xref: /freebsd/sys/fs/msdosfs/msdosfs_vfsops.c (revision a0ee8cc636cd5c2374ec44ca71226564ea0bca95)
1 /* $FreeBSD$ */
2 /*	$NetBSD: msdosfs_vfsops.c,v 1.51 1997/11/17 15:36:58 ws Exp $	*/
3 
4 /*-
5  * Copyright (C) 1994, 1995, 1997 Wolfgang Solfrank.
6  * Copyright (C) 1994, 1995, 1997 TooLs GmbH.
7  * All rights reserved.
8  * Original code by Paul Popelka (paulp@uts.amdahl.com) (see below).
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by TooLs GmbH.
21  * 4. The name of TooLs GmbH may not be used to endorse or promote products
22  *    derived from this software without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
25  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
26  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
27  * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
28  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
29  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
30  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
31  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
32  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
33  * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34  */
35 /*-
36  * Written by Paul Popelka (paulp@uts.amdahl.com)
37  *
38  * You can do anything you want with this software, just don't say you wrote
39  * it, and don't remove this notice.
40  *
41  * This software is provided "as is".
42  *
43  * The author supplies this software to be publicly redistributed on the
44  * understanding that the author is not responsible for the correct
45  * functioning of this software in any circumstances and is not liable for
46  * any damages caused by this software.
47  *
48  * October 1992
49  */
50 
51 #include <sys/param.h>
52 #include <sys/systm.h>
53 #include <sys/buf.h>
54 #include <sys/conf.h>
55 #include <sys/fcntl.h>
56 #include <sys/iconv.h>
57 #include <sys/kernel.h>
58 #include <sys/lock.h>
59 #include <sys/malloc.h>
60 #include <sys/mount.h>
61 #include <sys/mutex.h>
62 #include <sys/namei.h>
63 #include <sys/priv.h>
64 #include <sys/proc.h>
65 #include <sys/stat.h>
66 #include <sys/vnode.h>
67 
68 #include <geom/geom.h>
69 #include <geom/geom_vfs.h>
70 
71 #include <fs/msdosfs/bootsect.h>
72 #include <fs/msdosfs/bpb.h>
73 #include <fs/msdosfs/direntry.h>
74 #include <fs/msdosfs/denode.h>
75 #include <fs/msdosfs/fat.h>
76 #include <fs/msdosfs/msdosfsmount.h>
77 
78 static const char msdosfs_lock_msg[] = "fatlk";
79 
80 /* Mount options that we support. */
81 static const char *msdosfs_opts[] = {
82 	"async", "noatime", "noclusterr", "noclusterw",
83 	"export", "force", "from", "sync",
84 	"cs_dos", "cs_local", "cs_win", "dirmask",
85 	"gid", "kiconv", "large", "longname",
86 	"longnames", "mask", "shortname", "shortnames",
87 	"uid", "win95", "nowin95",
88 	NULL
89 };
90 
91 #if 1 /*def PC98*/
92 /*
93  * XXX - The boot signature formatted by NEC PC-98 DOS looks like a
94  *       garbage or a random value :-{
95  *       If you want to use that broken-signatured media, define the
96  *       following symbol even though PC/AT.
97  *       (ex. mount PC-98 DOS formatted FD on PC/AT)
98  */
99 #define	MSDOSFS_NOCHECKSIG
100 #endif
101 
102 MALLOC_DEFINE(M_MSDOSFSMNT, "msdosfs_mount", "MSDOSFS mount structure");
103 static MALLOC_DEFINE(M_MSDOSFSFAT, "msdosfs_fat", "MSDOSFS file allocation table");
104 
105 struct iconv_functions *msdosfs_iconv;
106 
107 static int	update_mp(struct mount *mp, struct thread *td);
108 static int	mountmsdosfs(struct vnode *devvp, struct mount *mp);
109 static vfs_fhtovp_t	msdosfs_fhtovp;
110 static vfs_mount_t	msdosfs_mount;
111 static vfs_root_t	msdosfs_root;
112 static vfs_statfs_t	msdosfs_statfs;
113 static vfs_sync_t	msdosfs_sync;
114 static vfs_unmount_t	msdosfs_unmount;
115 
116 /* Maximum length of a character set name (arbitrary). */
117 #define	MAXCSLEN	64
118 
119 static int
120 update_mp(struct mount *mp, struct thread *td)
121 {
122 	struct msdosfsmount *pmp = VFSTOMSDOSFS(mp);
123 	void *dos, *win, *local;
124 	int error, v;
125 
126 	if (!vfs_getopt(mp->mnt_optnew, "kiconv", NULL, NULL)) {
127 		if (msdosfs_iconv != NULL) {
128 			error = vfs_getopt(mp->mnt_optnew,
129 			    "cs_win", &win, NULL);
130 			if (!error)
131 				error = vfs_getopt(mp->mnt_optnew,
