xref: /netbsd-src/sys/fs/msdosfs/msdosfs_vfsops.c (revision e77448e07be3174235c13f58032a0d6d0ab7638d)
1 /*	$NetBSD: msdosfs_vfsops.c,v 1.67 2008/05/16 09:21:59 hannken Exp $	*/
2 
3 /*-
4  * Copyright (C) 1994, 1995, 1997 Wolfgang Solfrank.
5  * Copyright (C) 1994, 1995, 1997 TooLs GmbH.
6  * All rights reserved.
7  * Original code by Paul Popelka (paulp@uts.amdahl.com) (see below).
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 3. All advertising materials mentioning features or use of this software
18  *    must display the following acknowledgement:
19  *	This product includes software developed by TooLs GmbH.
20  * 4. The name of TooLs GmbH may not be used to endorse or promote products
21  *    derived from this software without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
24  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
25  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
26  * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
27  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
28  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
29  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
30  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
31  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
32  * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33  */
34 /*
35  * Written by Paul Popelka (paulp@uts.amdahl.com)
36  *
37  * You can do anything you want with this software, just don't say you wrote
38  * it, and don't remove this notice.
39  *
40  * This software is provided "as is".
41  *
42  * The author supplies this software to be publicly redistributed on the
43  * understanding that the author is not responsible for the correct
44  * functioning of this software in any circumstances and is not liable for
45  * any damages caused by this software.
46  *
47  * October 1992
48  */
49 
50 #include <sys/cdefs.h>
51 __KERNEL_RCSID(0, "$NetBSD: msdosfs_vfsops.c,v 1.67 2008/05/16 09:21:59 hannken Exp $");
52 
53 #if defined(_KERNEL_OPT)
54 #include "opt_quota.h"
55 #include "opt_compat_netbsd.h"
56 #endif
57 
58 #include <sys/param.h>
59 #include <sys/systm.h>
60 #include <sys/sysctl.h>
61 #include <sys/namei.h>
62 #include <sys/proc.h>
63 #include <sys/kernel.h>
64 #include <sys/vnode.h>
65 #include <miscfs/genfs/genfs.h>
66 #include <miscfs/specfs/specdev.h> /* XXX */	/* defines v_rdev */
67 #include <sys/mount.h>
68 #include <sys/buf.h>
69 #include <sys/file.h>
70 #include <sys/device.h>
71 #include <sys/disklabel.h>
72 #include <sys/disk.h>
73 #include <sys/ioctl.h>
74 #include <sys/malloc.h>
75 #include <sys/dirent.h>
76 #include <sys/stat.h>
77 #include <sys/conf.h>
78 #include <sys/kauth.h>
79 #include <sys/module.h>
80 
81 #include <fs/msdosfs/bpb.h>
82 #include <fs/msdosfs/bootsect.h>
83 #include <fs/msdosfs/direntry.h>
84 #include <fs/msdosfs/denode.h>
85 #include <fs/msdosfs/msdosfsmount.h>
86 #include <fs/msdosfs/fat.h>
87 
88 MODULE(MODULE_CLASS_VFS, msdosfs, NULL);
89 
90 #ifdef MSDOSFS_DEBUG
91 #define DPRINTF(a) uprintf a
92 #else
93 #define DPRINTF(a)
94 #endif
95 
96 #define MSDOSFS_NAMEMAX(pmp) \
97 	(pmp)->pm_flags & MSDOSFSMNT_LONGNAME ? WIN_MAXLEN : 12
98 
99 VFS_PROTOS(msdosfs);
100 
101 int msdosfs_mountfs(struct vnode *, struct mount *, struct lwp *,
102     struct msdosfs_args *);
103 
104 static int update_mp(struct mount *, struct msdosfs_args *);
105 
106 MALLOC_JUSTDEFINE(M_MSDOSFSMNT, "MSDOSFS mount", "MSDOS FS mount structure");
107 MALLOC_JUSTDEFINE(M_MSDOSFSFAT, "MSDOSFS fat", "MSDOS FS fat table");
108 MALLOC_JUSTDEFINE(M_MSDOSFSTMP, "MSDOSFS temp", "MSDOS FS temp. structures");
109 
110 #define ROOTNAME "root_device"
111 
112 extern const struct vnodeopv_desc msdosfs_vnodeop_opv_desc;
113 
114 const struct vnodeopv_desc * const msdosfs_vnodeopv_descs[] = {
115 	&msdosfs_vnodeop_opv_desc,
116 	NULL,
117 };
118 
119 struct vfsops msdosfs_vfsops = {
120 	MOUNT_MSDOS,
121 	sizeof (struct msdosfs_args),
122 	msdosfs_mount,
123 	msdosfs_start,
124 	msdosfs_unmount,
125 	msdosfs_root,
126 	(void *)eopnotsupp,		/* vfs_quotactl */
127 	msdosfs_statvfs,
128 	msdosfs_sync,
129 	msdosfs_vget,
130 	msdosfs_fhtovp,
131 	msdosfs_vptofh,
132 	msdosfs_init,
133 	msdosfs_reinit,
134 	msdosfs_done,
135 	msdosfs_mountroot,
136 	(int (*)(struct mount *, struct vnode *, struct timespec *)) eopnotsupp,
137 	vfs_stdextattrctl,
138 	(void *)eopnotsupp,		/* vfs_suspendctl */
139 	genfs_renamelock_enter,
140 	genfs_renamelock_exit,
