1 /* $OpenBSD: msdosfs_vnops.c,v 1.64 2007/06/02 02:04:21 deraadt 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/kernel.h>
57 #include <sys/file.h> /* define FWRITE ... */
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/signalvar.h>
64 #include <miscfs/specfs/specdev.h> /* XXX */ /* defines v_rdev */
65 #include <sys/malloc.h>
66 #include <sys/pool.h>
67 #include <sys/dirent.h> /* defines dirent structure */
68 #include <sys/lockf.h>
69 #include <sys/poll.h>
70
71 #include <uvm/uvm_extern.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 /*
82 * Some general notes:
83 *
84 * In the ufs filesystem the inodes, superblocks, and indirect blocks are
85 * read/written using the vnode for the filesystem. Blocks that represent
86 * the contents of a file are read/written using the vnode for the file
87 * (including directories when they are read/written as files). This
88 * presents problems for the dos filesystem because data that should be in
89 * an inode (if dos had them) resides in the directory itself. Since we
90 * must update directory entries without the benefit of having the vnode
91 * for the directory we must use the vnode for the filesystem. This means
92 * that when a directory is actually read/written (via read, write, or
93 * readdir, or seek) we must use the vnode for the filesystem instead of
94 * the vnode for the directory as would happen in ufs. This is to insure we
95 * retrieve the correct block from the buffer cache since the hash value is
96 * based upon the vnode address and the desired block number.
97 */
98
99 /*
100 * Create a regular file. On entry the directory to contain the file being
101 * created is locked. We must release before we return. We must also free
102 * the pathname buffer pointed at by cnp->cn_pnbuf, always on error, or
103 * only if the SAVESTART bit in cn_flags is clear on success.
104 */
105 int
106 msdosfs_create(v)
107 void *v;
108 {
109 struct vop_create_args *ap = v;
110 struct componentname *cnp = ap->a_cnp;
111 struct denode ndirent;
112 struct denode *dep;
113 struct denode *pdep = VTODE(ap->a_dvp);
114 int error;
115 struct timespec ts;
116
117 #ifdef MSDOSFS_DEBUG
118 printf("msdosfs_create(cnp %08x, vap %08x\n", cnp, ap->a_vap);
119 #endif
120
121 /*
122 * If this is the root directory and there is no space left we
123 * can't do anything. This is because the root directory can not
124 * change size.
125 */
126 if (pdep->de_StartCluster == MSDOSFSROOT
127 && pdep->de_fndoffset >= pdep->de_FileSize) {
128 error = ENOSPC;
129 goto bad;
130 }
131
132 /*
133 * Create a directory entry for the file, then call createde() to
134 * have it installed. NOTE: DOS files are always executable. We
135 * use the absence of the owner write bit to make the file
136 * readonly.
137 */
138 #ifdef DIAGNOSTIC
139 if ((cnp->cn_flags & HASBUF) == 0)
140 panic("msdosfs_create: no name");
141 #endif
142 bzero(&ndirent, sizeof(ndirent));
143 if ((error = uniqdosname(pdep, cnp, ndirent.de_Name)) != 0)
144 goto bad;
145
146 ndirent.de_Attributes = (ap->a_vap->va_mode & VWRITE) ?
