xref: /freebsd/sys/fs/msdosfs/msdosfs_vfsops.c (revision 058ac3e8063366dafa634d9107642e12b038bf09)
1 /* $FreeBSD$ */
2 /*	$NetBSD: msdosfs_vfsops.c,v 1.51 1997/11/17 15:36:58 ws Exp $	*/
3 
4 /*-
5  * SPDX-License-Identifier: BSD-4-Clause
6  *
7  * Copyright (C) 1994, 1995, 1997 Wolfgang Solfrank.
8  * Copyright (C) 1994, 1995, 1997 TooLs GmbH.
9  * All rights reserved.
10  * Original code by Paul Popelka (paulp@uts.amdahl.com) (see below).
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. All advertising materials mentioning features or use of this software
21  *    must display the following acknowledgement:
22  *	This product includes software developed by TooLs GmbH.
23  * 4. The name of TooLs GmbH may not be used to endorse or promote products
24  *    derived from this software without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
27  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
28  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
29  * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
30  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
31  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
32  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
33  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
34  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
35  * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
36  */
37 /*-
38  * Written by Paul Popelka (paulp@uts.amdahl.com)
39  *
40  * You can do anything you want with this software, just don't say you wrote
41  * it, and don't remove this notice.
42  *
43  * This software is provided "as is".
44  *
45  * The author supplies this software to be publicly redistributed on the
46  * understanding that the author is not responsible for the correct
47  * functioning of this software in any circumstances and is not liable for
48  * any damages caused by this software.
49  *
50  * October 1992
51  */
52 
53 #include <sys/param.h>
54 #include <sys/systm.h>
55 #include <sys/buf.h>
56 #include <sys/bufobj.h>
57 #include <sys/conf.h>
58 #include <sys/fcntl.h>
59 #include <sys/iconv.h>
60 #include <sys/kernel.h>
61 #include <sys/lock.h>
62 #include <sys/malloc.h>
63 #include <sys/mount.h>
64 #include <sys/mutex.h>
65 #include <sys/namei.h>
66 #include <sys/priv.h>
67 #include <sys/proc.h>
68 #include <sys/rwlock.h>
69 #include <sys/stat.h>
70 #include <sys/taskqueue.h>
71 #include <sys/vnode.h>
72 
73 #include <geom/geom.h>
74 #include <geom/geom_vfs.h>
75 
76 #include <fs/msdosfs/bootsect.h>
77 #include <fs/msdosfs/bpb.h>
78 #include <fs/msdosfs/direntry.h>
79 #include <fs/msdosfs/denode.h>
80 #include <fs/msdosfs/fat.h>
81 #include <fs/msdosfs/msdosfsmount.h>
82 
83 #ifdef MSDOSFS_DEBUG
84 #include <sys/rwlock.h>
85 #endif
86 
87 static const char msdosfs_lock_msg[] = "fatlk";
88 
89 /* Mount options that we support. */
90 static const char *msdosfs_opts[] = {
91 	"async", "noatime", "noclusterr", "noclusterw",
92 	"export", "force", "from", "sync",
93 	"cs_dos", "cs_local", "cs_win", "dirmask",
94 	"gid", "kiconv", "longname",
95 	"longnames", "mask", "shortname", "shortnames",
96 	"uid", "win95", "nowin95",
97 	NULL
98 };
99 
100 #if 1 /*def PC98*/
101 /*
102  * XXX - The boot signature formatted by NEC PC-98 DOS looks like a
103  *       garbage or a random value :-{
104  *       If you want to use that broken-signatured media, define the
105  *       following symbol even though PC/AT.
106  *       (ex. mount PC-98 DOS formatted FD on PC/AT)
107  */
108 #define	MSDOSFS_NOCHECKSIG
109 #endif
110 
111 MALLOC_DEFINE(M_MSDOSFSMNT, "msdosfs_mount", "MSDOSFS mount structure");
112 static MALLOC_DEFINE(M_MSDOSFSFAT, "msdosfs_fat", "MSDOSFS file allocation table");
113 
114 struct iconv_functions *msdosfs_iconv;
115 
116 static int	update_mp(struct mount *mp, struct thread *td);
117 static int	mountmsdosfs(struct vnode *devvp, struct mount *mp);
118 static void	msdosfs_remount_ro(void *arg, int pending);
119 static vfs_fhtovp_t	msdosfs_fhtovp;
120 static vfs_mount_t	msdosfs_mount;
121 static vfs_root_t	msdosfs_root;
122 static vfs_statfs_t	msdosfs_statfs;
123 static vfs_sync_t	msdosfs_sync;
124 static vfs_unmount_t	msdosfs_unmount;
125 
126 /* Maximum length of a character set name (arbitrary). */
127 #define	MAXCSLEN	64
128 
129 static int
130 update_mp(struct mount *mp, struct thread *td)
131 {
132 	struct msdosfsmount *pmp = VFSTOMSDOSFS(mp);
133 	void *dos, *win, *local;
134 	int error, v;
135 
136 	if (!vfs_getopt(mp->mnt_optnew, "kiconv", NULL, NULL)) {
137 		if (msdosfs_iconv != NULL) {
138 			error = vfs_getopt(mp->mnt_optnew,
139 			    "cs_win", &win, NULL);
140 			if (!error)
141 				error = vfs_getopt(mp->mnt_optnew,
142 				    "cs_local", &local, NULL);
143 			if (!error)
144 				error = vfs_getopt(mp->mnt_optnew,
145 				    "cs_dos", &dos, NULL);