132 				    "cs_local", &local, NULL);
133 			if (!error)
134 				error = vfs_getopt(mp->mnt_optnew,
135 				    "cs_dos", &dos, NULL);
136 			if (!error) {
137 				msdosfs_iconv->open(win, local, &pmp->pm_u2w);
138 				msdosfs_iconv->open(local, win, &pmp->pm_w2u);
139 				msdosfs_iconv->open(dos, local, &pmp->pm_u2d);
140 				msdosfs_iconv->open(local, dos, &pmp->pm_d2u);
141 			}
142 			if (error != 0)
143 				return (error);
144 		} else {
145 			pmp->pm_w2u = NULL;
146 			pmp->pm_u2w = NULL;
147 			pmp->pm_d2u = NULL;
148 			pmp->pm_u2d = NULL;
149 		}
150 	}
151 
152 	if (vfs_scanopt(mp->mnt_optnew, "gid", "%d", &v) == 1)
153 		pmp->pm_gid = v;
154 	if (vfs_scanopt(mp->mnt_optnew, "uid", "%d", &v) == 1)
155 		pmp->pm_uid = v;
156 	if (vfs_scanopt(mp->mnt_optnew, "mask", "%d", &v) == 1)
157 		pmp->pm_mask = v & ALLPERMS;
158 	if (vfs_scanopt(mp->mnt_optnew, "dirmask", "%d", &v) == 1)
159 		pmp->pm_dirmask = v & ALLPERMS;
160 	vfs_flagopt(mp->mnt_optnew, "shortname",
161 	    &pmp->pm_flags, MSDOSFSMNT_SHORTNAME);
162 	vfs_flagopt(mp->mnt_optnew, "shortnames",
163 	    &pmp->pm_flags, MSDOSFSMNT_SHORTNAME);
164 	vfs_flagopt(mp->mnt_optnew, "longname",
165 	    &pmp->pm_flags, MSDOSFSMNT_LONGNAME);
166 	vfs_flagopt(mp->mnt_optnew, "longnames",
167 	    &pmp->pm_flags, MSDOSFSMNT_LONGNAME);
168 	vfs_flagopt(mp->mnt_optnew, "kiconv",
169 	    &pmp->pm_flags, MSDOSFSMNT_KICONV);
170 
171 	if (vfs_getopt(mp->mnt_optnew, "nowin95", NULL, NULL) == 0)
172 		pmp->pm_flags |= MSDOSFSMNT_NOWIN95;
173 	else
174 		pmp->pm_flags &= ~MSDOSFSMNT_NOWIN95;
175 
176 	if (pmp->pm_flags & MSDOSFSMNT_NOWIN95)
177 		pmp->pm_flags |= MSDOSFSMNT_SHORTNAME;
178 	else if (!(pmp->pm_flags &
179 	    (MSDOSFSMNT_SHORTNAME | MSDOSFSMNT_LONGNAME))) {
180 		struct vnode *rootvp;
181 
182 		/*
183 		 * Try to divine whether to support Win'95 long filenames
184 		 */
185 		if (FAT32(pmp))
186 			pmp->pm_flags |= MSDOSFSMNT_LONGNAME;
187 		else {
188 			if ((error =
189 			    msdosfs_root(mp, LK_EXCLUSIVE, &rootvp)) != 0)
190 				return error;
191 			pmp->pm_flags |= findwin95(VTODE(rootvp)) ?
192 			    MSDOSFSMNT_LONGNAME : MSDOSFSMNT_SHORTNAME;
193 			vput(rootvp);
194 		}
195 	}
196 	return 0;
197 }
198 
199 static int
200 msdosfs_cmount(struct mntarg *ma, void *data, uint64_t flags)
201 {
202 	struct msdosfs_args args;
203 	struct export_args exp;
204 	int error;
205 
206 	if (data == NULL)
207 		return (EINVAL);
208 	error = copyin(data, &args, sizeof args);
209 	if (error)
210 		return (error);
211 	vfs_oexport_conv(&args.export, &exp);
212 
213 	ma = mount_argsu(ma, "from", args.fspec, MAXPATHLEN);
214 	ma = mount_arg(ma, "export", &exp, sizeof(exp));
215 	ma = mount_argf(ma, "uid", "%d", args.uid);
216 	ma = mount_argf(ma, "gid", "%d", args.gid);
217 	ma = mount_argf(ma, "mask", "%d", args.mask);
218 	ma = mount_argf(ma, "dirmask", "%d", args.dirmask);
219 
220 	ma = mount_argb(ma, args.flags & MSDOSFSMNT_SHORTNAME, "noshortname");
221 	ma = mount_argb(ma, args.flags & MSDOSFSMNT_LONGNAME, "nolongname");
222 	ma = mount_argb(ma, !(args.flags & MSDOSFSMNT_NOWIN95), "nowin95");
223 	ma = mount_argb(ma, args.flags & MSDOSFSMNT_KICONV, "nokiconv");
224 
225 	ma = mount_argsu(ma, "cs_win", args.cs_win, MAXCSLEN);
226 	ma = mount_argsu(ma, "cs_dos", args.cs_dos, MAXCSLEN);
227 	ma = mount_argsu(ma, "cs_local", args.cs_local, MAXCSLEN);
228 
229 	error = kernel_mount(ma, flags);
230 
231 	return (error);
232 }
233 
234 /*
235  * mp - path - addr in user space of mount point (ie /usr or whatever)
236  * data - addr in user space of mount params including the name of the block
237  * special file to treat as a filesystem.
238  */
239 static int
240 msdosfs_mount(struct mount *mp)
241 {
242 	struct vnode *devvp;	  /* vnode for blk device to mount */
243 	struct thread *td;
244 	/* msdosfs specific mount control block */
245 	struct msdosfsmount *pmp = NULL;
246 	struct nameidata ndp;
247 	int error, flags;
248 	accmode_t accmode;
249 	char *from;
250 
251 	td = curthread;
252 	if (vfs_filteropt(mp->mnt_optnew, msdosfs_opts))
253 		return (EINVAL);
254 
255 	/*
256 	 * If updating, check whether changing from read-only to
257 	 * read/write; if there is no device name, that's all we do.