147 ATTR_ARCHIVE : ATTR_ARCHIVE | ATTR_READONLY;
148 ndirent.de_StartCluster = 0;
149 ndirent.de_FileSize = 0;
150 ndirent.de_dev = pdep->de_dev;
151 ndirent.de_devvp = pdep->de_devvp;
152 ndirent.de_pmp = pdep->de_pmp;
153 ndirent.de_flag = DE_ACCESS | DE_CREATE | DE_UPDATE;
154 getnanotime(&ts);
155 DETIMES(&ndirent, &ts, &ts, &ts);
156 if ((error = createde(&ndirent, pdep, &dep, cnp)) != 0)
157 goto bad;
158 if ((cnp->cn_flags & SAVESTART) == 0)
159 pool_put(&namei_pool, cnp->cn_pnbuf);
160 vput(ap->a_dvp);
161 *ap->a_vpp = DETOV(dep);
162 return (0);
163
164 bad:
165 pool_put(&namei_pool, cnp->cn_pnbuf);
166 vput(ap->a_dvp);
167 return (error);
168 }
169
170 int
171 msdosfs_mknod(v)
172 void *v;
173 {
174 struct vop_mknod_args *ap = v;
175
176 pool_put(&namei_pool, ap->a_cnp->cn_pnbuf);
177 vput(ap->a_dvp);
178 return (EINVAL);
179 }
180
181 int
182 msdosfs_open(v)
183 void *v;
184 {
185 #if 0
186 struct vop_open_args /* {
187 struct vnode *a_vp;
188 int a_mode;
189 struct ucred *a_cred;
190 struct proc *a_p;
191 } */ *ap;
192 #endif
193
194 return (0);
195 }
196
197 int
198 msdosfs_close(v)
199 void *v;
200 {
201 struct vop_close_args *ap = v;
202 struct vnode *vp = ap->a_vp;
203 struct denode *dep = VTODE(vp);
204 struct timespec ts;
205
206 if (vp->v_usecount > 1 && !VOP_ISLOCKED(vp)) {
207 getnanotime(&ts);
208 DETIMES(dep, &ts, &ts, &ts);
209 }
210 return (0);
211 }
212
213 int
214 msdosfs_access(v)
215 void *v;
216 {
217 struct vop_access_args *ap = v;
218 struct denode *dep = VTODE(ap->a_vp);
219 struct msdosfsmount *pmp = dep->de_pmp;
220 mode_t dosmode;
221
222 dosmode = (S_IXUSR|S_IXGRP|S_IXOTH) | (S_IRUSR|S_IRGRP|S_IROTH);
223 if ((dep->de_Attributes & ATTR_READONLY) == 0)
224 dosmode |= (S_IWUSR|S_IWGRP|S_IWOTH);
225 dosmode &= pmp->pm_mask;
226 if (dep->de_Attributes & ATTR_DIRECTORY
227 && pmp->pm_flags & MSDOSFSMNT_ALLOWDIRX) {
228 dosmode |= (dosmode & S_IRUSR) ? S_IXUSR : 0;
229 dosmode |= (dosmode & S_IRGRP) ? S_IXGRP : 0;
230 dosmode |= (dosmode & S_IROTH) ? S_IXOTH : 0;
231 }
232
233 return (vaccess(dosmode, pmp->pm_uid, pmp->pm_gid, ap->a_mode,
234 ap->a_cred));
235 }
236
237 int
238 msdosfs_getattr(v)
239 void *v;
240 {
241 struct vop_getattr_args *ap = v;
242 struct denode *dep = VTODE(ap->a_vp);
243 struct msdosfsmount *pmp = dep->de_pmp;
244 struct vattr *vap = ap->a_vap;
245 struct timespec ts;
246 uint32_t fileid;
247
248 getnanotime(&ts);
249 DETIMES(dep, &ts, &ts, &ts);
250 vap->va_fsid = dep->de_dev;
251
252 /*
253 * The following computation of the fileid must be the same as
254 * that used in msdosfs_readdir() to compute d_fileno. If not,
255 * pwd doesn't work.
256 *
257 * We now use the starting cluster number as the fileid/fileno.
258 * This works for both files and directories (including the root
259 * directory, on FAT32). Even on FAT32, this will at most be a
260 * 28-bit number, as the high 4 bits of FAT32 cluster numbers
261 * are reserved.
262 *
263 * However, we do need to do something for 0-length files, which
264 * will not have a starting cluster number.
265 *
266 * These files cannot be directories, since (except for /, which
267 * is special-cased anyway) directories contain entries for . and
268 * .., so must have non-zero length.