146 			if (!error) {
147 				msdosfs_iconv->open(win, local, &pmp->pm_u2w);
148 				msdosfs_iconv->open(local, win, &pmp->pm_w2u);
149 				msdosfs_iconv->open(dos, local, &pmp->pm_u2d);
150 				msdosfs_iconv->open(local, dos, &pmp->pm_d2u);
151 			}
152 			if (error != 0)
153 				return (error);
154 		} else {
155 			pmp->pm_w2u = NULL;
156 			pmp->pm_u2w = NULL;
157 			pmp->pm_d2u = NULL;
158 			pmp->pm_u2d = NULL;
159 		}
160 	}
161 
162 	if (vfs_scanopt(mp->mnt_optnew, "gid", "%d", &v) == 1)
163 		pmp->pm_gid = v;
164 	if (vfs_scanopt(mp->mnt_optnew, "uid", "%d", &v) == 1)
165 		pmp->pm_uid = v;
166 	if (vfs_scanopt(mp->mnt_optnew, "mask", "%d", &v) == 1)
167 		pmp->pm_mask = v & ALLPERMS;
168 	if (vfs_scanopt(mp->mnt_optnew, "dirmask", "%d", &v) == 1)
169 		pmp->pm_dirmask = v & ALLPERMS;
170 	vfs_flagopt(mp->mnt_optnew, "shortname",
171 	    &pmp->pm_flags, MSDOSFSMNT_SHORTNAME);
172 	vfs_flagopt(mp->mnt_optnew, "shortnames",
173 	    &pmp->pm_flags, MSDOSFSMNT_SHORTNAME);
174 	vfs_flagopt(mp->mnt_optnew, "longname",
175 	    &pmp->pm_flags, MSDOSFSMNT_LONGNAME);
176 	vfs_flagopt(mp->mnt_optnew, "longnames",
177 	    &pmp->pm_flags, MSDOSFSMNT_LONGNAME);
178 	vfs_flagopt(mp->mnt_optnew, "kiconv",
179 	    &pmp->pm_flags, MSDOSFSMNT_KICONV);
180 
181 	if (vfs_getopt(mp->mnt_optnew, "nowin95", NULL, NULL) == 0)
182 		pmp->pm_flags |= MSDOSFSMNT_NOWIN95;
183 	else
184 		pmp->pm_flags &= ~MSDOSFSMNT_NOWIN95;
185 
186 	if (pmp->pm_flags & MSDOSFSMNT_NOWIN95)
187 		pmp->pm_flags |= MSDOSFSMNT_SHORTNAME;
188 	else
189 		pmp->pm_flags |= MSDOSFSMNT_LONGNAME;
190 	return 0;
191 }
192 
193 static int
194 msdosfs_cmount(struct mntarg *ma, void *data, uint64_t flags)
195 {
196 	struct msdosfs_args args;
197 	int error;
198 
199 	if (data == NULL)
200 		return (EINVAL);
201 	error = copyin(data, &args, sizeof args);
202 	if (error)
203 		return (error);
204 
205 	ma = mount_argsu(ma, "from", args.fspec, MAXPATHLEN);
206 	ma = mount_arg(ma, "export", &args.export, sizeof(args.export));
207 	ma = mount_argf(ma, "uid", "%d", args.uid);
208 	ma = mount_argf(ma, "gid", "%d", args.gid);
209 	ma = mount_argf(ma, "mask", "%d", args.mask);
210 	ma = mount_argf(ma, "dirmask", "%d", args.dirmask);
211 
212 	ma = mount_argb(ma, args.flags & MSDOSFSMNT_SHORTNAME, "noshortname");
213 	ma = mount_argb(ma, args.flags & MSDOSFSMNT_LONGNAME, "nolongname");
214 	ma = mount_argb(ma, !(args.flags & MSDOSFSMNT_NOWIN95), "nowin95");
215 	ma = mount_argb(ma, args.flags & MSDOSFSMNT_KICONV, "nokiconv");
216 
217 	ma = mount_argsu(ma, "cs_win", args.cs_win, MAXCSLEN);
218 	ma = mount_argsu(ma, "cs_dos", args.cs_dos, MAXCSLEN);
219 	ma = mount_argsu(ma, "cs_local", args.cs_local, MAXCSLEN);
220 
221 	error = kernel_mount(ma, flags);
222 
223 	return (error);
224 }
225 
226 /*
227  * mp - path - addr in user space of mount point (ie /usr or whatever)
228  * data - addr in user space of mount params including the name of the block
229  * special file to treat as a filesystem.
230  */
231 static int
232 msdosfs_mount(struct mount *mp)
233 {
234 	struct vnode *devvp, *odevvp;	  /* vnode for blk device to mount */
235 	struct thread *td;
236 	/* msdosfs specific mount control block */
237 	struct msdosfsmount *pmp = NULL;
238 	struct nameidata ndp;
239 	int error, flags;
240 	accmode_t accmode;
241 	char *from;
242 
243 	td = curthread;
244 	if (vfs_filteropt(mp->mnt_optnew, msdosfs_opts))
245 		return (EINVAL);
246 
247 	/*
248 	 * If updating, check whether changing from read-only to
249 	 * read/write; if there is no device name, that's all we do.
250 	 */
251 	if (mp->mnt_flag & MNT_UPDATE) {
252 		pmp = VFSTOMSDOSFS(mp);
253 		if (!(pmp->pm_flags & MSDOSFSMNT_RONLY) &&
254 		    vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) {
255 			if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
256 				return (error);
257 			error = vfs_write_suspend_umnt(mp);
258 			if (error != 0)
259 				return (error);
260 
261 			flags = WRITECLOSE;
262 			if (mp->mnt_flag & MNT_FORCE)
263 				flags |= FORCECLOSE;
264 			error = vflush(mp, 0, flags, td);
265 			if (error != 0) {
266 				vfs_write_resume(mp, 0);
267 				return (error);
268 			}
269 
270 			/*
271 			 * Now the volume is clean.  Mark it so while the
272 			 * device is still rw.
273 			 */
274 			error = markvoldirty(pmp, 0);
275 			if (error != 0) {
276 				vfs_write_resume(mp, 0);
277 				(void)markvoldirty(pmp, 1);
278 				return (error);
279 			}
280 
281 			/* Downgrade the device from rw to ro. */
282 			g_topology_lock();
283 			error = g_access(pmp->pm_cp, 0, -1, 0);
284 			g_topology_unlock();
285 			if (error) {
286 				vfs_write_resume(mp, 0);
287 				(void)markvoldirty(pmp, 1);
288 				return (error);
289 			}
290 
291 			/*
292 			 * Backing out after an error was painful in the
293 			 * above.  Now we are committed to succeeding.
