xref: /freebsd/sys/fs/msdosfs/msdosfs_vfsops.c (revision c03c5b1c)
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(&ndp, NDF_ONLY_PNBUF);
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 	VOP_UNLOCK(odevvp);
435 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
436 	dev = devvp->v_rdev;
437 	if (atomic_cmpset_acq_ptr((uintptr_t *)&dev->si_mountpt, 0,
438 	    (uintptr_t)mp) == 0) {
439 		mntfs_freevp(devvp);
440 		return (EBUSY);
441 	}
442 	g_topology_lock();
443 	error = g_vfs_open(devvp, &cp, "msdosfs", ronly ? 0 : 1);
444 	g_topology_unlock();
445 	if (error != 0) {
446 		atomic_store_rel_ptr((uintptr_t *)&dev->si_mountpt, 0);
447 		mntfs_freevp(devvp);
448 		return (error);
449 	}
450 	dev_ref(dev);
451 	bo = &devvp->v_bufobj;
452 	BO_LOCK(&odevvp->v_bufobj);
453 	odevvp->v_bufobj.bo_flag |= BO_NOBUFS;
454 	BO_UNLOCK(&odevvp->v_bufobj);
455 	VOP_UNLOCK(devvp);
456 	if (dev->si_iosize_max != 0)
457 		mp->mnt_iosize_max = dev->si_iosize_max;
458 	if (mp->mnt_iosize_max > maxphys)
459 		mp->mnt_iosize_max = maxphys;
460 
461 	/*
462 	 * Read the boot sector of the filesystem, and then check the
463 	 * boot signature.  If not a dos boot sector then error out.
464 	 *
465 	 * NOTE: 8192 is a magic size that works for ffs.
466 	 */
467 	error = bread(devvp, 0, 8192, NOCRED, &bp);
468 	if (error)
469 		goto error_exit;
470 	bp->b_flags |= B_AGE;
471 	bsp = (union bootsector *)bp->b_data;
472 	b33 = (struct byte_bpb33 *)bsp->bs33.bsBPB;
473 	b50 = (struct byte_bpb50 *)bsp->bs50.bsBPB;
474 	b710 = (struct byte_bpb710 *)bsp->bs710.bsBPB;
475 
476 #ifndef MSDOSFS_NOCHECKSIG
477 	if (bsp->bs50.bsBootSectSig0 != BOOTSIG0 ||
478 	    bsp->bs50.bsBootSectSig1 != BOOTSIG1) {
479 		error = EINVAL;
480 		goto error_exit;
481 	}
482 #endif
483 
484 	pmp = malloc(sizeof(*pmp), M_MSDOSFSMNT, M_WAITOK | M_ZERO);
485 	pmp->pm_mountp = mp;
486 	pmp->pm_cp = cp;
487 	pmp->pm_bo = bo;
488 
489 	lockinit(&pmp->pm_fatlock, 0, msdosfs_lock_msg, 0, 0);
490 	lockinit(&pmp->pm_checkpath_lock, 0, "msdoscp", 0, 0);
491 
492 	TASK_INIT(&pmp->pm_rw2ro_task, 0, msdosfs_remount_ro, pmp);
493 
494 	/*
495 	 * Initialize ownerships and permissions, since nothing else will
496 	 * initialize them iff we are mounting root.
497 	 */
498 	pmp->pm_uid = UID_ROOT;
499 	pmp->pm_gid = GID_WHEEL;
500 	pmp->pm_mask = pmp->pm_dirmask = S_IXUSR | S_IXGRP | S_IXOTH |
501 	    S_IRUSR | S_IRGRP | S_IROTH | S_IWUSR;
502 
503 	/*
504 	 * Compute several useful quantities from the bpb in the
505 	 * bootsector.  Copy in the dos 5 variant of the bpb then fix up
506 	 * the fields that are different between dos 5 and dos 3.3.