141 	(void *)eopnotsupp,
142 	msdosfs_vnodeopv_descs,
143 	0,
144 	{ NULL, NULL },
145 };
146 
147 static int
148 msdosfs_modcmd(modcmd_t cmd, void *arg)
149 {
150 
151 	switch (cmd) {
152 	case MODULE_CMD_INIT:
153 		return vfs_attach(&msdosfs_vfsops);
154 	case MODULE_CMD_FINI:
155 		return vfs_detach(&msdosfs_vfsops);
156 	default:
157 		return ENOTTY;
158 	}
159 }
160 
161 static int
162 update_mp(mp, argp)
163 	struct mount *mp;
164 	struct msdosfs_args *argp;
165 {
166 	struct msdosfsmount *pmp = VFSTOMSDOSFS(mp);
167 	int error;
168 
169 	pmp->pm_gid = argp->gid;
170 	pmp->pm_uid = argp->uid;
171 	pmp->pm_mask = argp->mask & ALLPERMS;
172 	pmp->pm_dirmask = argp->dirmask & ALLPERMS;
173 	pmp->pm_gmtoff = argp->gmtoff;
174 	pmp->pm_flags |= argp->flags & MSDOSFSMNT_MNTOPT;
175 
176 	/*
177 	 * GEMDOS knows nothing (yet) about win95
178 	 */
179 	if (pmp->pm_flags & MSDOSFSMNT_GEMDOSFS)
180 		pmp->pm_flags |= MSDOSFSMNT_NOWIN95;
181 
182 	if (pmp->pm_flags & MSDOSFSMNT_NOWIN95)
183 		pmp->pm_flags |= MSDOSFSMNT_SHORTNAME;
184 	else if (!(pmp->pm_flags &
185 	    (MSDOSFSMNT_SHORTNAME | MSDOSFSMNT_LONGNAME))) {
186 		struct vnode *rtvp;
187 
188 		/*
189 		 * Try to divine whether to support Win'95 long filenames
190 		 */
191 		if (FAT32(pmp))
192 			pmp->pm_flags |= MSDOSFSMNT_LONGNAME;
193 		else {
194 			if ((error = msdosfs_root(mp, &rtvp)) != 0)
195 				return error;
196 			pmp->pm_flags |= findwin95(VTODE(rtvp))
197 				? MSDOSFSMNT_LONGNAME
198 					: MSDOSFSMNT_SHORTNAME;
199 			vput(rtvp);
200 		}
201 	}
202 
203 	mp->mnt_stat.f_namemax = MSDOSFS_NAMEMAX(pmp);
204 
205 	return 0;
206 }
207 
208 int
209 msdosfs_mountroot()
210 {
211 	struct mount *mp;
212 	struct lwp *l = curlwp;	/* XXX */
213 	int error;
214 	struct msdosfs_args args;
215 
216 	if (device_class(root_device) != DV_DISK)
217 		return (ENODEV);
218 
219 	if ((error = vfs_rootmountalloc(MOUNT_MSDOS, "root_device", &mp))) {
220 		vrele(rootvp);
221 		return (error);
222 	}
223 
224 	args.flags = MSDOSFSMNT_VERSIONED;
225 	args.uid = 0;
226 	args.gid = 0;
227 	args.mask = 0777;
228 	args.version = MSDOSFSMNT_VERSION;
229 	args.dirmask = 0777;
230 
231 	if ((error = msdosfs_mountfs(rootvp, mp, l, &args)) != 0) {
232 		vfs_unbusy(mp, false, NULL);
233 		vfs_destroy(mp);
234 		return (error);
235 	}
236 
237 	if ((error = update_mp(mp, &args)) != 0) {
238 		(void)msdosfs_unmount(mp, 0);
239 		vfs_unbusy(mp, false, NULL);
240 		vfs_destroy(mp);
241 		vrele(rootvp);
242 		return (error);
243 	}
244 
245 	mutex_enter(&mountlist_lock);
246 	CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list);
247 	mutex_exit(&mountlist_lock);
248 	(void)msdosfs_statvfs(mp, &mp->mnt_stat);
249 	vfs_unbusy(mp, false, NULL);
250 	return (0);
251 }
252 
253 /*
254  * mp - path - addr in user space of mount point (ie /usr or whatever)
255  * data - addr in user space of mount params including the name of the block
256  * special file to treat as a filesystem.
257  */
258 int
259 msdosfs_mount(mp, path, data, data_len)
260 	struct mount *mp;
261 	const char *path;
262 	void *data;
263 	size_t *data_len;
264 {
265 	struct lwp *l = curlwp;
266 	struct nameidata nd;
267 	struct vnode *devvp;	  /* vnode for blk device to mount */
268 	struct msdosfs_args *args = data; /* holds data from mount request */
269 	/* msdosfs specific mount control block */
270 	struct msdosfsmount *pmp = NULL;
271 	int error, flags;
272 	mode_t accessmode;
273 
274 	if (*data_len < sizeof *args)
275 		return EINVAL;
276 
277 	if (mp->mnt_flag & MNT_GETARGS) {
278 		pmp = VFSTOMSDOSFS(mp);
279 		if (pmp == NULL)
280 			return EIO;
281 		args->fspec = NULL;
282 		args->uid = pmp->pm_uid;
283 		args->gid = pmp->pm_gid;
284 		args->mask = pmp->pm_mask;
285 		args->flags = pmp->pm_flags;
286 		args->version = MSDOSFSMNT_VERSION;
287 		args->dirmask = pmp->pm_dirmask;
288 		args->gmtoff = pmp->pm_gmtoff;
289 		*data_len = sizeof *args;
290 		return 0;
291 	}
292 
293 	/*
294 	 * If not versioned (i.e. using old mount_msdos(8)), fill in
295 	 * the additional structure items with suitable defaults.
296 	 */
297 	if ((args->flags & MSDOSFSMNT_VERSIONED) == 0) {
298 		args->version = 1;
299 		args->dirmask = args->mask;
300 	}
301 
302 	/*
303 	 * Reset GMT offset for pre-v3 mount structure args.
304 	 */
305 	if (args->version < 3)
306 		args->gmtoff = 0;
307 
308 	/*
309 	 * If updating, check whether changing from read-only to
310 	 * read/write; if there is no device name, that's all we do.
