xref: /freebsd/sys/fs/msdosfs/msdosfs_vfsops.c (revision 675be9115aae86ad6b3d877155d4fd7822892105)
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 (1 == vfs_scanopt(mp->mnt_optnew, "gid", "%d", &v))
153 		pmp->pm_gid = v;
154 	if (1 == vfs_scanopt(mp->mnt_optnew, "uid", "%d", &v))
155 		pmp->pm_uid = v;
156 	if (1 == vfs_scanopt(mp->mnt_optnew, "mask", "%d", &v))
157 		pmp->pm_mask = v & ALLPERMS;
158 	if (1 == vfs_scanopt(mp->mnt_optnew, "dirmask", "%d", &v))
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, int 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 	vfs_mountedfrom(mp, from);
405 #ifdef MSDOSFS_DEBUG
406 	printf("msdosfs_mount(): mp %p, pmp %p, inusemap %p\n", mp, pmp, pmp->pm_inusemap);
407 #endif
408 	return (0);
409 }
410 
411 static int
412 mountmsdosfs(struct vnode *devvp, struct mount *mp)
413 {
414 	struct msdosfsmount *pmp;
415 	struct buf *bp;
416 	struct cdev *dev;
417 	union bootsector *bsp;
418 	struct byte_bpb33 *b33;
419 	struct byte_bpb50 *b50;
420 	struct byte_bpb710 *b710;
421 	u_int8_t SecPerClust;
422 	u_long clusters;
423 	int ronly, error;
424 	struct g_consumer *cp;
425 	struct bufobj *bo;
426 
427 	bp = NULL;		/* This and pmp both used in error_exit. */
428 	pmp = NULL;
429 	ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
430 
431 	dev = devvp->v_rdev;
432 	dev_ref(dev);
433 	DROP_GIANT();
434 	g_topology_lock();
435 	error = g_vfs_open(devvp, &cp, "msdosfs", ronly ? 0 : 1);
436 	g_topology_unlock();
437 	PICKUP_GIANT();
438 	VOP_UNLOCK(devvp, 0);
439 	if (error)
440 		goto error_exit;
441 
442 	bo = &devvp->v_bufobj;
443 
444 	/*
445 	 * Read the boot sector of the filesystem, and then check the
446 	 * boot signature.  If not a dos boot sector then error out.
447 	 *
448 	 * NOTE: 8192 is a magic size that works for ffs.
449 	 */
450 	error = bread(devvp, 0, 8192, NOCRED, &bp);
451 	if (error)
452 		goto error_exit;
453 	bp->b_flags |= B_AGE;
454 	bsp = (union bootsector *)bp->b_data;
455 	b33 = (struct byte_bpb33 *)bsp->bs33.bsBPB;
456 	b50 = (struct byte_bpb50 *)bsp->bs50.bsBPB;
457 	b710 = (struct byte_bpb710 *)bsp->bs710.bsBPB;
458 
459 #ifndef MSDOSFS_NOCHECKSIG
460 	if (bsp->bs50.bsBootSectSig0 != BOOTSIG0
461 	    || bsp->bs50.bsBootSectSig1 != BOOTSIG1) {
462 		error = EINVAL;
463 		goto error_exit;
464 	}
465 #endif
466 
467 	pmp = malloc(sizeof *pmp, M_MSDOSFSMNT, M_WAITOK | M_ZERO);
468 	pmp->pm_mountp = mp;
469 	pmp->pm_cp = cp;
470 	pmp->pm_bo = bo;
471 
472 	lockinit(&pmp->pm_fatlock, 0, msdosfs_lock_msg, 0, 0);
473 
474 	/*
475 	 * Initialize ownerships and permissions, since nothing else will
476 	 * initialize them iff we are mounting root.