507 	 */
508 	SecPerClust = b50->bpbSecPerClust;
509 	pmp->pm_BytesPerSec = getushort(b50->bpbBytesPerSec);
510 	if (pmp->pm_BytesPerSec < DEV_BSIZE) {
511 		error = EINVAL;
512 		goto error_exit;
513 	}
514 	pmp->pm_ResSectors = getushort(b50->bpbResSectors);
515 	pmp->pm_FATs = b50->bpbFATs;
516 	pmp->pm_RootDirEnts = getushort(b50->bpbRootDirEnts);
517 	pmp->pm_Sectors = getushort(b50->bpbSectors);
518 	pmp->pm_FATsecs = getushort(b50->bpbFATsecs);
519 	pmp->pm_SecPerTrack = getushort(b50->bpbSecPerTrack);
520 	pmp->pm_Heads = getushort(b50->bpbHeads);
521 	pmp->pm_Media = b50->bpbMedia;
522 
523 	/* calculate the ratio of sector size to DEV_BSIZE */
524 	pmp->pm_BlkPerSec = pmp->pm_BytesPerSec / DEV_BSIZE;
525 
526 	/*
527 	 * We don't check pm_Heads nor pm_SecPerTrack, because
528 	 * these may not be set for EFI file systems. We don't
529 	 * use these anyway, so we're unaffected if they are
530 	 * invalid.
531 	 */
532 	if (pmp->pm_BytesPerSec == 0 || SecPerClust == 0) {
533 		error = EINVAL;
534 		goto error_exit;
535 	}
536 
537 	if (pmp->pm_Sectors == 0) {
538 		pmp->pm_HiddenSects = getulong(b50->bpbHiddenSecs);
539 		pmp->pm_HugeSectors = getulong(b50->bpbHugeSectors);
540 	} else {
541 		pmp->pm_HiddenSects = getushort(b33->bpbHiddenSecs);
542 		pmp->pm_HugeSectors = pmp->pm_Sectors;
543 	}
544 
545 	if (pmp->pm_RootDirEnts == 0) {
546 		if (pmp->pm_FATsecs != 0 || getushort(b710->bpbFSVers) != 0) {
547 			error = EINVAL;
548 #ifdef MSDOSFS_DEBUG
549 			printf("mountmsdosfs(): bad FAT32 filesystem\n");
550 #endif
551 			goto error_exit;
552 		}
553 		pmp->pm_fatmask = FAT32_MASK;
554 		pmp->pm_fatmult = 4;
555 		pmp->pm_fatdiv = 1;
556 		pmp->pm_FATsecs = getulong(b710->bpbBigFATsecs);
557 		if ((getushort(b710->bpbExtFlags) & FATMIRROR) != 0)
558 			pmp->pm_curfat = getushort(b710->bpbExtFlags) & FATNUM;
559 		else
560 			pmp->pm_flags |= MSDOSFS_FATMIRROR;
561 	} else
562 		pmp->pm_flags |= MSDOSFS_FATMIRROR;
563 
564 	/*
565 	 * Check a few values (could do some more):
566 	 * - logical sector size: power of 2, >= block size
567 	 * - sectors per cluster: power of 2, >= 1
568 	 * - number of sectors:   >= 1, <= size of partition
569 	 * - number of FAT sectors: >= 1
570 	 */
571 	if (SecPerClust == 0 || (SecPerClust & (SecPerClust - 1)) != 0 ||
572 	    pmp->pm_BytesPerSec < DEV_BSIZE ||
573 	    (pmp->pm_BytesPerSec & (pmp->pm_BytesPerSec - 1)) != 0 ||
574 	    pmp->pm_HugeSectors == 0 || pmp->pm_FATsecs == 0 ||
575 	    SecPerClust * pmp->pm_BlkPerSec > MAXBSIZE / DEV_BSIZE) {
576 		error = EINVAL;
577 		goto error_exit;
578 	}
579 
580 	if ((off_t)pmp->pm_HugeSectors * pmp->pm_BytesPerSec <