269 *
270 * In this case, we just create a non-cryptographic hash of the
271 * original fileid calculation, and set the top bit.
272 *
273 * This algorithm has the benefit that all directories, and all
274 * non-zero-length files, will have fileids that are persistent
275 * across mounts and reboots, and that cannot collide (as long
276 * as the filesystem is not corrupt). Zero-length files will
277 * have fileids that are persistent, but that may collide. We
278 * will just have to live with that.
279 */
280 fileid = dep->de_StartCluster;
281
282 if (dep->de_Attributes & ATTR_DIRECTORY) {
283 /* Special-case root */
284 if (dep->de_StartCluster == MSDOSFSROOT)
285 fileid = FAT32(pmp) ? pmp->pm_rootdirblk : 1;
286 } else {
287 if (dep->de_FileSize == 0) {
288 uint32_t dirsperblk;
289 uint64_t fileid64;
290
291 dirsperblk = pmp->pm_BytesPerSec /
292 sizeof(struct direntry);
293
294 fileid64 = (dep->de_dirclust == MSDOSFSROOT) ?
295 roottobn(pmp, 0) : cntobn(pmp, dep->de_dirclust);
296 fileid64 *= dirsperblk;
297 fileid64 += dep->de_diroffset / sizeof(struct direntry);
298
299 fileid = fileidhash(fileid64);
300 }
301 }
302
303 vap->va_fileid = fileid;
304 vap->va_mode = (S_IXUSR|S_IXGRP|S_IXOTH) | (S_IRUSR|S_IRGRP|S_IROTH) |
305 ((dep->de_Attributes & ATTR_READONLY) ? 0 : (S_IWUSR|S_IWGRP|S_IWOTH));
306 vap->va_mode &= dep->de_pmp->pm_mask;
307 if (dep->de_Attributes & ATTR_DIRECTORY) {
308 vap->va_mode |= S_IFDIR;
309 if (pmp->pm_flags & MSDOSFSMNT_ALLOWDIRX) {
310 vap->va_mode |= (vap->va_mode & S_IRUSR) ? S_IXUSR : 0;
311 vap->va_mode |= (vap->va_mode & S_IRGRP) ? S_IXGRP : 0;
312 vap->va_mode |= (vap->va_mode & S_IROTH) ? S_IXOTH : 0;
313 }
314 }
315 vap->va_nlink = 1;
316 vap->va_gid = dep->de_pmp->pm_gid;
317 vap->va_uid = dep->de_pmp->pm_uid;
318 vap->va_rdev = 0;
319 vap->va_size = dep->de_FileSize;
320 dos2unixtime(dep->de_MDate, dep->de_MTime, 0, &vap->va_mtime);
321 if (dep->de_pmp->pm_flags & MSDOSFSMNT_LONGNAME) {
322 dos2unixtime(dep->de_ADate, 0, 0, &vap->va_atime);
323 dos2unixtime(dep->de_CDate, dep->de_CTime, dep->de_CTimeHundredth, &vap->va_ctime);
324 } else {
325 vap->va_atime = vap->va_mtime;
326 vap->va_ctime = vap->va_mtime;
327 }
328 vap->va_flags = 0;
329 if ((dep->de_Attributes & ATTR_ARCHIVE) == 0)
330 vap->va_flags |= SF_ARCHIVED;
331 vap->va_gen = 0;
332 vap->va_blocksize = dep->de_pmp->pm_bpcluster;
333 vap->va_bytes = (dep->de_FileSize + dep->de_pmp->pm_crbomask) &
334 ~(dep->de_pmp->pm_crbomask);
335 vap->va_type = ap->a_vp->v_type;
336 return (0);
337 }
338
339 int
340 msdosfs_setattr(v)
341 void *v;
342 {
343 struct vop_setattr_args *ap = v;
344 int error = 0;
345 struct denode *dep = VTODE(ap->a_vp);
346 struct vattr *vap = ap->a_vap;
347 struct ucred *cred = ap->a_cred;
348
349 #ifdef MSDOSFS_DEBUG
350 printf("msdosfs_setattr(): vp %08x, vap %08x, cred %08x, p %08x\n",
351 ap->a_vp, vap, cred, ap->a_p);
352 #endif
353 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
