xref: /dragonfly/sys/vfs/msdosfs/msdosfs_vnops.c (revision 7c4f4eee)
1 /* $FreeBSD$ */
2 /*	$NetBSD: msdosfs_vnops.c,v 1.68 1998/02/10 14:10:04 mrg Exp $	*/
3 
4 /*-
5  * SPDX-License-Identifier: BSD-4-Clause
6  *
7  * Copyright (C) 1994, 1995, 1997 Wolfgang Solfrank.
8  * Copyright (C) 1994, 1995, 1997 TooLs GmbH.
9  * All rights reserved.
10  * Original code by Paul Popelka (paulp@uts.amdahl.com) (see below).
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. All advertising materials mentioning features or use of this software
21  *    must display the following acknowledgement:
22  *	This product includes software developed by TooLs GmbH.
23  * 4. The name of TooLs GmbH may not be used to endorse or promote products
24  *    derived from this software without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
27  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
28  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
29  * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
30  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
31  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
32  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
33  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
34  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
35  * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
36  */
37 /*-
38  * Written by Paul Popelka (paulp@uts.amdahl.com)
39  *
40  * You can do anything you want with this software, just don't say you wrote
41  * it, and don't remove this notice.
42  *
43  * This software is provided "as is".
44  *
45  * The author supplies this software to be publicly redistributed on the
46  * understanding that the author is not responsible for the correct
47  * functioning of this software in any circumstances and is not liable for
48  * any damages caused by this software.
49  *
50  * October 1992
51  */
52 
53 #include <sys/param.h>
54 #include <sys/systm.h>
55 #include <sys/resourcevar.h>	/* defines plimit structure in proc struct */
56 #include <sys/kernel.h>
57 #include <sys/stat.h>
58 #include <sys/buf.h>
59 #include <sys/proc.h>
60 #include <sys/priv.h>
61 #include <sys/namei.h>
62 #include <sys/mount.h>
63 #include <sys/unistd.h>
64 #include <sys/vnode.h>
65 #include <sys/malloc.h>
66 #include <sys/dirent.h>
67 #include <sys/signalvar.h>
68 
69 #include <vm/vm.h>
70 #include <vm/vm_extern.h>
71 #include <vm/vm_zone.h>
72 #include <vm/vnode_pager.h>
73 
74 #include <sys/buf2.h>
75 
76 #include <machine/inttypes.h>
77 
78 #include "bpb.h"
79 #include "direntry.h"
80 #include "denode.h"
81 #include "msdosfsmount.h"
82 #include "fat.h"
83 
84 #define	DOS_FILESIZE_MAX	0xffffffff
85 
86 /*
87  * Some general notes:
88  *
89  * In the ufs filesystem the inodes, superblocks, and indirect blocks are
90  * read/written using the vnode for the filesystem. Blocks that represent
91  * the contents of a file are read/written using the vnode for the file
92  * (including directories when they are read/written as files). This
93  * presents problems for the dos filesystem because data that should be in
94  * an inode (if dos had them) resides in the directory itself.  Since we
95  * must update directory entries without the benefit of having the vnode
96  * for the directory we must use the vnode for the filesystem.  This means
97  * that when a directory is actually read/written (via read, write, or
98  * readdir, or seek) we must use the vnode for the filesystem instead of
99  * the vnode for the directory as would happen in ufs. This is to insure we
100  * retrieve the correct block from the buffer cache since the hash value is
101  * based upon the vnode address and the desired block number.
102  */
103 
104 /*
105  * Create a regular file. On entry the directory to contain the file being
106  * created is locked.  We must release before we return. We must also free
107  * the pathname buffer pointed at by cnp->cn_pnbuf, always on error, or
108  * only if the SAVESTART bit in cn_flags is clear on success.
109  */
110 static int
111 msdosfs_create(struct vop_old_create_args *ap)
112 {
113 	struct componentname *cnp = ap->a_cnp;
114 	struct denode ndirent;
115 	struct denode *dep;
116 	struct denode *pdep = VTODE(ap->a_dvp);
117 	struct timespec ts;
118 	int error;
119 
120 	mprintf("msdosfs_create(cnp %p, vap %p\n", cnp, ap->a_vap);
121 
122 	/*
123 	 * If this is the root directory and there is no space left we
124 	 * can't do anything.  This is because the root directory can not
125 	 * change size.
126 	 */
127 	if (pdep->de_StartCluster == MSDOSFSROOT
128 	    && pdep->de_fndoffset >= pdep->de_FileSize) {
129 		error = ENOSPC;
130 		goto bad;
131 	}
132 
133 	/*
134 	 * Create a directory entry for the file, then call createde() to
135 	 * have it installed. NOTE: DOS files are always executable.  We
136 	 * use the absence of the owner write bit to make the file
137 	 * readonly.
138 	 */
139 	bzero(&ndirent, sizeof(ndirent));
140 	error = uniqdosname(pdep, cnp, ndirent.de_Name);
141 	if (error)
142 		goto bad;
143 
144 	ndirent.de_Attributes = (ap->a_vap->va_mode & VWRITE) ?
145 				ATTR_ARCHIVE : ATTR_ARCHIVE | ATTR_READONLY;
146 	ndirent.de_LowerCase = 0;
147 	ndirent.de_StartCluster = 0;
148 	ndirent.de_FileSize = 0;
149 	ndirent.de_dev = pdep->de_dev;
150 	ndirent.de_devvp = pdep->de_devvp;
151 	ndirent.de_pmp = pdep->de_pmp;
152 	ndirent.de_flag = DE_ACCESS | DE_CREATE | DE_UPDATE;
153 	getnanotime(&ts);
154 	DETIMES(&ndirent, &ts, &ts, &ts);
155 	error = createde(&ndirent, pdep, &dep, cnp);
156 	if (error)
157 		goto bad;
158 	*ap->a_vpp = DETOV(dep);
159 	return (0);
160 
161 bad:
162 	return (error);
163 }
164 
165 static int msdosfs_mkdir(struct vop_old_mkdir_args *);
166 
167 static int
168 msdosfs_mknod(struct vop_old_mknod_args *ap)
169 {
170 	switch (ap->a_vap->va_type) {
171 	case VDIR:
172 		return (msdosfs_mkdir((struct vop_old_mkdir_args *)ap));
173 	case VREG:
174 		return (msdosfs_create((struct vop_old_create_args *)ap));
175 	default:
176 		return (EINVAL);
177 	}
178 	/* NOTREACHED */
179 }
180 
181 static int
182 msdosfs_open(struct vop_open_args *ap)
183 {
184 	return(vop_stdopen(ap));
185 }
186 
187 static int
188 msdosfs_close(struct vop_close_args *ap)
189 {
190 	struct vnode *vp = ap->a_vp;
191 	struct denode *dep = VTODE(vp);
192 	struct timespec ts;
193 
194 	if (VREFCNT(vp) > 1) {
195 		getnanotime(&ts);
196 		DETIMES(dep, &ts, &ts, &ts);
197 	}
198 	return (vop_stdclose(ap));
199 }
200 
201 static int
202 msdosfs_access(struct vop_access_args *ap)
203 {
204 	struct denode *dep = VTODE(ap->a_vp);
205 	struct msdosfsmount *pmp = dep->de_pmp;
206 	mode_t file_mode;
207 
208 	file_mode = (S_IXUSR|S_IXGRP|S_IXOTH) | (S_IRUSR|S_IRGRP|S_IROTH) |
209 	    ((dep->de_Attributes & ATTR_READONLY) ?
210 	    0 : (S_IWUSR|S_IWGRP|S_IWOTH));
211 	file_mode &= (ap->a_vp->v_type == VDIR ? pmp->pm_dirmask : pmp->pm_mask);
212 
213 	return (vop_helper_access(ap, pmp->pm_uid, pmp->pm_gid, file_mode, 0));
214 }
215 
216 static int
217 msdosfs_getattr(struct vop_getattr_args *ap)
218 {
219 	struct denode *dep = VTODE(ap->a_vp);
220 	struct msdosfsmount *pmp = dep->de_pmp;
221 	struct vattr *vap = ap->a_vap;
222 	mode_t mode;
223 	struct timespec ts;
224 	u_long dirsperblk = pmp->pm_BytesPerSec / sizeof(struct direntry);
225 	u_long fileid;
226 
227 	getnanotime(&ts);
228 	DETIMES(dep, &ts, &ts, &ts);
229 	vap->va_fsid = dev2udev(dep->de_dev);
230 	/*
231 	 * The following computation of the fileid must be the same as that
232 	 * used in msdosfs_readdir() to compute d_fileno. If not, pwd
233 	 * doesn't work.