581 	    pmp->pm_HugeSectors /* overflow */ ||
582 	    (off_t)pmp->pm_HugeSectors * pmp->pm_BytesPerSec >
583 	    cp->provider->mediasize /* past end of vol */) {
584 		error = EINVAL;
585 		goto error_exit;
586 	}
587 
588 	pmp->pm_HugeSectors *= pmp->pm_BlkPerSec;
589 	pmp->pm_HiddenSects *= pmp->pm_BlkPerSec;	/* XXX not used? */
590 	pmp->pm_FATsecs     *= pmp->pm_BlkPerSec;
591 	SecPerClust         *= pmp->pm_BlkPerSec;
592 
593 	pmp->pm_fatblk = pmp->pm_ResSectors * pmp->pm_BlkPerSec;
594 
595 	if (FAT32(pmp)) {
596 		pmp->pm_rootdirblk = getulong(b710->bpbRootClust);
597 		pmp->pm_firstcluster = pmp->pm_fatblk +
598 		    pmp->pm_FATs * pmp->pm_FATsecs;
599 		pmp->pm_fsinfo = getushort(b710->bpbFSInfo) * pmp->pm_BlkPerSec;
600 	} else {
601 		pmp->pm_rootdirblk = pmp->pm_fatblk +
602 		    pmp->pm_FATs * pmp->pm_FATsecs;
603 		pmp->pm_rootdirsize = howmany(pmp->pm_RootDirEnts *
604 		    sizeof(struct direntry), DEV_BSIZE); /* in blocks */
605 		pmp->pm_firstcluster = pmp->pm_rootdirblk + pmp->pm_rootdirsize;
606 	}
607 
608 	if (pmp->pm_HugeSectors <= pmp->pm_firstcluster) {
609 		error = EINVAL;
610 		goto error_exit;
611 	}
612 	pmp->pm_maxcluster = (pmp->pm_HugeSectors - pmp->pm_firstcluster) /
613 	    SecPerClust + 1;
614 	pmp->pm_fatsize = pmp->pm_FATsecs * DEV_BSIZE;	/* XXX not used? */
615 
616 	if (pmp->pm_fatmask == 0) {
617 		if (pmp->pm_maxcluster <= ((CLUST_RSRVD - CLUST_FIRST) &
618 		    FAT12_MASK)) {
619 			/*
620 			 * This will usually be a floppy disk. This size makes
621 			 * sure that one FAT entry will not be split across
622 			 * multiple blocks.
623 			 */
624 			pmp->pm_fatmask = FAT12_MASK;
625 			pmp->pm_fatmult = 3;
626 			pmp->pm_fatdiv = 2;
627 		} else {
628 			pmp->pm_fatmask = FAT16_MASK;
629 			pmp->pm_fatmult = 2;
630 			pmp->pm_fatdiv = 1;
631 		}
632 	}
633 
634 	clusters = (pmp->pm_fatsize / pmp->pm_fatmult) * pmp->pm_fatdiv;
635 	if (pmp->pm_maxcluster >= clusters) {
636 #ifdef MSDOSFS_DEBUG
637 		printf("Warning: number of clusters (%ld) exceeds FAT "
638 		    "capacity (%ld)\n", pmp->pm_maxcluster + 1, clusters);
639 #endif
640 		pmp->pm_maxcluster = clusters - 1;
641 	}
642 
643 	if (FAT12(pmp))
644 		pmp->pm_fatblocksize = 3 * 512;
645 	else
646 		pmp->pm_fatblocksize = PAGE_SIZE;
647 	pmp->pm_fatblocksize = roundup(pmp->pm_fatblocksize,
648 	    pmp->pm_BytesPerSec);
649 	pmp->pm_fatblocksec = pmp->pm_fatblocksize / DEV_BSIZE;
650 	pmp->pm_bnshift = ffs(DEV_BSIZE) - 1;
651 
652 	/*
653 	 * Compute mask and shift value for isolating cluster relative byte
654 	 * offsets and cluster numbers from a file offset.