477 	 */
478 	pmp->pm_uid = UID_ROOT;
479 	pmp->pm_gid = GID_WHEEL;
480 	pmp->pm_mask = pmp->pm_dirmask = S_IXUSR | S_IXGRP | S_IXOTH |
481 	    S_IRUSR | S_IRGRP | S_IROTH | S_IWUSR;
482 
483 	/*
484 	 * Experimental support for large MS-DOS filesystems.
485 	 * WARNING: This uses at least 32 bytes of kernel memory (which is not
486 	 * reclaimed until the FS is unmounted) for each file on disk to map
487 	 * between the 32-bit inode numbers used by VFS and the 64-bit
488 	 * pseudo-inode numbers used internally by msdosfs. This is only
489 	 * safe to use in certain controlled situations (e.g. read-only FS
490 	 * with less than 1 million files).
491 	 * Since the mappings do not persist across unmounts (or reboots), these
492 	 * filesystems are not suitable for exporting through NFS, or any other
493 	 * application that requires fixed inode numbers.
494 	 */
495 	vfs_flagopt(mp->mnt_optnew, "large", &pmp->pm_flags, MSDOSFS_LARGEFS);
496 
497 	/*
498 	 * Compute several useful quantities from the bpb in the
499 	 * bootsector.  Copy in the dos 5 variant of the bpb then fix up
500 	 * the fields that are different between dos 5 and dos 3.3.
501 	 */
502 	SecPerClust = b50->bpbSecPerClust;
503 	pmp->pm_BytesPerSec = getushort(b50->bpbBytesPerSec);
504 	if (pmp->pm_BytesPerSec < DEV_BSIZE) {
505 		error = EINVAL;
506 		goto error_exit;
507 	}
508 	pmp->pm_ResSectors = getushort(b50->bpbResSectors);
509 	pmp->pm_FATs = b50->bpbFATs;
510 	pmp->pm_RootDirEnts = getushort(b50->bpbRootDirEnts);
511 	pmp->pm_Sectors = getushort(b50->bpbSectors);
512 	pmp->pm_FATsecs = getushort(b50->bpbFATsecs);
513 	pmp->pm_SecPerTrack = getushort(b50->bpbSecPerTrack);
514 	pmp->pm_Heads = getushort(b50->bpbHeads);
515 	pmp->pm_Media = b50->bpbMedia;
516 
517 	/* calculate the ratio of sector size to DEV_BSIZE */
518 	pmp->pm_BlkPerSec = pmp->pm_BytesPerSec / DEV_BSIZE;
519 
520 	/*
521 	 * We don't check pm_Heads nor pm_SecPerTrack, because
522 	 * these may not be set for EFI file systems. We don't
523 	 * use these anyway, so we're unaffected if they are
524 	 * invalid.
525 	 */
526 	if (!pmp->pm_BytesPerSec || !SecPerClust) {
527 		error = EINVAL;
528 		goto error_exit;
529 	}
530 
531 	if (pmp->pm_Sectors == 0) {
532 		pmp->pm_HiddenSects = getulong(b50->bpbHiddenSecs);
533 		pmp->pm_HugeSectors = getulong(b50->bpbHugeSectors);
534 	} else {
535 		pmp->pm_HiddenSects = getushort(b33->bpbHiddenSecs);
536 		pmp->pm_HugeSectors = pmp->pm_Sectors;
537 	}
538 	if (!(pmp->pm_flags & MSDOSFS_LARGEFS)) {
539 		if (pmp->pm_HugeSectors > 0xffffffff /
540 		    (pmp->pm_BytesPerSec / sizeof(struct direntry)) + 1) {
541 			/*
542 			 * We cannot deal currently with this size of disk
543 			 * due to fileid limitations (see msdosfs_getattr and