311 	 */
312 	if (mp->mnt_flag & MNT_UPDATE) {
313 		pmp = VFSTOMSDOSFS(mp);
314 		error = 0;
315 		if (!(pmp->pm_flags & MSDOSFSMNT_RONLY) && (mp->mnt_flag & MNT_RDONLY)) {
316 			flags = WRITECLOSE;
317 			if (mp->mnt_flag & MNT_FORCE)
318 				flags |= FORCECLOSE;
319 			error = vflush(mp, NULLVP, flags);
320 		}
321 		if (!error && (mp->mnt_flag & MNT_RELOAD))
322 			/* not yet implemented */
323 			error = EOPNOTSUPP;
324 		if (error) {
325 			DPRINTF(("vflush %d\n", error));
326 			return (error);
327 		}
328 		if ((pmp->pm_flags & MSDOSFSMNT_RONLY) && (mp->mnt_iflag & IMNT_WANTRDWR)) {
329 			/*
330 			 * If upgrade to read-write by non-root, then verify
331 			 * that user has necessary permissions on the device.
332 			 */
333 			if (kauth_authorize_generic(l->l_cred,
334 			    KAUTH_GENERIC_ISSUSER, NULL) != 0) {
335 				devvp = pmp->pm_devvp;
336 				vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
337 				error = VOP_ACCESS(devvp, VREAD | VWRITE,
338 						   l->l_cred);
339 				VOP_UNLOCK(devvp, 0);
340 				DPRINTF(("VOP_ACCESS %d\n", error));
341 				if (error)
342 					return (error);
343 			}
344 			pmp->pm_flags &= ~MSDOSFSMNT_RONLY;
345 		}
346 		if (args->fspec == NULL) {
347 			DPRINTF(("missing fspec\n"));
348 			return EINVAL;
349 		}
350 	}
351 	/*
352 	 * Not an update, or updating the name: look up the name
353 	 * and verify that it refers to a sensible block device.
354 	 */
355 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, args->fspec);
356 	if ((error = namei(&nd)) != 0) {
357 		DPRINTF(("namei %d\n", error));
358 		return (error);
359 	}
360 	devvp = nd.ni_vp;
361 
362 	if (devvp->v_type != VBLK) {
363 		DPRINTF(("not block\n"));
364 		vrele(devvp);
365 		return (ENOTBLK);
366 	}
367 	if (bdevsw_lookup(devvp->v_rdev) == NULL) {
368 		DPRINTF(("no block switch\n"));
369 		vrele(devvp);
370 		return (ENXIO);
371 	}
372 	/*
373 	 * If mount by non-root, then verify that user has necessary
374 	 * permissions on the device.
375 	 */
376 	if (kauth_authorize_generic(l->l_cred, KAUTH_GENERIC_ISSUSER, NULL) != 0) {
377 		accessmode = VREAD;
378 		if ((mp->mnt_flag & MNT_RDONLY) == 0)
379 			accessmode |= VWRITE;
380 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
381 		error = VOP_ACCESS(devvp, accessmode, l->l_cred);
382 		VOP_UNLOCK(devvp, 0);
383 		if (error) {
384 			DPRINTF(("VOP_ACCESS2 %d\n", error));
385 			vrele(devvp);
386 			return (error);
387 		}
388 	}
389 	if ((mp->mnt_flag & MNT_UPDATE) == 0) {
390 		int xflags;
391 
392 		if (mp->mnt_flag & MNT_RDONLY)
393 			xflags = FREAD;
394 		else
395 			xflags = FREAD|FWRITE;
396 		error = VOP_OPEN(devvp, xflags, FSCRED);
397 		if (error) {
398 			DPRINTF(("VOP_OPEN %d\n", error));
399 			goto fail;
400 		}
401 		error = msdosfs_mountfs(devvp, mp, l, args);
402 		if (error) {
403 			DPRINTF(("msdosfs_mountfs %d\n", error));
404 			vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
405 			(void) VOP_CLOSE(devvp, xflags, NOCRED);
406 			VOP_UNLOCK(devvp, 0);
407 			goto fail;
408 		}
409 #ifdef MSDOSFS_DEBUG		/* only needed for the printf below */
410 		pmp = VFSTOMSDOSFS(mp);
411 #endif
412 	} else {
413 		vrele(devvp);
414 		if (devvp != pmp->pm_devvp) {
415 			DPRINTF(("devvp %p pmp %p\n",
416 			    devvp, pmp->pm_devvp));
417 			return (EINVAL);	/* needs translation */
418 		}
419 	}
420 	if ((error = update_mp(mp, args)) != 0) {
421 		msdosfs_unmount(mp, MNT_FORCE);
422 		DPRINTF(("update_mp %d\n", error));
423 		return error;
424 	}
425 
426 #ifdef MSDOSFS_DEBUG
427 	printf("msdosfs_mount(): mp %p, pmp %p, inusemap %p\n", mp, pmp, pmp->pm_inusemap);
428 #endif
429 	return set_statvfs_info(path, UIO_USERSPACE, args->fspec, UIO_USERSPACE,
430 	    mp->mnt_op->vfs_name, mp, l);
431 
432 fail:
433 	vrele(devvp);
434 	return (error);
435 }
436 
437 int
438 msdosfs_mountfs(devvp, mp, l, argp)
439 	struct vnode *devvp;
440 	struct mount *mp;
441 	struct lwp *l;
442 	struct msdosfs_args *argp;
443 {
444 	struct msdosfsmount *pmp;
445 	struct buf *bp;
446 	dev_t dev = devvp->v_rdev;
447 	struct partinfo dpart;
448 	union bootsector *bsp;
449 	struct byte_bpb33 *b33;
450 	struct byte_bpb50 *b50;
451 	struct byte_bpb710 *b710;
452 	u_int8_t SecPerClust;
453 	int	ronly, error, tmp;
454 	int	bsize, dtype, fstype, secsize;
455 	u_int64_t psize;
456 
457 	/* Flush out any old buffers remaining from a previous use. */
458 	if ((error = vinvalbuf(devvp, V_SAVE, l->l_cred, l, 0, 0)) != 0)
459 		return (error);
460 
461 	ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
462 
463 	bp  = NULL; /* both used in error_exit */
464 	pmp = NULL;
465 
466 	/*
467  	 * We need the disklabel to calculate the size of a FAT entry
468 	 * later on. Also make sure the partition contains a filesystem
469 	 * of type FS_MSDOS. This doesn't work for floppies, so we have
470 	 * to check for them too.