354 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
355 (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
356 (vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL) ||
357 (vap->va_uid != VNOVAL) || (vap->va_gid != VNOVAL)) {
358 #ifdef MSDOSFS_DEBUG
359 printf("msdosfs_setattr(): returning EINVAL\n");
360 printf(" va_type %d, va_nlink %x, va_fsid %x, va_fileid %x\n",
361 vap->va_type, vap->va_nlink, vap->va_fsid, vap->va_fileid);
362 printf(" va_blocksize %x, va_rdev %x, va_bytes %x, va_gen %x\n",
363 vap->va_blocksize, vap->va_rdev, vap->va_bytes, vap->va_gen);
364 printf(" va_uid %x, va_gid %x\n",
365 vap->va_uid, vap->va_gid);
366 #endif
367 return (EINVAL);
368 }
369 /*
370 * Directories must not ever get their attributes modified
371 */
372 if (ap->a_vp->v_type == VDIR)
373 return EISDIR;
374
375 if (vap->va_size != VNOVAL) {
376 error = detrunc(dep, (uint32_t)vap->va_size, 0, cred, ap->a_p);
377 if (error)
378 return (error);
379 }
380 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
381 if (cred->cr_uid != dep->de_pmp->pm_uid &&
382 (error = suser_ucred(cred)) &&
383 ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
384 (error = VOP_ACCESS(ap->a_vp, VWRITE, cred, ap->a_p))))
385 return (error);
386 if (!(dep->de_pmp->pm_flags & MSDOSFSMNT_NOWIN95)
387 && vap->va_atime.tv_sec != VNOVAL)
388 unix2dostime(&vap->va_atime, &dep->de_ADate, NULL, NULL);
389 if (vap->va_mtime.tv_sec != VNOVAL)
390 unix2dostime(&vap->va_mtime, &dep->de_MDate, &dep->de_MTime, NULL);
391 dep->de_Attributes |= ATTR_ARCHIVE;
392 dep->de_flag |= DE_MODIFIED;
393 }
394 /*
395 * DOS files only have the ability to have their writability
396 * attribute set, so we use the owner write bit to set the readonly
397 * attribute.
398 */
399 if (vap->va_mode != (mode_t)VNOVAL) {
400 if (cred->cr_uid != dep->de_pmp->pm_uid &&
401 (error = suser_ucred(cred)))
402 return (error);
403 /* We ignore the read and execute bits. */
404 if (vap->va_mode & VWRITE)
405 dep->de_Attributes &= ~ATTR_READONLY;
406 else
407 dep->de_Attributes |= ATTR_READONLY;
408 dep->de_flag |= DE_MODIFIED;
409 }
410 /*
411 * Allow the `archived' bit to be toggled.
412 */
413 if (vap->va_flags != VNOVAL) {
414 if (cred->cr_uid != dep->de_pmp->pm_uid &&
415 (error = suser_ucred(cred)))
416 return (error);
417 if (vap->va_flags & SF_ARCHIVED)
418 dep->de_Attributes &= ~ATTR_ARCHIVE;
419 else
420 dep->de_Attributes |= ATTR_ARCHIVE;
421 dep->de_flag |= DE_MODIFIED;
422 }
423 return (deupdat(dep, 1));
424 }
425
426 int
427 msdosfs_read(v)
428 void *v;
429 {
430 struct vop_read_args *ap = v;
431 int error = 0;
432 uint32_t diff;
433 int blsize;
434 int isadir;
435 uint32_t n;
436 long on;
437 daddr64_t lbn, rablock, rablkno;
438 struct buf *bp;
439 struct vnode *vp = ap->a_vp;
440 struct denode *dep = VTODE(vp);
441 struct msdosfsmount *pmp = dep->de_pmp;
442 struct uio *uio = ap->a_uio;
443
444 /*
445 * If they didn't ask for any data, then we are done.