655 	 */
656 	pmp->pm_bpcluster = SecPerClust * DEV_BSIZE;
657 	pmp->pm_crbomask = pmp->pm_bpcluster - 1;
658 	pmp->pm_cnshift = ffs(pmp->pm_bpcluster) - 1;
659 
660 	/*
661 	 * Check for valid cluster size
662 	 * must be a power of 2
663 	 */
664 	if ((pmp->pm_bpcluster ^ (1 << pmp->pm_cnshift)) != 0) {
665 		error = EINVAL;
666 		goto error_exit;
667 	}
668 
669 	/*
670 	 * Release the bootsector buffer.
671 	 */
672 	brelse(bp);
673 	bp = NULL;
674 
675 	/*
676 	 * Check the fsinfo sector if we have one.  Silently fix up our
677 	 * in-core copy of fp->fsinxtfree if it is unknown (0xffffffff)
678 	 * or too large.  Ignore fp->fsinfree for now, since we need to
679 	 * read the entire FAT anyway to fill the inuse map.
680 	 */
681 	if (pmp->pm_fsinfo) {
682 		struct fsinfo *fp;
683 
684 		if ((error = bread(devvp, pmp->pm_fsinfo, pmp->pm_BytesPerSec,
685 		    NOCRED, &bp)) != 0)
686 			goto error_exit;
687 		fp = (struct fsinfo *)bp->b_data;
688 		if (!bcmp(fp->fsisig1, "RRaA", 4) &&
689 		    !bcmp(fp->fsisig2, "rrAa", 4) &&
690 		    !bcmp(fp->fsisig3, "\0\0\125\252", 4)) {
691 			pmp->pm_nxtfree = getulong(fp->fsinxtfree);
692 			if (pmp->pm_nxtfree > pmp->pm_maxcluster)
693 				pmp->pm_nxtfree = CLUST_FIRST;
694 		} else
695 			pmp->pm_fsinfo = 0;
696 		brelse(bp);
697 		bp = NULL;
698 	}
699 
700 	/*
701 	 * Finish initializing pmp->pm_nxtfree (just in case the first few
702 	 * sectors aren't properly reserved in the FAT).  This completes
703 	 * the fixup for fp->fsinxtfree, and fixes up the zero-initialized
704 	 * value if there is no fsinfo.  We will use pmp->pm_nxtfree
705 	 * internally even if there is no fsinfo.
706 	 */
707 	if (pmp->pm_nxtfree < CLUST_FIRST)
708 		pmp->pm_nxtfree = CLUST_FIRST;
709 
710 	/*
711 	 * Allocate memory for the bitmap of allocated clusters, and then
712 	 * fill it in.
713 	 */
714 	pmp->pm_inusemap = malloc(howmany(pmp->pm_maxcluster + 1,
715 	    N_INUSEBITS) * sizeof(*pmp->pm_inusemap), M_MSDOSFSFAT, M_WAITOK);
716 
717 	/*
718 	 * fillinusemap() needs pm_devvp.
719 	 */
720 	pmp->pm_devvp = devvp;
721 	pmp->pm_odevvp = odevvp;
722 	pmp->pm_dev = dev;
723 
724 	/*
725 	 * Have the inuse map filled in.
726 	 */
727 	MSDOSFS_LOCK_MP(pmp);
728 	error = fillinusemap(pmp);
729 	MSDOSFS_UNLOCK_MP(pmp);
730 	if (error != 0)
731 		goto error_exit;
732 
733 	/*
734 	 * If they want FAT updates to be synchronous then let them suffer
735 	 * the performance degradation in exchange for the on disk copy of
736 	 * the FAT being correct just about all the time.  I suppose this
737 	 * would be a good thing to turn on if the kernel is still flakey.
