwdi->cwdi_cdir == olddp) { 596 rele1 = cwdi->cwdi_cdir; 597 vref(newdp); 598 cwdi->cwdi_cdir = newdp; 599 } 600 if (cwdi->cwdi_rdir == olddp) { 601 rele2 = cwdi->cwdi_rdir; 602 vref(newdp); 603 cwdi->cwdi_rdir = newdp; 604 } 605 rw_exit(&cwdi->cwdi_lock); 606 cwdfree(cwdi); 607 if (rele1 != NULL) 608 vrele(rele1); 609 if (rele2 != NULL) 610 vrele(rele2); 611 mutex_enter(proc_lock); 612 break; 613 } 614 mutex_exit(proc_lock); 615 } while (retry); 616 617 if (rootvnode == olddp) { 618 vrele(rootvnode); 619 vref(newdp); 620 rootvnode = newdp; 621 } 622 vput(newdp); 623 } 624 625 int 626 mount_domount(struct lwp *l, vnode_t **vpp, struct vfsops *vfsops, 627 const char *path, int flags, void *data, size_t *data_len) 628 { 629 vnode_t *vp = *vpp; 630 struct mount *mp; 631 struct pathbuf *pb; 632 struct nameidata nd; 633 int error; 634 635 error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_MOUNT, 636 KAUTH_REQ_SYSTEM_MOUNT_NEW, vp, KAUTH_ARG(flags), data); 637 if (error) { 638 vfs_delref(vfsops); 639 return error; 640 } 641 642 /* Cannot make a non-dir a mount-point (from here anyway). */ 643 if (vp->v_type != VDIR) { 644 vfs_delref(vfsops); 645 return ENOTDIR; 646 } 647 648 if (flags & MNT_EXPORTED) { 649 vfs_delref(vfsops); 650 return EINVAL; 651 } 652 653 if ((mp = vfs_mountalloc(vfsops, vp)) == NULL) { 654 vfs_delref(vfsops); 655 return ENOMEM; 656 } 657 658 mp->mnt_stat.f_owner = kauth_cred_geteuid(l->l_cred); 659 660 /* 661 * The underlying file system may refuse the mount for 662 * various reasons. Allow the user to force it to happen. 663 * 664 * Set the mount level flags. 665 */ 666 mp->mnt_flag = flags & (MNT_BASIC_FLAGS | MNT_FORCE | MNT_IGNORE); 667 668 mutex_enter(&mp->mnt_updating); 669 error = VFS_MOUNT(mp, path, data, data_len); 670 mp->mnt_flag &= ~MNT_OP_FLAGS; 671 672 if (error != 0) 673 goto err_unmounted; 674 675 /* 676 * Validate and prepare the mount point. 677 */ 678 error = pathbuf_copyin(path, &pb); 679 if (error != 0) { 680 goto err_mounted; 681 } 682 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb); 683 error = namei(&nd); 684 pathbuf_destroy(pb); 685 if (error != 0) { 686 goto err_mounted; 687 } 688 if (nd.ni_vp != vp) { 689 vput(nd.ni_vp); 690 error = EINVAL; 691 goto err_mounted; 692 } 693 if (vp->v_mountedhere != NULL) { 694 vput(nd.ni_vp); 695 error = EBUSY; 696 goto err_mounted; 697 } 698 error = vinvalbuf(vp, V_SAVE, l->l_cred, l, 0, 0); 699 if (error != 0) { 700 vput(nd.ni_vp); 701 goto err_mounted; 702 } 703 704 /* 705 * Put the new filesystem on the mount list after root. 706 */ 707 cache_purge(vp); 708 mp->mnt_iflag &= ~IMNT_WANTRDWR; 709 710 mutex_enter(&mountlist_lock); 711 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 712 mutex_exit(&mountlist_lock); 713 if ((mp->mnt_flag & (MNT_RDONLY | MNT_ASYNC)) == 0) 714 error = vfs_allocate_syncvnode(mp); 715 if (error == 0) 716 vp->v_mountedhere = mp; 717 vput(nd.ni_vp); 718 if (error != 0) 719 goto err_onmountlist; 720 721 mount_checkdirs(vp); 722 mutex_exit(&mp->mnt_updating); 723 724 /* Hold an additional reference to the mount across VFS_START(). */ 725 vfs_unbusy(mp, true, NULL); 726 (void) VFS_STATVFS(mp, &mp->mnt_stat); 727 error = VFS_START(mp, 0); 728 if (error) { 729 vrele(vp); 730 } else if (flags & MNT_EXTATTR) { 731 error = VFS_EXTATTRCTL(vp->v_mountedhere, 732 EXTATTR_CMD_START, NULL, 0, NULL); 733 if (error) 734 printf("%s: failed to start extattr: error = %d\n", 735 vp->v_mountedhere->mnt_stat.f_mntonname, error); 736 } 737 /* Drop reference held for VFS_START(). */ 738 vfs_destroy(mp); 739 *vpp = NULL; 740 return error; 741 742 err_onmountlist: 743 mutex_enter(&mountlist_lock); 744 TAILQ_REMOVE(&mountlist, mp, mnt_list); 745 mp->mnt_iflag |= IMNT_GONE; 746 mutex_exit(&mountlist_lock); 747 748 err_mounted: 749 if (VFS_UNMOUNT(mp, MNT_FORCE) != 0) 750 panic("Unmounting fresh file system failed"); 751 752 err_unmounted: 753 vp->v_mountedhere = NULL; 754 mutex_exit(&mp->mnt_updating); 755 vfs_unbusy(mp, false, NULL); 756 vfs_destroy(mp); 757 758 return error; 759 } 760 761 /* 762 * Do the actual file system unmount. File system is assumed to have 763 * been locked by the caller. 764 * 765 * => Caller hold reference to the mount, explicitly for dounmount(). 766 */ 767 int 768 dounmount(struct mount *mp, int flags, struct lwp *l) 769 { 770 vnode_t *coveredvp; 771 int error, async, used_syncer; 772 773 #if NVERIEXEC > 0 774 error = veriexec_unmountchk(mp); 775 if (error) 776 return (error); 777 #endif /* NVERIEXEC > 0 */ 778 779 /* 780 * XXX Freeze syncer. Must do this before locking the 781 * mount point. See dounmount() for details. 782 */ 783 mutex_enter(&syncer_mutex); 784 785 /* 786 * Abort unmount attempt when the filesystem is in use 787 */ 788 mutex_enter(&mp->mnt_unmounting); 789 if (mp->mnt_busynest != 0) { 790 mutex_exit(&mp->mnt_unmounting); 791 mutex_exit(&syncer_mutex); 792 return EBUSY; 793 } 794 795 /* 796 * Abort unmount attempt when the filesystem is not mounted 797 */ 798 if ((mp->mnt_iflag & IMNT_GONE) != 0) { 799 mutex_exit(&mp->mnt_unmounting); 800 mutex_exit(&syncer_mutex); 801 return ENOENT; 802 } 803 804 used_syncer = (mp->mnt_syncer != NULL); 805 806 /* 807 * XXX Syncer must be frozen when we get here. This should really 808 * be done on a per-mountpoint basis, but the syncer doesn't work 809 * like that. 810 * 811 * The caller of dounmount() must acquire syncer_mutex because 812 * the syncer itself acquires locks in syncer_mutex -> vfs_busy 813 * order, and we must preserve that order to avoid deadlock. 814 * 815 * So, if the file system did not use the syncer, now is 816 * the time to release the syncer_mutex. 817 */ 818 if (used_syncer == 0) { 819 mutex_exit(&syncer_mutex); 820 } 821 mp->mnt_iflag |= IMNT_UNMOUNT; 822 mutex_enter(&mp->mnt_updating); 823 async = mp->mnt_flag & MNT_ASYNC; 824 mp->mnt_flag &= ~MNT_ASYNC; 825 cache_purgevfs(mp); /* remove cache entries for this file sys */ 826 if (mp->mnt_syncer != NULL) 827 vfs_deallocate_syncvnode(mp); 828 error = 0; 829 if ((mp->mnt_flag & MNT_RDONLY) == 0) { 830 error = VFS_SYNC(mp, MNT_WAIT, l->l_cred); 831 } 832 vfs_scrubvnlist(mp); 833 if (error == 0 || (flags & MNT_FORCE)) { 834 error = VFS_UNMOUNT(mp, flags); 835 } 836 if (error) { 837 mp->mnt_iflag &= ~IMNT_UNMOUNT; 838 mutex_exit(&mp->mnt_unmounting); 839 if ((mp->mnt_flag & (MNT_RDONLY | MNT_ASYNC)) == 0) 840 (void) vfs_allocate_syncvnode(mp); 841 mp->mnt_flag |= async; 842 mutex_exit(&mp->mnt_updating); 843 if (used_syncer) 844 mutex_exit(&syncer_mutex); 845 return (error); 846 } 847 mutex_exit(&mp->mnt_updating); 848 vfs_scrubvnlist(mp); 849 850 /* 851 * release mnt_umounting lock here, because other code calls 852 * vfs_busy() while holding the mountlist_lock. 853 * 854 * mark filesystem as gone to prevent further umounts 855 * after mnt_umounting lock is gone, this also prevents 856 * vfs_busy() from succeeding. 857 */ 858 mp->mnt_iflag |= IMNT_GONE; 859 mutex_exit(&mp->mnt_unmounting); 860 861 if ((coveredvp = mp->mnt_vnodecovered) != NULLVP) { 862 vn_lock(coveredvp, LK_EXCLUSIVE | LK_RETRY); 863 coveredvp->v_mountedhere = NULL; 864 VOP_UNLOCK(coveredvp); 865 } 866 mutex_enter(&mountlist_lock); 867 TAILQ_REMOVE(&mountlist, mp, mnt_list); 868 mutex_exit(&mountlist_lock); 869 if (TAILQ_FIRST(&mp->mnt_vnodelist) != NULL) 870 panic("unmount: dangling vnode"); 871 if (used_syncer) 872 mutex_exit(&syncer_mutex); 873 vfs_hooks_unmount(mp); 874 875 vfs_destroy(mp); /* reference from mount() */ 876 if (coveredvp != NULLVP) { 877 vrele(coveredvp); 878 } 879 return (0); 880 } 881 882 /* 883 * Unmount all file systems. 884 * We traverse the list in reverse order under the assumption that doing so 885 * will avoid needing to worry about dependencies. 886 */ 887 bool 888 vfs_unmountall(struct lwp *l) 889 { 890 891 printf("unmounting file systems...\n"); 892 return vfs_unmountall1(l, true, true); 893 } 894 895 static void 896 vfs_unmount_print(struct mount *mp, const char *pfx) 897 { 898 899 aprint_verbose("%sunmounted %s on %s type %s\n", pfx, 900 mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname, 901 mp->mnt_stat.f_fstypename); 902 } 903 904 bool 905 vfs_unmount_forceone(struct lwp *l) 906 { 907 struct mount *mp, *nmp; 908 int error; 909 910 nmp = NULL; 911 912 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 913 if (nmp == NULL || mp->mnt_gen > nmp->mnt_gen) { 914 nmp = mp; 915 } 916 } 917 if (nmp == NULL) { 918 return false; 919 } 920 921 #ifdef DEBUG 922 printf("forcefully unmounting %s (%s)...\n", 923 nmp->mnt_stat.f_mntonname, nmp->mnt_stat.f_mntfromname); 924 #endif 925 atomic_inc_uint(&nmp->mnt_refcnt); 926 if ((error = dounmount(nmp, MNT_FORCE, l)) == 0) { 927 vfs_unmount_print(nmp, "forcefully "); 928 return true; 929 } else { 930 vfs_destroy(nmp); 931 } 932 933 #ifdef DEBUG 934 printf("forceful unmount of %s failed with error %d\n", 935 nmp->mnt_stat.f_mntonname, error); 936 #endif 937 938 return false; 939 } 940 941 bool 942 vfs_unmountall1(struct lwp *l, bool force, bool verbose) 943 { 944 struct mount *mp, *nmp; 945 bool any_error = false, progress = false; 946 int error; 947 948 TAILQ_FOREACH_REVERSE_SAFE(mp, &mountlist, mntlist, mnt_list, nmp) { 949 #ifdef DEBUG 950 printf("unmounting %p %s (%s)...