446 */
447 if (uio->uio_resid == 0)
448 return (0);
449 if (uio->uio_offset < 0)
450 return (EINVAL);
451
452 isadir = dep->de_Attributes & ATTR_DIRECTORY;
453 do {
454 if (uio->uio_offset >= dep->de_FileSize)
455 return (0);
456
457 lbn = de_cluster(pmp, uio->uio_offset);
458 on = uio->uio_offset & pmp->pm_crbomask;
459 n = min((uint32_t) (pmp->pm_bpcluster - on), uio->uio_resid);
460
461 /*
462 * de_FileSize is uint32_t, and we know that uio_offset <
463 * de_FileSize, so uio->uio_offset < 2^32. Therefore
464 * the cast to uint32_t on the next line is safe.
465 */
466 diff = dep->de_FileSize - (uint32_t)uio->uio_offset;
467 if (diff < n)
468 n = diff;
469
470 /* convert cluster # to block # if a directory */
471 if (isadir) {
472 error = pcbmap(dep, lbn, &lbn, 0, &blsize);
473 if (error)
474 return (error);
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 error = bread(pmp->pm_devvp, lbn, blsize, NOCRED, &bp);
483 } else {
484 rablock = lbn + 1;
485 rablkno = de_cn2bn(pmp, rablock);
486 if (dep->de_lastr + 1 == lbn &&
487 de_cn2off(pmp, rablock) < dep->de_FileSize)
488 error = breadn(vp, de_cn2bn(pmp, lbn),
489 pmp->pm_bpcluster, &rablkno,
490 &pmp->pm_bpcluster, 1, NOCRED, &bp);
491 else
492 error = bread(vp, de_cn2bn(pmp, lbn),
493 pmp->pm_bpcluster, NOCRED, &bp);
494 dep->de_lastr = lbn;
495 }
496 n = min(n, pmp->pm_bpcluster - bp->b_resid);
497 if (error) {
498 brelse(bp);
499 return (error);
500 }
501 error = uiomove(bp->b_data + on, (int) n, uio);
502 brelse(bp);
503 } while (error == 0 && uio->uio_resid > 0 && n != 0);
504 if (!isadir && !(vp->v_mount->mnt_flag & MNT_NOATIME))
505 dep->de_flag |= DE_ACCESS;
506 return (error);
507 }
508
509 /*
510 * Write data to a file or directory.
511 */
512 int
513 msdosfs_write(v)
514 void *v;
515 {
516 struct vop_write_args *ap = v;
517 int n;
518 int croffset;
519 int resid;
520 uint32_t osize;
521 int error = 0;
522 uint32_t count, lastcn;
523 daddr64_t bn;
524 struct buf *bp;
525 int ioflag = ap->a_ioflag;
526 struct uio *uio = ap->a_uio;
527 struct proc *p = uio->uio_procp;
528 struct vnode *vp = ap->a_vp;
529 struct vnode *thisvp;
530 struct denode *dep = VTODE(vp);
531 struct msdosfsmount *pmp = dep->de_pmp;
532 struct ucred *cred = ap->a_cred;
533
534 #ifdef MSDOSFS_DEBUG
535 printf("msdosfs_write(vp %08x, uio %08x, ioflag %08x, cred %08x\n",
536 vp, uio, ioflag, cred);
537 printf("msdosfs_write(): diroff %d, dirclust %d, startcluster %d\n",
538 dep->de_diroffset, dep->de_dirclust, dep->de_StartCluster);
539 #endif
540
541 switch (vp->v_type) {
542 case VREG:
543 if (ioflag & IO_APPEND)
544 uio->uio_offset = dep->de_FileSize;
545 thisvp = vp;
546 break;
547 case VDIR:
548 return EISDIR;
549 default:
550 panic("msdosfs_write(): bad file type");
551 }
552
553 if (uio->uio_offset < 0)
554 return (EINVAL);
555
556 if (uio->uio_resid == 0)
557 return (0);
558
559 /* Don't bother to try to write files larger than the f/s limit */
560 if (uio->uio_offset + uio->uio_resid > MSDOSFS_FILESIZE_MAX)
561 return (EFBIG);
562
563 /*
564 * If they've exceeded their filesize limit, tell them about it.