738 	 */
739 	if (mp->mnt_flag & MNT_SYNCHRONOUS)
740 		pmp->pm_flags |= MSDOSFSMNT_WAITONFAT;
741 
742 	/*
743 	 * Finish up.
744 	 */
745 	if (ronly)
746 		pmp->pm_flags |= MSDOSFSMNT_RONLY;
747 	else {
748 		if ((error = markvoldirty(pmp, 1)) != 0)
749 			goto error_exit;
750 		pmp->pm_fmod = 1;
751 	}
752 	mp->mnt_data =  pmp;
753 	mp->mnt_stat.f_fsid.val[0] = dev2udev(dev);
754 	mp->mnt_stat.f_fsid.val[1] = mp->mnt_vfc->vfc_typenum;
755 	MNT_ILOCK(mp);
756 	mp->mnt_flag |= MNT_LOCAL;
757 	mp->mnt_kern_flag |= MNTK_USES_BCACHE | MNTK_NO_IOPF;
758 	MNT_IUNLOCK(mp);
759 
760 	return (0);
761 
762 error_exit:
763 	if (bp != NULL)
764 		brelse(bp);
765 	if (cp != NULL) {
766 		g_topology_lock();
767 		g_vfs_close(cp);
768 		g_topology_unlock();
769 	}
770 	if (pmp != NULL) {
771 		lockdestroy(&pmp->pm_fatlock);
772 		lockdestroy(&pmp->pm_checkpath_lock);
773 		free(pmp->pm_inusemap, M_MSDOSFSFAT);
774 		free(pmp, M_MSDOSFSMNT);
775 		mp->mnt_data = NULL;
776 	}
777 	BO_LOCK(&odevvp->v_bufobj);
778 	odevvp->v_bufobj.bo_flag &= ~BO_NOBUFS;
779 	BO_UNLOCK(&odevvp->v_bufobj);
780 	atomic_store_rel_ptr((uintptr_t *)&dev->si_mountpt, 0);
781 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
782 	mntfs_freevp(devvp);
783 	dev_rel(dev);
784 	return (error);
785 }
786 
787 /*
788  * Unmount the filesystem described by mp.
789  */
790 static int
791 msdosfs_unmount(struct mount *mp, int mntflags)
792 {
793 	struct msdosfsmount *pmp;
794 	int error, flags;
795 	bool susp;
796 
797 	error = flags = 0;
798 	pmp = VFSTOMSDOSFS(mp);
799 	susp = (pmp->pm_flags & MSDOSFSMNT_RONLY) == 0;
800 
801 	if (susp) {
802 		error = vfs_write_suspend_umnt(mp);
803 		if (error != 0)
804 			return (error);
805 	}
806 
807 	if ((mntflags & MNT_FORCE) != 0)
808 		flags |= FORCECLOSE;
809 	error = vflush(mp, 0, flags, curthread);
810 	if (error != 0 && error != ENXIO) {
811 		if (susp)
812 			vfs_write_resume(mp, VR_START_WRITE);
813 		return (error);
814 	}
815 	if (susp) {
816 		error = markvoldirty(pmp, 0);
817 		if (error != 0 && error != ENXIO) {
818 			if (susp)
819 				vfs_write_resume(mp, VR_START_WRITE);
820 			(void)markvoldirty(pmp, 1);
821 			return (error);
822 		}
823 	}
824 	if (pmp->pm_flags & MSDOSFSMNT_KICONV && msdosfs_iconv) {
825 		if (pmp->pm_w2u)
826 			msdosfs_iconv->close(pmp->pm_w2u);
827 		if (pmp->pm_u2w)
828 			msdosfs_iconv->close(pmp->pm_u2w);
829 		if (pmp->pm_d2u)
830 			msdosfs_iconv->close(pmp->pm_d2u);
831 		if (pmp->pm_u2d)
832 			msdosfs_iconv->close(pmp->pm_u2d);
833 	}
834 