\n", 951 (void *)mp, mp->mnt_stat.f_mntonname, 952 mp->mnt_stat.f_mntfromname); 953 #endif 954 atomic_inc_uint(&mp->mnt_refcnt); 955 if ((error = dounmount(mp, force ? MNT_FORCE : 0, l)) == 0) { 956 vfs_unmount_print(mp, ""); 957 progress = true; 958 } else { 959 vfs_destroy(mp); 960 if (verbose) { 961 printf("unmount of %s failed with error %d\n", 962 mp->mnt_stat.f_mntonname, error); 963 } 964 any_error = true; 965 } 966 } 967 if (verbose) { 968 printf("unmounting done\n"); 969 } 970 if (any_error && verbose) { 971 printf("WARNING: some file systems would not unmount\n"); 972 } 973 return progress; 974 } 975 976 void 977 vfs_sync_all(struct lwp *l) 978 { 979 printf("syncing disks... "); 980 981 /* remove user processes from run queue */ 982 suspendsched(); 983 (void)spl0(); 984 985 /* avoid coming back this way again if we panic. */ 986 doing_shutdown = 1; 987 988 do_sys_sync(l); 989 990 /* Wait for sync to finish. */ 991 if (buf_syncwait() != 0) { 992 #if defined(DDB) && defined(DEBUG_HALT_BUSY) 993 Debugger(); 994 #endif 995 printf("giving up\n"); 996 return; 997 } else 998 printf("done\n"); 999 } 1000 1001 /* 1002 * Sync and unmount file systems before shutting down. 1003 */ 1004 void 1005 vfs_shutdown(void) 1006 { 1007 lwp_t *l = curlwp; 1008 1009 vfs_sync_all(l); 1010 1011 /* 1012 * If we have paniced - do not make the situation potentially 1013 * worse by unmounting the file systems. 1014 */ 1015 if (panicstr != NULL) { 1016 return; 1017 } 1018 1019 /* Unmount file systems. */ 1020 vfs_unmountall(l); 1021 } 1022 1023 /* 1024 * Print a list of supported file system types (used by vfs_mountroot) 1025 */ 1026 static void 1027 vfs_print_fstypes(void) 1028 { 1029 struct vfsops *v; 1030 int cnt = 0; 1031 1032 mutex_enter(&vfs_list_lock); 1033 LIST_FOREACH(v, &vfs_list, vfs_list) 1034 ++cnt; 1035 mutex_exit(&vfs_list_lock); 1036 1037 if (cnt == 0) { 1038 printf("WARNING: No file system modules have been loaded.\n"); 1039 return; 1040 } 1041 1042 printf("Supported file systems:"); 1043 mutex_enter(&vfs_list_lock); 1044 LIST_FOREACH(v, &vfs_list, vfs_list) { 1045 printf(" %s", v->vfs_name); 1046 } 1047 mutex_exit(&vfs_list_lock); 1048 printf("\n"); 1049 } 1050 1051 /* 1052 * Mount the root file system. If the operator didn't specify a 1053 * file system to use, try all possible file systems until one 1054 * succeeds. 1055 */ 1056 int 1057 vfs_mountroot(void) 1058 { 1059 struct vfsops *v; 1060 int error = ENODEV; 1061 1062 if (root_device == NULL) 1063 panic("vfs_mountroot: root device unknown"); 1064 1065 switch (device_class(root_device)) { 1066 case DV_IFNET: 1067 if (rootdev != NODEV) 1068 panic("vfs_mountroot: rootdev set for DV_IFNET " 1069 "(0x%llx -> %llu,%llu)", 1070 (unsigned long long)rootdev, 1071 (unsigned long long)major(rootdev), 1072 (unsigned long long)minor(rootdev)); 1073 break; 1074 1075 case DV_DISK: 1076 if (rootdev == NODEV) 1077 panic("vfs_mountroot: rootdev not set for DV_DISK"); 1078 if (bdevvp(rootdev, &rootvp)) 1079 panic("vfs_mountroot: can't get vnode for rootdev"); 1080 error = VOP_OPEN(rootvp, FREAD, FSCRED); 1081 if (error) { 1082 printf("vfs_mountroot: can't open root device\n"); 1083 return (error); 1084 } 1085 break; 1086 1087 case DV_VIRTUAL: 1088 break; 1089 1090 default: 1091 printf("%s: inappropriate for root file system\n", 1092 device_xname(root_device)); 1093 return (ENODEV); 1094 } 1095 1096 /* 1097 * If user specified a root fs type, use it. Make sure the 1098 * specified type exists and has a mount_root() 1099 */ 1100 if (strcmp(rootfstype, ROOT_FSTYPE_ANY) != 0) { 1101 v = vfs_getopsbyname(rootfstype); 1102 error = EFTYPE; 1103 if (v != NULL) { 1104 if (v->vfs_mountroot != NULL) { 1105 error = (v->vfs_mountroot)(); 1106 } 1107 v->vfs_refcount--; 1108 } 1109 goto done; 1110 } 1111 1112 /* 1113 * Try each file system currently configured into the kernel. 