234 	 */
235 	if (dep->de_Attributes & ATTR_DIRECTORY) {
236 		fileid = xcntobn(pmp, dep->de_StartCluster) * dirsperblk;
237 		if (dep->de_StartCluster == MSDOSFSROOT)
238 			fileid = 1;
239 	} else {
240 		fileid = xcntobn(pmp, dep->de_dirclust) * dirsperblk;
241 		if (dep->de_dirclust == MSDOSFSROOT)
242 			fileid = roottobn(pmp, 0) * dirsperblk;
243 		fileid += dep->de_diroffset / sizeof(struct direntry);
244 	}
245 	vap->va_fileid = fileid;
246 	if ((dep->de_Attributes & ATTR_READONLY) == 0)
247 		mode = S_IRWXU|S_IRWXG|S_IRWXO;
248 	else
249 		mode = S_IRUSR|S_IXUSR|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH;
250 	vap->va_mode = mode &
251 		(ap->a_vp->v_type == VDIR ? pmp->pm_dirmask : pmp->pm_mask);
252 	vap->va_uid = pmp->pm_uid;
253 	vap->va_gid = pmp->pm_gid;
254 	vap->va_nlink = 1;
255 	vap->va_rmajor = VNOVAL;
256 	vap->va_rminor = VNOVAL;
257 	vap->va_size = dep->de_FileSize;
258 	dos2unixtime(dep->de_MDate, dep->de_MTime, 0, &vap->va_mtime);
259 	if (pmp->pm_flags & MSDOSFSMNT_LONGNAME) {
260 		dos2unixtime(dep->de_ADate, 0, 0, &vap->va_atime);
261 		dos2unixtime(dep->de_CDate, dep->de_CTime, dep->de_CHun,
262 			     &vap->va_ctime);
263 	} else {
264 		vap->va_atime = vap->va_mtime;
265 		vap->va_ctime = vap->va_mtime;
266 	}
267 	vap->va_flags = 0;
268 	if ((dep->de_Attributes & ATTR_ARCHIVE) == 0)
269 		vap->va_flags |= SF_ARCHIVED;
270 	vap->va_gen = 0;
271 	vap->va_blocksize = pmp->pm_bpcluster;
272 	vap->va_bytes =
273 	    (dep->de_FileSize + pmp->pm_crbomask) & ~pmp->pm_crbomask;
274 	vap->va_type = ap->a_vp->v_type;
275 	vap->va_filerev = dep->de_modrev;
276 	return (0);
277 }
278 
279 static int
280 msdosfs_setattr(struct vop_setattr_args *ap)
281 {
282 	struct vnode *vp = ap->a_vp;
283 	struct denode *dep = VTODE(vp);
284 	struct msdosfsmount *pmp = dep->de_pmp;
285 	struct vattr *vap = ap->a_vap;
286 	struct ucred *cred = ap->a_cred;
287 	int error = 0;
288 
289 	mprintf("msdosfs_setattr(): vp %p, vap %p, cred %p\n", vp, vap, cred);
290 
291 	/*
292 	 * Check for unsettable attributes.
293 	 */
294 	if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
295 	    (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
296 	    (vap->va_blocksize != VNOVAL) || (vap->va_rmajor != VNOVAL) ||
297 	    (vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
298 		mprintf("msdosfs_setattr(): returning EINVAL\n");
299 		mprintf("    va_type %u, va_nlink %"PRIx64", va_fsid %x, va_fileid %"PRIx64"\n",
300 		    vap->va_type, vap->va_nlink, vap->va_fsid, vap->va_fileid);
301 		mprintf("    va_blocksize %lx, va_rmajor %x, va_bytes %"PRIx64", va_gen %"PRIx64"\n",
302 		    vap->va_blocksize, vap->va_rmajor, vap->va_bytes,
303 		    vap->va_gen);
304 		mprintf("    va_uid %x, va_gid %x\n",
305 		    vap->va_uid, vap->va_gid);
306 		return (EINVAL);
307 	}
308 	if (vap->va_flags != VNOVAL) {
309 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
310 			return (EROFS);
311 		if (cred->cr_uid != pmp->pm_uid &&
312 		    (error = priv_check_cred(cred, PRIV_VFS_SETATTR, 0)))
313 			return (error);
314 		/*
315 		 * We are very inconsistent about handling unsupported
316 		 * attributes.  We ignored the access time and the
317 		 * read and execute bits.  We were strict for the other
318 		 * attributes.
319 		 *
320 		 * Here we are strict, stricter than ufs in not allowing
321 		 * users to attempt to set SF_SETTABLE bits or anyone to
322 		 * set unsupported bits.  However, we ignore attempts to
323 		 * set ATTR_ARCHIVE for directories `cp -pr' from a more
324 		 * sensible file system attempts it a lot.
325 		 */
326 		if (cred->cr_uid != 0) {
327 			if (vap->va_flags & SF_SETTABLE)
328 				return EPERM;
329 		}
330 		if (vap->va_flags & ~SF_ARCHIVED)
331 			return EOPNOTSUPP;
332 		if (vap->va_flags & SF_ARCHIVED)
333 			dep->de_Attributes &= ~ATTR_ARCHIVE;
334 		else if (!(dep->de_Attributes & ATTR_DIRECTORY))
335 			dep->de_Attributes |= ATTR_ARCHIVE;
336 		dep->de_flag |= DE_MODIFIED;
337 	}
338 
339 	if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
340 		uid_t uid;
341 		gid_t gid;
342 
343 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
344 			return (EROFS);
345 		uid = vap->va_uid;
346 		if (uid == (uid_t)VNOVAL)
347 			uid = pmp->pm_uid;
348 		gid = vap->va_gid;
349 		if (gid == (gid_t)VNOVAL)
350 			gid = pmp->pm_gid;
351 		if ((cred->cr_uid != pmp->pm_uid || uid != pmp->pm_uid ||
352 		    (gid != pmp->pm_gid && !groupmember(gid, cred))) &&
353 		    (error = priv_check_cred(cred, PRIV_VFS_SETATTR, 0)))
354 			return error;
355 		if (uid != pmp->pm_uid || gid != pmp->pm_gid)
356 			return EINVAL;
357 	}
358 
359 	if (vap->va_size != VNOVAL) {
360 		/*
361 		 * Disallow write attempts on read-only file systems;
362 		 * unless the file is a socket, fifo, or a block or
363 		 * character device resident on the file system.
364 		 */
365 		switch (vp->v_type) {
366 		case VDIR:
367 			return (EISDIR);
368 			/* NOT REACHED */
369 		case VLNK:
370 		case VREG:
371 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
372 				return (EROFS);
373 			break;
374 		default:
375 			break;
376 		}
377 		error = detrunc(dep, vap->va_size, 0);
378 		if (error)
379 			return error;
380 	}
381 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
382 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
383 			return (EROFS);
384 		if (cred->cr_uid != pmp->pm_uid &&
385 		    (error = priv_check_cred(cred, PRIV_VFS_SETATTR, 0)) &&
386 		    ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
387 		    (error = VOP_EACCESS(vp, VWRITE, cred))))
388 			return (error);
389 		if (vp->v_type != VDIR) {
390 			if ((pmp->pm_flags & MSDOSFSMNT_NOWIN95) == 0 &&
391 			    vap->va_atime.tv_sec != VNOVAL) {
392 				dep->de_flag &= ~DE_ACCESS;
393 				unix2dostime(&vap->va_atime, &dep->de_ADate,
394 				    NULL, NULL);
395 			}
396 			if (vap->va_mtime.tv_sec != VNOVAL) {
397 				dep->de_flag &= ~DE_UPDATE;
398 				unix2dostime(&vap->va_mtime, &dep->de_MDate,
399 				    &dep->de_MTime, NULL);
400 			}
401 			dep->de_Attributes |= ATTR_ARCHIVE;
402 			dep->de_flag |= DE_MODIFIED;
403 		}
404 	}
405 	/*
406 	 * DOS files only have the ability to have their writability
407 	 * attribute set, so we use the owner write bit to set the readonly
408 	 * attribute.
409 	 */
410 	if (vap->va_mode != (mode_t)VNOVAL) {
411 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
412 			return (EROFS);
413 		if (cred->cr_uid != pmp->pm_uid &&
414 		    (error = priv_check_cred(cred, PRIV_VFS_SETATTR, 0)))
415 			return (error);
416 		if (vp->v_type != VDIR) {
417 			/* We ignore the read and execute bits. */
418 			if (vap->va_mode & VWRITE)
419 				dep->de_Attributes &= ~ATTR_READONLY;
420 			else
421 				dep->de_Attributes |= ATTR_READONLY;
422 			dep->de_Attributes |= ATTR_ARCHIVE;
423 			dep->de_flag |= DE_MODIFIED;
424 		}
425 	}
426 	return (deupdat(dep, 1));
427 }
428 
429 static int
430 msdosfs_read(struct vop_read_args *ap)
431 {
432 	int error = 0;
433 	int blsize;
434 	int isadir;
435 	size_t orig_resid;
436 	u_int n;
437 	u_long diff;
438 	u_long on;
439 	daddr_t lbn;
440 	daddr_t rablock;
441 	off_t raoffset;
442 	off_t loffset;
443 	int rasize;
444 	int seqcount;
445 	struct buf *bp;
446 	struct vnode *vp = ap->a_vp;
447 	struct denode *dep = VTODE(vp);
448 	struct msdosfsmount *pmp = dep->de_pmp;
449 	struct uio *uio = ap->a_uio;
450 
451 	if (uio->uio_offset < 0)
452 		return (EINVAL);
453 
454 	if ((uoff_t)uio->uio_offset > DOS_FILESIZE_MAX)
455                 return (0);
456 	/*
457 	 * If they didn't ask for any data, then we are done.