565 */
566 if (p &&
567 ((uio->uio_offset + uio->uio_resid) >
568 p->p_rlimit[RLIMIT_FSIZE].rlim_cur)) {
569 psignal(p, SIGXFSZ);
570 return (EFBIG);
571 }
572
573 /*
574 * If the offset we are starting the write at is beyond the end of
575 * the file, then they've done a seek. Unix filesystems allow
576 * files with holes in them, DOS doesn't so we must fill the hole
577 * with zeroed blocks.
578 */
579 if (uio->uio_offset > dep->de_FileSize) {
580 if ((error = deextend(dep, uio->uio_offset, cred)) != 0)
581 return (error);
582 }
583
584 /*
585 * Remember some values in case the write fails.
586 */
587 resid = uio->uio_resid;
588 osize = dep->de_FileSize;
589
590 /*
591 * If we write beyond the end of the file, extend it to its ultimate
592 * size ahead of the time to hopefully get a contiguous area.
593 */
594 if (uio->uio_offset + resid > osize) {
595 count = de_clcount(pmp, uio->uio_offset + resid) -
596 de_clcount(pmp, osize);
597 if ((error = extendfile(dep, count, NULL, NULL, 0)) &&
598 (error != ENOSPC || (ioflag & IO_UNIT)))
599 goto errexit;
600 lastcn = dep->de_fc[FC_LASTFC].fc_frcn;
601 } else
602 lastcn = de_clcount(pmp, osize) - 1;
603
604 do {
605 if (de_cluster(pmp, uio->uio_offset) > lastcn) {
606 error = ENOSPC;
607 break;
608 }
609
610 bn = de_blk(pmp, uio->uio_offset);
611 if ((uio->uio_offset & pmp->pm_crbomask) == 0
612 && (de_blk(pmp, uio->uio_offset + uio->uio_resid) > de_blk(pmp, uio->uio_offset)
613 || uio->uio_offset + uio->uio_resid >= dep->de_FileSize)) {
614 /*
615 * If either the whole cluster gets written,
616 * or we write the cluster from its start beyond EOF,
617 * then no need to read data from disk.
618 */
619 bp = getblk(thisvp, bn, pmp->pm_bpcluster, 0, 0);
620 clrbuf(bp);
621 /*
622 * Do the bmap now, since pcbmap needs buffers
623 * for the fat table. (see msdosfs_strategy)
624 */
625 if (bp->b_blkno == bp->b_lblkno) {
626 error = pcbmap(dep,
627 de_bn2cn(pmp, bp->b_lblkno),
628 &bp->b_blkno, 0, 0);
629 if (error)
630 bp->b_blkno = -1;
631 }
632 if (bp->b_blkno == -1) {
633 brelse(bp);
634 if (!error)
635 error = EIO; /* XXX */
636 break;
637 }
638 } else {
639 /*
640 * The block we need to write into exists, so read it in.