471  	 *
472  	 * There might still be parts of the msdos fs driver which assume
473 	 * that the size of a disk block will always be 512 bytes.
474 	 * Let's root them out...
475 	 */
476 	error = VOP_IOCTL(devvp, DIOCGPART, &dpart, FREAD, NOCRED);
477 	if (error == 0) {
478 		secsize = dpart.disklab->d_secsize;
479 		dtype = dpart.disklab->d_type;
480 		fstype = dpart.part->p_fstype;
481 		psize = dpart.part->p_size;
482 	} else {
483 		struct dkwedge_info dkw;
484 		error = VOP_IOCTL(devvp, DIOCGWEDGEINFO, &dkw, FREAD, NOCRED);
485 		secsize = 512;	/* XXX */
486 		dtype = DTYPE_FLOPPY; /* XXX */
487 		fstype = FS_MSDOS;
488 		psize = -1;
489 		if (error) {
490 			if (error != ENOTTY) {
491 				DPRINTF(("Error getting partition info %d\n",
492 				    error));
493 				goto error_exit;
494 			}
495 		} else {
496 			fstype = strcmp(dkw.dkw_ptype, DKW_PTYPE_FAT) == 0 ?
497 			    FS_MSDOS : -1;
498 			psize = dkw.dkw_size;
499 		}
500 	}
501 	if (argp->flags & MSDOSFSMNT_GEMDOSFS) {
502 		bsize = secsize;
503 		if (bsize != 512 ||
504 		    (dtype != DTYPE_FLOPPY && fstype != FS_MSDOS)) {
505 			DPRINTF(("bsize %d dtype %d fstype %d\n", bsize, dtype,
506 			    fstype));
507 			error = EINVAL;
508 			goto error_exit;
509 		}
510 	} else
511 		bsize = 0;
512 
513 	/*
514 	 * Read the boot sector of the filesystem, and then check the
515 	 * boot signature.  If not a dos boot sector then error out.
516 	 */
517 	if ((error = bread(devvp, 0, secsize, NOCRED, 0, &bp)) != 0)
518 		goto error_exit;
519 	bsp = (union bootsector *)bp->b_data;
520 	b33 = (struct byte_bpb33 *)bsp->bs33.bsBPB;
521 	b50 = (struct byte_bpb50 *)bsp->bs50.bsBPB;
522 	b710 = (struct byte_bpb710 *)bsp->bs710.bsBPB;
523 
524 	if (!(argp->flags & MSDOSFSMNT_GEMDOSFS)) {
525 		if (bsp->bs50.bsBootSectSig0 != BOOTSIG0
526 		    || bsp->bs50.bsBootSectSig1 != BOOTSIG1) {
527 			DPRINTF(("bootsig0 %d bootsig1 %d\n",
528 			    bsp->bs50.bsBootSectSig0,
529 			    bsp->bs50.bsBootSectSig1));
530 			error = EINVAL;
531 			goto error_exit;
532 		}
533 	}
534 
535 	pmp = malloc(sizeof *pmp, M_MSDOSFSMNT, M_WAITOK);
536 	memset(pmp, 0, sizeof *pmp);
537 	pmp->pm_mountp = mp;
538 
539 	/*
540 	 * Compute several useful quantities from the bpb in the
541 	 * bootsector.  Copy in the dos 5 variant of the bpb then fix up
542 	 * the fields that are different between dos 5 and dos 3.3.
543 	 */
544 	SecPerClust = b50->bpbSecPerClust;
545 	pmp->pm_BytesPerSec = getushort(b50->bpbBytesPerSec);
546 	pmp->pm_ResSectors = getushort(b50->bpbResSectors);
547 	pmp->pm_FATs = b50->bpbFATs;
548 	pmp->pm_RootDirEnts = getushort(b50->bpbRootDirEnts);
549 	pmp->pm_Sectors = getushort(b50->bpbSectors);
550 	pmp->pm_FATsecs = getushort(b50->bpbFATsecs);
551 	pmp->pm_SecPerTrack = getushort(b50->bpbSecPerTrack);
552 	pmp->pm_Heads = getushort(b50->bpbHeads);
553 	pmp->pm_Media = b50->bpbMedia;
554 
555 	if (!(argp->flags & MSDOSFSMNT_GEMDOSFS)) {
556 		/* XXX - We should probably check more values here */
557     		if (!pmp->pm_BytesPerSec || !SecPerClust
558 	    		|| pmp->pm_Heads > 255 || pmp->pm_SecPerTrack > 63) {
559 			DPRINTF(("bytespersec %d secperclust %d "
560 			    "heads %d secpertrack %d\n",
561 			    pmp->pm_BytesPerSec, SecPerClust,
562 			    pmp->pm_Heads, pmp->pm_SecPerTrack));
563 			error = EINVAL;
564 			goto error_exit;
565 		}
566 	}
567 
568 	if (pmp->pm_Sectors == 0) {
569 		pmp->pm_HiddenSects = getulong(b50->bpbHiddenSecs);
570 		pmp->pm_HugeSectors = getulong(b50->bpbHugeSectors);
571 	} else {
572 		pmp->pm_HiddenSects = getushort(b33->bpbHiddenSecs);
573 		pmp->pm_HugeSectors = pmp->pm_Sectors;
574 	}
575 
576 	if (pmp->pm_RootDirEnts == 0) {
577 		unsigned short vers = getushort(b710->bpbFSVers);
578 		/*
579 		 * Some say that bsBootSectSig[23] must be zero, but
580 		 * Windows does not require this and some digital cameras
581 		 * do not set these to zero.  Therefore, do not insist.