544 			 * msdosfs_readdir)
545 			 */
546 			error = EINVAL;
547 			vfs_mount_error(mp,
548 			    "Disk too big, try '-o large' mount option");
549 			goto error_exit;
550 		}
551 	}
552 
553 	if (pmp->pm_RootDirEnts == 0) {
554 		if (pmp->pm_Sectors
555 		    || pmp->pm_FATsecs
556 		    || getushort(b710->bpbFSVers)) {
557 			error = EINVAL;
558 #ifdef MSDOSFS_DEBUG
559 			printf("mountmsdosfs(): bad FAT32 filesystem\n");
560 #endif
561 			goto error_exit;
562 		}
563 		pmp->pm_fatmask = FAT32_MASK;
564 		pmp->pm_fatmult = 4;
565 		pmp->pm_fatdiv = 1;
566 		pmp->pm_FATsecs = getulong(b710->bpbBigFATsecs);
567 		if (getushort(b710->bpbExtFlags) & FATMIRROR)
568 			pmp->pm_curfat = getushort(b710->bpbExtFlags) & FATNUM;
569 		else
570 			pmp->pm_flags |= MSDOSFS_FATMIRROR;
571 	} else
572 		pmp->pm_flags |= MSDOSFS_FATMIRROR;
573 
574 	/*
575 	 * Check a few values (could do some more):
576 	 * - logical sector size: power of 2, >= block size
577 	 * - sectors per cluster: power of 2, >= 1
578 	 * - number of sectors:   >= 1, <= size of partition
579 	 * - number of FAT sectors: >= 1
580 	 */
581 	if ( (SecPerClust == 0)
582 	  || (SecPerClust & (SecPerClust - 1))
583 	  || (pmp->pm_BytesPerSec < DEV_BSIZE)
584 	  || (pmp->pm_BytesPerSec & (pmp->pm_BytesPerSec - 1))
585 	  || (pmp->pm_HugeSectors == 0)
586 	  || (pmp->pm_FATsecs == 0)
587 	  || (SecPerClust * pmp->pm_BlkPerSec > MAXBSIZE / DEV_BSIZE)
588 	) {
589 		error = EINVAL;
590 		goto error_exit;
591 	}
592 
593 	pmp->pm_HugeSectors *= pmp->pm_BlkPerSec;
594 	pmp->pm_HiddenSects *= pmp->pm_BlkPerSec;	/* XXX not used? */
595 	pmp->pm_FATsecs     *= pmp->pm_BlkPerSec;
596 	SecPerClust         *= pmp->pm_BlkPerSec;
597 
598 	pmp->pm_fatblk = pmp->pm_ResSectors * pmp->pm_BlkPerSec;
599 
600 	if (FAT32(pmp)) {
601 		pmp->pm_rootdirblk = getulong(b710->bpbRootClust);
602 		pmp->pm_firstcluster = pmp->pm_fatblk
603 			+ (pmp->pm_FATs * pmp->pm_FATsecs);
604 		pmp->pm_fsinfo = getushort(b710->bpbFSInfo) * pmp->pm_BlkPerSec;
605 	} else {
606 		pmp->pm_rootdirblk = pmp->pm_fatblk +
607 			(pmp->pm_FATs * pmp->pm_FATsecs);
608 		pmp->pm_rootdirsize = (pmp->pm_RootDirEnts * sizeof(struct direntry)
609 				       + DEV_BSIZE - 1)
610 			/ DEV_BSIZE; /* in blocks */
611 		pmp->pm_firstcluster = pmp->pm_rootdirblk + pmp->pm_rootdirsize;
612 	}
613 
614 	pmp->pm_maxcluster = (pmp->pm_HugeSectors - pmp->pm_firstcluster) /
615 	    SecPerClust + 1;
616 	pmp->pm_fatsize = pmp->pm_FATsecs * DEV_BSIZE;	/* XXX not used? */
617 
618 	if (pmp->pm_fatmask == 0) {
619 		if (pmp->pm_maxcluster
620 		    <= ((CLUST_RSRVD - CLUST_FIRST) & FAT12_MASK)) {
621 			/*
622 			 * This will usually be a floppy disk. This size makes
623 			 * sure that one fat entry will not be split across
624 			 * multiple blocks.