458 	 */
459 	orig_resid = uio->uio_resid;
460 	if (orig_resid == 0)
461 		return (0);
462 
463 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
464 
465 	isadir = dep->de_Attributes & ATTR_DIRECTORY;
466 	do {
467 		if (uio->uio_offset >= dep->de_FileSize)
468 			break;
469 
470 		/*
471 		 * note: lbn is a cluster number, not a device block number.
472 		 */
473 		lbn = de_off2cn(pmp, uio->uio_offset);
474 		loffset = de_cn2doff(pmp, lbn);
475 
476 		/*
477 		 * If we are operating on a directory file then be sure to
478 		 * do i/o with the vnode for the filesystem instead of the
479 		 * vnode for the directory.
480 		 */
481 		if (isadir) {
482 			/*
483 			 * convert cluster # to block #.  lbn is a
484 			 * device block number after this.
485 			 */
486 			error = pcbmap(dep, lbn, &lbn, NULL, &blsize);
487 			loffset = de_bntodoff(pmp, lbn);
488 			if (error == E2BIG) {
489 				error = EINVAL;
490 				break;
491 			} else if (error)
492 				break;
493 			error = bread(pmp->pm_devvp, loffset, blsize, &bp);
494 		} else {
495 			blsize = pmp->pm_bpcluster;
496 			rablock = lbn + 1;
497 			raoffset = de_cn2doff(pmp, rablock);
498 			if (seqcount > 1 &&
499 			    raoffset < dep->de_FileSize) {
500 				rasize = pmp->pm_bpcluster;
501 				error = breadn(vp, loffset, blsize,
502 					       &raoffset, &rasize, 1, &bp);
503 			} else {
504 				error = bread(vp, loffset, blsize, &bp);
505 			}
506 		}
507 		if (error) {
508 			brelse(bp);
509 			break;
510 		}
511 		on = uio->uio_offset & pmp->pm_crbomask;
512 		diff = pmp->pm_bpcluster - on;
513 		n = szmin(uio->uio_resid, diff);
514 		diff = dep->de_FileSize - uio->uio_offset;
515 		if (diff < n)
516 			n = diff;
517 		diff = blsize - bp->b_resid;
518 		if (diff < n)
519 			n = diff;
520 		error = uiomovebp(bp, bp->b_data + on, (size_t)n, uio);
521 		brelse(bp);
522 	} while (error == 0 && uio->uio_resid > 0 && n != 0);
523 
524 	if (!isadir && (error == 0 || uio->uio_resid != orig_resid) &&
525 	    (vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
526 		dep->de_flag |= DE_ACCESS;
527 	return (error);
528 }
529 
530 /*
531  * Write data to a file or directory.
532  */
533 static int
534 msdosfs_write(struct vop_write_args *ap)
535 {
536 	int n;
537 	int croffset;
538 	size_t resid;
539 	u_long osize;
540 	int error = 0;
541 	u_long count;
542 	daddr_t cn, lastcn;
543 	struct buf *bp;
544 	int ioflag = ap->a_ioflag;
545 	struct uio *uio = ap->a_uio;
546 	struct thread *td = uio->uio_td;
547 	struct vnode *vp = ap->a_vp;
548 	struct vnode *thisvp;
549 	struct denode *dep = VTODE(vp);
550 	struct msdosfsmount *pmp = dep->de_pmp;
551 	struct proc *p = (td ? td->td_proc : NULL);
552 	struct lwp *lp = (td ? td->td_lwp : NULL);
553 
554 	mprintf("msdosfs_write(vp %p, uio %p, ioflag %x, cred %p\n",
555 	    vp, uio, ioflag, ap->a_cred);
556 	mprintf("msdosfs_write(): diroff %lu, dirclust %lu, startcluster %lu\n",
557 	    dep->de_diroffset, dep->de_dirclust, dep->de_StartCluster);
558 
559 	switch (vp->v_type) {
560 	case VREG:
561 		if (ioflag & IO_APPEND)
562 			uio->uio_offset = dep->de_FileSize;
563 		thisvp = vp;
564 		break;
565 	case VDIR:
566 		return EISDIR;
567 	default:
568 		panic("msdosfs_write(): bad file type");
569 	}
570 
571 	if (uio->uio_offset < 0)
572 		return (EFBIG);
573 
574 	if (uio->uio_resid == 0)
575 		return (0);
576 
577 	/*
578 	 * If they've exceeded their filesize limit, tell them about it.
579 	 */
580 	if (p &&
581 	    ((uoff_t)uio->uio_offset + uio->uio_resid >
582 	    p->p_rlimit[RLIMIT_FSIZE].rlim_cur)) {
583 		lwpsignal(p, lp, SIGXFSZ);
584 		return (EFBIG);
585 	}
586 
587 	if ((uoff_t)uio->uio_offset > DOS_FILESIZE_MAX)
588                 return (EFBIG);
589 	if ((uoff_t)uio->uio_offset + uio->uio_resid > DOS_FILESIZE_MAX)
590                 return (EFBIG);
591 
592 	/*
593 	 * If the offset we are starting the write at is beyond the end of
594 	 * the file, then they've done a seek.  Unix filesystems allow
595 	 * files with holes in them, DOS doesn't so we must fill the hole
596 	 * with zeroed blocks.
597 	 */
598 	if (uio->uio_offset > dep->de_FileSize) {
599 		error = deextend(dep, uio->uio_offset);
600 		if (error)
601 			return (error);
602 	}
603 
604 	/*
605 	 * Remember some values in case the write fails.
606 	 */
607 	resid = uio->uio_resid;
608 	osize = dep->de_FileSize;
609 
610 	/*
611 	 * If we write beyond the end of the file, extend it to its ultimate
612 	 * size ahead of the time to hopefully get a contiguous area.
613 	 */
614 	if (uio->uio_offset + resid > osize) {
615 		count = de_clcount(pmp, uio->uio_offset + resid) -
616 			de_clcount(pmp, osize);
617 		error = extendfile(dep, count, NULL, NULL, 0);
618 		if (error &&  (error != ENOSPC || (ioflag & IO_UNIT)))
619 			goto errexit;
620 		lastcn = dep->de_fc[FC_LASTFC].fc_frcn;
621 	} else
622 		lastcn = de_clcount(pmp, osize) - 1;
623 
624 	do {
625 		if (de_off2cn(pmp, uio->uio_offset) > lastcn) {
626 			error = ENOSPC;
627 			break;
628 		}
629 
630 		croffset = uio->uio_offset & pmp->pm_crbomask;
631 		n = (int)szmin(uio->uio_resid, pmp->pm_bpcluster - croffset);
632 		if (uio->uio_offset + n > dep->de_FileSize) {
633 			dep->de_FileSize = uio->uio_offset + n;
634 			/* The object size needs to be set before buffer is allocated */
635 			vnode_pager_setsize(vp, dep->de_FileSize);
636 		}
637 
638 		/*
639 		 * If either the whole cluster gets written, or we write
640 		 * the cluster from its start beyond EOF, then no need to
641 		 * read data from disk.
642 		 *
643 		 * If UIO_NOCOPY is set we have to do a read-before-write
644 		 * to fill in any missing pieces of the buffer since no
645 		 * actual overwrite will occur.
646 		 */
647 		cn = de_off2cn(pmp, uio->uio_offset);
648 		if ((uio->uio_offset & pmp->pm_crbomask) == 0
649 		    && uio->uio_segflg != UIO_NOCOPY
650 		    && (de_off2cn(pmp, uio->uio_offset + uio->uio_resid)
651 		        > de_off2cn(pmp, uio->uio_offset)
652 			|| uio->uio_offset + uio->uio_resid >= dep->de_FileSize)) {
653 			bp = getblk(thisvp, de_cn2doff(pmp, cn),
654 				    pmp->pm_bpcluster, 0, 0);
655 			clrbuf(bp);
656 			/*
657 			 * Do the bmap now, since pcbmap needs buffers
658 			 * for the FAT table. (see msdosfs_strategy)
659 			 */
660 			if (bp->b_bio2.bio_offset == NOOFFSET) {
661 				daddr_t lblkno = de_off2cn(pmp, bp->b_loffset);
662 				daddr_t dblkno;
663 
664 				error = pcbmap(dep, lblkno,
665 					       &dblkno, NULL, NULL);
666 				if (error || dblkno == (daddr_t)-1) {
667 					bp->b_bio2.bio_offset = NOOFFSET;
668 				} else {
669 					bp->b_bio2.bio_offset = de_bntodoff(pmp,
670 					    dblkno);
671 				}
672 			}
673 			if (bp->b_bio2.bio_offset == NOOFFSET) {
674 				brelse(bp);
675 				if (!error)
676 					error = EIO;		/* XXX */
677 				break;
678 			}
679 		} else {
680 			/*
681 			 * The block we need to write into exists, so read
682 			 * it in.