641 */
642 error = bread(thisvp, bn, pmp->pm_bpcluster,
643 NOCRED, &bp);
644 if (error) {
645 brelse(bp);
646 break;
647 }
648 }
649
650 croffset = uio->uio_offset & pmp->pm_crbomask;
651 n = min(uio->uio_resid, pmp->pm_bpcluster - croffset);
652 if (uio->uio_offset + n > dep->de_FileSize) {
653 dep->de_FileSize = uio->uio_offset + n;
654 uvm_vnp_setsize(vp, dep->de_FileSize);
655 }
656 uvm_vnp_uncache(vp);
657 /*
658 * Should these vnode_pager_* functions be done on dir
659 * files?
660 */
661
662 /*
663 * Copy the data from user space into the buf header.
664 */
665 error = uiomove(bp->b_data + croffset, n, uio);
666
667 /*
668 * If they want this synchronous then write it and wait for
669 * it. Otherwise, if on a cluster boundary write it
670 * asynchronously so we can move on to the next block
671 * without delay. Otherwise do a delayed write because we
672 * may want to write somemore into the block later.
673 */
674 if (ioflag & IO_SYNC)
675 (void) bwrite(bp);
676 else if (n + croffset == pmp->pm_bpcluster)
677 bawrite(bp);
678 else
679 bdwrite(bp);
680 dep->de_flag |= DE_UPDATE;
681 } while (error == 0 && uio->uio_resid > 0);
682
683 /*
684 * If the write failed and they want us to, truncate the file back
685 * to the size it was before the write was attempted.
686 */
687 errexit:
688 if (error) {
689 if (ioflag & IO_UNIT) {
690 detrunc(dep, osize, ioflag & IO_SYNC, NOCRED, NULL);
691 uio->uio_offset -= resid - uio->uio_resid;
692 uio->uio_resid = resid;
693 } else {
694 detrunc(dep, dep->de_FileSize, ioflag & IO_SYNC, NOCRED, NULL);
695 if (uio->uio_resid != resid)
696 error = 0;
697 }
698 } else if (ioflag & IO_SYNC)
699 error = deupdat(dep, 1);
700 return (error);
701 }
702
703 int
704 msdosfs_ioctl(v)
705 void *v;
706 {
707 #if 0
708 struct vop_ioctl_args /* {
709 struct vnode *a_vp;
710 uint32_t a_command;
711 caddr_t a_data;
712 int a_fflag;
713 struct ucred *a_cred;
714 struct proc *a_p;
715 } */ *ap;
716 #endif
717
718 return (ENOTTY);
719 }
720
721 int
722 msdosfs_poll(v)
723 void *v;
724 {
725 struct vop_poll_args *ap = v;
726
727 return (ap->a_events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
728 }
729
730 /*
731 * Flush the blocks of a file to disk.
732 *
733 * This function is worthless for vnodes that represent directories. Maybe we
734 * could just do a sync if they try an fsync on a directory file.
735 */
736 int
737 msdosfs_fsync(v)
738 void *v;
739 {
740 struct vop_fsync_args *ap = v;
741 struct vnode *vp = ap->a_vp;
742
743 vflushbuf(vp, ap->a_waitfor == MNT_WAIT);
744 return (deupdat(VTODE(vp), ap->a_waitfor == MNT_WAIT));
745 }
746
747 /*
748 * Flush the blocks of a file to disk.
749 *
750 * This function is worthless for vnodes that represent directories. Maybe we
751 * could just do a sync if they try an fsync on a directory file.
752 */
753 int
754 msdosfs_remove(v)
755 void *v;
756 {
757 struct vop_remove_args *ap = v;
758 struct denode *dep = VTODE(ap->a_vp);
759 struct denode *ddep = VTODE(ap->a_dvp);
760 int error;
761
762 if (ap->a_vp->v_type == VDIR)
763 error = EPERM;
764 else
765 error = removede(ddep, dep);
766 #ifdef MSDOSFS_DEBUG
767 printf("msdosfs_remove(), dep %08x, v_usecount %d\n", dep, ap->a_vp->v_usecount);
768 #endif
769 if (ddep == dep)
770 vrele(ap->a_vp);
771 else
772 vput(ap->a_vp); /* causes msdosfs_inactive() to be called
773 * via vrele() */
774 vput(ap->a_dvp);
775 return (error);
776 }
777
778 /*
779 * DOS filesystems don't know what links are. But since we already called
780 * msdosfs_lookup() with create and lockparent, the parent is locked so we
781 * have to free it before we return the error.