258 	 */
259 	if (mp->mnt_flag & MNT_UPDATE) {
260 		pmp = VFSTOMSDOSFS(mp);
261 		if (vfs_flagopt(mp->mnt_optnew, "export", NULL, 0)) {
262 			/*
263 			 * Forbid export requests if filesystem has
264 			 * MSDOSFS_LARGEFS flag set.
265 			 */
266 			if ((pmp->pm_flags & MSDOSFS_LARGEFS) != 0) {
267 				vfs_mount_error(mp,
268 				    "MSDOSFS_LARGEFS flag set, cannot export");
269 				return (EOPNOTSUPP);
270 			}
271 		}
272 		if (!(pmp->pm_flags & MSDOSFSMNT_RONLY) &&
273 		    vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) {
274 			error = VFS_SYNC(mp, MNT_WAIT);
275 			if (error)
276 				return (error);
277 			flags = WRITECLOSE;
278 			if (mp->mnt_flag & MNT_FORCE)
279 				flags |= FORCECLOSE;
280 			error = vflush(mp, 0, flags, td);
281 			if (error)
282 				return (error);
283 
284 			/*
285 			 * Now the volume is clean.  Mark it so while the
286 			 * device is still rw.
287 			 */
288 			error = markvoldirty(pmp, 0);
289 			if (error) {
290 				(void)markvoldirty(pmp, 1);
291 				return (error);
292 			}
293 
294 			/* Downgrade the device from rw to ro. */
295 			DROP_GIANT();
296 			g_topology_lock();
297 			error = g_access(pmp->pm_cp, 0, -1, 0);
298 			g_topology_unlock();
299 			PICKUP_GIANT();
300 			if (error) {
301 				(void)markvoldirty(pmp, 1);
302 				return (error);
303 			}
304 
305 			/*
306 			 * Backing out after an error was painful in the
307 			 * above.  Now we are committed to succeeding.
308 			 */
309 			pmp->pm_fmod = 0;
310 			pmp->pm_flags |= MSDOSFSMNT_RONLY;
311 			MNT_ILOCK(mp);
312 			mp->mnt_flag |= MNT_RDONLY;
313 			MNT_IUNLOCK(mp);
314 		} else if ((pmp->pm_flags & MSDOSFSMNT_RONLY) &&
315 		    !vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) {
316 			/*
317 			 * If upgrade to read-write by non-root, then verify
318 			 * that user has necessary permissions on the device.
319 			 */
320 			devvp = pmp->pm_devvp;
321 			vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
322 			error = VOP_ACCESS(devvp, VREAD | VWRITE,
323 			    td->td_ucred, td);
324 			if (error)
325 				error = priv_check(td, PRIV_VFS_MOUNT_PERM);
326 			if (error) {
327 				VOP_UNLOCK(devvp, 0);
328 				return (error);
329 			}
330 			VOP_UNLOCK(devvp, 0);
331 			DROP_GIANT();
332 			g_topology_lock();
333 			error = g_access(pmp->pm_cp, 0, 1, 0);
334 			g_topology_unlock();
335 			PICKUP_GIANT();
336 			if (error)
337 				return (error);
338 
339 			pmp->pm_fmod = 1;
340 			pmp->pm_flags &= ~MSDOSFSMNT_RONLY;
341 			MNT_ILOCK(mp);
342 			mp->mnt_flag &= ~MNT_RDONLY;
343 			MNT_IUNLOCK(mp);
344 
345 			/* Now that the volume is modifiable, mark it dirty. */
346 			error = markvoldirty(pmp, 1);
347 			if (error)
348 				return (error);
349 		}
350 	}
351 	/*
352 	 * Not an update, or updating the name: look up the name
353 	 * and verify that it refers to a sensible disk device.
354 	 */
355 	if (vfs_getopt(mp->mnt_optnew, "from", (void **)&from, NULL))
356 		return (EINVAL);
357 	NDINIT(&ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, from, td);
358 	error = namei(&ndp);
359 	if (error)
360 		return (error);
361 	devvp = ndp.ni_vp;
362 	NDFREE(&ndp, NDF_ONLY_PNBUF);
363 
364 	if (!vn_isdisk(devvp, &error)) {
365 		vput(devvp);
366 		return (error);
367 	}
368 	/*
369 	 * If mount by non-root, then verify that user has necessary
370 	 * permissions on the device.
371 	 */
372 	accmode = VREAD;
373 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
374 		accmode |= VWRITE;
375 	error = VOP_ACCESS(devvp, accmode, td->td_ucred, td);
376 	if (error)
377 		error = priv_check(td, PRIV_VFS_MOUNT_PERM);
378 	if (error) {
379 		vput(devvp);
380 		return (error);
381 	}
382 	if ((mp->mnt_flag & MNT_UPDATE) == 0) {
383 		error = mountmsdosfs(devvp, mp);
384 #ifdef MSDOSFS_DEBUG		/* only needed for the printf below */
385 		pmp = VFSTOMSDOSFS(mp);
386 #endif
387 	} else {
388 		vput(devvp);
389 		if (devvp != pmp->pm_devvp)
390 			return (EINVAL);	/* XXX needs translation */
391 	}
392 	if (error) {
393 		vrele(devvp);
394 		return (error);
395 	}
396 
397 	error = update_mp(mp, td);
398 	if (error) {
399 		if ((mp->mnt_flag & MNT_UPDATE) == 0)
400 			msdosfs_unmount(mp, MNT_FORCE);
401 		return error;
402 	}
403 
404 	if (devvp->v_type == VCHR && devvp->v_rdev != NULL)
405 		devvp->v_rdev->si_mountpt = mp;
406 	vfs_mountedfrom(mp, from);
407 #ifdef MSDOSFS_DEBUG
408 	printf("msdosfs_mount(): mp %p, pmp %p, inusemap %p\n", mp, pmp, pmp->pm_inusemap);
409 #endif
410 	return (0);
411 }
412 
413 static int
414 mountmsdosfs(struct vnode *devvp, struct mount *mp)
415 {
416 	struct msdosfsmount *pmp;
417 	struct buf *bp;
418 	struct cdev *dev;
419 	union bootsector *bsp;
420 	struct byte_bpb33 *b33;
421 	struct byte_bpb50 *b50;
422 	struct byte_bpb710 *b710;
423 	u_int8_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 	dev = devvp->v_rdev;
434 	dev_ref(dev);
435 	DROP_GIANT();
436 	g_topology_lock();
437 	error = g_vfs_open(devvp, &cp, "msdosfs", ronly ? 0 : 1);
438 	g_topology_unlock();
439 	PICKUP_GIANT();
440 	VOP_UNLOCK(devvp, 0);
441 	if (error)
442 		goto error_exit;
443 
444 	bo = &devvp->v_bufobj;
445 
446 	/*
447 	 * Read the boot sector of the filesystem, and then check the
448 	 * boot signature.  If not a dos boot sector then error out.