683 			 */
684 			error = bread(thisvp, de_cn2doff(pmp, cn),
685 				      pmp->pm_bpcluster, &bp);
686 			if (error) {
687 				brelse(bp);
688 				break;
689 			}
690 		}
691 
692 		/*
693 		 * Should these vnode_pager_* functions be done on dir
694 		 * files?
695 		 */
696 
697 		/*
698 		 * Copy the data from user space into the buf header.
699 		 */
700 		error = uiomovebp(bp, bp->b_data + croffset, (size_t)n, uio);
701 		if (error) {
702 			brelse(bp);
703 			break;
704 		}
705 
706 		/*
707 		 * If they want this synchronous then write it and wait for
708 		 * it.  Otherwise, if on a cluster boundary write it
709 		 * asynchronously so we can move on to the next block
710 		 * without delay.  Otherwise do a delayed write because we
711 		 * may want to write somemore into the block later.
712 		 */
713 		if (ioflag & IO_SYNC)
714 			bwrite(bp);
715 		else if (n + croffset == pmp->pm_bpcluster)
716 			bawrite(bp);
717 		else
718 			bdwrite(bp);
719 		dep->de_flag |= DE_UPDATE;
720 	} while (error == 0 && uio->uio_resid > 0);
721 
722 	/*
723 	 * If the write failed and they want us to, truncate the file back
724 	 * to the size it was before the write was attempted.
725 	 */
726 errexit:
727 	if (error) {
728 		if (ioflag & IO_UNIT) {
729 			detrunc(dep, osize, ioflag & IO_SYNC);
730 			uio->uio_offset -= resid - uio->uio_resid;
731 			uio->uio_resid = resid;
732 		} else {
733 			detrunc(dep, dep->de_FileSize, ioflag & IO_SYNC);
734 			if (uio->uio_resid != resid)
735 				error = 0;
736 		}
737 	} else if (ioflag & IO_SYNC)
738 		error = deupdat(dep, 1);
739 	return (error);
740 }
741 
742 /*
743  * Flush the blocks of a file to disk.
744  *
745  * This function is worthless for vnodes that represent directories. Maybe we
746  * could just do a sync if they try an fsync on a directory file.
747  */
748 static int
749 msdosfs_fsync(struct vop_fsync_args *ap)
750 {
751 	struct vnode *vp = ap->a_vp;
752 
753 	/*
754 	 * Flush all dirty buffers associated with a vnode.
755 	 */
756 #ifdef DIAGNOSTIC
757 loop:
758 #endif
759 	vfsync(vp, ap->a_waitfor, 0, NULL, NULL);
760 #ifdef DIAGNOSTIC
761 	if (ap->a_waitfor == MNT_WAIT && !RB_EMPTY(&vp->v_rbdirty_tree)) {
762 		vprint("msdosfs_fsync: dirty", vp);
763 		goto loop;
764 	}
765 #endif
766 	return (deupdat(VTODE(vp), ap->a_waitfor == MNT_WAIT));
767 }
768 
769 static int
770 msdosfs_remove(struct vop_old_remove_args *ap)
771 {
772 	struct denode *dep = VTODE(ap->a_vp);
773 	struct denode *ddep = VTODE(ap->a_dvp);
774 	int error;
775 
776 	if (ap->a_vp->v_type == VDIR)
777 		error = EPERM;
778 	else
779 		error = removede(ddep, dep);
780 	mprintf("msdosfs_remove(), dep %p, v_refcnt 0x%08x\n",
781 		dep, ap->a_vp->v_refcnt);
782 	return (error);
783 }
784 
785 /*
786  * DOS filesystems don't know what links are. But since we already called
787  * msdosfs_lookup() with create and lockparent, the parent is locked so we
788  * have to free it before we return the error.
789  */
790 static int
791 msdosfs_link(struct vop_old_link_args *ap)
792 {
793 	return (EOPNOTSUPP);
794 }
795 
796 /*
797  * Renames on files require moving the denode to a new hash queue since the
798  * denode's location is used to compute which hash queue to put the file
799  * in. Unless it is a rename in place.  For example "mv a b".
800  *
801  * What follows is the basic algorithm:
802  *
803  * if (file move) {
804  *	if (dest file exists) {
805  *		remove dest file
806  *	}
807  *	if (dest and src in same directory) {
808  *		rewrite name in existing directory slot
809  *	} else {
810  *		write new entry in dest directory
811  *		update offset and dirclust in denode
812  *		move denode to new hash chain
813  *		clear old directory entry
814  *	}
815  * } else {
816  *	directory move
817  *	if (dest directory exists) {
818  *		if (dest is not empty) {
819  *			return ENOTEMPTY
820  *		}
821  *		remove dest directory
822  *	}
823  *	if (dest and src in same directory) {
824  *		rewrite name in existing entry
825  *	} else {
826  *		be sure dest is not a child of src directory
827  *		write entry in dest directory
828  *		update "." and ".." in moved directory
829  *		clear old directory entry for moved directory
830  *	}
831  * }
832  *
833  * On entry:
834  *	source's parent directory is unlocked
835  *	source file or directory is unlocked
836  *	destination's parent directory is locked
837  *	destination file or directory is locked if it exists
838  *
839  * On exit:
840  *	all denodes should be released
841  *
842  * Notes:
843  * I'm not sure how the memory containing the pathnames pointed at by the
844  * componentname structures is freed, there may be some memory bleeding
845  * for each rename done.
846  */
847 static int
848 msdosfs_rename(struct vop_old_rename_args *ap)
849 {
850 	struct vnode *tdvp = ap->a_tdvp;
851 	struct vnode *fvp = ap->a_fvp;
852 	struct vnode *fdvp = ap->a_fdvp;
853 	struct vnode *tvp = ap->a_tvp;
854 	struct componentname *tcnp = ap->a_tcnp;
855 	struct componentname *fcnp = ap->a_fcnp;
856 	struct denode *ip, *xp, *dp, *zp;
857 	u_char toname[11], oldname[11];
858 	u_long from_diroffset, to_diroffset;
859 	u_char to_count;
860 	int doingdirectory = 0, newparent = 0;
861 	int error;
862 	u_long cn;
863 	daddr_t bn;
864 	struct msdosfsmount *pmp;
865 	struct direntry *dotdotp;
866 	struct buf *bp;
867 
868 	pmp = VFSTOMSDOSFS(fdvp->v_mount);
869 
870 	/*
871 	 * Check for cross-device rename.
872 	 */
873 	if ((fvp->v_mount != tdvp->v_mount) ||
874 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
875 		error = EXDEV;
876 abortit:
877 		if (tdvp == tvp)
878 			vrele(tdvp);
879 		else
880 			vput(tdvp);
881 		if (tvp)
882 			vput(tvp);
883 		vrele(fdvp);
884 		vrele(fvp);
885 		return (error);
886 	}
887 
888 	/*
889 	 * If source and dest are the same, do nothing.
890 	 */
891 	if (tvp == fvp) {
892 		error = 0;
893 		goto abortit;
894 	}
895 
896 	/*
897 	 * fvp, fdvp are unlocked, tvp, tdvp are locked.  Lock fvp and note
898 	 * that we have to unlock it to use the abortit target.
899 	 */
900 	error = vn_lock(fvp, LK_EXCLUSIVE | LK_FAILRECLAIM);
901 	if (error)
902 		goto abortit;
903 	dp = VTODE(fdvp);
904 	ip = VTODE(fvp);
905 
906 	/*
907 	 * Be sure we are not renaming ".", "..", or an alias of ".". This
908 	 * leads to a crippled directory tree.  It's pretty tough to do a
909 	 * "ls" or "pwd" with the "." directory entry missing, and "cd .."
910 	 * doesn't work if the ".." entry is missing.
911 	 */
912 	if (ip->de_Attributes & ATTR_DIRECTORY) {
913 		/*
914 		 * Avoid ".", "..", and aliases of "." for obvious reasons.
915 		 */
916 		if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
917 		    dp == ip ||
918 		    (fcnp->cn_flags & CNP_ISDOTDOT) ||
919 		    (tcnp->cn_flags & CNP_ISDOTDOT) ||
920 		    (ip->de_flag & DE_RENAME)) {
921 			vn_unlock(fvp);
922 			error = EINVAL;
923 			goto abortit;
924 		}
925 		ip->de_flag |= DE_RENAME;
926 		doingdirectory++;
927 	}
928 
929 	/*
930 	 * fvp locked, fdvp unlocked, tvp, tdvp locked.  DE_RENAME only
931 	 * set if doingdirectory.  We will get fvp unlocked in fairly
932 	 * short order.  dp and xp must be setup and fvp must be unlocked
933 	 * for the out and bad targets to work properly.
