xref: /openbsd/sys/msdosfs/msdosfs_vnops.c (revision 73471bf0)
1 /*	$OpenBSD: msdosfs_vnops.c,v 1.137 2021/12/12 09:14:59 visa Exp $	*/
2 /*	$NetBSD: msdosfs_vnops.c,v 1.63 1997/10/17 11:24:19 ws Exp $	*/
3 
4 /*-
5  * Copyright (C) 2005 Thomas Wang.
6  * Copyright (C) 1994, 1995, 1997 Wolfgang Solfrank.
7  * Copyright (C) 1994, 1995, 1997 TooLs GmbH.
8  * All rights reserved.
9  * Original code by Paul Popelka (paulp@uts.amdahl.com) (see below).
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. All advertising materials mentioning features or use of this software
20  *    must display the following acknowledgement:
21  *	This product includes software developed by TooLs GmbH.
22  * 4. The name of TooLs GmbH may not be used to endorse or promote products
23  *    derived from this software without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
26  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
27  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
28  * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
29  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
30  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
31  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
32  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
33  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
34  * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35  */
36 /*
37  * Written by Paul Popelka (paulp@uts.amdahl.com)
38  *
39  * You can do anything you want with this software, just don't say you wrote
40  * it, and don't remove this notice.
41  *
42  * This software is provided "as is".
43  *
44  * The author supplies this software to be publicly redistributed on the
45  * understanding that the author is not responsible for the correct
46  * functioning of this software in any circumstances and is not liable for
47  * any damages caused by this software.
48  *
49  * October 1992
50  */
51 
52 #include <sys/param.h>
53 #include <sys/systm.h>
54 #include <sys/namei.h>
55 #include <sys/resourcevar.h>	/* defines plimit structure in proc struct */
56 #include <sys/fcntl.h>		/* define FWRITE ... */
57 #include <sys/file.h>
58 #include <sys/stat.h>
59 #include <sys/buf.h>
60 #include <sys/proc.h>
61 #include <sys/mount.h>
62 #include <sys/vnode.h>
63 #include <sys/lock.h>
64 #include <sys/signalvar.h>
65 #include <sys/specdev.h> /* XXX */	/* defines v_rdev */
66 #include <sys/malloc.h>
67 #include <sys/pool.h>
68 #include <sys/dirent.h>		/* defines dirent structure */
69 #include <sys/lockf.h>
70 #include <sys/poll.h>
71 #include <sys/unistd.h>
72 
73 #include <msdosfs/bpb.h>
74 #include <msdosfs/direntry.h>
75 #include <msdosfs/denode.h>
76 #include <msdosfs/msdosfsmount.h>
77 #include <msdosfs/fat.h>
78 
79 static uint32_t fileidhash(uint64_t);
80 
81 int msdosfs_bmaparray(struct vnode *, uint32_t, daddr_t *, int *);
82 int msdosfs_kqfilter(void *);
83 int filt_msdosfsread(struct knote *, long);
84 int filt_msdosfswrite(struct knote *, long);
85 int filt_msdosfsvnode(struct knote *, long);
86 void filt_msdosfsdetach(struct knote *);
87 
88 /*
89  * Some general notes:
90  *
91  * In the ufs filesystem the inodes, superblocks, and indirect blocks are
92  * read/written using the vnode for the filesystem. Blocks that represent
93  * the contents of a file are read/written using the vnode for the file
94  * (including directories when they are read/written as files). This
95  * presents problems for the dos filesystem because data that should be in
96  * an inode (if dos had them) resides in the directory itself.  Since we
97  * must update directory entries without the benefit of having the vnode
98  * for the directory we must use the vnode for the filesystem.  This means
99  * that when a directory is actually read/written (via read, write, or
100  * readdir, or seek) we must use the vnode for the filesystem instead of
101  * the vnode for the directory as would happen in ufs. This is to insure we
102  * retrieve the correct block from the buffer cache since the hash value is
103  * based upon the vnode address and the desired block number.
104  */
105 
106 /*
107  * Create a regular file. On entry the directory to contain the file being
108  * created is locked.  We must release before we return. We must also free
109  * the pathname buffer pointed at by cnp->cn_pnbuf, always on error, or
110  * only if the SAVESTART bit in cn_flags is clear on success.
111  */
112 int
113 msdosfs_create(void *v)
114 {
115 	struct vop_create_args *ap = v;
116 	struct componentname *cnp = ap->a_cnp;
117 	struct denode ndirent;
118 	struct denode *dep;
119 	struct denode *pdep = VTODE(ap->a_dvp);
120 	int error;
121 	struct timespec ts;
122 
123 #ifdef MSDOSFS_DEBUG
124 	printf("msdosfs_create(cnp %p, vap %p\n", cnp, ap->a_vap);
125 #endif
126 
127 	/*
128 	 * If this is the root directory and there is no space left we
129 	 * can't do anything.  This is because the root directory can not
130 	 * change size.
131 	 */
132 	if (pdep->de_StartCluster == MSDOSFSROOT
133 	    && pdep->de_fndoffset >= pdep->de_FileSize) {
134 		error = ENOSPC;
135 		goto bad;
136 	}
137 
138 	/*
139 	 * Create a directory entry for the file, then call createde() to
140 	 * have it installed. NOTE: DOS files are always executable.  We
141 	 * use the absence of the owner write bit to make the file
142 	 * readonly.
143 	 */
144 #ifdef DIAGNOSTIC
145 	if ((cnp->cn_flags & HASBUF) == 0)
146 		panic("msdosfs_create: no name");
147 #endif
148 	bzero(&ndirent, sizeof(ndirent));
149 	if ((error = uniqdosname(pdep, cnp, ndirent.de_Name)) != 0)
150 		goto bad;
151 
152 	ndirent.de_Attributes = (ap->a_vap->va_mode & VWRITE) ?
153 				ATTR_ARCHIVE : ATTR_ARCHIVE | ATTR_READONLY;
154 	ndirent.de_StartCluster = 0;
155 	ndirent.de_FileSize = 0;
156 	ndirent.de_dev = pdep->de_dev;
157 	ndirent.de_devvp = pdep->de_devvp;
158 	ndirent.de_pmp = pdep->de_pmp;
159 	ndirent.de_flag = DE_ACCESS | DE_CREATE | DE_UPDATE;
160 	getnanotime(&ts);
161 	DETIMES(&ndirent, &ts, &ts, &ts);
162 	if ((error = createde(&ndirent, pdep, &dep, cnp)) != 0)
163 		goto bad;
164 	if ((cnp->cn_flags & SAVESTART) == 0)
165 		pool_put(&namei_pool, cnp->cn_pnbuf);
166 	VN_KNOTE(ap->a_dvp, NOTE_WRITE);
167 	*ap->a_vpp = DETOV(dep);
168 	return (0);
169 
170 bad:
171 	pool_put(&namei_pool, cnp->cn_pnbuf);
172 	return (error);
173 }
174 
175 int
176 msdosfs_mknod(void *v)
177 {
178 	struct vop_mknod_args *ap = v;
179 
180 	pool_put(&namei_pool, ap->a_cnp->cn_pnbuf);
181 	VN_KNOTE(ap->a_dvp, NOTE_WRITE);
182 	return (EINVAL);
183 }
184 
185 int
186 msdosfs_open(void *v)
187 {
188 	return (0);
189 }
190 
191 int
192 msdosfs_close(void *v)
193 {
194 	struct vop_close_args *ap = v;
195 	struct vnode *vp = ap->a_vp;
196 	struct denode *dep = VTODE(vp);
197 	struct timespec ts;
198 
199 	if (vp->v_usecount > 1 && !VOP_ISLOCKED(vp)) {
200 		getnanotime(&ts);
201 		DETIMES(dep, &ts, &ts, &ts);
202 	}
203 	return (0);
204 }
205 
206 int
207 msdosfs_access(void *v)
208 {
209 	struct vop_access_args *ap = v;
210 	struct denode *dep = VTODE(ap->a_vp);
211 	struct msdosfsmount *pmp = dep->de_pmp;
212 	mode_t dosmode;
213 
214 	dosmode = (S_IRUSR | S_IRGRP | S_IROTH);
215 	if ((dep->de_Attributes & ATTR_READONLY) == 0)
216 		dosmode |= (S_IWUSR | S_IWGRP | S_IWOTH);
217 	if (dep->de_Attributes & ATTR_DIRECTORY)
218 		dosmode |= (S_IXUSR | S_IXGRP | S_IXOTH);
219 	dosmode &= pmp->pm_mask;
220 
221 	return (vaccess(ap->a_vp->v_type, dosmode, pmp->pm_uid, pmp->pm_gid,
222 	    ap->a_mode, ap->a_cred));
223 }
224 
225 int
226 msdosfs_getattr(void *v)
227 {
228 	struct vop_getattr_args *ap = v;
229 	struct denode *dep = VTODE(ap->a_vp);
230 	struct msdosfsmount *pmp = dep->de_pmp;
231 	struct vattr *vap = ap->a_vap;
232 	struct timespec ts;
233 	uint32_t fileid;
234 
235 	getnanotime(&ts);
236 	DETIMES(dep, &ts, &ts, &ts);
237 	vap->va_fsid = dep->de_dev;
238 
239 	/*
240 	 * The following computation of the fileid must be the same as
241 	 * that used in msdosfs_readdir() to compute d_fileno. If not,
242 	 * pwd doesn't work.
243 	 *
244 	 * We now use the starting cluster number as the fileid/fileno.
245 	 * This works for both files and directories (including the root
246 	 * directory, on FAT32).  Even on FAT32, this will at most be a
247 	 * 28-bit number, as the high 4 bits of FAT32 cluster numbers
248 	 * are reserved.
249 	 *
250 	 * However, we do need to do something for 0-length files, which
251 	 * will not have a starting cluster number.
252 	 *
253 	 * These files cannot be directories, since (except for /, which
254 	 * is special-cased anyway) directories contain entries for . and
255 	 * .., so must have non-zero length.
256 	 *
257 	 * In this case, we just create a non-cryptographic hash of the
258 	 * original fileid calculation, and set the top bit.