835 #ifdef MSDOSFS_DEBUG
836 	{
837 		struct vnode *vp = pmp->pm_devvp;
838 		struct bufobj *bo;
839 
840 		bo = &vp->v_bufobj;
841 		BO_LOCK(bo);
842 		VI_LOCK(vp);
843 		vn_printf(vp,
844 		    "msdosfs_umount(): just before calling VOP_CLOSE()\n");
845 		printf("freef %p, freeb %p, mount %p\n",
846 		    TAILQ_NEXT(vp, v_vnodelist), vp->v_vnodelist.tqe_prev,
847 		    vp->v_mount);
848 		printf("cleanblkhd %p, dirtyblkhd %p, numoutput %d, type %d\n",
849 		    TAILQ_FIRST(&vp->v_bufobj.bo_clean.bv_hd),
850 		    TAILQ_FIRST(&vp->v_bufobj.bo_dirty.bv_hd),
851 		    vp->v_bufobj.bo_numoutput, vp->v_type);
852 		VI_UNLOCK(vp);
853 		BO_UNLOCK(bo);
854 	}
855 #endif
856 	if (susp)
857 		vfs_write_resume(mp, VR_START_WRITE);
858 
859 	vn_lock(pmp->pm_devvp, LK_EXCLUSIVE | LK_RETRY);
860 	g_topology_lock();
861 	g_vfs_close(pmp->pm_cp);
862 	g_topology_unlock();
863 	BO_LOCK(&pmp->pm_odevvp->v_bufobj);
864 	pmp->pm_odevvp->v_bufobj.bo_flag &= ~BO_NOBUFS;
865 	BO_UNLOCK(&pmp->pm_odevvp->v_bufobj);
866 	atomic_store_rel_ptr((uintptr_t *)&pmp->pm_dev->si_mountpt, 0);
867 	mntfs_freevp(pmp->pm_devvp);
868 	vrele(pmp->pm_odevvp);
869 	dev_rel(pmp->pm_dev);
870 	free(pmp->pm_inusemap, M_MSDOSFSFAT);
871 	lockdestroy(&pmp->pm_fatlock);
872 	lockdestroy(&pmp->pm_checkpath_lock);
873 	free(pmp, M_MSDOSFSMNT);
874 	mp->mnt_data = NULL;
875 	MNT_ILOCK(mp);
876 	mp->mnt_flag &= ~MNT_LOCAL;
877 	MNT_IUNLOCK(mp);
878 	return (error);
879 }
880 
881 static void
882 msdosfs_remount_ro(void *arg, int pending)
883 {
884 	struct msdosfsmount *pmp;
885 	int error;
886 
887 	pmp = arg;
888 
889 	MSDOSFS_LOCK_MP(pmp);
890 	if ((pmp->pm_flags & MSDOSFS_ERR_RO) != 0) {
891 		while ((pmp->pm_flags & MSDOSFS_ERR_RO) != 0)
892 			msleep(&pmp->pm_flags, &pmp->pm_fatlock, PVFS,
893 			    "msdoserrro", hz);
894 	} else if ((pmp->pm_mountp->mnt_flag & MNT_RDONLY) == 0) {
895 		pmp->pm_flags |= MSDOSFS_ERR_RO;
896 		MSDOSFS_UNLOCK_MP(pmp);
897 		printf("%s: remounting read-only due to corruption\n",
898 		    pmp->pm_mountp->mnt_stat.f_mntfromname);
899 		error = vfs_remount_ro(pmp->pm_mountp);
900 		if (error != 0)
901 			printf("%s: remounting read-only failed: error %d\n",
902 			    pmp->pm_mountp->mnt_stat.f_mntfromname, error);
903 		else
904 			printf("remounted %s read-only\n",
905 			    pmp->pm_mountp->mnt_stat.f_mntfromname);
906 		MSDOSFS_LOCK_MP(pmp);
907 		pmp->pm_flags &= ~MSDOSFS_ERR_RO;
908 		wakeup(&pmp->pm_flags);
909 	}
910 	MSDOSFS_UNLOCK_MP(pmp);
911 
912 	vfs_unbusy(pmp->pm_mountp);
913 }
914 
915 void