934 	 */
935 	dp = VTODE(tdvp);
936 	xp = tvp ? VTODE(tvp) : NULL;
937 
938 	/*
939 	 * Remember direntry place to use for destination
940 	 */
941 	to_diroffset = dp->de_fndoffset;
942 	to_count = dp->de_fndcnt;
943 
944 	/*
945 	 * If ".." must be changed (ie the directory gets a new
946 	 * parent) then the source directory must not be in the
947 	 * directory hierarchy above the target, as this would
948 	 * orphan everything below the source directory. Also
949 	 * the user must have write permission in the source so
950 	 * as to be able to change "..". We must repeat the call
951 	 * to namei, as the parent directory is unlocked by the
952 	 * call to doscheckpath().
953 	 */
954 	error = VOP_EACCESS(fvp, VWRITE, tcnp->cn_cred);
955 	vn_unlock(fvp);
956 	if (VTODE(fdvp)->de_StartCluster != VTODE(tdvp)->de_StartCluster)
957 		newparent = 1;
958 
959 	/*
960 	 * ok. fvp, fdvp unlocked, tvp, tdvp locked. tvp may be NULL.
961 	 * DE_RENAME only set if doingdirectory.
962 	 */
963 	if (doingdirectory && newparent) {
964 		if (error)	/* write access check above */
965 			goto bad;
966 		if (xp != NULL) {
967 			vput(tvp);
968 			xp = NULL;
969 		}
970 		/*
971 		 * checkpath vput's tdvp (VTOI(dp)) on return no matter what,
972 		 * get an extra ref so we wind up with just an unlocked, ref'd
973 		 * tdvp.  The 'out' target skips tvp and tdvp cleanups (tdvp
974 		 * isn't locked so we can't use the out target).
975 		 */
976 		vref(tdvp);
977 		error = doscheckpath(ip, dp);
978 		tcnp->cn_flags |= CNP_PDIRUNLOCK;
979 		if (error) {
980 			vrele(tdvp);
981 			goto out;
982 		}
983 		/*
984 		 * relookup no longer messes with the ref count.  tdvp must
985 		 * be unlocked on entry and on success will be locked on
986 		 * return.
987 		 */
988 		error = relookup(tdvp, &tvp, tcnp);
989 		if (error) {
990 			if (tcnp->cn_flags & CNP_PDIRUNLOCK)
991 				vrele(tdvp);
992 			else
993 				vput(tdvp);
994 			goto out;
995 		}
996 
997 		/*
998 		 * tvp and tdvp are now locked again.
999 		 */
1000 		dp = VTODE(tdvp);
1001 		xp = tvp ? VTODE(tvp) : NULL;
1002 	}
1003 
1004 	/*
1005 	 * tvp and tdvp are now locked again, the 'bad' target can be used
1006 	 * to clean them up again.  Delete an existant target and clean
1007 	 * up tvp.  Set xp to NULL to indicate that tvp has been cleaned up.
1008 	 */
1009 	if (xp != NULL) {
1010 		/*
1011 		 * Target must be empty if a directory and have no links
1012 		 * to it. Also, ensure source and target are compatible
1013 		 * (both directories, or both not directories).
1014 		 */
1015 		if (xp->de_Attributes & ATTR_DIRECTORY) {
1016 			if (!dosdirempty(xp)) {
1017 				error = ENOTEMPTY;
1018 				goto bad;
1019 			}
1020 			if (!doingdirectory) {
1021 				error = ENOTDIR;
1022 				goto bad;
1023 			}
1024 		} else if (doingdirectory) {
1025 			error = EISDIR;
1026 			goto bad;
1027 		}
1028 		error = removede(dp, xp);
1029 		if (error)
1030 			goto bad;
1031 		vput(tvp);
1032 		xp = NULL;
1033 		tvp = NULL;
1034 	}
1035 
1036 	/*
1037 	 * Convert the filename in tcnp into a dos filename. We copy this
1038 	 * into the denode and directory entry for the destination
1039 	 * file/directory.
1040 	 */
1041 	error = uniqdosname(VTODE(tdvp), tcnp, toname);
1042 	if (error)
1043 		goto bad;
1044 
1045 	/*
1046 	 * Since from wasn't locked at various places above, we have to do
1047 	 * a relookup here.  If the target and source are the same directory
1048 	 * we have to unlock the target directory in order to safely relookup
1049 	 * the source, because relookup expects its directory to be unlocked.
1050 	 *
1051 	 * Note that ap->a_fvp is still valid and ref'd.  Any cleanup must
1052 	 * now take that into account.
1053 	 *
1054 	 * The tdvp locking issues make this a real mess.
1055 	 */
1056 	fcnp->cn_flags &= ~CNP_MODMASK;
1057 	fcnp->cn_flags |= CNP_LOCKPARENT;
1058 	if (newparent == 0)
1059 		vn_unlock(tdvp);
1060 	error = relookup(fdvp, &fvp, fcnp);
1061 	if (error || fvp == NULL) {
1062 		/*
1063 		 * From name has disappeared.  Note: fdvp might == tdvp.
1064 		 *
1065 		 * DE_RENAME is only set if doingdirectory.
1066 		 */
1067 		if (doingdirectory)
1068 			panic("rename: lost dir entry");
1069 		if (fcnp->cn_flags & CNP_PDIRUNLOCK)
1070 			vrele(fdvp);
1071 		else
1072 			vput(fdvp);
1073 		if (newparent == 0)
1074 			vrele(tdvp);
1075 		else
1076 			vput(tdvp);
1077 		vrele(ap->a_fvp);
1078 		return(0);
1079 	}
1080 
1081 	/*
1082 	 * No error occured.  tdvp, fdvp and fvp are all locked.  If
1083 	 * newparent was 0 be aware that fdvp == tdvp.  tvp has been cleaned
1084 	 * up.  ap->a_fvp is still refd.
1085 	 */
1086 	xp = VTODE(fvp);
1087 	zp = VTODE(fdvp);
1088 	from_diroffset = zp->de_fndoffset;
1089 
1090 	/*
1091 	 * Ensure that the directory entry still exists and has not
1092 	 * changed till now. If the source is a file the entry may
1093 	 * have been unlinked or renamed. In either case there is
1094 	 * no further work to be done. If the source is a directory
1095 	 * then it cannot have been rmdir'ed or renamed; this is
1096 	 * prohibited by the DE_RENAME flag.
1097 	 *
1098 	 * DE_RENAME is only set if doingdirectory.
1099 	 */
1100 	if (xp != ip) {
1101 		if (doingdirectory)
1102 			panic("rename: lost dir entry");
1103 		goto done;
1104 	} else {
1105 		u_long new_dirclust;
1106 		u_long new_diroffset;
1107 
1108 		/*
1109 		 * First write a new entry in the destination
1110 		 * directory and mark the entry in the source directory
1111 		 * as deleted.  Then move the denode to the correct hash
1112 		 * chain for its new location in the filesystem.  And, if
1113 		 * we moved a directory, then update its .. entry to point
1114 		 * to the new parent directory.
1115 		 */
1116 		bcopy(ip->de_Name, oldname, 11);
1117 		bcopy(toname, ip->de_Name, 11);	/* update denode */
1118 		dp->de_fndoffset = to_diroffset;
1119 		dp->de_fndcnt = to_count;
1120 		error = createde(ip, dp, NULL, tcnp);
1121 		if (error) {
1122 			bcopy(oldname, ip->de_Name, 11);
1123 			goto done;
1124 		}
1125 		ip->de_refcnt++;
1126 		zp->de_fndoffset = from_diroffset;
1127 		error = removede(zp, ip);
1128 		if (error) {
1129 			/* XXX should really panic here, fs is corrupt */
1130 			goto done;
1131 		}
1132 		if (!doingdirectory) {
1133 			error = pcbmap(dp, de_cluster(pmp, to_diroffset),
1134 				       NULL, &new_dirclust, NULL);
1135 			if (error) {
1136 				/* XXX should really panic here, fs is corrupt */
1137 				goto done;
1138 			}
1139 			if (new_dirclust == MSDOSFSROOT)
1140 				new_diroffset = to_diroffset;
1141 			else
1142 				new_diroffset = to_diroffset & pmp->pm_crbomask;
1143 			msdosfs_reinsert(ip, new_dirclust, new_diroffset);
1144 		}
1145 	}
1146 
1147 	/*
1148 	 * If we moved a directory to a new parent directory, then we must
1149 	 * fixup the ".." entry in the moved directory.