449 	 *
450 	 * NOTE: 8192 is a magic size that works for ffs.
451 	 */
452 	error = bread(devvp, 0, 8192, NOCRED, &bp);
453 	if (error)
454 		goto error_exit;
455 	bp->b_flags |= B_AGE;
456 	bsp = (union bootsector *)bp->b_data;
457 	b33 = (struct byte_bpb33 *)bsp->bs33.bsBPB;
458 	b50 = (struct byte_bpb50 *)bsp->bs50.bsBPB;
459 	b710 = (struct byte_bpb710 *)bsp->bs710.bsBPB;
460 
461 #ifndef MSDOSFS_NOCHECKSIG
462 	if (bsp->bs50.bsBootSectSig0 != BOOTSIG0
463 	    || bsp->bs50.bsBootSectSig1 != BOOTSIG1) {
464 		error = EINVAL;
465 		goto error_exit;
466 	}
467 #endif
468 
469 	pmp = malloc(sizeof *pmp, M_MSDOSFSMNT, M_WAITOK | M_ZERO);
470 	pmp->pm_mountp = mp;
471 	pmp->pm_cp = cp;
472 	pmp->pm_bo = bo;
473 
474 	lockinit(&pmp->pm_fatlock, 0, msdosfs_lock_msg, 0, 0);
475 
476 	/*
477 	 * Initialize ownerships and permissions, since nothing else will
478 	 * initialize them iff we are mounting root.
479 	 */
480 	pmp->pm_uid = UID_ROOT;
481 	pmp->pm_gid = GID_WHEEL;
482 	pmp->pm_mask = pmp->pm_dirmask = S_IXUSR | S_IXGRP | S_IXOTH |
483 	    S_IRUSR | S_IRGRP | S_IROTH | S_IWUSR;
484 
485 	/*
486 	 * Experimental support for large MS-DOS filesystems.
487 	 * WARNING: This uses at least 32 bytes of kernel memory (which is not
488 	 * reclaimed until the FS is unmounted) for each file on disk to map
489 	 * between the 32-bit inode numbers used by VFS and the 64-bit
490 	 * pseudo-inode numbers used internally by msdosfs. This is only
491 	 * safe to use in certain controlled situations (e.g. read-only FS
492 	 * with less than 1 million files).
493 	 * Since the mappings do not persist across unmounts (or reboots), these
494 	 * filesystems are not suitable for exporting through NFS, or any other
495 	 * application that requires fixed inode numbers.
496 	 */
497 	vfs_flagopt(mp->mnt_optnew, "large", &pmp->pm_flags, MSDOSFS_LARGEFS);
498 
499 	/*
500 	 * Compute several useful quantities from the bpb in the
501 	 * bootsector.  Copy in the dos 5 variant of the bpb then fix up
502 	 * the fields that are different between dos 5 and dos 3.3.
503 	 */
504 	SecPerClust = b50->bpbSecPerClust;
505 	pmp->pm_BytesPerSec = getushort(b50->bpbBytesPerSec);
506 	if (pmp->pm_BytesPerSec < DEV_BSIZE) {
507 		error = EINVAL;
508 		goto error_exit;
509 	}
510 	pmp->pm_ResSectors = getushort(b50->bpbResSectors);
511 	pmp->pm_FATs = b50->bpbFATs;
512 	pmp->pm_RootDirEnts = getushort(b50->bpbRootDirEnts);
513 	pmp->pm_Sectors = getushort(b50->bpbSectors);
514 	pmp->pm_FATsecs = getushort(b50->bpbFATsecs);
515 	pmp->pm_SecPerTrack = getushort(b50->bpbSecPerTrack);
516 	pmp->pm_Heads = getushort(b50->bpbHeads);
517 	pmp->pm_Media = b50->bpbMedia;
518 
519 	/* calculate the ratio of sector size to DEV_BSIZE */
520 	pmp->pm_BlkPerSec = pmp->pm_BytesPerSec / DEV_BSIZE;
521 
522 	/*
523 	 * We don't check pm_Heads nor pm_SecPerTrack, because
524 	 * these may not be set for EFI file systems. We don't
525 	 * use these anyway, so we're unaffected if they are
526 	 * invalid.