259 	 *
260 	 * This algorithm has the benefit that all directories, and all
261 	 * non-zero-length files, will have fileids that are persistent
262 	 * across mounts and reboots, and that cannot collide (as long
263 	 * as the filesystem is not corrupt).  Zero-length files will
264 	 * have fileids that are persistent, but that may collide.  We
265 	 * will just have to live with that.
266 	 */
267 	fileid = dep->de_StartCluster;
268 
269 	if (dep->de_Attributes & ATTR_DIRECTORY) {
270 		/* Special-case root */
271 		if (dep->de_StartCluster == MSDOSFSROOT)
272 			fileid = FAT32(pmp) ? pmp->pm_rootdirblk : 1;
273 	} else {
274 		if (dep->de_FileSize == 0) {
275 			uint32_t dirsperblk;
276 			uint64_t fileid64;
277 
278 			dirsperblk = pmp->pm_BytesPerSec /
279 			    sizeof(struct direntry);
280 
281 			fileid64 = (dep->de_dirclust == MSDOSFSROOT) ?
282 			    roottobn(pmp, 0) : cntobn(pmp, dep->de_dirclust);
283 			fileid64 *= dirsperblk;
284 			fileid64 += dep->de_diroffset / sizeof(struct direntry);
285 
286 			fileid = fileidhash(fileid64);
287 		}
288 	}
289 
290 	vap->va_fileid = fileid;
291 	vap->va_mode = (S_IRUSR|S_IRGRP|S_IROTH);
292 	if ((dep->de_Attributes & ATTR_READONLY) == 0)
293 		vap->va_mode |= (S_IWUSR|S_IWGRP|S_IWOTH);
294 	if (dep->de_Attributes & ATTR_DIRECTORY) {
295 		vap->va_mode |= S_IFDIR;
296 		vap->va_mode |= (vap->va_mode & S_IRUSR) ? S_IXUSR : 0;
297 		vap->va_mode |= (vap->va_mode & S_IRGRP) ? S_IXGRP : 0;
298 		vap->va_mode |= (vap->va_mode & S_IROTH) ? S_IXOTH : 0;
299 	}
300 	vap->va_mode &= dep->de_pmp->pm_mask;
301 	vap->va_nlink = 1;
302 	vap->va_gid = dep->de_pmp->pm_gid;
303 	vap->va_uid = dep->de_pmp->pm_uid;
304 	vap->va_rdev = 0;
305 	vap->va_size = dep->de_FileSize;
306 	dos2unixtime(dep->de_MDate, dep->de_MTime, 0, &vap->va_mtime);
307 	if (dep->de_pmp->pm_flags & MSDOSFSMNT_LONGNAME) {
308 		dos2unixtime(dep->de_ADate, 0, 0, &vap->va_atime);
309 		dos2unixtime(dep->de_CDate, dep->de_CTime, dep->de_CTimeHundredth, &vap->va_ctime);
310 	} else {
311 		vap->va_atime = vap->va_mtime;
312 		vap->va_ctime = vap->va_mtime;
313 	}
314 	vap->va_flags = 0;
315 	if ((dep->de_Attributes & ATTR_ARCHIVE) == 0)
316 		vap->va_flags |= SF_ARCHIVED;
317 	vap->va_gen = 0;
318 	vap->va_blocksize = dep->de_pmp->pm_bpcluster;
319 	vap->va_bytes = (dep->de_FileSize + dep->de_pmp->pm_crbomask) &
320 				~(dep->de_pmp->pm_crbomask);
321 	vap->va_type = ap->a_vp->v_type;
322 	return (0);
323 }
324 
325 int
326 msdosfs_setattr(void *v)
327 {
328 	struct vop_setattr_args *ap = v;
329 	struct vnode *vp = ap->a_vp;
330 	struct denode *dep = VTODE(ap->a_vp);
331 	struct msdosfsmount *pmp = dep->de_pmp;
332 	struct vattr *vap = ap->a_vap;
333 	struct ucred *cred = ap->a_cred;
334 	int error = 0;
335 
336 #ifdef MSDOSFS_DEBUG
337 	printf("msdosfs_setattr(): vp %p, vap %p, cred %p, p %p\n",
338 	    ap->a_vp, vap, cred, ap->a_p);
339 #endif
340 	if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
341 	    (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
342 	    (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
343 	    (vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
344 #ifdef MSDOSFS_DEBUG
345 		printf("msdosfs_setattr(): returning EINVAL\n");
346 		printf("    va_type %d, va_nlink %x, va_fsid %ld, "
347 		    "va_fileid %llx\n", vap->va_type, vap->va_nlink,
348 		    vap->va_fsid, (unsigned long long)vap->va_fileid);
349 		printf("    va_blocksize %lx, va_rdev %x, va_bytes %llx, "
350 		    "va_gen %lx\n", vap->va_blocksize,
351 		    (unsigned int)vap->va_rdev,
352 		    (unsigned long long)vap->va_bytes, vap->va_gen);
353 #endif
354 		return (EINVAL);
355 	}
356 	if (vap->va_flags != VNOVAL) {
357 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
358 			return (EINVAL);
359 		if (cred->cr_uid != pmp->pm_uid) {
360 			error = suser_ucred(cred);
361 			if (error)
362 				return (error);
363 		}
364 		/*
365 		 * We are very inconsistent about handling unsupported
366 		 * attributes.  We ignored the access time and the
367 		 * read and execute bits.  We were strict for the other
368 		 * attributes.
369 		 *
370 		 * Here we are strict, stricter than ufs in not allowing
371 		 * users to attempt to set SF_SETTABLE bits or anyone to
372 		 * set unsupported bits.  However, we ignore attempts to
373 		 * set ATTR_ARCHIVE for directories `cp -pr' from a more
374 		 * sensible filesystem attempts it a lot.
375 		 */
376 		if (vap->va_flags & SF_SETTABLE) {
377 			error = suser_ucred(cred);
378 			if (error)
379 				return (error);
380 		}
381 		if (vap->va_flags & ~SF_ARCHIVED)
382 			return EOPNOTSUPP;
383 		if (vap->va_flags & SF_ARCHIVED)
384 			dep->de_Attributes &= ~ATTR_ARCHIVE;
385 		else if (!(dep->de_Attributes & ATTR_DIRECTORY))
386 			dep->de_Attributes |= ATTR_ARCHIVE;
387 		dep->de_flag |= DE_MODIFIED;
388 	}
389 
390 	if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
391 		uid_t uid;
392 		gid_t gid;
393 
394 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
395 			return (EINVAL);
396 		uid = vap->va_uid;
397 		if (uid == (uid_t)VNOVAL)
398 			uid = pmp->pm_uid;
399 		gid = vap->va_gid;
400 		if (gid == (gid_t)VNOVAL)
401 			gid = pmp->pm_gid;
402 		if (cred->cr_uid != pmp->pm_uid || uid != pmp->pm_uid ||
403 		    (gid != pmp->pm_gid && !groupmember(gid, cred))) {
404 			error = suser_ucred(cred);
405 			if (error)
406 				return (error);
407 		}
408 		if (uid != pmp->pm_uid || gid != pmp->pm_gid)
409 			return EINVAL;
410 	}
411 
412 	if (vap->va_size != VNOVAL) {
413 		switch (vp->v_type) {
414 		case VDIR:
415 			return (EISDIR);
416 		case VREG:
417 			/*
418 			 * Truncation is only supported for regular files,
419 			 * Disallow it if the filesystem is read-only.
420 			 */
421 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
422 				return (EINVAL);
423 			break;
424 		default:
425 			/*
426 			 * According to POSIX, the result is unspecified
427 			 * for file types other than regular files,
428 			 * directories and shared memory objects.  We
429 			 * don't support any file types except regular
430 			 * files and directories in this file system, so
431 			 * this (default) case is unreachable and can do
432 			 * anything.  Keep falling through to detrunc()
433 			 * for now.
434 			 */
435 			break;
436 		}
437 		error = detrunc(dep, vap->va_size, 0, cred, ap->a_p);
438 		if (error)
439 			return error;
440 	}
441 	if ((vap->va_vaflags & VA_UTIMES_CHANGE) ||
442 	    vap->va_atime.tv_nsec != VNOVAL ||
443 	    vap->va_mtime.tv_nsec != VNOVAL) {
444 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
445 			return (EINVAL);
446 		if (cred->cr_uid != pmp->pm_uid &&
447 		    (error = suser_ucred(cred)) &&
448 		    ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
449 		    (error = VOP_ACCESS(ap->a_vp, VWRITE, cred, ap->a_p))))
450 			return (error);
451 		if (vp->v_type != VDIR) {
452 			if ((pmp->pm_flags & MSDOSFSMNT_NOWIN95) == 0 &&
453 			    vap->va_atime.tv_nsec != VNOVAL) {
454 				dep->de_flag &= ~DE_ACCESS;
455 				unix2dostime(&vap->va_atime, &dep->de_ADate,
456 				    NULL, NULL);
457 			}
458 			if (vap->va_mtime.tv_nsec != VNOVAL) {
459 				dep->de_flag &= ~DE_UPDATE;
460 				unix2dostime(&vap->va_mtime, &dep->de_MDate,
461 				    &dep->de_MTime, NULL);
462 			}
463 			dep->de_Attributes |= ATTR_ARCHIVE;
464 			dep->de_flag |= DE_MODIFIED;
465 		}
466 	}
467 	/*
468 	 * DOS files only have the ability to have their writability
469 	 * attribute set, so we use the owner write bit to set the readonly
470 	 * attribute.