916 msdosfs_integrity_error(struct msdosfsmount *pmp)
917 {
918 	int error;
919 
920 	error = vfs_busy(pmp->pm_mountp, MBF_NOWAIT);
921 	if (error == 0)
922 		taskqueue_enqueue(taskqueue_thread, &pmp->pm_rw2ro_task);
923 	else
924 		printf("%s: integrity error busying failed, error %d\n",
925 		    pmp->pm_mountp->mnt_stat.f_mntfromname, error);
926 }
927 
928 static int
929 msdosfs_root(struct mount *mp, int flags, struct vnode **vpp)
930 {
931 	struct msdosfsmount *pmp = VFSTOMSDOSFS(mp);
932 	struct denode *ndep;
933 	int error;
934 
935 #ifdef MSDOSFS_DEBUG
936 	printf("msdosfs_root(); mp %p, pmp %p\n", mp, pmp);
937 #endif
938 	error = deget(pmp, MSDOSFSROOT, MSDOSFSROOT_OFS, LK_EXCLUSIVE, &ndep);
939 	if (error)
940 		return (error);
941 	*vpp = DETOV(ndep);
942 	return (0);
943 }
944 
945 static int
946 msdosfs_statfs(struct mount *mp, struct statfs *sbp)
947 {
948 	struct msdosfsmount *pmp;
949 
950 	pmp = VFSTOMSDOSFS(mp);
951 	sbp->f_bsize = pmp->pm_bpcluster;
952 	sbp->f_iosize = pmp->pm_bpcluster;
953 	sbp->f_blocks = pmp->pm_maxcluster + 1;
954 	sbp->f_bfree = pmp->pm_freeclustercount;
955 	sbp->f_bavail = pmp->pm_freeclustercount;
956 	sbp->f_files = pmp->pm_RootDirEnts;	/* XXX */
957 	sbp->f_ffree = 0;	/* what to put in here? */
958 	return (0);
959 }
960 
961 /*
962  * If we have an FSInfo block, update it.
963  */
964 static int
965 msdosfs_fsiflush(struct msdosfsmount *pmp, int waitfor)
966 {
967 	struct fsinfo *fp;
968 	struct buf *bp;
969 	int error;
970 
971 	MSDOSFS_LOCK_MP(pmp);
972 	if (pmp->pm_fsinfo == 0 || (pmp->pm_flags & MSDOSFS_FSIMOD) == 0) {
973 		error = 0;
974 		goto unlock;
975 	}
976 	error = bread(pmp->pm_devvp, pmp->pm_fsinfo, pmp->pm_BytesPerSec,
977 	    NOCRED, &bp);
978 	if (error != 0) {
979 		goto unlock;
980 	}
981 	fp = (struct fsinfo *)bp->b_data;
982 	putulong(fp->fsinfree, pmp->pm_freeclustercount);
983 	putulong(fp->fsinxtfree, pmp->pm_nxtfree);
984 	pmp->pm_flags &= ~MSDOSFS_FSIMOD;
985 	if (waitfor == MNT_WAIT)
986 		error = bwrite(bp);
987 	else
988 		bawrite(bp);
989 unlock:
990 	MSDOSFS_UNLOCK_MP(pmp);
991 	return (error);
992 }
993 
994 static int
995 msdosfs_sync(struct mount *mp, int waitfor)
996 {
997 	struct vnode *vp, *nvp;
998 	struct thread *td;
999 	struct denode *dep;
1000 	struct msdosfsmount *pmp = VFSTOMSDOSFS(mp);
1001 	int error, allerror = 0;
1002 
1003 	td = curthread;
1004 
1005 	/*
1006 	 * If we ever switch to not updating all of the FATs all the time,
1007 	 * this would be the place to update them from the first one.