582 		 */
583 		if (pmp->pm_Sectors || pmp->pm_FATsecs || vers) {
584 			DPRINTF(("sectors %d fatsecs %lu vers %d\n",
585 			    pmp->pm_Sectors, pmp->pm_FATsecs, vers));
586 			error = EINVAL;
587 			goto error_exit;
588 		}
589 		pmp->pm_fatmask = FAT32_MASK;
590 		pmp->pm_fatmult = 4;
591 		pmp->pm_fatdiv = 1;
592 		pmp->pm_FATsecs = getulong(b710->bpbBigFATsecs);
593 
594 		/* mirrorring is enabled if the FATMIRROR bit is not set */
595 		if ((getushort(b710->bpbExtFlags) & FATMIRROR) == 0)
596 			pmp->pm_flags |= MSDOSFS_FATMIRROR;
597 		else
598 			pmp->pm_curfat = getushort(b710->bpbExtFlags) & FATNUM;
599 	} else
600 		pmp->pm_flags |= MSDOSFS_FATMIRROR;
601 
602 	if (argp->flags & MSDOSFSMNT_GEMDOSFS) {
603 		if (FAT32(pmp)) {
604 			DPRINTF(("fat32 for gemdos\n"));
605 			/*
606 			 * GEMDOS doesn't know fat32.
607 			 */
608 			error = EINVAL;
609 			goto error_exit;
610 		}
611 
612 		/*
613 		 * Check a few values (could do some more):
614 		 * - logical sector size: power of 2, >= block size
615 		 * - sectors per cluster: power of 2, >= 1
616 		 * - number of sectors:   >= 1, <= size of partition
617 		 */
618 		if ( (SecPerClust == 0)
619 		  || (SecPerClust & (SecPerClust - 1))
620 		  || (pmp->pm_BytesPerSec < bsize)
621 		  || (pmp->pm_BytesPerSec & (pmp->pm_BytesPerSec - 1))
622 		  || (pmp->pm_HugeSectors == 0)
623 		  || (pmp->pm_HugeSectors * (pmp->pm_BytesPerSec / bsize)
624 		      > psize)) {
625 			DPRINTF(("consistency checks for gemdos\n"));
626 			error = EINVAL;
627 			goto error_exit;
628 		}
629 		/*
630 		 * XXX - Many parts of the msdos fs driver seem to assume that
631 		 * the number of bytes per logical sector (BytesPerSec) will
632 		 * always be the same as the number of bytes per disk block
633 		 * Let's pretend it is.
634 		 */
635 		tmp = pmp->pm_BytesPerSec / bsize;
636 		pmp->pm_BytesPerSec  = bsize;
637 		pmp->pm_HugeSectors *= tmp;
638 		pmp->pm_HiddenSects *= tmp;
639 		pmp->pm_ResSectors  *= tmp;
640 		pmp->pm_Sectors     *= tmp;
641 		pmp->pm_FATsecs     *= tmp;
642 		SecPerClust         *= tmp;
643 	}
644 
645 	/* Check that fs has nonzero FAT size */
646 	if (pmp->pm_FATsecs == 0) {
647 		DPRINTF(("FATsecs is 0\n"));
648 		error = EINVAL;
649 		goto error_exit;
650 	}
651 
652 	pmp->pm_fatblk = pmp->pm_ResSectors;
653 	if (FAT32(pmp)) {
654 		pmp->pm_rootdirblk = getulong(b710->bpbRootClust);
655 		pmp->pm_firstcluster = pmp->pm_fatblk
656 			+ (pmp->pm_FATs * pmp->pm_FATsecs);
657 		pmp->pm_fsinfo = getushort(b710->bpbFSInfo);
658 	} else {
659 		pmp->pm_rootdirblk = pmp->pm_fatblk +
660 			(pmp->pm_FATs * pmp->pm_FATsecs);
661 		pmp->pm_rootdirsize = (pmp->pm_RootDirEnts * sizeof(struct direntry)
662 				       + pmp->pm_BytesPerSec - 1)
663 			/ pmp->pm_BytesPerSec;/* in sectors */
664 		pmp->pm_firstcluster = pmp->pm_rootdirblk + pmp->pm_rootdirsize;
665 	}
666 
667 	pmp->pm_nmbrofclusters = (pmp->pm_HugeSectors - pmp->pm_firstcluster) /
668 	    SecPerClust;
669 	pmp->pm_maxcluster = pmp->pm_nmbrofclusters + 1;
670 	pmp->pm_fatsize = pmp->pm_FATsecs * pmp->pm_BytesPerSec;
671 
672 	if (argp->flags & MSDOSFSMNT_GEMDOSFS) {
673 		if (pmp->pm_nmbrofclusters <= (0xff0 - 2)
674 		      && (dtype == DTYPE_FLOPPY
675 			  || (dtype == DTYPE_VND
676 				&& (pmp->pm_Heads == 1 || pmp->pm_Heads == 2)))
677 		    ) {
678 			pmp->pm_fatmask = FAT12_MASK;
679 			pmp->pm_fatmult = 3;
680 			pmp->pm_fatdiv = 2;
681 		} else {
682 			pmp->pm_fatmask = FAT16_MASK;
683 			pmp->pm_fatmult = 2;
684 			pmp->pm_fatdiv = 1;
685 		}
686 	} else if (pmp->pm_fatmask == 0) {
687 		if (pmp->pm_maxcluster
688 		    <= ((CLUST_RSRVD - CLUST_FIRST) & FAT12_MASK)) {
689 			/*
690 			 * This will usually be a floppy disk. This size makes
691 			 * sure that one fat entry will not be split across
692 			 * multiple blocks.