625 			 */
626 			pmp->pm_fatmask = FAT12_MASK;
627 			pmp->pm_fatmult = 3;
628 			pmp->pm_fatdiv = 2;
629 		} else {
630 			pmp->pm_fatmask = FAT16_MASK;
631 			pmp->pm_fatmult = 2;
632 			pmp->pm_fatdiv = 1;
633 		}
634 	}
635 
636 	clusters = (pmp->pm_fatsize / pmp->pm_fatmult) * pmp->pm_fatdiv;
637 	if (pmp->pm_maxcluster >= clusters) {
638 #ifdef MSDOSFS_DEBUG
639 		printf("Warning: number of clusters (%ld) exceeds FAT "
640 		    "capacity (%ld)\n", pmp->pm_maxcluster + 1, clusters);
641 #endif
642 		pmp->pm_maxcluster = clusters - 1;
643 	}
644 
645 	if (FAT12(pmp))
646 		pmp->pm_fatblocksize = 3 * 512;
647 	else
648 		pmp->pm_fatblocksize = PAGE_SIZE;
649 	pmp->pm_fatblocksize = roundup(pmp->pm_fatblocksize,
650 	    pmp->pm_BytesPerSec);
651 	pmp->pm_fatblocksec = pmp->pm_fatblocksize / DEV_BSIZE;
652 	pmp->pm_bnshift = ffs(DEV_BSIZE) - 1;
653 
654 	/*
655 	 * Compute mask and shift value for isolating cluster relative byte
656 	 * offsets and cluster numbers from a file offset.
657 	 */
658 	pmp->pm_bpcluster = SecPerClust * DEV_BSIZE;
659 	pmp->pm_crbomask = pmp->pm_bpcluster - 1;
660 	pmp->pm_cnshift = ffs(pmp->pm_bpcluster) - 1;
661 
662 	/*
663 	 * Check for valid cluster size
664 	 * must be a power of 2
665 	 */
666 	if (pmp->pm_bpcluster ^ (1 << pmp->pm_cnshift)) {
667 		error = EINVAL;
668 		goto error_exit;
669 	}
670 
671 	/*
672 	 * Release the bootsector buffer.
673 	 */
674 	brelse(bp);
675 	bp = NULL;
676 
677 	/*
678 	 * Check the fsinfo sector if we have one.  Silently fix up our
679 	 * in-core copy of fp->fsinxtfree if it is unknown (0xffffffff)
680 	 * or too large.  Ignore fp->fsinfree for now, since we need to
681 	 * read the entire FAT anyway to fill the inuse map.
682 	 */
683 	if (pmp->pm_fsinfo) {
684 		struct fsinfo *fp;
685 
686 		if ((error = bread(devvp, pmp->pm_fsinfo, pmp->pm_BytesPerSec,
687 		    NOCRED, &bp)) != 0)
688 			goto error_exit;
689 		fp = (struct fsinfo *)bp->b_data;
690 		if (!bcmp(fp->fsisig1, "RRaA", 4)
691 		    && !bcmp(fp->fsisig2, "rrAa", 4)
692 		    && !bcmp(fp->fsisig3, "\0\0\125\252", 4)) {
693 			pmp->pm_nxtfree = getulong(fp->fsinxtfree);
694 			if (pmp->pm_nxtfree > pmp->pm_maxcluster)
695 				pmp->pm_nxtfree = CLUST_FIRST;
696 		} else
697 			pmp->pm_fsinfo = 0;
698 		brelse(bp);
699 		bp = NULL;
700 	}
701 
702 	/*
703 	 * Finish initializing pmp->pm_nxtfree (just in case the first few
704 	 * sectors aren't properly reserved in the FAT).  This completes
705 	 * the fixup for fp->fsinxtfree, and fixes up the zero-initialized
706 	 * value if there is no fsinfo.  We will use pmp->pm_nxtfree
707 	 * internally even if there is no fsinfo.