294 			 */
295 			pmp->pm_fmod = 0;
296 			pmp->pm_flags |= MSDOSFSMNT_RONLY;
297 			MNT_ILOCK(mp);
298 			mp->mnt_flag |= MNT_RDONLY;
299 			MNT_IUNLOCK(mp);
300 			vfs_write_resume(mp, 0);
301 		} else if ((pmp->pm_flags & MSDOSFSMNT_RONLY) &&
302 		    !vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) {
303 			/*
304 			 * If upgrade to read-write by non-root, then verify
305 			 * that user has necessary permissions on the device.
306 			 */
307 			odevvp = pmp->pm_odevvp;
308 			vn_lock(odevvp, LK_EXCLUSIVE | LK_RETRY);
309 			error = VOP_ACCESS(odevvp, VREAD | VWRITE,
310 			    td->td_ucred, td);
311 			if (error)
312 				error = priv_check(td, PRIV_VFS_MOUNT_PERM);
313 			if (error) {
314 				VOP_UNLOCK(odevvp);
315 				return (error);
316 			}
317 			VOP_UNLOCK(odevvp);
318 			g_topology_lock();
319 			error = g_access(pmp->pm_cp, 0, 1, 0);
320 			g_topology_unlock();
321 			if (error)
322 				return (error);
323 
324 			/* Now that the volume is modifiable, mark it dirty. */
325 			error = markvoldirty_upgrade(pmp, true, true);
326 			if (error) {
327 				/*
328 				 * If dirtying the superblock failed, drop GEOM
329 				 * 'w' refs (we're still RO).
330 				 */
331 				g_topology_lock();
332 				(void)g_access(pmp->pm_cp, 0, -1, 0);
333 				g_topology_unlock();
334 
335 				return (error);
336 			}
337 
338 			pmp->pm_fmod = 1;
339 			pmp->pm_flags &= ~MSDOSFSMNT_RONLY;
340 			MNT_ILOCK(mp);
341 			mp->mnt_flag &= ~MNT_RDONLY;
342 			MNT_IUNLOCK(mp);
343 		}
344 
345 		/*
346 		 * Avoid namei() below.  The "from" option is not set.
347 		 * Update of the devvp is pointless for this case.
348 		 */
349 		if ((pmp->pm_flags & MSDOSFS_ERR_RO) != 0)
350 			return (0);
351 	}
352 	/*
353 	 * Not an update, or updating the name: look up the name
354 	 * and verify that it refers to a sensible disk device.
355 	 */
356 	if (vfs_getopt(mp->mnt_optnew, "from", (void **)&from, NULL))
357 		return (EINVAL);
358 	NDINIT(&ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, from);
359 	error = namei(&ndp);
360 	if (error)
361 		return (error);
362 	devvp = ndp.ni_vp;
363 	NDFREE_PNBUF(&ndp);
364 
365 	if (!vn_isdisk_error(devvp, &error)) {
366 		vput(devvp);
367 		return (error);
368 	}
369 	/*
370 	 * If mount by non-root, then verify that user has necessary
371 	 * permissions on the device.
372 	 */
373 	accmode = VREAD;
374 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
375 		accmode |= VWRITE;
376 	error = VOP_ACCESS(devvp, accmode, td->td_ucred, td);
377 	if (error)
378 		error = priv_check(td, PRIV_VFS_MOUNT_PERM);
379 	if (error) {
380 		vput(devvp);
381 		return (error);
382 	}
383 	if ((mp->mnt_flag & MNT_UPDATE) == 0) {
384 		error = mountmsdosfs(devvp, mp);
385 #ifdef MSDOSFS_DEBUG		/* only needed for the printf below */
386 		pmp = VFSTOMSDOSFS(mp);
387 #endif
388 	} else {
389 		vput(devvp);
390 		if (devvp != pmp->pm_odevvp)
391 			return (EINVAL);	/* XXX needs translation */
392 	}
393 	if (error) {
394 		vrele(devvp);
395 		return (error);
396 	}
397 
398 	error = update_mp(mp, td);
399 	if (error) {
400 		if ((mp->mnt_flag & MNT_UPDATE) == 0)
401 			msdosfs_unmount(mp, MNT_FORCE);
402 		return error;
403 	}
404 
405 	vfs_mountedfrom(mp, from);
406 #ifdef MSDOSFS_DEBUG
407 	printf("msdosfs_mount(): mp %p, pmp %p, inusemap %p\n", mp, pmp, pmp->pm_inusemap);
408 #endif
409 	return (0);
410 }
411 
412 static int
413 mountmsdosfs(struct vnode *odevvp, struct mount *mp)
414 {
415 	struct msdosfsmount *pmp;
416 	struct buf *bp;
417 	struct cdev *dev;
418 	struct vnode *devvp;
419 	union bootsector *bsp;
420 	struct byte_bpb33 *b33;
421 	struct byte_bpb50 *b50;
422 	struct byte_bpb710 *b710;
423 	uint8_t SecPerClust;
424 	u_long clusters;
425 	int ronly, error;
426 	struct g_consumer *cp;
427 	struct bufobj *bo;
428 
429 	bp = NULL;		/* This and pmp both used in error_exit. */
430 	pmp = NULL;
431 	ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
432 
433 	devvp = mntfs_allocvp(mp, odevvp);
434 	dev = devvp->v_rdev;
435 	if (atomic_cmpset_acq_ptr((uintptr_t *)&dev->si_mountpt, 0,
436 	    (uintptr_t)mp) == 0) {
437 		mntfs_freevp(devvp);
438 		return (EBUSY);
439 	}
440 	g_topology_lock();
441 	error = g_vfs_open(devvp, &cp, "msdosfs", ronly ? 0 : 1);
442 	g_topology_unlock();
443 	if (error != 0) {
444 		atomic_store_rel_ptr((uintptr_t *)&dev->si_mountpt, 0);
445 		mntfs_freevp(devvp);
446 		return (error);
447 	}
448 	dev_ref(dev);
449 	bo = &devvp->v_bufobj;
450 	BO_LOCK(&odevvp->v_bufobj);
451 	odevvp->v_bufobj.bo_flag |= BO_NOBUFS;
452 	BO_UNLOCK(&odevvp->v_bufobj);
453 	VOP_UNLOCK(devvp);
454 	if (dev->si_iosize_max != 0)
455 		mp->mnt_iosize_max = dev->si_iosize_max;
456 	if (mp->mnt_iosize_max > maxphys)
457 		mp->mnt_iosize_max = maxphys;
458 
459 	/*
460 	 * Read the boot sector of the filesystem, and then check the
461 	 * boot signature.  If not a dos boot sector then error out.