693 			 */
694 			pmp->pm_fatmask = FAT12_MASK;
695 			pmp->pm_fatmult = 3;
696 			pmp->pm_fatdiv = 2;
697 		} else {
698 			pmp->pm_fatmask = FAT16_MASK;
699 			pmp->pm_fatmult = 2;
700 			pmp->pm_fatdiv = 1;
701 		}
702 	}
703 	if (FAT12(pmp))
704 		pmp->pm_fatblocksize = 3 * pmp->pm_BytesPerSec;
705 	else
706 		pmp->pm_fatblocksize = MAXBSIZE;
707 
708 	pmp->pm_fatblocksec = pmp->pm_fatblocksize / pmp->pm_BytesPerSec;
709 	pmp->pm_bnshift = ffs(pmp->pm_BytesPerSec) - 1;
710 
711 	/*
712 	 * Compute mask and shift value for isolating cluster relative byte
713 	 * offsets and cluster numbers from a file offset.
714 	 */
715 	pmp->pm_bpcluster = SecPerClust * pmp->pm_BytesPerSec;
716 	pmp->pm_crbomask = pmp->pm_bpcluster - 1;
717 	pmp->pm_cnshift = ffs(pmp->pm_bpcluster) - 1;
718 
719 	/*
720 	 * Check for valid cluster size
721 	 * must be a power of 2
722 	 */
723 	if (pmp->pm_bpcluster ^ (1 << pmp->pm_cnshift)) {
724 		DPRINTF(("bpcluster %lu cnshift %lu\n",
725 		    pmp->pm_bpcluster, pmp->pm_cnshift));
726 		error = EINVAL;
727 		goto error_exit;
728 	}
729 
730 	/*
731 	 * Release the bootsector buffer.
732 	 */
733 	brelse(bp, BC_AGE);
734 	bp = NULL;
735 
736 	/*
737 	 * Check FSInfo.
738 	 */
739 	if (pmp->pm_fsinfo) {
740 		struct fsinfo *fp;
741 
742 		/*
743 		 * XXX	If the fsinfo block is stored on media with
744 		 *	2KB or larger sectors, is the fsinfo structure
745 		 *	padded at the end or in the middle?
746 		 */
747 		if ((error = bread(devvp, de_bn2kb(pmp, pmp->pm_fsinfo),
748 		    pmp->pm_BytesPerSec, NOCRED, 0, &bp)) != 0)
749 			goto error_exit;
750 		fp = (struct fsinfo *)bp->b_data;
751 		if (!memcmp(fp->fsisig1, "RRaA", 4)
752 		    && !memcmp(fp->fsisig2, "rrAa", 4)
753 		    && !memcmp(fp->fsisig3, "\0\0\125\252", 4)
754 		    && !memcmp(fp->fsisig4, "\0\0\125\252", 4))
755 			pmp->pm_nxtfree = getulong(fp->fsinxtfree);
756 		else
757 			pmp->pm_fsinfo = 0;
758 		brelse(bp, 0);
759 		bp = NULL;
760 	}
761 
762 	/*
763 	 * Check and validate (or perhaps invalidate?) the fsinfo structure?
764 	 * XXX
765 	 */
766 	if (pmp->pm_fsinfo) {
767 		if (pmp->pm_nxtfree == (u_long)-1)
768 			pmp->pm_fsinfo = 0;
769 	}
770 
771 	/*
772 	 * Allocate memory for the bitmap of allocated clusters, and then
773 	 * fill it in.
774 	 */
775 	pmp->pm_inusemap = malloc(((pmp->pm_maxcluster + N_INUSEBITS - 1)
776 				   / N_INUSEBITS)
777 				  * sizeof(*pmp->pm_inusemap),
778 				  M_MSDOSFSFAT, M_WAITOK);
779 
780 	/*
781 	 * fillinusemap() needs pm_devvp.
782 	 */
783 	pmp->pm_dev = dev;
784 	pmp->pm_devvp = devvp;
785 
786 	/*
787 	 * Have the inuse map filled in.
788 	 */
789 	if ((error = fillinusemap(pmp)) != 0) {
790 		DPRINTF(("fillinusemap %d\n", error));
791 		goto error_exit;
792 	}
793 
794 	/*
795 	 * If they want fat updates to be synchronous then let them suffer
796 	 * the performance degradation in exchange for the on disk copy of
797 	 * the fat being correct just about all the time.  I suppose this
798 	 * would be a good thing to turn on if the kernel is still flakey.
799 	 */
800 	if (mp->mnt_flag & MNT_SYNCHRONOUS)
801 		pmp->pm_flags |= MSDOSFSMNT_WAITONFAT;
802 
803 	/*
804 	 * Finish up.
805 	 */
806 	if (ronly)
807 		pmp->pm_flags |= MSDOSFSMNT_RONLY;
808 	else
809 		pmp->pm_fmod = 1;
810 	mp->mnt_data = pmp;
811 	mp->mnt_stat.f_fsidx.__fsid_val[0] = (long)dev;
812 	mp->mnt_stat.f_fsidx.__fsid_val[1] = makefstype(MOUNT_MSDOS);
813 	mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0];
814 	mp->mnt_stat.f_namemax = MSDOSFS_NAMEMAX(pmp);
815 	mp->mnt_flag |= MNT_LOCAL;
816 	mp->mnt_dev_bshift = pmp->pm_bnshift;
817 	mp->mnt_fs_bshift = pmp->pm_cnshift;
818 
819 #ifdef QUOTA
820 	/*
821 	 * If we ever do quotas for DOS filesystems this would be a place
822 	 * to fill in the info in the msdosfsmount structure. You dolt,
823 	 * quotas on dos filesystems make no sense because files have no
824 	 * owners on dos filesystems. of course there is some empty space
825 	 * in the directory entry where we could put uid's and gid's.