1150 	 */
1151 	if (doingdirectory && newparent) {
1152 		cn = ip->de_StartCluster;
1153 		if (cn == MSDOSFSROOT) {
1154 			/* this should never happen */
1155 			panic("msdosfs_rename(): updating .. in root directory?");
1156 		} else {
1157 			bn = xcntobn(pmp, cn);
1158 		}
1159 		error = bread(pmp->pm_devvp, de_bntodoff(pmp, bn),
1160 			      pmp->pm_bpcluster, &bp);
1161 		if (error) {
1162 			/* XXX should really panic here, fs is corrupt */
1163 			brelse(bp);
1164 			goto done;
1165 		}
1166 		dotdotp = (struct direntry *)bp->b_data + 1;
1167 		putushort(dotdotp->deStartCluster, dp->de_StartCluster);
1168 		if (FAT32(pmp))
1169 			putushort(dotdotp->deHighClust, dp->de_StartCluster >> 16);
1170 		error = bwrite(bp);
1171 		if (error) {
1172 			/* XXX should really panic here, fs is corrupt */
1173 			goto done;
1174 		}
1175 	}
1176 
1177 	/*
1178 	 * done case fvp, fdvp, tdvp are locked.  ap->a_fvp is refd
1179 	 */
1180 done:
1181 	if (doingdirectory)
1182 		ip->de_flag &= ~DE_RENAME;	/* XXX fvp not locked */
1183 	vput(fvp);
1184 	if (newparent)
1185 		vput(fdvp);
1186 	else
1187 		vrele(fdvp);
1188 	vput(tdvp);
1189 	vrele(ap->a_fvp);
1190 	return (error);
1191 
1192 	/*
1193 	 * 'bad' target: xp governs tvp.  tvp and tdvp arel ocked, fdvp and fvp
1194 	 * are not locked.  ip points to fvp's inode which may have DE_RENAME
1195 	 * set.
1196 	 */
1197 bad:
1198 	if (xp)
1199 		vput(tvp);
1200 	vput(tdvp);
1201 out:
1202 	/*
1203 	 * 'out' target: tvp and tdvp have already been cleaned up.
1204 	 */
1205 	if (doingdirectory)
1206 		ip->de_flag &= ~DE_RENAME;
1207 	vrele(fdvp);
1208 	vrele(fvp);
1209 	return (error);
1210 
1211 }
1212 
1213 static struct {
1214 	struct direntry dot;
1215 	struct direntry dotdot;
1216 } dosdirtemplate = {
1217 	{	".          ",				/* the . entry */
1218 		ATTR_DIRECTORY,				/* file attribute */
1219 		0,					/* reserved */
1220 		0, { 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 		0,					/* reserved */
1230 		0, { 0, 0 }, { 0, 0 },			/* create time & date */
1231 		{ 0, 0 },				/* access date */
1232 		{ 0, 0 },				/* high bits of start cluster */
1233 		{ 210, 4 }, { 210, 4 },			/* modify time & date */
1234 		{ 0, 0 },				/* startcluster */
1235 		{ 0, 0, 0, 0 }				/* filesize */
1236 	}
1237 };
1238 
1239 static int
1240 msdosfs_mkdir(struct vop_old_mkdir_args *ap)
1241 {
1242 	struct componentname *cnp = ap->a_cnp;
1243 	struct denode *dep;
1244 	struct denode *pdep = VTODE(ap->a_dvp);
1245 	struct direntry *denp;
1246 	struct msdosfsmount *pmp = pdep->de_pmp;
1247 	struct buf *bp;
1248 	u_long newcluster, pcl;
1249 	int bn;
1250 	int error;
1251 	struct denode ndirent;
1252 	struct timespec ts;
1253 
1254 	/*
1255 	 * If this is the root directory and there is no space left we
1256 	 * can't do anything.  This is because the root directory can not
1257 	 * change size.
1258 	 */
1259 	if (pdep->de_StartCluster == MSDOSFSROOT
1260 	    && pdep->de_fndoffset >= pdep->de_FileSize) {
1261 		error = ENOSPC;
1262 		goto bad2;
1263 	}
1264 
1265 	/*
1266 	 * Allocate a cluster to hold the about to be created directory.
1267 	 */
1268 	error = clusteralloc(pmp, 0, 1, CLUST_EOFE, &newcluster, NULL);
1269 	if (error)
1270 		goto bad2;
1271 
1272 	bzero(&ndirent, sizeof(ndirent));
1273 	ndirent.de_pmp = pmp;
1274 	ndirent.de_flag = DE_ACCESS | DE_CREATE | DE_UPDATE;
1275 	getnanotime(&ts);
1276 	DETIMES(&ndirent, &ts, &ts, &ts);
1277 
1278 	/*
1279 	 * Now fill the cluster with the "." and ".." entries. And write
1280 	 * the cluster to disk.  This way it is there for the parent
1281 	 * directory to be pointing at if there were a crash.
1282 	 */
1283 	bn = xcntobn(pmp, newcluster);
1284 	/* always succeeds */
1285 	bp = getblk(pmp->pm_devvp, de_bntodoff(pmp, bn),
1286 		    pmp->pm_bpcluster, 0, 0);
1287 	bzero(bp->b_data, pmp->pm_bpcluster);
1288 	bcopy(&dosdirtemplate, bp->b_data, sizeof(dosdirtemplate));
1289 	denp = (struct direntry *)bp->b_data;
1290 	putushort(denp[0].deStartCluster, newcluster);
1291 	putushort(denp[0].deCDate, ndirent.de_CDate);
1292 	putushort(denp[0].deCTime, ndirent.de_CTime);
1293 	denp[0].deCHundredth = ndirent.de_CHun;
1294 	putushort(denp[0].deADate, ndirent.de_ADate);
1295 	putushort(denp[0].deMDate, ndirent.de_MDate);
1296 	putushort(denp[0].deMTime, ndirent.de_MTime);
1297 	pcl = pdep->de_StartCluster;
1298 	if (FAT32(pmp) && pcl == pmp->pm_rootdirblk)
1299 		pcl = 0;
1300 	putushort(denp[1].deStartCluster, pcl);
1301 	putushort(denp[1].deCDate, ndirent.de_CDate);
1302 	putushort(denp[1].deCTime, ndirent.de_CTime);
1303 	denp[1].deCHundredth = ndirent.de_CHun;
1304 	putushort(denp[1].deADate, ndirent.de_ADate);
1305 	putushort(denp[1].deMDate, ndirent.de_MDate);
1306 	putushort(denp[1].deMTime, ndirent.de_MTime);
1307 	if (FAT32(pmp)) {
1308 		putushort(denp[0].deHighClust, newcluster >> 16);
1309 		putushort(denp[1].deHighClust, pdep->de_StartCluster >> 16);
1310 	}
1311 
1312 	error = bwrite(bp);
1313 	if (error)
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 	error = uniqdosname(pdep, cnp, ndirent.de_Name);
1322 	if (error)
1323 		goto bad;
1324 
1325 	ndirent.de_Attributes = ATTR_DIRECTORY;
1326 	ndirent.de_LowerCase = 0;
1327 	ndirent.de_StartCluster = newcluster;
1328 	ndirent.de_FileSize = 0;
1329 	ndirent.de_dev = pdep->de_dev;
1330 	ndirent.de_devvp = pdep->de_devvp;
1331 	error = createde(&ndirent, pdep, &dep, cnp);
1332 	if (error)
1333 		goto bad;
1334 	*ap->a_vpp = DETOV(dep);
1335 	return (0);
1336 
1337 bad:
1338 	clusterfree(pmp, newcluster, NULL);
1339 bad2:
1340 	return (error);
1341 }
1342 
1343 static int
1344 msdosfs_rmdir(struct vop_old_rmdir_args *ap)
1345 {
1346 	struct vnode *vp = ap->a_vp;
1347 	struct vnode *dvp = ap->a_dvp;
1348 	struct denode *ip, *dp;
1349 	int error;
1350 
1351 	ip = VTODE(vp);
1352 	dp = VTODE(dvp);
1353 
1354 	/*
1355 	 * Verify the directory is empty (and valid).
1356 	 * (Rmdir ".." won't be valid since
1357 	 *  ".." will contain a reference to
1358 	 *  the current directory and thus be
1359 	 *  non-empty.)
1360 	 */
1361 	error = 0;
1362 	if (!dosdirempty(ip) || ip->de_flag & DE_RENAME) {
1363 		error = ENOTEMPTY;
1364 		goto out;
1365 	}
1366 	/*
1367 	 * Delete the entry from the directory.  For dos filesystems this
1368 	 * gets rid of the directory entry on disk, the in memory copy
1369 	 * still exists but the de_refcnt is <= 0.  This prevents it from
1370 	 * being found by deget().  When the vput() on dep is done we give
1371 	 * up access and eventually msdosfs_reclaim() will be called which
1372 	 * will remove it from the denode cache.