462 	 *
463 	 * NOTE: 8192 is a magic size that works for ffs.
464 	 */
465 	error = bread(devvp, 0, 8192, NOCRED, &bp);
466 	if (error)
467 		goto error_exit;
468 	bp->b_flags |= B_AGE;
469 	bsp = (union bootsector *)bp->b_data;
470 	b33 = (struct byte_bpb33 *)bsp->bs33.bsBPB;
471 	b50 = (struct byte_bpb50 *)bsp->bs50.bsBPB;
472 	b710 = (struct byte_bpb710 *)bsp->bs710.bsBPB;
473 
474 #ifndef MSDOSFS_NOCHECKSIG
475 	if (bsp->bs50.bsBootSectSig0 != BOOTSIG0 ||
476 	    bsp->bs50.bsBootSectSig1 != BOOTSIG1) {
477 		error = EINVAL;
478 		goto error_exit;
479 	}
480 #endif
481 
482 	pmp = malloc(sizeof(*pmp), M_MSDOSFSMNT, M_WAITOK | M_ZERO);
483 	pmp->pm_mountp = mp;
484 	pmp->pm_cp = cp;
485 	pmp->pm_bo = bo;
486 
487 	lockinit(&pmp->pm_fatlock, 0, msdosfs_lock_msg, 0, 0);
488 	lockinit(&pmp->pm_checkpath_lock, 0, "msdoscp", 0, 0);
489 
490 	TASK_INIT(&pmp->pm_rw2ro_task, 0, msdosfs_remount_ro, pmp);
491 
492 	/*
493 	 * Initialize ownerships and permissions, since nothing else will
494 	 * initialize them iff we are mounting root.
495 	 */
496 	pmp->pm_uid = UID_ROOT;
497 	pmp->pm_gid = GID_WHEEL;
498 	pmp->pm_mask = pmp->pm_dirmask = S_IXUSR | S_IXGRP | S_IXOTH |
499 	    S_IRUSR | S_IRGRP | S_IROTH | S_IWUSR;
500 
501 	/*
502 	 * Compute several useful quantities from the bpb in the
503 	 * bootsector.  Copy in the dos 5 variant of the bpb then fix up
504 	 * the fields that are different between dos 5 and dos 3.3.
505 	 */
506 	SecPerClust = b50->bpbSecPerClust;
507 	pmp->pm_BytesPerSec = getushort(b50->bpbBytesPerSec);
508 	if (pmp->pm_BytesPerSec < DEV_BSIZE) {
509 		error = EINVAL;
510 		goto error_exit;
511 	}
512 	pmp->pm_ResSectors = getushort(b50->bpbResSectors);
513 	pmp->pm_FATs = b50->bpbFATs;
514 	pmp->pm_RootDirEnts = getushort(b50->bpbRootDirEnts);
515 	pmp->pm_Sectors = getushort(b50->bpbSectors);
516 	pmp->pm_FATsecs = getushort(b50->bpbFATsecs);
517 	pmp->pm_SecPerTrack = getushort(b50->bpbSecPerTrack);
518 	pmp->pm_Heads = getushort(b50->bpbHeads);
519 	pmp->pm_Media = b50->bpbMedia;
520 
521 	/* calculate the ratio of sector size to DEV_BSIZE */
522 	pmp->pm_BlkPerSec = pmp->pm_BytesPerSec / DEV_BSIZE;
523 
524 	/*
525 	 * We don't check pm_Heads nor pm_SecPerTrack, because
526 	 * these may not be set for EFI file systems. We don't
527 	 * use these anyway, so we're unaffected if they are
528 	 * invalid.