471 	 */
472 	if (vap->va_mode != (mode_t)VNOVAL) {
473 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
474 			return (EINVAL);
475 		if (cred->cr_uid != pmp->pm_uid) {
476 			error = suser_ucred(cred);
477 			if (error)
478 				return (error);
479 		}
480 		if (vp->v_type != VDIR) {
481 			/* We ignore the read and execute bits. */
482 			if (vap->va_mode & VWRITE)
483 				dep->de_Attributes &= ~ATTR_READONLY;
484 			else
485 				dep->de_Attributes |= ATTR_READONLY;
486 			dep->de_Attributes |= ATTR_ARCHIVE;
487 			dep->de_flag |= DE_MODIFIED;
488 		}
489 	}
490 	VN_KNOTE(ap->a_vp, NOTE_ATTRIB);
491 	return (deupdat(dep, 1));
492 }
493 
494 int
495 msdosfs_read(void *v)
496 {
497 	struct vop_read_args *ap = v;
498 	struct vnode *vp = ap->a_vp;
499 	struct denode *dep = VTODE(vp);
500 	struct msdosfsmount *pmp = dep->de_pmp;
501 	struct uio *uio = ap->a_uio;
502 	int isadir, error = 0;
503 	uint32_t n, diff, size, on;
504 	struct buf *bp;
505 	uint32_t cn;
506 	daddr_t bn;
507 
508 	/*
509 	 * If they didn't ask for any data, then we are done.
510 	 */
511 	if (uio->uio_resid == 0)
512 		return (0);
513 	if (uio->uio_offset < 0)
514 		return (EINVAL);
515 
516 	isadir = dep->de_Attributes & ATTR_DIRECTORY;
517 	do {
518 		if (uio->uio_offset >= dep->de_FileSize)
519 			return (0);
520 
521 		cn = de_cluster(pmp, uio->uio_offset);
522 		size = pmp->pm_bpcluster;
523 		on = uio->uio_offset & pmp->pm_crbomask;
524 		n = ulmin(pmp->pm_bpcluster - on, uio->uio_resid);
525 
526 		/*
527 		 * de_FileSize is uint32_t, and we know that uio_offset <
528 		 * de_FileSize, so uio->uio_offset < 2^32.  Therefore
529 		 * the cast to uint32_t on the next line is safe.
530 		 */
531 		diff = dep->de_FileSize - (uint32_t)uio->uio_offset;
532 		if (diff < n)
533 			n = diff;
534 
535 		/*
536 		 * If we are operating on a directory file then be sure to
537 		 * do i/o with the vnode for the filesystem instead of the
538 		 * vnode for the directory.
539 		 */
540 		if (isadir) {
541 			/* convert cluster # to block # */
542 			error = pcbmap(dep, cn, &bn, 0, &size);
543 			if (error)
544 				return (error);
545 			error = bread(pmp->pm_devvp, bn, size, &bp);
546 		} else {
547 			if (de_cn2off(pmp, cn + 1) >= dep->de_FileSize)
548 				error = bread(vp, cn, size, &bp);
549 			else
550 				error = bread_cluster(vp, cn, size, &bp);
551 		}
552 		n = min(n, pmp->pm_bpcluster - bp->b_resid);
553 		if (error) {
554 			brelse(bp);
555 			return (error);
556 		}
557 		error = uiomove(bp->b_data + on, n, uio);
558 		brelse(bp);
559 	} while (error == 0 && uio->uio_resid > 0 && n != 0);
560 	if (!isadir && !(vp->v_mount->mnt_flag & MNT_NOATIME))
561 		dep->de_flag |= DE_ACCESS;
562 	return (error);
563 }
564 
565 /*
566  * Write data to a file or directory.
567  */
568 int
569 msdosfs_write(void *v)
570 {
571 	struct vop_write_args *ap = v;
572 	uint32_t n, croffset;
573 	size_t resid;
574 	ssize_t overrun;
575 	int extended = 0;
576 	uint32_t osize;
577 	int error = 0;
578 	uint32_t count, lastcn, cn;
579 	struct buf *bp;
580 	int ioflag = ap->a_ioflag;
581 	struct uio *uio = ap->a_uio;
582 	struct vnode *vp = ap->a_vp;
583 	struct vnode *thisvp;
584 	struct denode *dep = VTODE(vp);
585 	struct msdosfsmount *pmp = dep->de_pmp;
586 	struct ucred *cred = ap->a_cred;
587 
588 #ifdef MSDOSFS_DEBUG
589 	printf("msdosfs_write(vp %p, uio %p, ioflag %08x, cred %p\n",
590 	    vp, uio, ioflag, cred);
591 	printf("msdosfs_write(): diroff %d, dirclust %d, startcluster %d\n",
592 	    dep->de_diroffset, dep->de_dirclust, dep->de_StartCluster);
593 #endif
594 
595 	switch (vp->v_type) {
596 	case VREG:
597 		if (ioflag & IO_APPEND)
598 			uio->uio_offset = dep->de_FileSize;
599 		thisvp = vp;
600 		break;
601 	case VDIR:
602 		return EISDIR;
603 	default:
604 		panic("msdosfs_write(): bad file type");
605 	}
606 
607 	if (uio->uio_offset < 0)
608 		return (EINVAL);
609 
610 	if (uio->uio_resid == 0)
611 		return (0);
612 
613 	/* Don't bother to try to write files larger than the f/s limit */
614 	if (uio->uio_offset > MSDOSFS_FILESIZE_MAX ||
615 	    uio->uio_resid > (MSDOSFS_FILESIZE_MAX - uio->uio_offset))
616 		return (EFBIG);
617 
618 	/* do the filesize rlimit check */
619 	if ((error = vn_fsizechk(vp, uio, ioflag, &overrun)))
620 		return (error);
621 
622 	/*
623 	 * If the offset we are starting the write at is beyond the end of
624 	 * the file, then they've done a seek.  Unix filesystems allow
625 	 * files with holes in them, DOS doesn't so we must fill the hole
626 	 * with zeroed blocks.
627 	 */
628 	if (uio->uio_offset > dep->de_FileSize) {
629 		if ((error = deextend(dep, uio->uio_offset, cred)) != 0)
630 			goto out;
631 	}
632 
633 	/*
634 	 * Remember some values in case the write fails.
635 	 */
636 	resid = uio->uio_resid;
637 	osize = dep->de_FileSize;
638 
639 	/*
640 	 * If we write beyond the end of the file, extend it to its ultimate
641 	 * size ahead of the time to hopefully get a contiguous area.
642 	 */
643 	if (uio->uio_offset + resid > osize) {
644 		extended = 1;
645 		count = de_clcount(pmp, uio->uio_offset + resid) -
646 			de_clcount(pmp, osize);
647 		if ((error = extendfile(dep, count, NULL, NULL, 0)) &&
648 		    (error != ENOSPC || (ioflag & IO_UNIT)))
649 			goto errexit;
650 		lastcn = dep->de_fc[FC_LASTFC].fc_frcn;
651 	} else
652 		lastcn = de_clcount(pmp, osize) - 1;
653 
654 	do {
655 		croffset = uio->uio_offset & pmp->pm_crbomask;
656 		cn = de_cluster(pmp, uio->uio_offset);
657 
658 		if (cn > lastcn) {
659 			error = ENOSPC;
660 			break;
661 		}
662 
663 		if (croffset == 0 &&
664 		    (de_cluster(pmp, uio->uio_offset + uio->uio_resid) > cn ||
665 		     (uio->uio_offset + uio->uio_resid) >= dep->de_FileSize)) {
666 			/*
667 			 * If either the whole cluster gets written,
668 			 * or we write the cluster from its start beyond EOF,
669 			 * then no need to read data from disk.
670 			 */
671 			bp = getblk(thisvp, cn, pmp->pm_bpcluster, 0, INFSLP);
672 			clrbuf(bp);
673 			/*
674 			 * Do the bmap now, since pcbmap needs buffers
675 			 * for the fat table. (see msdosfs_strategy)
676 			 */
677 			if (bp->b_blkno == bp->b_lblkno) {
678 				error = pcbmap(dep, bp->b_lblkno, &bp->b_blkno, 0, 0);
679 				if (error)
680 					bp->b_blkno = -1;
681 			}
682 			if (bp->b_blkno == -1) {
683 				brelse(bp);
684 				if (!error)
685 					error = EIO;		/* XXX */
686 				break;
687 			}
688 		} else {
689 			/*
690 			 * The block we need to write into exists, so
691 			 * read it in.
692 			 */
693 			error = bread(thisvp, cn, pmp->pm_bpcluster, &bp);
694 			if (error) {
695 				brelse(bp);
696 				break;
697 			}
698 		}
699 
700 		n = ulmin(uio->uio_resid, pmp->pm_bpcluster - croffset);
701 		if (uio->uio_offset + n > dep->de_FileSize) {
702 			dep->de_FileSize = uio->uio_offset + n;
703 			uvm_vnp_setsize(vp, dep->de_FileSize);
704 		}
705 		uvm_vnp_uncache(vp);
706 		/*
707 		 * Should these vnode_pager_* functions be done on dir
708 		 * files?
709 		 */
710 
711 		/*
712 		 * Copy the data from user space into the buf header.
713 		 */
714 		error = uiomove(bp->b_data + croffset, n, uio);
715 
716 		/*
717 		 * If they want this synchronous then write it and wait for
718 		 * it.  Otherwise, if on a cluster boundary write it
719 		 * asynchronously so we can move on to the next block
720 		 * without delay.  Otherwise do a delayed write because we
721 		 * may want to write somemore into the block later.
722 		 */
723 #if 0
724 		if (ioflag & IO_NOCACHE)
725 			bp->b_flags |= B_NOCACHE;
726 #endif
727 		if (ioflag & IO_SYNC)
728 			(void) bwrite(bp);
729 		else if (n + croffset == pmp->pm_bpcluster)
730 			bawrite(bp);
731 		else
732 			bdwrite(bp);
733 		dep->de_flag |= DE_UPDATE;
734 	} while (error == 0 && uio->uio_resid > 0);
735 
736 	if (resid > uio->uio_resid)
737 		 VN_KNOTE(ap->a_vp, NOTE_WRITE | (extended ? NOTE_EXTEND : 0));
738 
739 	if (dep->de_FileSize < osize)
740 		VN_KNOTE(ap->a_vp, NOTE_TRUNCATE);
741 
742 	/*
743 	 * If the write failed and they want us to, truncate the file back
744 	 * to the size it was before the write was attempted.