527 	 */
528 	if (!pmp->pm_BytesPerSec || !SecPerClust) {
529 		error = EINVAL;
530 		goto error_exit;
531 	}
532 
533 	if (pmp->pm_Sectors == 0) {
534 		pmp->pm_HiddenSects = getulong(b50->bpbHiddenSecs);
535 		pmp->pm_HugeSectors = getulong(b50->bpbHugeSectors);
536 	} else {
537 		pmp->pm_HiddenSects = getushort(b33->bpbHiddenSecs);
538 		pmp->pm_HugeSectors = pmp->pm_Sectors;
539 	}
540 	if (!(pmp->pm_flags & MSDOSFS_LARGEFS)) {
541 		if (pmp->pm_HugeSectors > 0xffffffff /
542 		    (pmp->pm_BytesPerSec / sizeof(struct direntry)) + 1) {
543 			/*
544 			 * We cannot deal currently with this size of disk
545 			 * due to fileid limitations (see msdosfs_getattr and
546 			 * msdosfs_readdir)
547 			 */
548 			error = EINVAL;
549 			vfs_mount_error(mp,
550 			    "Disk too big, try '-o large' mount option");
551 			goto error_exit;
552 		}
553 	}
554 
555 	if (pmp->pm_RootDirEnts == 0) {
556 		if (pmp->pm_FATsecs
557 		    || getushort(b710->bpbFSVers)) {
558 			error = EINVAL;
559 #ifdef MSDOSFS_DEBUG
560 			printf("mountmsdosfs(): bad FAT32 filesystem\n");
561 #endif
562 			goto error_exit;
563 		}
564 		pmp->pm_fatmask = FAT32_MASK;
565 		pmp->pm_fatmult = 4;
566 		pmp->pm_fatdiv = 1;
567 		pmp->pm_FATsecs = getulong(b710->bpbBigFATsecs);
568 		if (getushort(b710->bpbExtFlags) & FATMIRROR)
569 			pmp->pm_curfat = getushort(b710->bpbExtFlags) & FATNUM;
570 		else
571 			pmp->pm_flags |= MSDOSFS_FATMIRROR;
572 	} else
573 		pmp->pm_flags |= MSDOSFS_FATMIRROR;
574 
575 	/*
576 	 * Check a few values (could do some more):
577 	 * - logical sector size: power of 2, >= block size
578 	 * - sectors per cluster: power of 2, >= 1
579 	 * - number of sectors:   >= 1, <= size of partition
580 	 * - number of FAT sectors: >= 1
581 	 */
582 	if ( (SecPerClust == 0)
583 	  || (SecPerClust & (SecPerClust - 1))
584 	  || (pmp->pm_BytesPerSec < DEV_BSIZE)
585 	  || (pmp->pm_BytesPerSec & (pmp->pm_BytesPerSec - 1))
586 	  || (pmp->pm_HugeSectors == 0)
587 	  || (pmp->pm_FATsecs == 0)
588 	  || (SecPerClust * pmp->pm_BlkPerSec > MAXBSIZE / DEV_BSIZE)
589 	) {
590 		error = EINVAL;
591 		goto error_exit;
592 	}
593 
594 	pmp->pm_HugeSectors *= pmp->pm_BlkPerSec;
595 	pmp->pm_HiddenSects *= pmp->pm_BlkPerSec;	/* XXX not used? */
596 	pmp->pm_FATsecs     *= pmp->pm_BlkPerSec;
597 	SecPerClust         *= pmp->pm_BlkPerSec;
598 
599 	pmp->pm_fatblk = pmp->pm_ResSectors * pmp->pm_BlkPerSec;
600 
601 	if (FAT32(pmp)) {
602 		pmp->pm_rootdirblk = getulong(b710->bpbRootClust);
603 		pmp->pm_firstcluster = pmp->pm_fatblk
604 			+ (pmp->pm_FATs * pmp->pm_FATsecs);
605 		pmp->pm_fsinfo = getushort(b710->bpbFSInfo) * pmp->pm_BlkPerSec;
606 	} else {
607 		pmp->pm_rootdirblk = pmp->pm_fatblk +
608 			(pmp->pm_FATs * pmp->pm_FATsecs);
609 		pmp->pm_rootdirsize = (pmp->pm_RootDirEnts * sizeof(struct direntry)
610 				       + DEV_BSIZE - 1)
611 			/ DEV_BSIZE; /* in blocks */
612 		pmp->pm_firstcluster = pmp->pm_rootdirblk + pmp->pm_rootdirsize;
613 	}
614 
615 	pmp->pm_maxcluster = (pmp->pm_HugeSectors - pmp->pm_firstcluster) /
616 	    SecPerClust + 1;
617 	pmp->pm_fatsize = pmp->pm_FATsecs * DEV_BSIZE;	/* XXX not used? */
618 
619 	if (pmp->pm_fatmask == 0) {
620 		if (pmp->pm_maxcluster
621 		    <= ((CLUST_RSRVD - CLUST_FIRST) & FAT12_MASK)) {
622 			/*
623 			 * This will usually be a floppy disk. This size makes
624 			 * sure that one fat entry will not be split across
625 			 * multiple blocks.