529 	 */
530 	if (pmp->pm_BytesPerSec == 0 || SecPerClust == 0) {
531 		error = EINVAL;
532 		goto error_exit;
533 	}
534 
535 	if (pmp->pm_Sectors == 0) {
536 		pmp->pm_HiddenSects = getulong(b50->bpbHiddenSecs);
537 		pmp->pm_HugeSectors = getulong(b50->bpbHugeSectors);
538 	} else {
539 		pmp->pm_HiddenSects = getushort(b33->bpbHiddenSecs);
540 		pmp->pm_HugeSectors = pmp->pm_Sectors;
541 	}
542 
543 	if (pmp->pm_RootDirEnts == 0) {
544 		if (pmp->pm_FATsecs != 0 || getushort(b710->bpbFSVers) != 0) {
545 			error = EINVAL;
546 #ifdef MSDOSFS_DEBUG
547 			printf("mountmsdosfs(): bad FAT32 filesystem\n");
548 #endif
549 			goto error_exit;
550 		}
551 		pmp->pm_fatmask = FAT32_MASK;
552 		pmp->pm_fatmult = 4;
553 		pmp->pm_fatdiv = 1;
554 		pmp->pm_FATsecs = getulong(b710->bpbBigFATsecs);
555 		if ((getushort(b710->bpbExtFlags) & FATMIRROR) != 0)
556 			pmp->pm_curfat = getushort(b710->bpbExtFlags) & FATNUM;
557 		else
558 			pmp->pm_flags |= MSDOSFS_FATMIRROR;
559 	} else
560 		pmp->pm_flags |= MSDOSFS_FATMIRROR;
561 
562 	/*
563 	 * Check a few values (could do some more):
564 	 * - logical sector size: power of 2, >= block size
565 	 * - sectors per cluster: power of 2, >= 1
566 	 * - number of sectors:   >= 1, <= size of partition
567 	 * - number of FAT sectors: >= 1
568 	 */
569 	if (SecPerClust == 0 || (SecPerClust & (SecPerClust - 1)) != 0 ||
570 	    pmp->pm_BytesPerSec < DEV_BSIZE ||
571 	    (pmp->pm_BytesPerSec & (pmp->pm_BytesPerSec - 1)) != 0 ||
572 	    pmp->pm_HugeSectors == 0 || pmp->pm_FATsecs == 0 ||
573 	    SecPerClust * pmp->pm_BlkPerSec > MAXBSIZE / DEV_BSIZE) {
574 		error = EINVAL;
575 		goto error_exit;
576 	}
577 
578 	if ((off_t)pmp->pm_HugeSectors * pmp->pm_BytesPerSec <
579 	    pmp->pm_HugeSectors /* overflow */ ||
580 	    (off_t)pmp->pm_HugeSectors * pmp->pm_BytesPerSec >
581 	    cp->provider->mediasize /* past end of vol */) {
582 		error = EINVAL;
583 		goto error_exit;
584 	}
585 
586 	pmp->pm_HugeSectors *= pmp->pm_BlkPerSec;
587 	pmp->pm_HiddenSects *= pmp->pm_BlkPerSec;	/* XXX not used? */
588 	pmp->pm_FATsecs     *= pmp->pm_BlkPerSec;
589 	SecPerClust         *= pmp->pm_BlkPerSec;
590 
591 	pmp->pm_fatblk = pmp->pm_ResSectors * pmp->pm_BlkPerSec;
592 
593 	if (FAT32(pmp)) {
594 		pmp->pm_rootdirblk = getulong(b710->bpbRootClust);
595 		pmp->pm_firstcluster = pmp->pm_fatblk +
596 		    pmp->pm_FATs * pmp->pm_FATsecs;
597 		pmp->pm_fsinfo = getushort(b710->bpbFSInfo) * pmp->pm_BlkPerSec;
598 	} else {
599 		pmp->pm_rootdirblk = pmp->pm_fatblk +
600 		    pmp->pm_FATs * pmp->pm_FATsecs;
601 		pmp->pm_rootdirsize = howmany(pmp->pm_RootDirEnts *
602 		    sizeof(struct direntry), DEV_BSIZE); /* in blocks */
603 		pmp->pm_firstcluster = pmp->pm_rootdirblk + pmp->pm_rootdirsize;
604 	}
605 
606 	if (pmp->pm_HugeSectors <= pmp->pm_firstcluster) {
607 		error = EINVAL;
608 		goto error_exit;
609 	}
610 	pmp->pm_maxcluster = (pmp->pm_HugeSectors - pmp->pm_firstcluster) /
611 	    SecPerClust + 1;
612 	pmp->pm_fatsize = pmp->pm_FATsecs * DEV_BSIZE;	/* XXX not used? */
613 
614 	if (pmp->pm_fatmask == 0) {
615 		if (pmp->pm_maxcluster <= ((CLUST_RSRVD - CLUST_FIRST) &
616 		    FAT12_MASK)) {
617 			/*
618 			 * This will usually be a floppy disk. This size makes
619 			 * sure that one FAT entry will not be split across
620 			 * multiple blocks.
621 			 */
622 			pmp->pm_fatmask = FAT12_MASK;
623 			pmp->pm_fatmult = 3;
624 			pmp->pm_fatdiv = 2;
625 		} else {
626 			pmp->pm_fatmask = FAT16_MASK;
627 			pmp->pm_fatmult = 2;
628 			pmp->pm_fatdiv = 1;
629 		}
630 	}
631 
632 	clusters = (pmp->pm_fatsize / pmp->pm_fatmult) * pmp->pm_fatdiv;
633 	if (pmp->pm_maxcluster >= clusters) {
634 #ifdef MSDOSFS_DEBUG
635 		printf("Warning: number of clusters (%ld) exceeds FAT "
636 		    "capacity (%ld)\n", pmp->pm_maxcluster + 1, clusters);
637 #endif
638 		pmp->pm_maxcluster = clusters - 1;
639 	}
640 
641 	if (FAT12(pmp))
642 		pmp->pm_fatblocksize = 3 * 512;
643 	else
644 		pmp->pm_fatblocksize = PAGE_SIZE;
645 	pmp->pm_fatblocksize = roundup(pmp->pm_fatblocksize,
646 	    pmp->pm_BytesPerSec);
647 	pmp->pm_fatblocksec = pmp->pm_fatblocksize / DEV_BSIZE;
648 	pmp->pm_bnshift = ffs(DEV_BSIZE) - 1;
649 
650 	/*
651 	 * Compute mask and shift value for isolating cluster relative byte
652 	 * offsets and cluster numbers from a file offset.
653 	 */
654 	pmp->pm_bpcluster = SecPerClust * DEV_BSIZE;
655 	pmp->pm_crbomask = pmp->pm_bpcluster - 1;
656 	pmp->pm_cnshift = ffs(pmp->pm_bpcluster) - 1;
657 
658 	/*
659 	 * Check for valid cluster size
660 	 * must be a power of 2
661 	 */
662 	if ((pmp->pm_bpcluster ^ (1 << pmp->pm_cnshift)) != 0) {
663 		error = EINVAL;
664 		goto error_exit;
665 	}
666 
667 	/*
668 	 * Release the bootsector buffer.
669 	 */
670 	brelse(bp);
671 	bp = NULL;
672 
673 	/*
674 	 * Check the fsinfo sector if we have one.  Silently fix up our
675 	 * in-core copy of fp->fsinxtfree if it is unknown (0xffffffff)
676 	 * or too large.  Ignore fp->fsinfree for now, since we need to
677 	 * read the entire FAT anyway to fill the inuse map.