826 	 */
827 #endif
828 	devvp->v_specmountpoint = mp;
829 
830 	return (0);
831 
832 error_exit:;
833 	if (bp)
834 		brelse(bp, BC_AGE);
835 	if (pmp) {
836 		if (pmp->pm_inusemap)
837 			free(pmp->pm_inusemap, M_MSDOSFSFAT);
838 		free(pmp, M_MSDOSFSMNT);
839 		mp->mnt_data = NULL;
840 	}
841 	return (error);
842 }
843 
844 int
845 msdosfs_start(struct mount *mp, int flags)
846 {
847 
848 	return (0);
849 }
850 
851 /*
852  * Unmount the filesystem described by mp.
853  */
854 int
855 msdosfs_unmount(mp, mntflags)
856 	struct mount *mp;
857 	int mntflags;
858 {
859 	struct msdosfsmount *pmp;
860 	int error, flags;
861 
862 	flags = 0;
863 	if (mntflags & MNT_FORCE)
864 		flags |= FORCECLOSE;
865 #ifdef QUOTA
866 #endif
867 	if ((error = vflush(mp, NULLVP, flags)) != 0)
868 		return (error);
869 	pmp = VFSTOMSDOSFS(mp);
870 	if (pmp->pm_devvp->v_type != VBAD)
871 		pmp->pm_devvp->v_specmountpoint = NULL;
872 #ifdef MSDOSFS_DEBUG
873 	{
874 		struct vnode *vp = pmp->pm_devvp;
875 
876 		printf("msdosfs_umount(): just before calling VOP_CLOSE()\n");
877 		printf("flag %08x, usecount %d, writecount %d, holdcnt %d\n",
878 		    vp->v_vflag | vp->v_iflag | vp->v_uflag, vp->v_usecount,
879 		    vp->v_writecount, vp->v_holdcnt);
880 		printf("mount %p, op %p\n",
881 		    vp->v_mount, vp->v_op);
882 		printf("freef %p, freeb %p, mount %p\n",
883 		    vp->v_freelist.tqe_next, vp->v_freelist.tqe_prev,
884 		    vp->v_mount);
885 		printf("cleanblkhd %p, dirtyblkhd %p, numoutput %d, type %d\n",
886 		    vp->v_cleanblkhd.lh_first,
887 		    vp->v_dirtyblkhd.lh_first,
888 		    vp->v_numoutput, vp->v_type);
889 		printf("union %p, tag %d, data[0] %08x, data[1] %08x\n",
890 		    vp->v_socket, vp->v_tag,
891 		    ((u_int *)vp->v_data)[0],
892 		    ((u_int *)vp->v_data)[1]);
893 	}
894 #endif
895 	vn_lock(pmp->pm_devvp, LK_EXCLUSIVE | LK_RETRY);
896 	error = VOP_CLOSE(pmp->pm_devvp,
897 	    pmp->pm_flags & MSDOSFSMNT_RONLY ? FREAD : FREAD|FWRITE, NOCRED);
898 	vput(pmp->pm_devvp);
899 	free(pmp->pm_inusemap, M_MSDOSFSFAT);
900 	free(pmp, M_MSDOSFSMNT);
901 	mp->mnt_data = NULL;
902 	mp->mnt_flag &= ~MNT_LOCAL;
903 	return (error);
904 }
905 
906 int
907 msdosfs_root(mp, vpp)
908 	struct mount *mp;
909 	struct vnode **vpp;
910 {
911 	struct msdosfsmount *pmp = VFSTOMSDOSFS(mp);
912 	struct denode *ndep;
913 	int error;
914 
915 #ifdef MSDOSFS_DEBUG
916 	printf("msdosfs_root(); mp %p, pmp %p\n", mp, pmp);
917 #endif
918 	if ((error = deget(pmp, MSDOSFSROOT, MSDOSFSROOT_OFS, &ndep)) != 0)
919 		return (error);
920 	*vpp = DETOV(ndep);
921 	return (0);
922 }
923 
924 int
925 msdosfs_statvfs(struct mount *mp, struct statvfs *sbp)
926 {
927 	struct msdosfsmount *pmp;
928 
929 	pmp = VFSTOMSDOSFS(mp);
930 	sbp->f_bsize = pmp->pm_bpcluster;
931 	sbp->f_frsize = sbp->f_bsize;
932 	sbp->f_iosize = pmp->pm_bpcluster;
933 	sbp->f_blocks = pmp->pm_nmbrofclusters;
934 	sbp->f_bfree = pmp->pm_freeclustercount;
935 	sbp->f_bavail = pmp->pm_freeclustercount;
936 	sbp->f_bresvd = 0;
937 	sbp->f_files = pmp->pm_RootDirEnts;			/* XXX */
938 	sbp->f_ffree = 0;	/* what to put in here? */
939 	sbp->f_favail = 0;	/* what to put in here? */
940 	sbp->f_fresvd = 0;
941 	copy_statvfs_info(sbp, mp);
942 	return (0);
943 }
944 
945 int
946 msdosfs_sync(mp, waitfor, cred)
947 	struct mount *mp;
948 	int waitfor;
949 	kauth_cred_t cred;
950 {
951 	struct vnode *vp, *mvp;
952 	struct denode *dep;
953 	struct msdosfsmount *pmp = VFSTOMSDOSFS(mp);
954 	int error, allerror = 0;
955 
956 	/*
957 	 * If we ever switch to not updating all of the fats all the time,
958 	 * this would be the place to update them from the first one.