626 			 */
627 			pmp->pm_fatmask = FAT12_MASK;
628 			pmp->pm_fatmult = 3;
629 			pmp->pm_fatdiv = 2;
630 		} else {
631 			pmp->pm_fatmask = FAT16_MASK;
632 			pmp->pm_fatmult = 2;
633 			pmp->pm_fatdiv = 1;
634 		}
635 	}
636 
637 	clusters = (pmp->pm_fatsize / pmp->pm_fatmult) * pmp->pm_fatdiv;
638 	if (pmp->pm_maxcluster >= clusters) {
639 #ifdef MSDOSFS_DEBUG
640 		printf("Warning: number of clusters (%ld) exceeds FAT "
641 		    "capacity (%ld)\n", pmp->pm_maxcluster + 1, clusters);
642 #endif
643 		pmp->pm_maxcluster = clusters - 1;
644 	}
645 
646 	if (FAT12(pmp))
647 		pmp->pm_fatblocksize = 3 * 512;
648 	else
649 		pmp->pm_fatblocksize = PAGE_SIZE;
650 	pmp->pm_fatblocksize = roundup(pmp->pm_fatblocksize,
651 	    pmp->pm_BytesPerSec);
652 	pmp->pm_fatblocksec = pmp->pm_fatblocksize / DEV_BSIZE;
653 	pmp->pm_bnshift = ffs(DEV_BSIZE) - 1;
654 
655 	/*
656 	 * Compute mask and shift value for isolating cluster relative byte
657 	 * offsets and cluster numbers from a file offset.
658 	 */
659 	pmp->pm_bpcluster = SecPerClust * DEV_BSIZE;
660 	pmp->pm_crbomask = pmp->pm_bpcluster - 1;
661 	pmp->pm_cnshift = ffs(pmp->pm_bpcluster) - 1;
662 
663 	/*
664 	 * Check for valid cluster size
665 	 * must be a power of 2
666 	 */
667 	if (pmp->pm_bpcluster ^ (1 << pmp->pm_cnshift)) {
668 		error = EINVAL;
669 		goto error_exit;
670 	}
671 
672 	/*
673 	 * Release the bootsector buffer.
674 	 */
675 	brelse(bp);
676 	bp = NULL;
677 
678 	/*
679 	 * Check the fsinfo sector if we have one.  Silently fix up our
680 	 * in-core copy of fp->fsinxtfree if it is unknown (0xffffffff)
681 	 * or too large.  Ignore fp->fsinfree for now, since we need to
682 	 * read the entire FAT anyway to fill the inuse map.
683 	 */
684 	if (pmp->pm_fsinfo) {
685 		struct fsinfo *fp;
686 
687 		if ((error = bread(devvp, pmp->pm_fsinfo, pmp->pm_BytesPerSec,
688 		    NOCRED, &bp)) != 0)
689 			goto error_exit;
690 		fp = (struct fsinfo *)bp->b_data;
691 		if (!bcmp(fp->fsisig1, "RRaA", 4)
692 		    && !bcmp(fp->fsisig2, "rrAa", 4)
693 		    && !bcmp(fp->fsisig3, "\0\0\125\252", 4)) {
694 			pmp->pm_nxtfree = getulong(fp->fsinxtfree);
695 			if (pmp->pm_nxtfree > pmp->pm_maxcluster)
696 				pmp->pm_nxtfree = CLUST_FIRST;
697 		} else
698 			pmp->pm_fsinfo = 0;
699 		brelse(bp);
700 		bp = NULL;
701 	}
702 
703 	/*
704 	 * Finish initializing pmp->pm_nxtfree (just in case the first few
705 	 * sectors aren't properly reserved in the FAT).  This completes
706 	 * the fixup for fp->fsinxtfree, and fixes up the zero-initialized
707 	 * value if there is no fsinfo.  We will use pmp->pm_nxtfree
708 	 * internally even if there is no fsinfo.
709 	 */
710 	if (pmp->pm_nxtfree < CLUST_FIRST)
711 		pmp->pm_nxtfree = CLUST_FIRST;
712 
713 	/*
714 	 * Allocate memory for the bitmap of allocated clusters, and then
715 	 * fill it in.
716 	 */
717 	pmp->pm_inusemap = malloc(howmany(pmp->pm_maxcluster + 1, N_INUSEBITS)
718 				  * sizeof(*pmp->pm_inusemap),
719 				  M_MSDOSFSFAT, M_WAITOK);
720 
721 	/*
722 	 * fillinusemap() needs pm_devvp.
723 	 */
724 	pmp->pm_devvp = devvp;
725 	pmp->pm_dev = dev;
726 
727 	/*
728 	 * Have the inuse map filled in.
729 	 */
730 	MSDOSFS_LOCK_MP(pmp);
731 	error = fillinusemap(pmp);
732 	MSDOSFS_UNLOCK_MP(pmp);
733 	if (error != 0)
734 		goto error_exit;
735 
736 	/*
737 	 * If they want fat updates to be synchronous then let them suffer
738 	 * the performance degradation in exchange for the on disk copy of
739 	 * the fat being correct just about all the time.  I suppose this
740 	 * would be a good thing to turn on if the kernel is still flakey.
741 	 */
742 	if (mp->mnt_flag & MNT_SYNCHRONOUS)
743 		pmp->pm_flags |= MSDOSFSMNT_WAITONFAT;
744 
745 	/*
746 	 * Finish up.