678 	 */
679 	if (pmp->pm_fsinfo) {
680 		struct fsinfo *fp;
681 
682 		if ((error = bread(devvp, pmp->pm_fsinfo, pmp->pm_BytesPerSec,
683 		    NOCRED, &bp)) != 0)
684 			goto error_exit;
685 		fp = (struct fsinfo *)bp->b_data;
686 		if (!bcmp(fp->fsisig1, "RRaA", 4) &&
687 		    !bcmp(fp->fsisig2, "rrAa", 4) &&
688 		    !bcmp(fp->fsisig3, "\0\0\125\252", 4)) {
689 			pmp->pm_nxtfree = getulong(fp->fsinxtfree);
690 			if (pmp->pm_nxtfree > pmp->pm_maxcluster)
691 				pmp->pm_nxtfree = CLUST_FIRST;
692 		} else
693 			pmp->pm_fsinfo = 0;
694 		brelse(bp);
695 		bp = NULL;
696 	}
697 
698 	/*
699 	 * Finish initializing pmp->pm_nxtfree (just in case the first few
700 	 * sectors aren't properly reserved in the FAT).  This completes
701 	 * the fixup for fp->fsinxtfree, and fixes up the zero-initialized
702 	 * value if there is no fsinfo.  We will use pmp->pm_nxtfree
703 	 * internally even if there is no fsinfo.
704 	 */
705 	if (pmp->pm_nxtfree < CLUST_FIRST)
706 		pmp->pm_nxtfree = CLUST_FIRST;
707 
708 	/*
709 	 * Allocate memory for the bitmap of allocated clusters, and then
710 	 * fill it in.
711 	 */
712 	pmp->pm_inusemap = malloc(howmany(pmp->pm_maxcluster + 1,
713 	    N_INUSEBITS) * sizeof(*pmp->pm_inusemap), M_MSDOSFSFAT, M_WAITOK);
714 
715 	/*
716 	 * fillinusemap() needs pm_devvp.
717 	 */
718 	pmp->pm_devvp = devvp;
719 	pmp->pm_odevvp = odevvp;
720 	pmp->pm_dev = dev;
721 
722 	/*
723 	 * Have the inuse map filled in.
724 	 */
725 	MSDOSFS_LOCK_MP(pmp);
726 	error = fillinusemap(pmp);
727 	MSDOSFS_UNLOCK_MP(pmp);
728 	if (error != 0)
729 		goto error_exit;
730 
731 	/*
732 	 * If they want FAT updates to be synchronous then let them suffer
733 	 * the performance degradation in exchange for the on disk copy of
734 	 * the FAT being correct just about all the time.  I suppose this
735 	 * would be a good thing to turn on if the kernel is still flakey.
736 	 */
737 	if (mp->mnt_flag & MNT_SYNCHRONOUS)
738 		pmp->pm_flags |= MSDOSFSMNT_WAITONFAT;
739 
740 	/*
741 	 * Finish up.
742 	 */
743 	if (ronly)
744 		pmp->pm_flags |= MSDOSFSMNT_RONLY;
745 	else {
746 		if ((error = markvoldirty(pmp, 1)) != 0)
747 			goto error_exit;
748 		pmp->pm_fmod = 1;
749 	}
750 	mp->mnt_data =  pmp;
751 	mp->mnt_stat.f_fsid.val[0] = dev2udev(dev);
752 	mp->mnt_stat.f_fsid.val[1] = mp->mnt_vfc->vfc_typenum;
753 	MNT_ILOCK(mp);
754 	mp->mnt_flag |= MNT_LOCAL;
755 	mp->mnt_kern_flag |= MNTK_USES_BCACHE | MNTK_NO_IOPF;
756 	MNT_IUNLOCK(mp);
757 
758 	return (0);
759 
760 error_exit:
761 	if (bp != NULL)
762 		brelse(bp);
763 	if (cp != NULL) {
764 		g_topology_lock();
765 		g_vfs_close(cp);
766 		g_topology_unlock();
767 	}
768 	if (pmp != NULL) {
769 		lockdestroy(&pmp->pm_fatlock);
770 		lockdestroy(&pmp->pm_checkpath_lock);
771 		free(pmp->pm_inusemap, M_MSDOSFSFAT);
772 		free(pmp, M_MSDOSFSMNT);
773 		mp->mnt_data = NULL;
774 	}
775 	BO_LOCK(&odevvp->v_bufobj);
776 	odevvp->v_bufobj.bo_flag &= ~BO_NOBUFS;
777 	BO_UNLOCK(&odevvp->v_bufobj);
778 	atomic_store_rel_ptr((uintptr_t *)&dev->si_mountpt, 0);
779 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
780 	mntfs_freevp(devvp);
781 	dev_rel(dev);
782 	return (error);
783 }
784 
785 /*
786  * Unmount the filesystem described by mp.