959 	 */
960 	if (pmp->pm_fmod != 0) {
961 		if (pmp->pm_flags & MSDOSFSMNT_RONLY)
962 			panic("msdosfs_sync: rofs mod");
963 		else {
964 			/* update fats here */
965 		}
966 	}
967 	/* Allocate a marker vnode. */
968 	if ((mvp = vnalloc(mp)) == NULL)
969 		return ENOMEM;
970 	/*
971 	 * Write back each (modified) denode.
972 	 */
973 	mutex_enter(&mntvnode_lock);
974 loop:
975 	for (vp = TAILQ_FIRST(&mp->mnt_vnodelist); vp; vp = vunmark(mvp)) {
976 		vmark(mvp, vp);
977 		if (vp->v_mount != mp || vismarker(vp))
978 			continue;
979 		mutex_enter(&vp->v_interlock);
980 		dep = VTODE(vp);
981 		if (waitfor == MNT_LAZY || vp->v_type == VNON ||
982 		    (((dep->de_flag &
983 		    (DE_ACCESS | DE_CREATE | DE_UPDATE | DE_MODIFIED)) == 0) &&
984 		     (LIST_EMPTY(&vp->v_dirtyblkhd) &&
985 		      UVM_OBJ_IS_CLEAN(&vp->v_uobj)))) {
986 			mutex_exit(&vp->v_interlock);
987 			continue;
988 		}
989 		mutex_exit(&mntvnode_lock);
990 		error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK);
991 		if (error) {
992 			mutex_enter(&mntvnode_lock);
993 			if (error == ENOENT) {
994 				(void)vunmark(mvp);
995 				goto loop;
996 			}
997 			continue;
998 		}
999 		if ((error = VOP_FSYNC(vp, cred,
1000 		    waitfor == MNT_WAIT ? FSYNC_WAIT : 0, 0, 0)) != 0)
1001 			allerror = error;
1002 		vput(vp);
1003 		mutex_enter(&mntvnode_lock);
1004 	}
1005 	mutex_exit(&mntvnode_lock);
1006 	vnfree(mvp);
1007 
1008 	/*
1009 	 * Force stale file system control information to be flushed.
1010 	 */
1011 	if ((error = VOP_FSYNC(pmp->pm_devvp, cred,
1012 	    waitfor == MNT_WAIT ? FSYNC_WAIT : 0, 0, 0)) != 0)
1013 		allerror = error;
1014 #ifdef QUOTA
1015 	/* qsync(mp); */
1016 #endif
1017 	return (allerror);
1018 }
1019 
1020 int
1021 msdosfs_fhtovp(mp, fhp, vpp)
1022 	struct mount *mp;
1023 	struct fid *fhp;
1024 	struct vnode **vpp;
1025 {
1026 	struct msdosfsmount *pmp = VFSTOMSDOSFS(mp);
1027 	struct defid defh;
1028 	struct denode *dep;
1029 	int error;
1030 
1031 	if (fhp->fid_len != sizeof(struct defid)) {
1032 		DPRINTF(("fid_len %d %zd\n", fhp->fid_len,
1033 		    sizeof(struct defid)));
1034 		return EINVAL;
1035 	}
1036 
1037 	memcpy(&defh, fhp, sizeof(defh));
1038 	error = deget(pmp, defh.defid_dirclust, defh.defid_dirofs, &dep);
1039 	if (error) {
1040 		DPRINTF(("deget %d\n", error));
1041 		*vpp = NULLVP;
1042 		return (error);
1043 	}
1044 	*vpp = DETOV(dep);
1045 	return (0);
1046 }
1047 
1048 int
1049 msdosfs_vptofh(vp, fhp, fh_size)
1050 	struct vnode *vp;
1051 	struct fid *fhp;
1052 	size_t *fh_size;
1053 {
1054 	struct denode *dep;
1055 	struct defid defh;
1056 
1057 	if (*fh_size < sizeof(struct defid)) {
1058 		*fh_size = sizeof(struct defid);
1059 		return E2BIG;
1060 	}
1061 	*fh_size = sizeof(struct defid);
1062 	dep = VTODE(vp);
1063 	memset(&defh, 0, sizeof(defh));
1064 	defh.defid_len = sizeof(struct defid);
1065 	defh.defid_dirclust = dep->de_dirclust;
1066 	defh.defid_dirofs = dep->de_diroffset;
1067 	/* defh.defid_gen = dep->de_gen; */
1068 	memcpy(fhp, &defh, sizeof(defh));
1069 	return (0);
1070 }
1071 
1072 int
1073 msdosfs_vget(struct mount *mp, ino_t ino,
1074     struct vnode **vpp)
1075 {
1076 
1077 	return (EOPNOTSUPP);
1078 }
1079 
1080 SYSCTL_SETUP(sysctl_vfs_msdosfs_setup, "sysctl vfs.msdosfs subtree setup")
1081 {
1082 
1083 	sysctl_createv(clog, 0, NULL, NULL,
1084 		       CTLFLAG_PERMANENT,
1085 		       CTLTYPE_NODE, "vfs", NULL,
1086 		       NULL, 0, NULL, 0,
1087 		       CTL_VFS, CTL_EOL);
1088 	sysctl_createv(clog, 0, NULL, NULL,
1089 		       CTLFLAG_PERMANENT,
1090 		       CTLTYPE_NODE, "msdosfs",
1091 		       SYSCTL_DESCR("MS-DOS file system"),
1092 		       NULL, 0, NULL, 0,
1093 		       CTL_VFS, 4, CTL_EOL);
1094 	/*
1095 	 * XXX the "4" above could be dynamic, thereby eliminating one
1096 	 * more instance of the "number to vfs" mapping problem, but
1097 	 * "4" is the order as taken from sys/mount.h
1098 	 */
1099 }
1100