745 	 */
746 errexit:
747 	if (error) {
748 		if (ioflag & IO_UNIT) {
749 			detrunc(dep, osize, ioflag & IO_SYNC, NOCRED, NULL);
750 			uio->uio_offset -= resid - uio->uio_resid;
751 			uio->uio_resid = resid;
752 		} else {
753 			detrunc(dep, dep->de_FileSize, ioflag & IO_SYNC, NOCRED, NULL);
754 			if (uio->uio_resid != resid)
755 				error = 0;
756 		}
757 	} else if (ioflag & IO_SYNC)
758 		error = deupdat(dep, 1);
759 
760 out:
761 	/* correct the result for writes clamped by vn_fsizechk() */
762 	uio->uio_resid += overrun;
763 	return (error);
764 }
765 
766 int
767 msdosfs_ioctl(void *v)
768 {
769 	return (ENOTTY);
770 }
771 
772 int
773 msdosfs_poll(void *v)
774 {
775 	struct vop_poll_args *ap = v;
776 
777 	return (ap->a_events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
778 }
779 
780 /*
781  * Flush the blocks of a file to disk.
782  *
783  * This function is worthless for vnodes that represent directories. Maybe we
784  * could just do a sync if they try an fsync on a directory file.
785  */
786 int
787 msdosfs_fsync(void *v)
788 {
789 	struct vop_fsync_args *ap = v;
790 	struct vnode *vp = ap->a_vp;
791 
792 	vflushbuf(vp, ap->a_waitfor == MNT_WAIT);
793 	return (deupdat(VTODE(vp), ap->a_waitfor == MNT_WAIT));
794 }
795 
796 /*
797  * Flush the blocks of a file to disk.
798  *
799  * This function is worthless for vnodes that represent directories. Maybe we
800  * could just do a sync if they try an fsync on a directory file.
801  */
802 int
803 msdosfs_remove(void *v)
804 {
805 	struct vop_remove_args *ap = v;
806 	struct denode *dep = VTODE(ap->a_vp);
807 	struct denode *ddep = VTODE(ap->a_dvp);
808 	int error;
809 
810 	if (ap->a_vp->v_type == VDIR)
811 		error = EPERM;
812 	else
813 		error = removede(ddep, dep);
814 
815 	VN_KNOTE(ap->a_vp, NOTE_DELETE);
816 	VN_KNOTE(ap->a_dvp, NOTE_WRITE);
817 
818 #ifdef MSDOSFS_DEBUG
819 	printf("msdosfs_remove(), dep %p, v_usecount %d\n", dep,
820 	    ap->a_vp->v_usecount);
821 #endif
822 	if (ddep == dep)
823 		vrele(ap->a_vp);
824 	else
825 		vput(ap->a_vp);	/* causes msdosfs_inactive() to be called
826 				 * via vrele() */
827 	vput(ap->a_dvp);
828 	return (error);
829 }
830 
831 /*
832  * DOS filesystems don't know what links are. But since we already called
833  * msdosfs_lookup() with create and lockparent, the parent is locked so we
834  * have to free it before we return the error.
835  */
836 int
837 msdosfs_link(void *v)
838 {
839 	struct vop_link_args *ap = v;
840 
841 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
842 	vput(ap->a_dvp);
843 	return (EOPNOTSUPP);
844 }
845 
846 /*
847  * Renames on files require moving the denode to a new hash queue since the
848  * denode's location is used to compute which hash queue to put the file
849  * in. Unless it is a rename in place.  For example "mv a b".
850  *
851  * What follows is the basic algorithm:
852  *
853  * if (file move) {
854  *	if (dest file exists) {
855  *		remove dest file
856  *	}
857  *	if (dest and src in same directory) {
858  *		rewrite name in existing directory slot
859  *	} else {
860  *		write new entry in dest directory
861  *		update offset and dirclust in denode
862  *		move denode to new hash chain
863  *		clear old directory entry
864  *	}
865  * } else {
866  *	directory move
867  *	if (dest directory exists) {
868  *		if (dest is not empty) {
869  *			return ENOTEMPTY
870  *		}
871  *		remove dest directory
872  *	}
873  *	if (dest and src in same directory) {
874  *		rewrite name in existing entry
875  *	} else {
876  *		be sure dest is not a child of src directory
877  *		write entry in dest directory
878  *		update "." and ".." in moved directory
879  *		update offset and dirclust in denode
880  *		move denode to new hash chain
881  *		clear old directory entry for moved directory
882  *	}
883  * }
884  *
885  * On entry:
886  *	source's parent directory is unlocked
887  *	source file or directory is unlocked
888  *	destination's parent directory is locked
889  *	destination file or directory is locked if it exists
890  *
891  * On exit:
892  *	all denodes should be released
893  *
894  * Notes:
895  * I'm not sure how the memory containing the pathnames pointed at by the
896  * componentname structures is freed, there may be some memory bleeding
897  * for each rename done.
898  */
899 int
900 msdosfs_rename(void *v)
901 {
902 	struct vop_rename_args *ap = v;
903 	struct vnode *tvp = ap->a_tvp;
904 	struct vnode *tdvp = ap->a_tdvp;
905 	struct vnode *fvp = ap->a_fvp;
906 	struct vnode *fdvp = ap->a_fdvp;
907 	struct componentname *tcnp = ap->a_tcnp;
908 	struct componentname *fcnp = ap->a_fcnp;
909 	struct denode *ip, *xp, *dp, *zp;
910 	u_char toname[11], oldname[11];
911 	uint32_t from_diroffset, to_diroffset;
912 	u_char to_count;
913 	int doingdirectory = 0, newparent = 0;
914 	int error;
915 	uint32_t cn, pcl;
916 	daddr_t bn;
917 	struct msdosfsmount *pmp;
918 	struct direntry *dotdotp;
919 	struct buf *bp;
920 
921 	pmp = VFSTOMSDOSFS(fdvp->v_mount);
922 
923 #ifdef DIAGNOSTIC
924 	if ((tcnp->cn_flags & HASBUF) == 0 ||
925 	    (fcnp->cn_flags & HASBUF) == 0)
926 		panic("msdosfs_rename: no name");
927 #endif
928 	/*
929 	 * Check for cross-device rename.
930 	 */
931 	if ((fvp->v_mount != tdvp->v_mount) ||
932 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
933 		error = EXDEV;
934 abortit:
935 		VOP_ABORTOP(tdvp, tcnp);
936 		if (tdvp == tvp)
937 			vrele(tdvp);
938 		else
939 			vput(tdvp);
940 		if (tvp)
941 			vput(tvp);
942 		VOP_ABORTOP(fdvp, fcnp);
943 		vrele(fdvp);
944 		vrele(fvp);
945 		return (error);
946 	}
947 
948 	/*
949 	 * If source and dest are the same, do nothing.
950 	 */
951 	if (tvp == fvp) {
952 		error = 0;
953 		goto abortit;
954 	}
955 
956 	/* */
957 	if ((error = vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY)) != 0)
958 		goto abortit;
959 	dp = VTODE(fdvp);
960 	ip = VTODE(fvp);
961 
962 	/*
963 	 * Be sure we are not renaming ".", "..", or an alias of ".". This
964 	 * leads to a crippled directory tree.  It's pretty tough to do a
965 	 * "ls" or "pwd" with the "." directory entry missing, and "cd .."
966 	 * doesn't work if the ".." entry is missing.
967 	 */
968 	if (ip->de_Attributes & ATTR_DIRECTORY) {
969 		/*
970 		 * Avoid ".", "..", and aliases of "." for obvious reasons.
971 		 */
972 		if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
973 		    dp == ip ||
974 		    (fcnp->cn_flags & ISDOTDOT) ||
975 		    (tcnp->cn_flags & ISDOTDOT) ||
976 		    (ip->de_flag & DE_RENAME)) {
977 			VOP_UNLOCK(fvp);
978 			error = EINVAL;
979 			goto abortit;
980 		}
981 		ip->de_flag |= DE_RENAME;
982 		doingdirectory++;
983 	}
984 	VN_KNOTE(fdvp, NOTE_WRITE);	/* XXX right place? */
985 
986 	/*
987 	 * When the target exists, both the directory
988 	 * and target vnodes are returned locked.
989 	 */
990 	dp = VTODE(tdvp);
991 	xp = tvp ? VTODE(tvp) : NULL;
992 	/*
993 	 * Remember direntry place to use for destination
994 	 */
995 	to_diroffset = dp->de_fndoffset;
996 	to_count = dp->de_fndcnt;
997 
998 	/*
999 	 * If ".." must be changed (ie the directory gets a new
1000 	 * parent) then the source directory must not be in the
1001 	 * directory hierarchy above the target, as this would
1002 	 * orphan everything below the source directory. Also
1003 	 * the user must have write permission in the source so
1004 	 * as to be able to change "..". We must repeat the call
1005 	 * to namei, as the parent directory is unlocked by the
1006 	 * call to doscheckpath().
1007 	 */
1008 	error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_proc);
1009 	VOP_UNLOCK(fvp);
1010 	if (VTODE(fdvp)->de_StartCluster != VTODE(tdvp)->de_StartCluster)
1011 		newparent = 1;
1012 	vrele(fdvp);
1013 	if (doingdirectory && newparent) {
1014 		if (error)	/* write access check above */
1015 			goto bad1;
1016 		if (xp != NULL)
1017 			vput(tvp);
1018 		/*
1019 		 * doscheckpath() vput()'s dp,
1020 		 * so we have to do a relookup afterwards
1021 		 */
1022 		if ((error = doscheckpath(ip, dp)) != 0)
1023 			goto out;
1024 		if ((tcnp->cn_flags & SAVESTART) == 0)
1025 			panic("msdosfs_rename: lost to startdir");
1026 		if ((error = vfs_relookup(tdvp, &tvp, tcnp)) != 0)
1027 			goto out;
1028 		dp = VTODE(tdvp);
1029 		xp = tvp ? VTODE(tvp) : NULL;
1030 	}
1031 
1032 	VN_KNOTE(tdvp, NOTE_WRITE);
1033 
1034 	if (xp != NULL) {
1035 		/*
1036 		 * Target must be empty if a directory and have no links
1037 		 * to it. Also, ensure source and target are compatible
1038 		 * (both directories, or both not directories).