708 	 */
709 	if (pmp->pm_nxtfree < CLUST_FIRST)
710 		pmp->pm_nxtfree = CLUST_FIRST;
711 
712 	/*
713 	 * Allocate memory for the bitmap of allocated clusters, and then
714 	 * fill it in.
715 	 */
716 	pmp->pm_inusemap = malloc(howmany(pmp->pm_maxcluster + 1, N_INUSEBITS)
717 				  * sizeof(*pmp->pm_inusemap),
718 				  M_MSDOSFSFAT, M_WAITOK);
719 
720 	/*
721 	 * fillinusemap() needs pm_devvp.
722 	 */
723 	pmp->pm_devvp = devvp;
724 	pmp->pm_dev = dev;
725 
726 	/*
727 	 * Have the inuse map filled in.
728 	 */
729 	MSDOSFS_LOCK_MP(pmp);
730 	error = fillinusemap(pmp);
731 	MSDOSFS_UNLOCK_MP(pmp);
732 	if (error != 0)
733 		goto error_exit;
734 
735 	/*
736 	 * If they want fat updates to be synchronous then let them suffer
737 	 * the performance degradation in exchange for the on disk copy of
738 	 * the fat being correct just about all the time.  I suppose this
739 	 * would be a good thing to turn on if the kernel is still flakey.
740 	 */
741 	if (mp->mnt_flag & MNT_SYNCHRONOUS)
742 		pmp->pm_flags |= MSDOSFSMNT_WAITONFAT;
743 
744 	/*
745 	 * Finish up.
746 	 */
747 	if (ronly)
748 		pmp->pm_flags |= MSDOSFSMNT_RONLY;
749 	else {
750 		if ((error = markvoldirty(pmp, 1)) != 0) {
751 			(void)markvoldirty(pmp, 0);
752 			goto error_exit;
753 		}
754 		pmp->pm_fmod = 1;
755 	}
756 	mp->mnt_data =  pmp;
757 	mp->mnt_stat.f_fsid.val[0] = dev2udev(dev);
758 	mp->mnt_stat.f_fsid.val[1] = mp->mnt_vfc->vfc_typenum;
759 	MNT_ILOCK(mp);
760 	mp->mnt_flag |= MNT_LOCAL;
761 	mp->mnt_kern_flag |= MNTK_MPSAFE;
762 	MNT_IUNLOCK(mp);
763 
764 	if (pmp->pm_flags & MSDOSFS_LARGEFS)
765 		msdosfs_fileno_init(mp);
766 
767 	return 0;
768 
769 error_exit:
770 	if (bp)
771 		brelse(bp);
772 	if (cp != NULL) {
773 		DROP_GIANT();
774 		g_topology_lock();
775 		g_vfs_close(cp);
776 		g_topology_unlock();
777 		PICKUP_GIANT();
778 	}
779 	if (pmp) {
780 		lockdestroy(&pmp->pm_fatlock);
781 		if (pmp->pm_inusemap)
782 			free(pmp->pm_inusemap, M_MSDOSFSFAT);
783 		free(pmp, M_MSDOSFSMNT);
784 		mp->mnt_data = NULL;
785 	}
786 	dev_rel(dev);
787 	return (error);
788 }
789 
790 /*
791  * Unmount the filesystem described by mp.