747 	 */
748 	if (ronly)
749 		pmp->pm_flags |= MSDOSFSMNT_RONLY;
750 	else {
751 		if ((error = markvoldirty(pmp, 1)) != 0) {
752 			(void)markvoldirty(pmp, 0);
753 			goto error_exit;
754 		}
755 		pmp->pm_fmod = 1;
756 	}
757 	mp->mnt_data =  pmp;
758 	mp->mnt_stat.f_fsid.val[0] = dev2udev(dev);
759 	mp->mnt_stat.f_fsid.val[1] = mp->mnt_vfc->vfc_typenum;
760 	MNT_ILOCK(mp);
761 	mp->mnt_flag |= MNT_LOCAL;
762 	mp->mnt_kern_flag |= MNTK_USES_BCACHE;
763 	MNT_IUNLOCK(mp);
764 
765 	if (pmp->pm_flags & MSDOSFS_LARGEFS)
766 		msdosfs_fileno_init(mp);
767 
768 	return 0;
769 
770 error_exit:
771 	if (bp)
772 		brelse(bp);
773 	if (cp != NULL) {
774 		DROP_GIANT();
775 		g_topology_lock();
776 		g_vfs_close(cp);
777 		g_topology_unlock();
778 		PICKUP_GIANT();
779 	}
780 	if (pmp) {
781 		lockdestroy(&pmp->pm_fatlock);
782 		if (pmp->pm_inusemap)
783 			free(pmp->pm_inusemap, M_MSDOSFSFAT);
784 		free(pmp, M_MSDOSFSMNT);
785 		mp->mnt_data = NULL;
786 	}
787 	dev_rel(dev);
788 	return (error);
789 }
790 
791 /*
792  * Unmount the filesystem described by mp.
793  */
794 static int
795 msdosfs_unmount(struct mount *mp, int mntflags)
796 {
797 	struct msdosfsmount *pmp;
798 	int error, flags;
799 
800 	error = flags = 0;
801 	pmp = VFSTOMSDOSFS(mp);
802 	if ((pmp->pm_flags & MSDOSFSMNT_RONLY) == 0)
803 		error = msdosfs_sync(mp, MNT_WAIT);
804 	if ((mntflags & MNT_FORCE) != 0)
805 		flags |= FORCECLOSE;
806 	else if (error != 0)
807 		return (error);
808 	error = vflush(mp, 0, flags, curthread);
809 	if (error != 0 && error != ENXIO)
810 		return (error);
811 	if ((pmp->pm_flags & MSDOSFSMNT_RONLY) == 0) {
812 		error = markvoldirty(pmp, 0);
813 		if (error && error != ENXIO) {
814 			(void)markvoldirty(pmp, 1);
815 			return (error);
816 		}
817 	}
818 	if (pmp->pm_flags & MSDOSFSMNT_KICONV && msdosfs_iconv) {
819 		if (pmp->pm_w2u)
820 			msdosfs_iconv->close(pmp->pm_w2u);
821 		if (pmp->pm_u2w)
822 			msdosfs_iconv->close(pmp->pm_u2w);
823 		if (pmp->pm_d2u)
824 			msdosfs_iconv->close(pmp->pm_d2u);
825 		if (pmp->pm_u2d)
826 			msdosfs_iconv->close(pmp->pm_u2d);
827 	}
828 
829 #ifdef MSDOSFS_DEBUG
830 	{
831 		struct vnode *vp = pmp->pm_devvp;
832 		struct bufobj *bo;
833 
834 		bo = &vp->v_bufobj;
835 		BO_LOCK(bo);
836 		VI_LOCK(vp);
837 		vn_printf(vp,
838 		    "msdosfs_umount(): just before calling VOP_CLOSE()\n");
839 		printf("freef %p, freeb %p, mount %p\n",
840 		    TAILQ_NEXT(vp, v_actfreelist), vp->v_actfreelist.tqe_prev,
841 		    vp->v_mount);
842 		printf("cleanblkhd %p, dirtyblkhd %p, numoutput %ld, type %d\n",
843 		    TAILQ_FIRST(&vp->v_bufobj.bo_clean.bv_hd),
844 		    TAILQ_FIRST(&vp->v_bufobj.bo_dirty.bv_hd),
845 		    vp->v_bufobj.bo_numoutput, vp->v_type);
846 		VI_UNLOCK(vp);
847 		BO_UNLOCK(bo);
848 	}
849 #endif
850 	DROP_GIANT();
851 	if (pmp->pm_devvp->v_type == VCHR && pmp->pm_devvp->v_rdev != NULL)
852 		pmp->pm_devvp->v_rdev->si_mountpt = NULL;
853 	g_topology_lock();
854 	g_vfs_close(pmp->pm_cp);
855 	g_topology_unlock();
856 	PICKUP_GIANT();
857 	vrele(pmp->pm_devvp);
858 	dev_rel(pmp->pm_dev);
859 	free(pmp->pm_inusemap, M_MSDOSFSFAT);
860 	if (pmp->pm_flags & MSDOSFS_LARGEFS)
861 		msdosfs_fileno_free(mp);
862 	lockdestroy(&pmp->pm_fatlock);
863 	free(pmp, M_MSDOSFSMNT);
864 	mp->mnt_data = NULL;
865 	MNT_ILOCK(mp);
866 	mp->mnt_flag &= ~MNT_LOCAL;
867 	MNT_IUNLOCK(mp);
868 	return (error);
869 }
870 
871 static int
872 msdosfs_root(struct mount *mp, int flags, struct vnode **vpp)
873 {
874 	struct msdosfsmount *pmp = VFSTOMSDOSFS(mp);
875 	struct denode *ndep;
876 	int error;
877 
878 #ifdef MSDOSFS_DEBUG
879 	printf("msdosfs_root(); mp %p, pmp %p\n", mp, pmp);
880 #endif
881 	error = deget(pmp, MSDOSFSROOT, MSDOSFSROOT_OFS, &ndep);
882 	if (error)
883 		return (error);
884 	*vpp = DETOV(ndep);
885 	return (0);
886 }
887 
888 static int
889 msdosfs_statfs(struct mount *mp, struct statfs *sbp)
890 {
891 	struct msdosfsmount *pmp;
892 
893 	pmp = VFSTOMSDOSFS(mp);
894 	sbp->f_bsize = pmp->pm_bpcluster;
895 	sbp->f_iosize = pmp->pm_bpcluster;
896 	sbp->f_blocks = pmp->pm_maxcluster + 1;
897 	sbp->f_bfree = pmp->pm_freeclustercount;
898 	sbp->f_bavail = pmp->pm_freeclustercount;
899 	sbp->f_files = pmp->pm_RootDirEnts;	/* XXX */
900 	sbp->f_ffree = 0;	/* what to put in here? */
901 	return (0);
902 }
903 
904 /*
905  * If we have an FSInfo block, update it.