1039 		 */
1040 		if (xp->de_Attributes & ATTR_DIRECTORY) {
1041 			if (!dosdirempty(xp)) {
1042 				error = ENOTEMPTY;
1043 				goto bad1;
1044 			}
1045 			if (!doingdirectory) {
1046 				error = ENOTDIR;
1047 				goto bad1;
1048 			}
1049 			cache_purge(tdvp);
1050 		} else if (doingdirectory) {
1051 			error = EISDIR;
1052 			goto bad1;
1053 		}
1054 		if ((error = removede(dp, xp)) != 0)
1055 			goto bad1;
1056 		VN_KNOTE(tvp, NOTE_DELETE);
1057 		vput(tvp);
1058 		xp = NULL;
1059 	}
1060 
1061 	/*
1062 	 * Convert the filename in tcnp into a dos filename. We copy this
1063 	 * into the denode and directory entry for the destination
1064 	 * file/directory.
1065 	 */
1066 	if ((error = uniqdosname(VTODE(tdvp), tcnp, toname)) != 0)
1067 		goto bad1;
1068 
1069 	/*
1070 	 * Since from wasn't locked at various places above,
1071 	 * have to do a relookup here.
1072 	 */
1073 	fcnp->cn_flags &= ~MODMASK;
1074 	fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
1075 	if ((fcnp->cn_flags & SAVESTART) == 0)
1076 		panic("msdosfs_rename: lost from startdir");
1077 	if (!newparent)
1078 		VOP_UNLOCK(tdvp);
1079 	(void) vfs_relookup(fdvp, &fvp, fcnp);
1080 	if (fvp == NULL) {
1081 		/*
1082 		 * From name has disappeared.
1083 		 */
1084 		if (doingdirectory)
1085 			panic("rename: lost dir entry");
1086 		vrele(ap->a_fvp);
1087 		if (newparent)
1088 			VOP_UNLOCK(tdvp);
1089 		vrele(tdvp);
1090 		return 0;
1091 	}
1092 	xp = VTODE(fvp);
1093 	zp = VTODE(fdvp);
1094 	from_diroffset = zp->de_fndoffset;
1095 
1096 	/*
1097 	 * Ensure that the directory entry still exists and has not
1098 	 * changed till now. If the source is a file the entry may
1099 	 * have been unlinked or renamed. In either case there is
1100 	 * no further work to be done. If the source is a directory
1101 	 * then it cannot have been rmdir'ed or renamed; this is
1102 	 * prohibited by the DE_RENAME flag.
1103 	 */
1104 	if (xp != ip) {
1105 		if (doingdirectory)
1106 			panic("rename: lost dir entry");
1107 		vrele(ap->a_fvp);
1108 		if (newparent)
1109 			VOP_UNLOCK(fdvp);
1110 		xp = NULL;
1111 	} else {
1112 		vrele(fvp);
1113 		xp = NULL;
1114 
1115 		/*
1116 		 * First write a new entry in the destination
1117 		 * directory and mark the entry in the source directory
1118 		 * as deleted.  Then move the denode to the correct hash
1119 		 * chain for its new location in the filesystem.  And, if
1120 		 * we moved a directory, then update its .. entry to point
1121 		 * to the new parent directory.
1122 		 */
1123 		bcopy(ip->de_Name, oldname, 11);
1124 		bcopy(toname, ip->de_Name, 11);	/* update denode */
1125 		dp->de_fndoffset = to_diroffset;
1126 		dp->de_fndcnt = to_count;
1127 		error = createde(ip, dp, NULL, tcnp);
1128 		if (error) {
1129 			bcopy(oldname, ip->de_Name, 11);
1130 			if (newparent)
1131 				VOP_UNLOCK(fdvp);
1132 			goto bad;
1133 		}
1134 		ip->de_refcnt++;
1135 		zp->de_fndoffset = from_diroffset;
1136 		if ((error = removede(zp, ip)) != 0) {
1137 			/* XXX should really panic here, fs is corrupt */
1138 			if (newparent)
1139 				VOP_UNLOCK(fdvp);
1140 			goto bad;
1141 		}
1142 
1143 		cache_purge(fvp);
1144 
1145 		if (!doingdirectory) {
1146 			error = pcbmap(dp, de_cluster(pmp, to_diroffset), 0,
1147 				       &ip->de_dirclust, 0);
1148 			if (error) {
1149 				/* XXX should really panic here, fs is corrupt */
1150 				if (newparent)
1151 					VOP_UNLOCK(fdvp);
1152 				goto bad;
1153 			}
1154 			ip->de_diroffset = to_diroffset;
1155 			if (ip->de_dirclust != MSDOSFSROOT)
1156 				ip->de_diroffset &= pmp->pm_crbomask;
1157 		}
1158 		reinsert(ip);
1159 		if (newparent)
1160 			VOP_UNLOCK(fdvp);
1161 	}
1162 
1163 	/*
1164 	 * If we moved a directory to a new parent directory, then we must
1165 	 * fixup the ".." entry in the moved directory.
1166 	 */
1167 	if (doingdirectory && newparent) {
1168 		cn = ip->de_StartCluster;
1169 		if (cn == MSDOSFSROOT) {
1170 			/* this should never happen */
1171 			panic("msdosfs_rename: updating .. in root directory?");
1172 		} else
1173 			bn = cntobn(pmp, cn);
1174 		error = bread(pmp->pm_devvp, bn, pmp->pm_bpcluster, &bp);
1175 		if (error) {
1176 			/* XXX should really panic here, fs is corrupt */
1177 			brelse(bp);
1178 			goto bad;
1179 		}
1180 		dotdotp = (struct direntry *)bp->b_data;
1181 		putushort(dotdotp[0].deStartCluster, cn);
1182 		pcl = dp->de_StartCluster;
1183 		if (FAT32(pmp) && pcl == pmp->pm_rootdirblk)
1184 			pcl = 0;
1185 		putushort(dotdotp[1].deStartCluster, pcl);
1186 		if (FAT32(pmp)) {
1187 			putushort(dotdotp[0].deHighClust, cn >> 16);
1188 			putushort(dotdotp[1].deHighClust, pcl >> 16);
1189 		}
1190 		if ((error = bwrite(bp)) != 0) {
1191 			/* XXX should really panic here, fs is corrupt */
1192 			goto bad;
1193 		}
1194 	}
1195 
1196 	VN_KNOTE(fvp, NOTE_RENAME);
1197 
1198 bad:
1199 	VOP_UNLOCK(fvp);
1200 	vrele(fdvp);
1201 bad1:
1202 	if (xp)
1203 		vput(tvp);
1204 	vput(tdvp);
1205 out:
1206 	ip->de_flag &= ~DE_RENAME;
1207 	vrele(fvp);
1208 	return (error);
1209 
1210 }
1211 
1212 struct {
1213 	struct direntry dot;
1214 	struct direntry dotdot;
1215 } dosdirtemplate = {
1216 	{	".       ", "   ",			/* the . entry */
1217 		ATTR_DIRECTORY,				/* file attribute */
1218 		CASE_LOWER_BASE | CASE_LOWER_EXT,	/* lower case */
1219 		0,					/* create time 100ths */
1220 		{ 0, 0 }, { 0, 0 },			/* create time & date */
1221 		{ 0, 0 },				/* access date */
1222 		{ 0, 0 },				/* high bits of start cluster */
1223 		{ 210, 4 }, { 210, 4 },			/* modify time & date */
1224 		{ 0, 0 },				/* startcluster */
1225 		{ 0, 0, 0, 0 }				/* filesize */
1226 	},
1227 	{	"..      ", "   ",			/* the .. entry */
1228 		ATTR_DIRECTORY,				/* file attribute */
1229 		CASE_LOWER_BASE | CASE_LOWER_EXT,	/* lower case */
1230 		0,					/* create time 100ths */
1231 		{ 0, 0 }, { 0, 0 },			/* create time & date */
1232 		{ 0, 0 },				/* access date */
1233 		{ 0, 0 },				/* high bits of start cluster */
1234 		{ 210, 4 }, { 210, 4 },			/* modify time & date */
1235 		{ 0, 0 },				/* startcluster */
1236 		{ 0, 0, 0, 0 }				/* filesize */
1237 	}
1238 };
1239 
1240 int
1241 msdosfs_mkdir(void *v)
1242 {
1243 	struct vop_mkdir_args *ap = v;
1244 	struct componentname *cnp = ap->a_cnp;
1245 	struct denode ndirent;
1246 	struct denode *dep;
1247 	struct denode *pdep = VTODE(ap->a_dvp);
1248 	int error;
1249 	daddr_t bn;
1250 	uint32_t newcluster, pcl;
1251 	struct direntry *denp;
1252 	struct msdosfsmount *pmp = pdep->de_pmp;
1253 	struct buf *bp;
1254 	struct timespec ts;
1255 
1256 	/*
1257 	 * If this is the root directory and there is no space left we
1258 	 * can't do anything.  This is because the root directory can not
1259 	 * change size.
1260 	 */
1261 	if (pdep->de_StartCluster == MSDOSFSROOT
1262 	    && pdep->de_fndoffset >= pdep->de_FileSize) {
1263 		error = ENOSPC;
1264 		goto bad2;
1265 	}
1266 
1267 	/*
1268 	 * Allocate a cluster to hold the about to be created directory.
1269 	 */
1270 	error = clusteralloc(pmp, 0, 1, &newcluster, NULL);
1271 	if (error)
1272 		goto bad2;
1273 
1274 	bzero(&ndirent, sizeof(ndirent));
1275 	ndirent.de_pmp = pmp;
1276 	ndirent.de_flag = DE_ACCESS | DE_CREATE | DE_UPDATE;
1277 	getnanotime(&ts);
1278 	DETIMES(&ndirent, &ts, &ts, &ts);
1279 
1280 	/*
1281 	 * Now fill the cluster with the "." and ".." entries. And write
1282 	 * the cluster to disk.  This way it is there for the parent
1283 	 * directory to be pointing at if there were a crash.