1373 	 */
1374 	error = removede(dp, ip);
1375 	if (error)
1376 		goto out;
1377 	/*
1378 	 * This is where we decrement the link count in the parent
1379 	 * directory.  Since dos filesystems don't do this we just purge
1380 	 * the name cache.
1381 	 */
1382 	vn_unlock(dvp);
1383 	/*
1384 	 * Truncate the directory that is being deleted.
1385 	 */
1386 	error = detrunc(ip, (u_long)0, IO_SYNC);
1387 
1388 	vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1389 out:
1390 	return (error);
1391 }
1392 
1393 /*
1394  * DOS filesystems don't know what symlinks are.
1395  */
1396 static int
1397 msdosfs_symlink(struct vop_old_symlink_args *ap)
1398 {
1399 	return (EOPNOTSUPP);
1400 }
1401 
1402 static int
1403 msdosfs_readdir(struct vop_readdir_args *ap)
1404 {
1405 	struct mbnambuf nb;
1406 	int error = 0;
1407 	int diff;
1408 	long n;
1409 	int blsize;
1410 	long on;
1411 	u_long cn;
1412 	u_long dirsperblk;
1413 	long bias = 0;
1414 	daddr_t bn, lbn;
1415 	struct buf *bp;
1416 	struct denode *dep;
1417 	struct msdosfsmount *pmp;
1418 	struct direntry *dentp;
1419 	struct uio *uio = ap->a_uio;
1420 	off_t *cookies = NULL;
1421 	int ncookies = 0;
1422 	off_t offset, off;
1423 	int chksum = -1;
1424 	ino_t d_ino;
1425 	uint16_t d_namlen;
1426 	uint8_t d_type;
1427 	char *d_name_storage = NULL;
1428 	char *d_name = NULL;
1429 
1430 	error = vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY | LK_FAILRECLAIM);
1431 	if (error)
1432 		return (error);
1433 
1434 	dep = VTODE(ap->a_vp);
1435 	pmp = dep->de_pmp;
1436 
1437 	mprintf("msdosfs_readdir(): vp %p, uio %p, cred %p, eofflagp %p\n",
1438 	    ap->a_vp, uio, ap->a_cred, ap->a_eofflag);
1439 
1440 	/*
1441 	 * msdosfs_readdir() won't operate properly on regular files since
1442 	 * it does i/o only with the the filesystem vnode, and hence can
1443 	 * retrieve the wrong block from the buffer cache for a plain file.
1444 	 * So, fail attempts to readdir() on a plain file.
1445 	 */
1446 	if ((dep->de_Attributes & ATTR_DIRECTORY) == 0) {
1447 		error = ENOTDIR;
1448 		goto done;
1449 	}
1450 
1451 	/*
1452 	 * If the user buffer is smaller than the size of one dos directory
1453 	 * entry or the file offset is not a multiple of the size of a
1454 	 * directory entry, then we fail the read.
1455 	 */
1456 	off = offset = uio->uio_offset;
1457 	if (uio->uio_resid < sizeof(struct direntry) ||
1458 	    (offset & (sizeof(struct direntry) - 1))) {
1459 		error = EINVAL;
1460 		goto done;
1461 	}
1462 
1463 	if (ap->a_ncookies) {
1464 		ncookies = uio->uio_resid / 16 + 1;
1465 		if (ncookies > 1024)
1466 			ncookies = 1024;
1467 		cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
1468 		*ap->a_cookies = cookies;
1469 		*ap->a_ncookies = ncookies;
1470 	}
1471 
1472 	dirsperblk = pmp->pm_BytesPerSec / sizeof(struct direntry);
1473 
1474 	/*
1475 	 * If they are reading from the root directory then, we simulate
1476 	 * the . and .. entries since these don't exist in the root
1477 	 * directory.  We also set the offset bias to make up for having to
1478 	 * simulate these entries. By this I mean that at file offset 64 we
1479 	 * read the first entry in the root directory that lives on disk.
1480 	 */
1481 	if (dep->de_StartCluster == MSDOSFSROOT
1482 	    || (FAT32(pmp) && dep->de_StartCluster == pmp->pm_rootdirblk)) {
1483 		bias = 2 * sizeof(struct direntry);
1484 		if (offset < bias) {
1485 			for (n = (int)offset / sizeof(struct direntry); n < 2;
1486 			     n++) {
1487 				if (FAT32(pmp))
1488 					d_ino = xcntobn(pmp, pmp->pm_rootdirblk)
1489 					    * dirsperblk;
1490 				else
1491 					d_ino = 1;
1492 				d_type = DT_DIR;
1493 				if (n == 0) {
1494 					d_namlen = 1;
1495 					d_name = ".";
1496 				} else if (n == 1) {
1497 					d_namlen = 2;
1498 					d_name = "..";
1499 				}
1500 				if (vop_write_dirent(&error, uio, d_ino, d_type,
1501 				    d_namlen, d_name))
1502 					goto out;
1503 				if (error)
1504 					goto out;
1505 				offset += sizeof(struct direntry);
1506 				off = offset;
1507 				if (cookies) {
1508 					*cookies++ = offset;
1509 					if (--ncookies <= 0)
1510 						goto out;
1511 				}
1512 			}
1513 		}
1514 	}
1515 
1516 	d_name_storage = kmalloc(WIN_MAXLEN, M_TEMP, M_WAITOK);
1517 	mbnambuf_init(&nb);
1518 	off = offset;
1519 
1520 	while (uio->uio_resid > 0) {
1521 		lbn = de_off2cn(pmp, offset - bias);
1522 		on = (offset - bias) & pmp->pm_crbomask;
1523 		n = szmin(pmp->pm_bpcluster - on, uio->uio_resid);
1524 		diff = dep->de_FileSize - (offset - bias);
1525 		if (diff <= 0)
1526 			break;
1527 		n = min(n, diff);
1528 		error = pcbmap(dep, lbn, &bn, &cn, &blsize);
1529 		if (error)
1530 			break;
1531 		error = bread(pmp->pm_devvp, de_bntodoff(pmp, bn), blsize, &bp);
1532 		if (error) {
1533 			brelse(bp);
1534 			kfree(d_name_storage, M_TEMP);
1535 			goto done;
1536 		}
1537 		n = min(n, blsize - bp->b_resid);
1538 
1539 		/*
1540 		 * Convert from dos directory entries to fs-independent
1541 		 * directory entries.
1542 		 */
1543 		for (dentp = (struct direntry *)(bp->b_data + on);
1544 		     (char *)dentp < bp->b_data + on + n;
1545 		     dentp++, offset += sizeof(struct direntry)) {
1546 			d_name = d_name_storage;
1547 			d_namlen = 0;
1548 			/*
1549 			 * If this is an unused entry, we can stop.
1550 			 */
1551 			if (dentp->deName[0] == SLOT_EMPTY) {
1552 				brelse(bp);
1553 				goto out;
1554 			}
1555 			/*
1556 			 * Skip deleted entries.
1557 			 */
1558 			if (dentp->deName[0] == SLOT_DELETED) {
1559 				chksum = -1;
1560 				mbnambuf_init(&nb);
1561 				continue;
1562 			}
1563 			/*
1564 			 * Handle Win95 long directory entries
1565 			 */
1566 			if (dentp->deAttributes == ATTR_WIN95) {
1567 				if (pmp->pm_flags & MSDOSFSMNT_SHORTNAME)
1568 					continue;
1569 				chksum = win2unixfn(&nb,
1570 					(struct winentry *)dentp,
1571 					chksum, pmp);
1572 				continue;
1573 			}
1574 
1575 			/*
1576 			 * Skip volume labels
1577 			 */
1578 			if (dentp->deAttributes & ATTR_VOLUME) {
1579 				chksum = -1;
1580 				mbnambuf_init(&nb);
1581 				continue;
1582 			}
1583 			/*
1584 			 * This computation of d_ino must match
1585 			 * the computation of va_fileid in
1586 			 * msdosfs_getattr.
1587 			 */
1588 			if (dentp->deAttributes & ATTR_DIRECTORY) {
1589 				d_ino = getushort(dentp->deStartCluster);
1590 				if (FAT32(pmp))
1591 					d_ino |= getushort(dentp->deHighClust) << 16;
1592 				/* if this is the root directory */
1593 				if (d_ino != MSDOSFSROOT)
1594 					d_ino = xcntobn(pmp, d_ino) * dirsperblk;
1595 				else if (FAT32(pmp))
1596 					d_ino = xcntobn(pmp, pmp->pm_rootdirblk)
1597 					    * dirsperblk;
1598 				else
1599 					d_ino = 1;
1600 				d_type = DT_DIR;
1601 			} else {
1602 				d_ino = offset / sizeof(struct direntry);
1603 				d_type = DT_REG;
1604 			}
1605 			if (chksum != winChksum(dentp->deName)) {
1606 				d_namlen = dos2unixfn(dentp->deName,
1607 				    (u_char *)d_name,
1608 				    dentp->deLowerCase |
1609 					((pmp->pm_flags & MSDOSFSMNT_SHORTNAME) ?