906  */
907 static int
908 msdosfs_fsiflush(struct msdosfsmount *pmp, int waitfor)
909 {
910 	struct fsinfo *fp;
911 	struct buf *bp;
912 	int error;
913 
914 	MSDOSFS_LOCK_MP(pmp);
915 	if (pmp->pm_fsinfo == 0 || (pmp->pm_flags & MSDOSFS_FSIMOD) == 0) {
916 		error = 0;
917 		goto unlock;
918 	}
919 	error = bread(pmp->pm_devvp, pmp->pm_fsinfo, pmp->pm_BytesPerSec,
920 	    NOCRED, &bp);
921 	if (error != 0) {
922 		brelse(bp);
923 		goto unlock;
924 	}
925 	fp = (struct fsinfo *)bp->b_data;
926 	putulong(fp->fsinfree, pmp->pm_freeclustercount);
927 	putulong(fp->fsinxtfree, pmp->pm_nxtfree);
928 	pmp->pm_flags &= ~MSDOSFS_FSIMOD;
929 	if (waitfor == MNT_WAIT)
930 		error = bwrite(bp);
931 	else
932 		bawrite(bp);
933 unlock:
934 	MSDOSFS_UNLOCK_MP(pmp);
935 	return (error);
936 }
937 
938 static int
939 msdosfs_sync(struct mount *mp, int waitfor)
940 {
941 	struct vnode *vp, *nvp;
942 	struct thread *td;
943 	struct denode *dep;
944 	struct msdosfsmount *pmp = VFSTOMSDOSFS(mp);
945 	int error, allerror = 0;
946 
947 	td = curthread;
948 
949 	/*
950 	 * If we ever switch to not updating all of the fats all the time,
951 	 * this would be the place to update them from the first one.
952 	 */
953 	if (pmp->pm_fmod != 0) {
954 		if (pmp->pm_flags & MSDOSFSMNT_RONLY)
955 			panic("msdosfs_sync: rofs mod");
956 		else {
957 			/* update fats here */
958 		}
959 	}
960 	/*
961 	 * Write back each (modified) denode.
962 	 */
963 loop:
964 	MNT_VNODE_FOREACH_ALL(vp, mp, nvp) {
965 		if (vp->v_type == VNON) {
966 			VI_UNLOCK(vp);
967 			continue;
968 		}
969 		dep = VTODE(vp);
970 		if ((dep->de_flag &
971 		    (DE_ACCESS | DE_CREATE | DE_UPDATE | DE_MODIFIED)) == 0 &&
972 		    (vp->v_bufobj.bo_dirty.bv_cnt == 0 ||
973 		    waitfor == MNT_LAZY)) {
974 			VI_UNLOCK(vp);
975 			continue;
976 		}
977 		error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK, td);
978 		if (error) {
979 			if (error == ENOENT)
980 				goto loop;
981 			continue;
982 		}
983 		error = VOP_FSYNC(vp, waitfor, td);
984 		if (error)
985 			allerror = error;
986 		VOP_UNLOCK(vp, 0);
987 		vrele(vp);
988 	}
989 
990 	/*
991 	 * Flush filesystem control info.
992 	 */
993 	if (waitfor != MNT_LAZY) {
994 		vn_lock(pmp->pm_devvp, LK_EXCLUSIVE | LK_RETRY);
995 		error = VOP_FSYNC(pmp->pm_devvp, waitfor, td);
996 		if (error)
997 			allerror = error;
998 		VOP_UNLOCK(pmp->pm_devvp, 0);
999 	}
1000 
1001 	error = msdosfs_fsiflush(pmp, waitfor);
1002 	if (error != 0)
1003 		allerror = error;
1004 	return (allerror);
1005 }
1006 
1007 static int
1008 msdosfs_fhtovp(struct mount *mp, struct fid *fhp, int flags, struct vnode **vpp)
1009 {
1010 	struct msdosfsmount *pmp = VFSTOMSDOSFS(mp);
1011 	struct defid *defhp = (struct defid *) fhp;
1012 	struct denode *dep;
1013 	int error;
1014 
1015 	error = deget(pmp, defhp->defid_dirclust, defhp->defid_dirofs, &dep);
1016 	if (error) {
1017 		*vpp = NULLVP;
1018 		return (error);
1019 	}
1020 	*vpp = DETOV(dep);
1021 	vnode_create_vobject(*vpp, dep->de_FileSize, curthread);
1022 	return (0);
1023 }
1024 
1025 static struct vfsops msdosfs_vfsops = {
1026 	.vfs_fhtovp =		msdosfs_fhtovp,
1027 	.vfs_mount =		msdosfs_mount,
1028 	.vfs_cmount =		msdosfs_cmount,
1029 	.vfs_root =		msdosfs_root,
1030 	.vfs_statfs =		msdosfs_statfs,
1031 	.vfs_sync =		msdosfs_sync,
1032 	.vfs_unmount =		msdosfs_unmount,
1033 };
1034 
1035 VFS_SET(msdosfs_vfsops, msdosfs, 0);
1036 MODULE_VERSION(msdosfs, 1);
1037