1284 	 */
1285 	bn = cntobn(pmp, newcluster);
1286 	/* always succeeds */
1287 	bp = getblk(pmp->pm_devvp, bn, pmp->pm_bpcluster, 0, INFSLP);
1288 	bzero(bp->b_data, pmp->pm_bpcluster);
1289 	bcopy(&dosdirtemplate, bp->b_data, sizeof dosdirtemplate);
1290 	denp = (struct direntry *)bp->b_data;
1291 	putushort(denp[0].deStartCluster, newcluster);
1292 	putushort(denp[0].deCDate, ndirent.de_CDate);
1293 	putushort(denp[0].deCTime, ndirent.de_CTime);
1294 	denp[0].deCTimeHundredth = ndirent.de_CTimeHundredth;
1295 	putushort(denp[0].deADate, ndirent.de_ADate);
1296 	putushort(denp[0].deMDate, ndirent.de_MDate);
1297 	putushort(denp[0].deMTime, ndirent.de_MTime);
1298 	pcl = pdep->de_StartCluster;
1299 	if (FAT32(pmp) && pcl == pmp->pm_rootdirblk)
1300 		pcl = 0;
1301 	putushort(denp[1].deStartCluster, pcl);
1302 	putushort(denp[1].deCDate, ndirent.de_CDate);
1303 	putushort(denp[1].deCTime, ndirent.de_CTime);
1304 	denp[1].deCTimeHundredth = ndirent.de_CTimeHundredth;
1305 	putushort(denp[1].deADate, ndirent.de_ADate);
1306 	putushort(denp[1].deMDate, ndirent.de_MDate);
1307 	putushort(denp[1].deMTime, ndirent.de_MTime);
1308 	if (FAT32(pmp)) {
1309 		putushort(denp[0].deHighClust, newcluster >> 16);
1310 		putushort(denp[1].deHighClust, pdep->de_StartCluster >> 16);
1311 	}
1312 
1313 	if ((error = bwrite(bp)) != 0)
1314 		goto bad;
1315 
1316 	/*
1317 	 * Now build up a directory entry pointing to the newly allocated
1318 	 * cluster.  This will be written to an empty slot in the parent
1319 	 * directory.
1320 	 */
1321 #ifdef DIAGNOSTIC
1322 	if ((cnp->cn_flags & HASBUF) == 0)
1323 		panic("msdosfs_mkdir: no name");
1324 #endif
1325 	if ((error = uniqdosname(pdep, cnp, ndirent.de_Name)) != 0)
1326 		goto bad;
1327 
1328 	ndirent.de_Attributes = ATTR_DIRECTORY;
1329 	ndirent.de_StartCluster = newcluster;
1330 	ndirent.de_FileSize = 0;
1331 	ndirent.de_dev = pdep->de_dev;
1332 	ndirent.de_devvp = pdep->de_devvp;
1333 	if ((error = createde(&ndirent, pdep, &dep, cnp)) != 0)
1334 		goto bad;
1335 	if ((cnp->cn_flags & SAVESTART) == 0)
1336 		pool_put(&namei_pool, cnp->cn_pnbuf);
1337 	VN_KNOTE(ap->a_dvp, NOTE_WRITE | NOTE_LINK);
1338 	vput(ap->a_dvp);
1339 	*ap->a_vpp = DETOV(dep);
1340 	return (0);
1341 
1342 bad:
1343 	clusterfree(pmp, newcluster, NULL);
1344 bad2:
1345 	pool_put(&namei_pool, cnp->cn_pnbuf);
1346 	vput(ap->a_dvp);
1347 	return (error);
1348 }
1349 
1350 int
1351 msdosfs_rmdir(void *v)
1352 {
1353 	struct vop_rmdir_args *ap = v;
1354 	struct vnode *vp = ap->a_vp;
1355 	struct vnode *dvp = ap->a_dvp;
1356 	struct componentname *cnp = ap->a_cnp;
1357 	struct denode *ip, *dp;
1358 	int error;
1359 
1360 	ip = VTODE(vp);
1361 	dp = VTODE(dvp);
1362 	/*
1363 	 * Verify the directory is empty (and valid).
1364 	 * (Rmdir ".." won't be valid since
1365 	 *  ".." will contain a reference to
1366 	 *  the current directory and thus be
1367 	 *  non-empty.)
1368 	 */
1369 	error = 0;
1370 	if (!dosdirempty(ip) || ip->de_flag & DE_RENAME) {
1371 		error = ENOTEMPTY;
1372 		goto out;
1373 	}
1374 
1375 	VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
1376 
1377 	/*
1378 	 * Delete the entry from the directory.  For dos filesystems this
1379 	 * gets rid of the directory entry on disk, the in memory copy
1380 	 * still exists but the de_refcnt is <= 0.  This prevents it from
1381 	 * being found by deget().  When the vput() on dep is done we give
1382 	 * up access and eventually msdosfs_reclaim() will be called which
1383 	 * will remove it from the denode cache.
1384 	 */
1385 	if ((error = removede(dp, ip)) != 0)
1386 		goto out;
1387 	/*
1388 	 * This is where we decrement the link count in the parent
1389 	 * directory.  Since dos filesystems don't do this we just purge
1390 	 * the name cache and let go of the parent directory denode.
1391 	 */
1392 	cache_purge(dvp);
1393 	vput(dvp);
1394 	dvp = NULL;
1395 	/*
1396 	 * Truncate the directory that is being deleted.
1397 	 */
1398 	error = detrunc(ip, (uint32_t)0, IO_SYNC, cnp->cn_cred, cnp->cn_proc);
1399 	cache_purge(vp);
1400 out:
1401 	if (dvp)
1402 		vput(dvp);
1403 	VN_KNOTE(vp, NOTE_DELETE);
1404 	vput(vp);
1405 	return (error);
1406 }
1407 
1408 /*
1409  * DOS filesystems don't know what symlinks are.
1410  */
1411 int
1412 msdosfs_symlink(void *v)
1413 {
1414 	struct vop_symlink_args *ap = v;
1415 
1416 	VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
1417 	vput(ap->a_dvp);
1418 	return (EOPNOTSUPP);
1419 }
1420 
1421 int
1422 msdosfs_readdir(void *v)
1423 {
1424 	struct vop_readdir_args *ap = v;
1425 	int error = 0;
1426 	int diff;
1427 	long n;
1428 	int blsize;
1429 	long on;
1430 	long lost;
1431 	long count;
1432 	uint32_t dirsperblk;
1433 	uint32_t cn, lbn;
1434 	uint32_t fileno;
1435 	long bias = 0;
1436 	daddr_t bn;
1437 	struct buf *bp;
1438 	struct denode *dep = VTODE(ap->a_vp);
1439 	struct msdosfsmount *pmp = dep->de_pmp;
1440 	struct direntry *dentp;
1441 	struct dirent dirbuf;
1442 	struct uio *uio = ap->a_uio;
1443 	off_t offset, wlast = -1;
1444 	int chksum = -1;
1445 
1446 #ifdef MSDOSFS_DEBUG
1447 	printf("msdosfs_readdir(): vp %p, uio %p, cred %p, eofflagp %p\n",
1448 	    ap->a_vp, uio, ap->a_cred, ap->a_eofflag);
1449 #endif
1450 
1451 	/*
1452 	 * msdosfs_readdir() won't operate properly on regular files since
1453 	 * it does i/o only with the filesystem vnode, and hence can
1454 	 * retrieve the wrong block from the buffer cache for a plain file.
1455 	 * So, fail attempts to readdir() on a plain file.
1456 	 */
1457 	if ((dep->de_Attributes & ATTR_DIRECTORY) == 0)
1458 		return (ENOTDIR);
1459 
1460 	/*
1461 	 * To be safe, initialize dirbuf
1462 	 */
1463 	bzero(&dirbuf, sizeof(dirbuf));
1464 
1465 	/*
1466 	 * If the user buffer is smaller than the size of one dos directory
1467 	 * entry or the file offset is not a multiple of the size of a
1468 	 * directory entry, then we fail the read.
1469 	 */
1470 	count = uio->uio_resid & ~(sizeof(struct direntry) - 1);
1471 	offset = uio->uio_offset;
1472 	if (count < sizeof(struct direntry) ||
1473 	    (offset & (sizeof(struct direntry) - 1)))
1474 		return (EINVAL);
1475 	lost = uio->uio_resid - count;
1476 	uio->uio_resid = count;
1477 
1478 	dirsperblk = pmp->pm_BytesPerSec / sizeof(struct direntry);
1479 
1480 	/*
1481 	 * If they are reading from the root directory then, we simulate
1482 	 * the . and .. entries since these don't exist in the root
1483 	 * directory.  We also set the offset bias to make up for having to
1484 	 * simulate these entries. By this I mean that at file offset 64 we
1485 	 * read the first entry in the root directory that lives on disk.
1486 	 */
1487 	if (dep->de_StartCluster == MSDOSFSROOT
1488 	    || (FAT32(pmp) && dep->de_StartCluster == pmp->pm_rootdirblk)) {
1489 #if 0
1490 		printf("msdosfs_readdir(): going after . or .. in root dir, offset %d\n",
1491 		    offset);
1492 #endif
1493 		bias = 2 * sizeof(struct direntry);
1494 		if (offset < bias) {
1495 			for (n = (int)offset / sizeof(struct direntry);
1496 			     n < 2; n++) {
1497 			        if (FAT32(pmp))
1498 				        dirbuf.d_fileno = pmp->pm_rootdirblk;
1499 				else
1500 				        dirbuf.d_fileno = 1;
1501 				dirbuf.d_type = DT_DIR;
1502 				switch (n) {
1503 				case 0:
1504 					dirbuf.d_namlen = 1;
1505 					strlcpy(dirbuf.d_name, ".",
1506 					    sizeof dirbuf.d_name);
1507 					break;
1508 				case 1:
1509 					dirbuf.d_namlen = 2;
1510 					strlcpy(dirbuf.d_name, "..",
1511 					    sizeof dirbuf.d_name);
1512 					break;
1513 				}
1514 				dirbuf.d_reclen = DIRENT_SIZE(&dirbuf);
1515 				dirbuf.d_off = offset +
1516 				    sizeof(struct direntry);
1517 				if (uio->uio_resid < dirbuf.d_reclen)
1518 					goto out;
1519 				error = uiomove(&dirbuf, dirbuf.d_reclen, uio);
1520 				if (error)
1521 					goto out;
1522 				offset = dirbuf.d_off;
1523 			}
1524 		}
1525 	}
1526 
1527 	while (uio->uio_resid > 0) {
1528 		lbn = de_cluster(pmp, offset - bias);
1529 		on = (offset - bias) & pmp->pm_crbomask;
1530 		n = min(pmp->pm_bpcluster - on, uio->uio_resid);
1531 		diff = dep->de_FileSize - (offset - bias);
1532 		if (diff <= 0)
1533 			break;
1534 		n = min(n, diff);
1535 		if ((error = pcbmap(dep, lbn, &bn, &cn, &blsize)) != 0)
1536 			break;
1537 		error = bread(pmp->pm_devvp, bn, blsize, &bp);
1538 		if (error) {
1539 			brelse(bp);
1540 			return (error);
1541 		}
1542 		n = min(n, blsize - bp->b_resid);
1543 
1544 		/*
1545 		 * Convert from dos directory entries to fs-independent
1546 		 * directory entries.