1008 	 */
1009 	if (pmp->pm_fmod != 0) {
1010 		if (pmp->pm_flags & MSDOSFSMNT_RONLY)
1011 			panic("msdosfs_sync: rofs mod");
1012 		else {
1013 			/* update FATs here */
1014 		}
1015 	}
1016 	/*
1017 	 * Write back each (modified) denode.
1018 	 */
1019 loop:
1020 	MNT_VNODE_FOREACH_ALL(vp, mp, nvp) {
1021 		if (vp->v_type == VNON) {
1022 			VI_UNLOCK(vp);
1023 			continue;
1024 		}
1025 		dep = VTODE(vp);
1026 		if ((dep->de_flag &
1027 		    (DE_ACCESS | DE_CREATE | DE_UPDATE | DE_MODIFIED)) == 0 &&
1028 		    (vp->v_bufobj.bo_dirty.bv_cnt == 0 ||
1029 		    waitfor == MNT_LAZY)) {
1030 			VI_UNLOCK(vp);
1031 			continue;
1032 		}
1033 		error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK);
1034 		if (error) {
1035 			if (error == ENOENT) {
1036 				MNT_VNODE_FOREACH_ALL_ABORT(mp, nvp);
1037 				goto loop;
1038 			}
1039 			continue;
1040 		}
1041 		error = VOP_FSYNC(vp, waitfor, td);
1042 		if (error)
1043 			allerror = error;
1044 		VOP_UNLOCK(vp);
1045 		vrele(vp);
1046 	}
1047 
1048 	/*
1049 	 * Flush filesystem control info.
1050 	 */
1051 	if (waitfor != MNT_LAZY) {
1052 		vn_lock(pmp->pm_devvp, LK_EXCLUSIVE | LK_RETRY);
1053 		error = VOP_FSYNC(pmp->pm_devvp, waitfor, td);
1054 		if (error)
1055 			allerror = error;
1056 		VOP_UNLOCK(pmp->pm_devvp);
1057 	}
1058 
1059 	error = msdosfs_fsiflush(pmp, waitfor);
1060 	if (error != 0)
1061 		allerror = error;
1062 
1063 	if (allerror == 0 && waitfor == MNT_SUSPEND) {
1064 		MNT_ILOCK(mp);
1065 		mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED;
1066 		MNT_IUNLOCK(mp);
1067 	}
1068 	return (allerror);
1069 }
1070 
1071 static int
1072 msdosfs_fhtovp(struct mount *mp, struct fid *fhp, int flags, struct vnode **vpp)
1073 {
1074 	struct msdosfsmount *pmp = VFSTOMSDOSFS(mp);
1075 	struct defid *defhp = (struct defid *) fhp;
1076 	struct denode *dep;
1077 	int error;
1078 
1079 	error = deget(pmp, defhp->defid_dirclust, defhp->defid_dirofs,
1080 	    LK_EXCLUSIVE, &dep);
1081 	if (error) {
1082 		*vpp = NULLVP;
1083 		return (error);
1084 	}
1085 	*vpp = DETOV(dep);
1086 	vnode_create_vobject(*vpp, dep->de_FileSize, curthread);
1087 	return (0);
1088 }
1089 
1090 static struct vfsops msdosfs_vfsops = {
1091 	.vfs_fhtovp =		msdosfs_fhtovp,
1092 	.vfs_mount =		msdosfs_mount,
1093 	.vfs_cmount =		msdosfs_cmount,
1094 	.vfs_root =		msdosfs_root,
1095 	.vfs_statfs =		msdosfs_statfs,
1096 	.vfs_sync =		msdosfs_sync,
1097 	.vfs_unmount =		msdosfs_unmount,
1098 };
1099 
1100 VFS_SET(msdosfs_vfsops, msdosfs, 0);
1101 MODULE_VERSION(msdosfs, 1);
1102