1114 */ 1115 mutex_enter(&vfs_list_lock); 1116 LIST_FOREACH(v, &vfs_list, vfs_list) { 1117 if (v->vfs_mountroot == NULL) 1118 continue; 1119 #ifdef DEBUG 1120 aprint_normal("mountroot: trying %s...\n", v->vfs_name); 1121 #endif 1122 v->vfs_refcount++; 1123 mutex_exit(&vfs_list_lock); 1124 error = (*v->vfs_mountroot)(); 1125 mutex_enter(&vfs_list_lock); 1126 v->vfs_refcount--; 1127 if (!error) { 1128 aprint_normal("root file system type: %s\n", 1129 v->vfs_name); 1130 break; 1131 } 1132 } 1133 mutex_exit(&vfs_list_lock); 1134 1135 if (v == NULL) { 1136 vfs_print_fstypes(); 1137 printf("no file system for %s", device_xname(root_device)); 1138 if (device_class(root_device) == DV_DISK) 1139 printf(" (dev 0x%llx)", (unsigned long long)rootdev); 1140 printf("\n"); 1141 error = EFTYPE; 1142 } 1143 1144 done: 1145 if (error && device_class(root_device) == DV_DISK) { 1146 VOP_CLOSE(rootvp, FREAD, FSCRED); 1147 vrele(rootvp); 1148 } 1149 if (error == 0) { 1150 struct mount *mp; 1151 extern struct cwdinfo cwdi0; 1152 1153 mp = TAILQ_FIRST(&mountlist); 1154 mp->mnt_flag |= MNT_ROOTFS; 1155 mp->mnt_op->vfs_refcount++; 1156 1157 /* 1158 * Get the vnode for '/'. Set cwdi0.cwdi_cdir to 1159 * reference it. 1160 */ 1161 error = VFS_ROOT(mp, &rootvnode); 1162 if (error) 1163 panic("cannot find root vnode, error=%d", error); 1164 cwdi0.cwdi_cdir = rootvnode; 1165 vref(cwdi0.cwdi_cdir); 1166 VOP_UNLOCK(rootvnode); 1167 cwdi0.cwdi_rdir = NULL; 1168 1169 /* 1170 * Now that root is mounted, we can fixup initproc's CWD 1171 * info. All other processes are kthreads, which merely 1172 * share proc0's CWD info. 1173 */ 1174 initproc->p_cwdi->cwdi_cdir = rootvnode; 1175 vref(initproc->p_cwdi->cwdi_cdir); 1176 initproc->p_cwdi->cwdi_rdir = NULL; 1177 /* 1178 * Enable loading of modules from the filesystem 1179 */ 1180 module_load_vfs_init(); 1181 1182 } 1183 return (error); 1184 } 1185 1186 /* 1187 * mount_specific_key_create -- 1188 * Create a key for subsystem mount-specific data. 1189 */ 1190 int 1191 mount_specific_key_create(specificdata_key_t *keyp, specificdata_dtor_t dtor) 1192 { 1193 1194 return specificdata_key_create(mount_specificdata_domain, keyp, dtor); 1195 } 1196 1197 /* 1198 * mount_specific_key_delete -- 1199 * Delete a key for subsystem mount-specific data. 1200 */ 1201 void 1202 mount_specific_key_delete(specificdata_key_t key) 1203 { 1204 1205 specificdata_key_delete(mount_specificdata_domain, key); 1206 } 1207 1208 /* 1209 * mount_initspecific -- 1210 * Initialize a mount's specificdata container. 1211 */ 1212 void 1213 mount_initspecific(struct mount *mp) 1214 { 1215 int error __diagused; 1216 1217 error = specificdata_init(mount_specificdata_domain, 1218 &mp->mnt_specdataref); 1219 KASSERT(error == 0); 1220 } 1221 1222 /* 1223 * mount_finispecific -- 1224 * Finalize a mount's specificdata container. 