1547 		 */
1548 		for (dentp = (struct direntry *)(bp->b_data + on);
1549 		     (char *)dentp < bp->b_data + on + n;
1550 		     dentp++, offset += sizeof(struct direntry)) {
1551 #if 0
1552 			printf("rd: dentp %08x prev %08x crnt %08x deName %02x attr %02x\n",
1553 			    dentp, prev, crnt, dentp->deName[0], dentp->deAttributes);
1554 #endif
1555 			/*
1556 			 * If this is an unused entry, we can stop.
1557 			 */
1558 			if (dentp->deName[0] == SLOT_EMPTY) {
1559 				brelse(bp);
1560 				goto out;
1561 			}
1562 			/*
1563 			 * Skip deleted entries.
1564 			 */
1565 			if (dentp->deName[0] == SLOT_DELETED) {
1566 				chksum = -1;
1567 				wlast = -1;
1568 				continue;
1569 			}
1570 
1571 			/*
1572 			 * Handle Win95 long directory entries
1573 			 */
1574 			if (dentp->deAttributes == ATTR_WIN95) {
1575 				struct winentry *wep;
1576 				if (pmp->pm_flags & MSDOSFSMNT_SHORTNAME)
1577 					continue;
1578 				wep = (struct winentry *)dentp;
1579 				chksum = win2unixfn(wep, &dirbuf, chksum);
1580 				if (wep->weCnt & WIN_LAST)
1581 					wlast = offset;
1582 				continue;
1583 			}
1584 
1585 			/*
1586 			 * Skip volume labels
1587 			 */
1588 			if (dentp->deAttributes & ATTR_VOLUME) {
1589 				chksum = -1;
1590 				wlast = -1;
1591 				continue;
1592 			}
1593 
1594 			/*
1595 			 * This computation of d_fileno must match
1596 			 * the computation of va_fileid in
1597 			 * msdosfs_getattr.
1598 			 */
1599 			fileno = getushort(dentp->deStartCluster);
1600 			if (FAT32(pmp))
1601 			    fileno |= getushort(dentp->deHighClust) << 16;
1602 
1603 			if (dentp->deAttributes & ATTR_DIRECTORY) {
1604 				/* Special-case root */
1605 				if (fileno == MSDOSFSROOT)  {
1606 					fileno = FAT32(pmp) ?
1607 					    pmp->pm_rootdirblk : 1;
1608 				}
1609 
1610 				dirbuf.d_fileno = fileno;
1611 				dirbuf.d_type = DT_DIR;
1612 			} else {
1613 				if (getulong(dentp->deFileSize) == 0) {
1614 					uint64_t fileno64;
1615 
1616 					fileno64 = (cn == MSDOSFSROOT) ?
1617 					    roottobn(pmp, 0) : cntobn(pmp, cn);
1618 
1619 					fileno64 *= dirsperblk;
1620 					fileno64 += dentp -
1621 					    (struct direntry *)bp->b_data;
1622 
1623 					fileno = fileidhash(fileno64);
1624 				}
1625 
1626 				dirbuf.d_fileno = fileno;
1627 				dirbuf.d_type = DT_REG;
1628 			}
1629 
1630 			if (chksum != winChksum(dentp->deName))
1631 				dirbuf.d_namlen = dos2unixfn(dentp->deName,
1632 				    (u_char *)dirbuf.d_name,
1633 				    pmp->pm_flags & MSDOSFSMNT_SHORTNAME);
1634 			else
1635 				dirbuf.d_name[dirbuf.d_namlen] = 0;
1636 			chksum = -1;
1637 			dirbuf.d_reclen = DIRENT_SIZE(&dirbuf);
1638 			dirbuf.d_off = offset + sizeof(struct direntry);
1639 			if (uio->uio_resid < dirbuf.d_reclen) {
1640 				brelse(bp);
1641 				/* Remember long-name offset. */
1642 				if (wlast != -1)
1643 					offset = wlast;
1644 				goto out;
1645 			}
1646 			wlast = -1;
1647 			error = uiomove(&dirbuf, dirbuf.d_reclen, uio);
1648 			if (error) {
1649 				brelse(bp);
1650 				goto out;
1651 			}
1652 		}
1653 		brelse(bp);
1654 	}
1655 
1656 out:
1657 	uio->uio_offset = offset;
1658 	uio->uio_resid += lost;
1659 	if (dep->de_FileSize - (offset - bias) <= 0)
1660 		*ap->a_eofflag = 1;
1661 	else
1662 		*ap->a_eofflag = 0;
1663 	return (error);
1664 }
1665 
1666 /*
1667  * DOS filesystems don't know what symlinks are.
1668  */
1669 int
1670 msdosfs_readlink(void *v)
1671 {
1672 #if 0
1673 	struct vop_readlink_args /* {
1674 		struct vnode *a_vp;
1675 		struct uio *a_uio;
1676 		struct ucred *a_cred;
1677 	} */ *ap;
1678 #endif
1679 
1680 	return (EINVAL);
1681 }
1682 
1683 int
1684 msdosfs_lock(void *v)
1685 {
1686 	struct vop_lock_args *ap = v;
1687 	struct vnode *vp = ap->a_vp;
1688 
1689 	return rrw_enter(&VTODE(vp)->de_lock, ap->a_flags & LK_RWFLAGS);
1690 }
1691 
1692 int
1693 msdosfs_unlock(void *v)
1694 {
1695 	struct vop_unlock_args *ap = v;
1696 	struct vnode *vp = ap->a_vp;
1697 
1698 	rrw_exit(&VTODE(vp)->de_lock);
1699 	return 0;
1700 }
1701 
1702 int
1703 msdosfs_islocked(void *v)
1704 {
1705 	struct vop_islocked_args *ap = v;
1706 
1707 	return rrw_status(&VTODE(ap->a_vp)->de_lock);
1708 }
1709 
1710 /*
1711  * vp  - address of vnode file the file
1712  * bn  - which cluster we are interested in mapping to a filesystem block number
1713  * vpp - returns the vnode for the block special file holding the filesystem
1714  *	 containing the file of interest
1715  * bnp - address of where to return the filesystem relative block number
1716  */
1717 int
1718 msdosfs_bmap(void *v)
1719 {
1720 	struct vop_bmap_args *ap = v;
1721 	struct denode *dep = VTODE(ap->a_vp);
1722 	uint32_t cn;
1723 
1724 	if (ap->a_vpp != NULL)
1725 		*ap->a_vpp = dep->de_devvp;
1726 	if (ap->a_bnp == NULL)
1727 		return (0);
1728 
1729 	cn = ap->a_bn;
1730 	if (cn != ap->a_bn)
1731 		return (EFBIG);
1732 
1733 	return (msdosfs_bmaparray(ap->a_vp, cn, ap->a_bnp, ap->a_runp));
1734 }
1735 
1736 int
1737 msdosfs_bmaparray(struct vnode *vp, uint32_t cn, daddr_t *bnp, int *runp)
1738 {
1739 	struct denode *dep = VTODE(vp);
1740 	struct msdosfsmount *pmp = dep->de_pmp;
1741 	struct mount *mp;
1742 	int error, maxrun = 0, run;
1743 	daddr_t runbn;
1744 
1745 	mp = vp->v_mount;
1746 
1747 	if (runp) {
1748 		/*
1749 		 * XXX
1750 		 * If MAXBSIZE is the largest transfer the disks can handle,
1751 		 * we probably want maxrun to be 1 block less so that we
1752 		 * don't create a block larger than the device can handle.
1753 		 */
1754 		*runp = 0;
1755 		maxrun = min(MAXBSIZE / mp->mnt_stat.f_iosize - 1,
1756 		    pmp->pm_maxcluster - cn);
1757 	}
1758 
1759 	if ((error = pcbmap(dep, cn, bnp, 0, 0)) != 0)
1760 		return (error);
1761 
1762 	for (run = 1; run <= maxrun; run++) {
1763 		error = pcbmap(dep, cn + run, &runbn, 0, 0);
1764 		if (error != 0 || (runbn != *bnp + de_cn2bn(pmp, run)))
1765 			break;
1766 	}
1767 
1768 	if (runp)
1769 		*runp = run - 1;
1770 
1771 	return (0);
1772 }
1773 
1774 int
1775 msdosfs_strategy(void *v)
1776 {
1777 	struct vop_strategy_args *ap = v;
1778 	struct buf *bp = ap->a_bp;
1779 	struct denode *dep = VTODE(bp->b_vp);
1780 	struct vnode *vp;
1781 	int error = 0;
1782 	int s;
1783 
1784 	if (bp->b_vp->v_type == VBLK || bp->b_vp->v_type == VCHR)
1785 		panic("msdosfs_strategy: spec");
1786 	/*
1787 	 * If we don't already know the filesystem relative block number
1788 	 * then get it using pcbmap().  If pcbmap() returns the block
1789 	 * number as -1 then we've got a hole in the file.  DOS filesystems
1790 	 * don't allow files with holes, so we shouldn't ever see this.