1610 					(LCASE_BASE | LCASE_EXT) : 0),
1611 					pmp);
1612 				mbnambuf_init(&nb);
1613 			} else {
1614 				mbnambuf_flush(&nb, d_name, &d_namlen);
1615 			}
1616 			chksum = -1;
1617 			if (vop_write_dirent(&error, uio, d_ino, d_type,
1618 			    d_namlen, d_name)) {
1619 				brelse(bp);
1620 				goto out;
1621 			}
1622 			if (error) {
1623 				brelse(bp);
1624 				goto out;
1625 			}
1626 
1627 			if (cookies) {
1628 				*cookies++ = offset + sizeof(struct direntry);
1629 				if (--ncookies <= 0) {
1630 					brelse(bp);
1631 					goto out;
1632 				}
1633 			}
1634 			off = offset + sizeof(struct direntry);
1635 		}
1636 		brelse(bp);
1637 	}
1638 out:
1639 	if (d_name_storage != NULL)
1640 		kfree(d_name_storage, M_TEMP);
1641 
1642 	/* Subtract unused cookies */
1643 	if (ap->a_ncookies)
1644 		*ap->a_ncookies -= ncookies;
1645 
1646 	uio->uio_offset = off;
1647 
1648 	/*
1649 	 * Set the eofflag (NFS uses it)
1650 	 */
1651 	if (ap->a_eofflag) {
1652 		if (dep->de_FileSize - (offset - bias) <= 0)
1653 			*ap->a_eofflag = 1;
1654 		else
1655 			*ap->a_eofflag = 0;
1656 	}
1657 done:
1658 	vn_unlock(ap->a_vp);
1659 	return (error);
1660 }
1661 
1662 /*
1663  * vp  - address of vnode file the file
1664  * bn  - which cluster we are interested in mapping to a filesystem block number.
1665  * vpp - returns the vnode for the block special file holding the filesystem
1666  *	 containing the file of interest
1667  * bnp - address of where to return the filesystem relative block number
1668  */
1669 static int
1670 msdosfs_bmap(struct vop_bmap_args *ap)
1671 {
1672 	struct denode *dep = VTODE(ap->a_vp);
1673 	struct msdosfsmount *pmp = dep->de_pmp;
1674 	daddr_t lbn;
1675 	daddr_t dbn;
1676 	int error;
1677 
1678 	if (ap->a_doffsetp == NULL)
1679 		return (0);
1680 	if (ap->a_runp) {
1681 		/*
1682 		 * Sequential clusters should be counted here.
1683 		 */
1684 		*ap->a_runp = 0;
1685 	}
1686 	if (ap->a_runb) {
1687 		*ap->a_runb = 0;
1688 	}
1689 	KKASSERT(((int)ap->a_loffset & ((1 << pmp->pm_cnshift) - 1)) == 0);
1690 	lbn = de_off2cn(pmp, ap->a_loffset);
1691 	error = pcbmap(dep, lbn, &dbn, NULL, NULL);
1692 	if (error || dbn == (daddr_t)-1) {
1693 		*ap->a_doffsetp = NOOFFSET;
1694 	} else {
1695 		*ap->a_doffsetp = de_bntodoff(pmp, dbn);
1696 	}
1697 	return (error);
1698 }
1699 
1700 static int
1701 msdosfs_strategy(struct vop_strategy_args *ap)
1702 {
1703 	struct bio *bio = ap->a_bio;
1704 	struct bio *nbio;
1705 	struct buf *bp = bio->bio_buf;
1706 	struct vnode *vp = ap->a_vp;
1707 	struct denode *dep = VTODE(vp);
1708 	struct msdosfsmount *pmp = dep->de_pmp;
1709 	int error = 0;
1710 	daddr_t dblkno;
1711 
1712 	if (vp->v_type == VBLK || vp->v_type == VCHR)
1713 		panic("msdosfs_strategy: spec");
1714 	/*
1715 	 * If we don't already know the filesystem relative block number
1716 	 * then get it using pcbmap().  If pcbmap() returns the block
1717 	 * number as -1 then we've got a hole in the file.  DOS filesystems
1718 	 * don't allow files with holes, so we shouldn't ever see this.
1719 	 */
1720 	nbio = push_bio(bio);
1721 	if (nbio->bio_offset == NOOFFSET) {
1722 		error = pcbmap(dep, de_off2cn(pmp, bio->bio_offset),
1723 			       &dblkno, NULL, NULL);
1724 		if (error) {
1725 			bp->b_error = error;
1726 			bp->b_flags |= B_ERROR;
1727 			/* I/O was never started on nbio, must biodone(bio) */
1728 			biodone(bio);
1729 			return (error);
1730 		}
1731 		if (dblkno == (daddr_t)-1) {
1732 			nbio->bio_offset = NOOFFSET;
1733 			vfs_bio_clrbuf(bp);
1734 		} else {
1735 			nbio->bio_offset = de_bntodoff(pmp, dblkno);
1736 		}
1737 	}
1738 	if (nbio->bio_offset == NOOFFSET) {
1739 		/* I/O was never started on nbio, must biodone(bio) */
1740 		biodone(bio);
1741 		return (0);
1742 	}
1743 	/*
1744 	 * Read/write the block from/to the disk that contains the desired
1745 	 * file block.
1746 	 */
1747 	vn_strategy(dep->de_devvp, nbio);
1748 	return (0);
1749 }
1750 
1751 static int
1752 msdosfs_print(struct vop_print_args *ap)
1753 {
1754 	struct denode *dep = VTODE(ap->a_vp);
1755 
1756 	kprintf("tag VT_MSDOSFS, startcluster %lu, dircluster %lu, "
1757 		"diroffset %lu, dev %d, %d",
1758 		dep->de_StartCluster, dep->de_dirclust, dep->de_diroffset,
1759 		major(dep->de_dev), minor(dep->de_dev));
1760 	lockmgr_printinfo(&ap->a_vp->v_lock);
1761 	return (0);
1762 }
1763 
1764 static int
1765 msdosfs_pathconf(struct vop_pathconf_args *ap)
1766 {
1767 	struct msdosfsmount *pmp = VTODE(ap->a_vp)->de_pmp;
1768 
1769 	switch (ap->a_name) {
1770 	case _PC_CHOWN_RESTRICTED:
1771 		*ap->a_retval = 1;
1772 		return (0);
1773 	case _PC_FILESIZEBITS:
1774 		*ap->a_retval = FAT12(pmp) ? 32 : 33;
1775 		return (0);
1776 	case _PC_LINK_MAX:
1777 		*ap->a_retval = 1;
1778 		return (0);
1779 	case _PC_NAME_MAX:
1780 		*ap->a_retval =
1781 		    pmp->pm_flags & MSDOSFSMNT_LONGNAME ? WIN_MAXLEN : 12;
1782 		return (0);
1783 	case _PC_NO_TRUNC:
1784 		*ap->a_retval = 0;
1785 		return (0);
1786 	case _PC_PATH_MAX:
1787 		*ap->a_retval = PATH_MAX;
1788 		return (0);
1789 	default:
1790 		return (EINVAL);
1791 	}
1792 	/* NOTREACHED */
1793 }
1794 
1795 /* Global vfs data structures for msdosfs */
1796 struct vop_ops msdosfs_vnode_vops = {
1797 	.vop_default =		vop_defaultop,
1798 	.vop_access =		msdosfs_access,
1799 	.vop_bmap =		msdosfs_bmap,
1800 	.vop_old_lookup =	msdosfs_lookup,
1801 	.vop_open =		msdosfs_open,
1802 	.vop_close =		msdosfs_close,
1803 	.vop_old_create =	msdosfs_create,
1804 	.vop_fsync =		msdosfs_fsync,
1805 	.vop_getattr =		msdosfs_getattr,
1806 	.vop_inactive =		msdosfs_inactive,
1807 	.vop_old_link =		msdosfs_link,
1808 	.vop_old_mkdir =	msdosfs_mkdir,
1809 	.vop_old_mknod =	msdosfs_mknod,
1810 	.vop_pathconf =		msdosfs_pathconf,
1811 	.vop_print =		msdosfs_print,
1812 	.vop_read =		msdosfs_read,
1813 	.vop_readdir =		msdosfs_readdir,
1814 	.vop_reclaim =		msdosfs_reclaim,
1815 	.vop_old_remove =	msdosfs_remove,
1816 	.vop_old_rename =	msdosfs_rename,
1817 	.vop_old_rmdir =	msdosfs_rmdir,
1818 	.vop_setattr =		msdosfs_setattr,
1819 	.vop_strategy =		msdosfs_strategy,
1820 	.vop_old_symlink =	msdosfs_symlink,
1821 	.vop_write =		msdosfs_write,
1822 	.vop_getpages =		vop_stdgetpages,
1823 	.vop_putpages =		vop_stdputpages
1824 };
1825