787  */
788 static int
789 msdosfs_unmount(struct mount *mp, int mntflags)
790 {
791 	struct msdosfsmount *pmp;
792 	int error, flags;
793 	bool susp;
794 
795 	error = flags = 0;
796 	pmp = VFSTOMSDOSFS(mp);
797 	susp = (pmp->pm_flags & MSDOSFSMNT_RONLY) == 0;
798 
799 	if (susp) {
800 		error = vfs_write_suspend_umnt(mp);
801 		if (error != 0)
802 			return (error);
803 	}
804 
805 	if ((mntflags & MNT_FORCE) != 0)
806 		flags |= FORCECLOSE;
807 	error = vflush(mp, 0, flags, curthread);
808 	if (error != 0 && error != ENXIO) {
809 		if (susp)
810 			vfs_write_resume(mp, VR_START_WRITE);
811 		return (error);
812 	}
813 	if (susp) {
814 		error = markvoldirty(pmp, 0);
815 		if (error != 0 && error != ENXIO) {
816 			if (susp)
817 				vfs_write_resume(mp, VR_START_WRITE);
818 			(void)markvoldirty(pmp, 1);
819 			return (error);
820 		}
821 	}
822 	if (pmp->pm_flags & MSDOSFSMNT_KICONV && msdosfs_iconv) {
823 		if (pmp->pm_w2u)
824 			msdosfs_iconv->close(pmp->pm_w2u);
825 		if (pmp->pm_u2w)
826 			msdosfs_iconv->close(pmp->pm_u2w);
827 		if (pmp->pm_d2u)
828 			msdosfs_iconv->close(pmp->pm_d2u);
829 		if (pmp->pm_u2d)
830 			msdosfs_iconv->close(pmp->pm_u2d);
831 	}
832 
833 #ifdef MSDOSFS_DEBUG
834 	{
835 		struct vnode *vp = pmp->pm_devvp;
836 		struct bufobj *bo;
837 
838 		bo = &vp->v_bufobj;
839 		BO_LOCK(bo);
840 		VI_LOCK(vp);
841 		vn_printf(vp,
842 		    "msdosfs_umount(): just before calling VOP_CLOSE()\n");
843 		printf("freef %p, freeb %p, mount %p\n",
844 		    TAILQ_NEXT(vp, v_vnodelist), vp->v_vnodelist.tqe_prev,
845 		    vp->v_mount);
846 		printf("cleanblkhd %p, dirtyblkhd %p, numoutput %d, type %d\n",
847 		    TAILQ_FIRST(&vp->v_bufobj.bo_clean.bv_hd),
848 		    TAILQ_FIRST(&vp->v_bufobj.bo_dirty.bv_hd),
849 		    vp->v_bufobj.bo_numoutput, vp->v_type);
850 		VI_UNLOCK(vp);
851 		BO_UNLOCK(bo);
852 	}
853 #endif
854 	if (susp)
855 		vfs_write_resume(mp, VR_START_WRITE);
856 
857 	vn_lock(pmp->pm_devvp, LK_EXCLUSIVE | LK_RETRY);
858 	g_topology_lock();
859 	g_vfs_close(pmp->pm_cp);
860 	g_topology_unlock();
861 	BO_LOCK(&pmp->pm_odevvp->v_bufobj);
862 	pmp->pm_odevvp->v_bufobj.bo_flag &= ~BO_NOBUFS;
863 	BO_UNLOCK(&pmp->pm_odevvp->v_bufobj);
864 	atomic_store_rel_ptr((uintptr_t *)&pmp->pm_dev->si_mountpt, 0);
865 	mntfs_freevp(pmp->pm_devvp);
866 	vrele(pmp->pm_odevvp);
867 	dev_rel(pmp->pm_dev);
868 	free(pmp->pm_inusemap, M_MSDOSFSFAT);
869 	lockdestroy(&pmp->pm_fatlock);
870 	lockdestroy(&pmp->pm_checkpath_lock);
871 	free(pmp, M_MSDOSFSMNT);
872 	mp->mnt_data = NULL;
873 	MNT_ILOCK(mp);
874 	mp->mnt_flag &= ~MNT_LOCAL;
875 	MNT_IUNLOCK(mp);
876 	return (error);
877 }
878 
879 static void
880 msdosfs_remount_ro(void *arg, int pending)
881 {
882 	struct msdosfsmount *pmp;
883 	int error;
884 
885 	pmp = arg;
886 
887 	MSDOSFS_LOCK_MP(pmp);
888 	if ((pmp->pm_flags & MSDOSFS_ERR_RO) != 0) {
889 		while ((pmp->pm_flags & MSDOSFS_ERR_RO) != 0)
890 			msleep(&pmp->pm_flags, &pmp->pm_fatlock, PVFS,
891 			    "msdoserrro", hz);
892 	} else if ((pmp->pm_mountp->mnt_flag & MNT_RDONLY) == 0) {
893 		pmp->pm_flags |= MSDOSFS_ERR_RO;
894 		MSDOSFS_UNLOCK_MP(pmp);
895 		printf("%s: remounting read-only due to corruption\n",
896 		    pmp->pm_mountp->mnt_stat.f_mntfromname);
897 		error = vfs_remount_ro(pmp->pm_mountp);
898 		if (error != 0)
899 			printf("%s: remounting read-only failed: error %d\n",
900 			    pmp->pm_mountp->mnt_stat.f_mntfromname, error);
901 		else
902 			printf("remounted %s read-only\n",
903 			    pmp->pm_mountp->mnt_stat.f_mntfromname);
904 		MSDOSFS_LOCK_MP(pmp);
905 		pmp->pm_flags &= ~MSDOSFS_ERR_RO;
906 		wakeup(&pmp->pm_flags);
907 	}
908 	MSDOSFS_UNLOCK_MP(pmp);
909 
910 	vfs_unbusy(pmp->pm_mountp);
911 }
912 
913 void
914 msdosfs_integrity_error(struct msdosfsmount *pmp)
915 {
916 	int error;
917 
918 	error = vfs_busy(pmp->pm_mountp, MBF_NOWAIT);
919 	if (error == 0)
920 		taskqueue_enqueue(taskqueue_thread, &pmp->pm_rw2ro_task);
921 	else
922 		printf("%s: integrity error busying failed, error %d\n",
923 		    pmp->pm_mountp->mnt_stat.f_mntfromname, error);
924 }
925 
926 static int
927 msdosfs_root(struct mount *mp, int flags, struct vnode **vpp)
928 {
929 	struct msdosfsmount *pmp = VFSTOMSDOSFS(mp);
930 	struct denode *ndep;
931 	int error;
932 
933 #ifdef MSDOSFS_DEBUG
934 	printf("msdosfs_root(); mp %p, pmp %p\n", mp, pmp);
935 #endif
936 	error = deget(pmp, MSDOSFSROOT, MSDOSFSROOT_OFS, LK_EXCLUSIVE, &ndep);
937 	if (error)
938 		return (error);
939 	*vpp = DETOV(ndep);
940 	return (0);
941 }
942 
943 static int
944 msdosfs_statfs(struct mount *mp, struct statfs *sbp)
945 {
946 	struct msdosfsmount *pmp;
947 
948 	pmp = VFSTOMSDOSFS(mp);
949 	sbp->f_bsize = pmp->pm_bpcluster;
950 	sbp->f_iosize = pmp->pm_bpcluster;
951 	sbp->f_blocks = pmp->pm_maxcluster + 1;
952 	sbp->f_bfree = pmp->pm_freeclustercount;
953 	sbp->f_bavail = pmp->pm_freeclustercount;
954 	sbp->f_files = pmp->pm_RootDirEnts;	/* XXX */
955 	sbp->f_ffree = 0;	/* what to put in here? */
956 	return (0);
957 }
958 
959 /*
960  * If we have an FSInfo block, update it.