1791 	 */
1792 	if (bp->b_blkno == bp->b_lblkno) {
1793 		error = pcbmap(dep, bp->b_lblkno, &bp->b_blkno, 0, 0);
1794 		if (error)
1795 			bp->b_blkno = -1;
1796 		if (bp->b_blkno == -1)
1797 			clrbuf(bp);
1798 	}
1799 	if (bp->b_blkno == -1) {
1800 		s = splbio();
1801 		biodone(bp);
1802 		splx(s);
1803 		return (error);
1804 	}
1805 
1806 	/*
1807 	 * Read/write the block from/to the disk that contains the desired
1808 	 * file block.
1809 	 */
1810 
1811 	vp = dep->de_devvp;
1812 	bp->b_dev = vp->v_rdev;
1813 	VOP_STRATEGY(vp, bp);
1814 	return (0);
1815 }
1816 
1817 int
1818 msdosfs_print(void *v)
1819 {
1820 	struct vop_print_args *ap = v;
1821 	struct denode *dep = VTODE(ap->a_vp);
1822 
1823 	printf(
1824 	    "tag VT_MSDOSFS, startcluster %u, dircluster %u, diroffset %u ",
1825 	    dep->de_StartCluster, dep->de_dirclust, dep->de_diroffset);
1826 	printf(" dev %d, %d, %s\n",
1827 	    major(dep->de_dev), minor(dep->de_dev),
1828 	    VOP_ISLOCKED(ap->a_vp) ? "(LOCKED)" : "");
1829 #ifdef DIAGNOSTIC
1830 	printf("\n");
1831 #endif
1832 
1833 	return (0);
1834 }
1835 
1836 int
1837 msdosfs_advlock(void *v)
1838 {
1839 	struct vop_advlock_args *ap = v;
1840 	struct denode *dep = VTODE(ap->a_vp);
1841 
1842 	return (lf_advlock(&dep->de_lockf, dep->de_FileSize, ap->a_id, ap->a_op,
1843 	    ap->a_fl, ap->a_flags));
1844 }
1845 
1846 int
1847 msdosfs_pathconf(void *v)
1848 {
1849 	struct vop_pathconf_args *ap = v;
1850 	struct msdosfsmount *pmp = VTODE(ap->a_vp)->de_pmp;
1851 	int error = 0;
1852 
1853 	switch (ap->a_name) {
1854 	case _PC_LINK_MAX:
1855 		*ap->a_retval = 1;
1856 		break;
1857 	case _PC_NAME_MAX:
1858 		*ap->a_retval = pmp->pm_flags & MSDOSFSMNT_LONGNAME ? WIN_MAXLEN : 12;
1859 		break;
1860 	case _PC_CHOWN_RESTRICTED:
1861 		*ap->a_retval = 1;
1862 		break;
1863 	case _PC_NO_TRUNC:
1864 		*ap->a_retval = 0;
1865 		break;
1866 	case _PC_TIMESTAMP_RESOLUTION:
1867 		*ap->a_retval = 2000000000;	/* 2 billion nanoseconds */
1868 		break;
1869 	default:
1870 		error = EINVAL;
1871 		break;
1872 	}
1873 
1874 	return (error);
1875 }
1876 
1877 /*
1878  * Thomas Wang's hash function, severely hacked to always set the high
1879  * bit on the number it returns (so no longer a proper hash function).
1880  */
1881 static uint32_t
1882 fileidhash(uint64_t fileid)
1883 {
1884 	uint64_t c1 = 0x6e5ea73858134343LL;
1885 	uint64_t c2 = 0xb34e8f99a2ec9ef5LL;
1886 
1887 	/*
1888 	 * We now have the original fileid value, as 64-bit value.
1889 	 * We need to reduce it to 32-bits, with the top bit set.
1890 	 */
1891 	fileid ^= ((c1 ^ fileid) >> 32);
1892 	fileid *= c1;
1893 	fileid ^= ((c2 ^ fileid) >> 31);
1894 	fileid *= c2;
1895 	fileid ^= ((c1 ^ fileid) >> 32);
1896 
1897 	return (uint32_t)(fileid | 0x80000000);
1898 }
1899 
1900 /* Global vfs data structures for msdosfs */
1901 const struct vops msdosfs_vops = {
1902 	.vop_lookup	= msdosfs_lookup,
1903 	.vop_create	= msdosfs_create,
1904 	.vop_mknod	= msdosfs_mknod,
1905 	.vop_open	= msdosfs_open,
1906 	.vop_close	= msdosfs_close,
1907 	.vop_access	= msdosfs_access,
1908 	.vop_getattr	= msdosfs_getattr,
1909 	.vop_setattr	= msdosfs_setattr,
1910 	.vop_read	= msdosfs_read,
1911 	.vop_write	= msdosfs_write,
1912 	.vop_ioctl	= msdosfs_ioctl,
1913 	.vop_poll	= msdosfs_poll,
1914 	.vop_kqfilter	= msdosfs_kqfilter,
1915 	.vop_fsync	= msdosfs_fsync,
1916 	.vop_remove	= msdosfs_remove,
1917 	.vop_link	= msdosfs_link,
1918 	.vop_rename	= msdosfs_rename,
1919 	.vop_mkdir	= msdosfs_mkdir,
1920 	.vop_rmdir	= msdosfs_rmdir,
1921 	.vop_symlink	= msdosfs_symlink,
1922 	.vop_readdir	= msdosfs_readdir,
1923 	.vop_readlink	= msdosfs_readlink,
1924 	.vop_abortop	= vop_generic_abortop,
1925 	.vop_inactive	= msdosfs_inactive,
1926 	.vop_reclaim	= msdosfs_reclaim,
1927 	.vop_lock	= msdosfs_lock,
1928 	.vop_unlock	= msdosfs_unlock,
1929 	.vop_bmap	= msdosfs_bmap,
1930 	.vop_strategy	= msdosfs_strategy,
1931 	.vop_print	= msdosfs_print,
1932 	.vop_islocked	= msdosfs_islocked,
1933 	.vop_pathconf	= msdosfs_pathconf,
1934 	.vop_advlock	= msdosfs_advlock,
1935 	.vop_bwrite	= vop_generic_bwrite,
1936 	.vop_revoke	= vop_generic_revoke,
1937 };
1938 
1939 const struct filterops msdosfsread_filtops = {
1940 	.f_flags	= FILTEROP_ISFD,
1941 	.f_attach	= NULL,
1942 	.f_detach	= filt_msdosfsdetach,
1943 	.f_event	= filt_msdosfsread,
1944 };
1945 
1946 const struct filterops msdosfswrite_filtops = {
1947 	.f_flags	= FILTEROP_ISFD,
1948 	.f_attach	= NULL,
1949 	.f_detach	= filt_msdosfsdetach,
1950 	.f_event	= filt_msdosfswrite,
1951 };
1952 
1953 const struct filterops msdosfsvnode_filtops = {
1954 	.f_flags	= FILTEROP_ISFD,
1955 	.f_attach	= NULL,
1956 	.f_detach	= filt_msdosfsdetach,
1957 	.f_event	= filt_msdosfsvnode,
1958 };
1959 
1960 int
1961 msdosfs_kqfilter(void *v)
1962 {
1963 	struct vop_kqfilter_args *ap = v;
1964 	struct vnode *vp = ap->a_vp;
1965 	struct knote *kn = ap->a_kn;
1966 
1967 	switch (kn->kn_filter) {
1968 	case EVFILT_READ:
1969 		kn->kn_fop = &msdosfsread_filtops;
1970 		break;
1971 	case EVFILT_WRITE:
1972 		kn->kn_fop = &msdosfswrite_filtops;
1973 		break;
1974 	case EVFILT_VNODE:
1975 		kn->kn_fop = &msdosfsvnode_filtops;
1976 		break;
1977 	default:
1978 		return (EINVAL);
1979 	}
1980 
1981 	kn->kn_hook = (caddr_t)vp;
1982 
1983 	klist_insert_locked(&vp->v_selectinfo.si_note, kn);
1984 
1985 	return (0);
1986 }
1987 
1988 void
1989 filt_msdosfsdetach(struct knote *kn)
1990 {
1991 	struct vnode *vp = (struct vnode *)kn->kn_hook;
1992 
1993 	klist_remove_locked(&vp->v_selectinfo.si_note, kn);
1994 }
1995 
1996 int
1997 filt_msdosfsread(struct knote *kn, long hint)
1998 {
1999 	struct vnode *vp = (struct vnode *)kn->kn_hook;
2000 	struct denode *dep = VTODE(vp);
2001 
2002 	/*
2003 	 * filesystem is gone, so set the EOF flag and schedule
2004 	 * the knote for deletion.
2005 	 */
2006 	if (hint == NOTE_REVOKE) {
2007 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
2008 		return (1);
2009 	}
2010 
2011 	kn->kn_data = dep->de_FileSize - foffset(kn->kn_fp);
2012 	if (kn->kn_data == 0 && kn->kn_sfflags & NOTE_EOF) {
2013 		kn->kn_fflags |= NOTE_EOF;
2014 		return (1);
2015 	}
2016 
2017 	if (kn->kn_flags & (__EV_POLL | __EV_SELECT))
2018 		return (1);
2019 
2020 	return (kn->kn_data != 0);
2021 }
2022 
2023 int
2024 filt_msdosfswrite(struct knote *kn, long hint)
2025 {
2026 	/*
2027 	 * filesystem is gone, so set the EOF flag and schedule
2028 	 * the knote for deletion.
2029 	 */
2030 	if (hint == NOTE_REVOKE) {
2031 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
2032 		return (1);
2033 	}
2034 
2035 	kn->kn_data = 0;
2036 	return (1);
2037 }
2038 
2039 int
2040 filt_msdosfsvnode(struct knote *kn, long hint)
2041 {
2042 	if (kn->kn_sfflags & hint)
2043 		kn->kn_fflags |= hint;
2044 	if (hint == NOTE_REVOKE) {
2045 		kn->kn_flags |= EV_EOF;
2046 		return (1);
2047 	}
2048 	return (kn->kn_fflags != 0);
2049 }
2050