1225 */ 1226 void 1227 mount_finispecific(struct mount *mp) 1228 { 1229 1230 specificdata_fini(mount_specificdata_domain, &mp->mnt_specdataref); 1231 } 1232 1233 /* 1234 * mount_getspecific -- 1235 * Return mount-specific data corresponding to the specified key. 1236 */ 1237 void * 1238 mount_getspecific(struct mount *mp, specificdata_key_t key) 1239 { 1240 1241 return specificdata_getspecific(mount_specificdata_domain, 1242 &mp->mnt_specdataref, key); 1243 } 1244 1245 /* 1246 * mount_setspecific -- 1247 * Set mount-specific data corresponding to the specified key. 1248 */ 1249 void 1250 mount_setspecific(struct mount *mp, specificdata_key_t key, void *data) 1251 { 1252 1253 specificdata_setspecific(mount_specificdata_domain, 1254 &mp->mnt_specdataref, key, data); 1255 } 1256 1257 /* 1258 * Check to see if a filesystem is mounted on a block device. 1259 */ 1260 int 1261 vfs_mountedon(vnode_t *vp) 1262 { 1263 vnode_t *vq; 1264 int error = 0; 1265 1266 if (vp->v_type != VBLK) 1267 return ENOTBLK; 1268 if (spec_node_getmountedfs(vp) != NULL) 1269 return EBUSY; 1270 if (spec_node_lookup_by_dev(vp->v_type, vp->v_rdev, &vq) == 0) { 1271 if (spec_node_getmountedfs(vq) != NULL) 1272 error = EBUSY; 1273 vrele(vq); 1274 } 1275 1276 return error; 1277 } 1278 1279 /* 1280 * Check if a device pointed to by vp is mounted. 1281 * 1282 * Returns: 1283 * EINVAL if it's not a disk 1284 * EBUSY if it's a disk and mounted 1285 * 0 if it's a disk and not mounted 1286 */ 1287 int 1288 rawdev_mounted(vnode_t *vp, vnode_t **bvpp) 1289 { 1290 vnode_t *bvp; 1291 dev_t dev; 1292 int d_type; 1293 1294 bvp = NULL; 1295 d_type = D_OTHER; 1296 1297 if (iskmemvp(vp)) 1298 return EINVAL; 1299 1300 switch (vp->v_type) { 1301 case VCHR: { 1302 const struct cdevsw *cdev; 1303 1304 dev = vp->v_rdev; 1305 cdev = cdevsw_lookup(dev); 1306 if (cdev != NULL) { 1307 dev_t blkdev; 1308 1309 blkdev = devsw_chr2blk(dev); 1310 if (blkdev != NODEV) { 1311 if (vfinddev(blkdev, VBLK, &bvp) != 0) { 1312 d_type = (cdev->d_flag & D_TYPEMASK); 1313 /* XXX: what if bvp disappears? */ 1314 vrele(bvp); 1315 } 1316 } 1317 } 1318 1319 break; 1320 } 1321 1322 case VBLK: { 1323 const struct bdevsw *bdev; 1324 1325 dev = vp->v_rdev; 1326 bdev = bdevsw_lookup(dev); 1327 if (bdev != NULL) 1328 d_type = (bdev->d_flag & D_TYPEMASK); 1329 1330 bvp = vp; 1331 1332 break; 1333 } 1334 1335 default: 1336 break; 1337 } 1338 1339 if (d_type != D_DISK) 1340 return EINVAL; 1341 1342 if (bvpp != NULL) 1343 *bvpp = bvp; 1344 1345 /* 1346 * XXX: This is bogus. We should be failing the request 1347 * XXX: not only if this specific slice is mounted, but 1348 * XXX: if it's on a disk with any other mounted slice. 1349 */ 1350 if (vfs_mountedon(bvp)) 1351 return EBUSY; 1352 1353 return 0; 1354 } 1355 1356 /* 1357 * Make a 'unique' number from a mount type name. 1358 */ 1359 long 1360 makefstype(const char *type) 1361 { 1362 long rv; 1363 1364 for (rv = 0; *type; type++) { 1365 rv <<= 2; 1366 rv ^= *type; 1367 } 1368 return rv; 1369 } 1370 1371 void 1372 mountlist_append(struct mount *mp) 1373 { 1374 mutex_enter(&mountlist_lock); 1375 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1376 mutex_exit(&mountlist_lock); 1377 } 1378