961  */
962 static int
963 msdosfs_fsiflush(struct msdosfsmount *pmp, int waitfor)
964 {
965 	struct fsinfo *fp;
966 	struct buf *bp;
967 	int error;
968 
969 	MSDOSFS_LOCK_MP(pmp);
970 	if (pmp->pm_fsinfo == 0 || (pmp->pm_flags & MSDOSFS_FSIMOD) == 0) {
971 		error = 0;
972 		goto unlock;
973 	}
974 	error = bread(pmp->pm_devvp, pmp->pm_fsinfo, pmp->pm_BytesPerSec,
975 	    NOCRED, &bp);
976 	if (error != 0) {
977 		goto unlock;
978 	}
979 	fp = (struct fsinfo *)bp->b_data;
980 	putulong(fp->fsinfree, pmp->pm_freeclustercount);
981 	putulong(fp->fsinxtfree, pmp->pm_nxtfree);
982 	pmp->pm_flags &= ~MSDOSFS_FSIMOD;
983 	if (waitfor == MNT_WAIT)
984 		error = bwrite(bp);
985 	else
986 		bawrite(bp);
987 unlock:
988 	MSDOSFS_UNLOCK_MP(pmp);
989 	return (error);
990 }
991 
992 static int
993 msdosfs_sync(struct mount *mp, int waitfor)
994 {
995 	struct vnode *vp, *nvp;
996 	struct thread *td;
997 	struct denode *dep;
998 	struct msdosfsmount *pmp = VFSTOMSDOSFS(mp);
999 	int error, allerror = 0;
1000 
1001 	td = curthread;
1002 
1003 	/*
1004 	 * If we ever switch to not updating all of the FATs all the time,
1005 	 * this would be the place to update them from the first one.
1006 	 */
1007 	if (pmp->pm_fmod != 0) {
1008 		if (pmp->pm_flags & MSDOSFSMNT_RONLY)
1009 			panic("msdosfs_sync: rofs mod");
1010 		else {
1011 			/* update FATs here */
1012 		}
1013 	}
1014 	/*
1015 	 * Write back each (modified) denode.
1016 	 */
1017 loop:
1018 	MNT_VNODE_FOREACH_ALL(vp, mp, nvp) {
1019 		if (vp->v_type == VNON) {
1020 			VI_UNLOCK(vp);
1021 			continue;
1022 		}
1023 		dep = VTODE(vp);
1024 		if ((dep->de_flag &
1025 		    (DE_ACCESS | DE_CREATE | DE_UPDATE | DE_MODIFIED)) == 0 &&
1026 		    (vp->v_bufobj.bo_dirty.bv_cnt == 0 ||
1027 		    waitfor == MNT_LAZY)) {
1028 			VI_UNLOCK(vp);
1029 			continue;
1030 		}
1031 		error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK);
1032 		if (error) {
1033 			if (error == ENOENT) {
1034 				MNT_VNODE_FOREACH_ALL_ABORT(mp, nvp);
1035 				goto loop;
1036 			}
1037 			continue;
1038 		}
1039 		error = VOP_FSYNC(vp, waitfor, td);
1040 		if (error)
1041 			allerror = error;
1042 		vput(vp);
1043 	}
1044 
1045 	/*
1046 	 * Flush filesystem control info.
1047 	 */
1048 	if (waitfor != MNT_LAZY) {
1049 		vn_lock(pmp->pm_devvp, LK_EXCLUSIVE | LK_RETRY);
1050 		error = VOP_FSYNC(pmp->pm_devvp, waitfor, td);
1051 		if (error)
1052 			allerror = error;
1053 		VOP_UNLOCK(pmp->pm_devvp);
1054 	}
1055 
1056 	error = msdosfs_fsiflush(pmp, waitfor);
1057 	if (error != 0)
1058 		allerror = error;
1059 
1060 	if (allerror == 0 && waitfor == MNT_SUSPEND) {
1061 		MNT_ILOCK(mp);
1062 		mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED;
1063 		MNT_IUNLOCK(mp);
1064 	}
1065 	return (allerror);
1066 }
1067 
1068 static int
1069 msdosfs_fhtovp(struct mount *mp, struct fid *fhp, int flags, struct vnode **vpp)
1070 {
1071 	struct msdosfsmount *pmp = VFSTOMSDOSFS(mp);
1072 	struct defid *defhp = (struct defid *) fhp;
1073 	struct denode *dep;
1074 	int error;
1075 
1076 	error = deget(pmp, defhp->defid_dirclust, defhp->defid_dirofs,
1077 	    LK_EXCLUSIVE, &dep);
1078 	if (error) {
1079 		*vpp = NULLVP;
1080 		return (error);
1081 	}
1082 	*vpp = DETOV(dep);
1083 	vnode_create_vobject(*vpp, dep->de_FileSize, curthread);
1084 	return (0);
1085 }
1086 
1087 static struct vfsops msdosfs_vfsops = {
1088 	.vfs_fhtovp =		msdosfs_fhtovp,
1089 	.vfs_mount =		msdosfs_mount,
1090 	.vfs_cmount =		msdosfs_cmount,
1091 	.vfs_root =		msdosfs_root,
1092 	.vfs_statfs =		msdosfs_statfs,
1093 	.vfs_sync =		msdosfs_sync,
1094 	.vfs_unmount =		msdosfs_unmount,
1095 };
1096 
1097 VFS_SET(msdosfs_vfsops, msdosfs, 0);
1098 MODULE_VERSION(msdosfs, 1);
1099