792  */
793 static int
794 msdosfs_unmount(struct mount *mp, int mntflags)
795 {
796 	struct msdosfsmount *pmp;
797 	int error, flags;
798 
799 	flags = 0;
800 	if (mntflags & MNT_FORCE)
801 		flags |= FORCECLOSE;
802 	error = vflush(mp, 0, flags, curthread);
803 	if (error && error != ENXIO)
804 		return error;
805 	pmp = VFSTOMSDOSFS(mp);
806 	if ((pmp->pm_flags & MSDOSFSMNT_RONLY) == 0) {
807 		error = markvoldirty(pmp, 0);
808 		if (error && error != ENXIO) {
809 			(void)markvoldirty(pmp, 1);
810 			return (error);
811 		}
812 	}
813 	if (pmp->pm_flags & MSDOSFSMNT_KICONV && msdosfs_iconv) {
814 		if (pmp->pm_w2u)
815 			msdosfs_iconv->close(pmp->pm_w2u);
816 		if (pmp->pm_u2w)
817 			msdosfs_iconv->close(pmp->pm_u2w);
818 		if (pmp->pm_d2u)
819 			msdosfs_iconv->close(pmp->pm_d2u);
820 		if (pmp->pm_u2d)
821 			msdosfs_iconv->close(pmp->pm_u2d);
822 	}
823 
824 #ifdef MSDOSFS_DEBUG
825 	{
826 		struct vnode *vp = pmp->pm_devvp;
827 		struct bufobj *bo;
828 
829 		bo = &vp->v_bufobj;
830 		BO_LOCK(bo);
831 		VI_LOCK(vp);
832 		vn_printf(vp,
833 		    "msdosfs_umount(): just before calling VOP_CLOSE()\n");
834 		printf("freef %p, freeb %p, mount %p\n",
835 		    TAILQ_NEXT(vp, v_freelist), vp->v_freelist.tqe_prev,
836 		    vp->v_mount);
837 		printf("cleanblkhd %p, dirtyblkhd %p, numoutput %ld, type %d\n",
838 		    TAILQ_FIRST(&vp->v_bufobj.bo_clean.bv_hd),
839 		    TAILQ_FIRST(&vp->v_bufobj.bo_dirty.bv_hd),
840 		    vp->v_bufobj.bo_numoutput, vp->v_type);
841 		VI_UNLOCK(vp);
842 		BO_UNLOCK(bo);
843 	}
844 #endif
845 	DROP_GIANT();
846 	g_topology_lock();
847 	g_vfs_close(pmp->pm_cp);
848 	g_topology_unlock();
849 	PICKUP_GIANT();
850 	vrele(pmp->pm_devvp);
851 	dev_rel(pmp->pm_dev);
852 	free(pmp->pm_inusemap, M_MSDOSFSFAT);
853 	if (pmp->pm_flags & MSDOSFS_LARGEFS)
854 		msdosfs_fileno_free(mp);
855 	lockdestroy(&pmp->pm_fatlock);
856 	free(pmp, M_MSDOSFSMNT);
857 	mp->mnt_data = NULL;
858 	MNT_ILOCK(mp);
859 	mp->mnt_flag &= ~MNT_LOCAL;
860 	MNT_IUNLOCK(mp);
861 	return (error);
862 }
863 
864 static int
865 msdosfs_root(struct mount *mp, int flags, struct vnode **vpp)
866 {
867 	struct msdosfsmount *pmp = VFSTOMSDOSFS(mp);
868 	struct denode *ndep;
869 	int error;
870 
871 #ifdef MSDOSFS_DEBUG
872 	printf("msdosfs_root(); mp %p, pmp %p\n", mp, pmp);
873 #endif
874 	error = deget(pmp, MSDOSFSROOT, MSDOSFSROOT_OFS, &ndep);
875 	if (error)
876 		return (error);
877 	*vpp = DETOV(ndep);
878 	return (0);
879 }
880 
881 static int
882 msdosfs_statfs(struct mount *mp, struct statfs *sbp)
883 {
884 	struct msdosfsmount *pmp;
885 
886 	pmp = VFSTOMSDOSFS(mp);
887 	sbp->f_bsize = pmp->pm_bpcluster;
888 	sbp->f_iosize = pmp->pm_bpcluster;
889 	sbp->f_blocks = pmp->pm_maxcluster + 1;
890 	sbp->f_bfree = pmp->pm_freeclustercount;
891 	sbp->f_bavail = pmp->pm_freeclustercount;
892 	sbp->f_files = pmp->pm_RootDirEnts;	/* XXX */
893 	sbp->f_ffree = 0;	/* what to put in here? */
894 	return (0);
895 }
896 
897 static int
898 msdosfs_sync(struct mount *mp, int waitfor)
899 {
900 	struct vnode *vp, *nvp;
901 	struct thread *td;
902 	struct denode *dep;
903 	struct msdosfsmount *pmp = VFSTOMSDOSFS(mp);
904 	int error, allerror = 0;
905 
906 	td = curthread;
907 
908 	/*
909 	 * If we ever switch to not updating all of the fats all the time,
910 	 * this would be the place to update them from the first one.
911 	 */
912 	if (pmp->pm_fmod != 0) {
913 		if (pmp->pm_flags & MSDOSFSMNT_RONLY)
914 			panic("msdosfs_sync: rofs mod");
915 		else {
916 			/* update fats here */
917 		}
918 	}
919 	/*
920 	 * Write back each (modified) denode.
921 	 */
922 	MNT_ILOCK(mp);
923 loop:
924 	MNT_VNODE_FOREACH(vp, mp, nvp) {
925 		VI_LOCK(vp);
926 		if (vp->v_type == VNON || (vp->v_iflag & VI_DOOMED)) {
927 			VI_UNLOCK(vp);
928 			continue;
929 		}
930 		MNT_IUNLOCK(mp);
931 		dep = VTODE(vp);
932 		if ((dep->de_flag &
933 		    (DE_ACCESS | DE_CREATE | DE_UPDATE | DE_MODIFIED)) == 0 &&
934 		    (vp->v_bufobj.bo_dirty.bv_cnt == 0 ||
935 		    waitfor == MNT_LAZY)) {
936 			VI_UNLOCK(vp);
937 			MNT_ILOCK(mp);
938 			continue;
939 		}
940 		error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK, td);
941 		if (error) {
942 			MNT_ILOCK(mp);
943 			if (error == ENOENT)
944 				goto loop;
945 			continue;
946 		}
947 		error = VOP_FSYNC(vp, waitfor, td);
948 		if (error)
949 			allerror = error;
950 		VOP_UNLOCK(vp, 0);
951 		vrele(vp);
952 		MNT_ILOCK(mp);
953 	}
954 	MNT_IUNLOCK(mp);
955 
956 	/*
957 	 * Flush filesystem control info.
958 	 */
959 	if (waitfor != MNT_LAZY) {
960 		vn_lock(pmp->pm_devvp, LK_EXCLUSIVE | LK_RETRY);
961 		error = VOP_FSYNC(pmp->pm_devvp, waitfor, td);
962 		if (error)
963 			allerror = error;
964 		VOP_UNLOCK(pmp->pm_devvp, 0);
965 	}
966 	return (allerror);
967 }
968 
969 static int
970 msdosfs_fhtovp(struct mount *mp, struct fid *fhp, int flags, struct vnode **vpp)
971 {
972 	struct msdosfsmount *pmp = VFSTOMSDOSFS(mp);
973 	struct defid *defhp = (struct defid *) fhp;
974 	struct denode *dep;
975 	int error;
976 
977 	error = deget(pmp, defhp->defid_dirclust, defhp->defid_dirofs, &dep);
978 	if (error) {
979 		*vpp = NULLVP;
980 		return (error);
981 	}
982 	*vpp = DETOV(dep);
983 	vnode_create_vobject(*vpp, dep->de_FileSize, curthread);
984 	return (0);
985 }
986 
987 static struct vfsops msdosfs_vfsops = {
988 	.vfs_fhtovp =		msdosfs_fhtovp,
989 	.vfs_mount =		msdosfs_mount,
990 	.vfs_cmount =		msdosfs_cmount,
991 	.vfs_root =		msdosfs_root,
992 	.vfs_statfs =		msdosfs_statfs,
993 	.vfs_sync =		msdosfs_sync,
994 	.vfs_unmount =		msdosfs_unmount,
995 };
996 
997 VFS_SET(msdosfs_vfsops, msdosfs, 0);
998 MODULE_VERSION(msdosfs, 1);
999