782 */
783 int
784 msdosfs_link(v)
785 void *v;
786 {
787 struct vop_link_args *ap = v;
788
789 VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
790 vput(ap->a_dvp);
791 return (EOPNOTSUPP);
792 }
793
794 /*
795 * Renames on files require moving the denode to a new hash queue since the
796 * denode's location is used to compute which hash queue to put the file
797 * in. Unless it is a rename in place. For example "mv a b".
798 *
799 * What follows is the basic algorithm:
800 *
801 * if (file move) {
802 * if (dest file exists) {
803 * remove dest file
804 * }
805 * if (dest and src in same directory) {
806 * rewrite name in existing directory slot
807 * } else {
808 * write new entry in dest directory
809 * update offset and dirclust in denode
810 * move denode to new hash chain
811 * clear old directory entry
812 * }
813 * } else {
814 * directory move
815 * if (dest directory exists) {
816 * if (dest is not empty) {
817 * return ENOTEMPTY
818 * }
819 * remove dest directory
820 * }
821 * if (dest and src in same directory) {
822 * rewrite name in existing entry
823 * } else {
824 * be sure dest is not a child of src directory
825 * write entry in dest directory
826 * update "." and ".." in moved directory
827 * update offset and dirclust in denode
828 * move denode to new hash chain
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 int
848 msdosfs_rename(v)
849 void *v;
850 {
851 struct vop_rename_args *ap = v;
852 struct vnode *tvp = ap->a_tvp;
853 register struct vnode *tdvp = ap->a_tdvp;
854 struct vnode *fvp = ap->a_fvp;
855 register struct vnode *fdvp = ap->a_fdvp;
856 register struct componentname *tcnp = ap->a_tcnp;
857 register struct componentname *fcnp = ap->a_fcnp;
858 struct proc *p = curproc; /* XXX */
859 register struct denode *ip, *xp, *dp, *zp;
860 u_char toname[11], oldname[11];
861 uint32_t from_diroffset, to_diroffset;
862 u_char to_count;
863 int doingdirectory = 0, newparent = 0;
864 int error;
865 uint32_t cn, pcl;
866 daddr64_t bn;
867 struct msdosfsmount *pmp;
868 struct direntry *dotdotp;
869 struct buf *bp;
870
871 pmp = VFSTOMSDOSFS(fdvp->v_mount);
872
873 #ifdef DIAGNOSTIC
874 if ((tcnp->cn_flags & HASBUF) == 0 ||
875 (fcnp->cn_flags & HASBUF) == 0)
876 panic("msdosfs_rename: no name");
877 #endif
878 /*
879 * Check for cross-device rename.
880 */
881 if ((fvp->v_mount != tdvp->v_mount) ||
882 (tvp && (fvp->v_mount != tvp->v_mount))) {
883 error = EXDEV;
884 abortit:
885 VOP_ABORTOP(tdvp, tcnp);
886 if (tdvp == tvp)
887 vrele(tdvp);
888 else
889 vput(tdvp);
890 if (tvp)
891 vput(tvp);
892 VOP_ABORTOP(fdvp, fcnp);
893 vrele(fdvp);
894 vrele(fvp);
895 return (error);
896 }
897
898 /*
899 * If source and dest are the same, do nothing.
900 */
901 if (tvp == fvp) {
902 error = 0;
903 goto abortit;
904 }
905
906 /* */
907 if ((error = vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY, p)) != 0)
908 goto abortit;
909 dp = VTODE(fdvp);
910 ip = VTODE(fvp);
911
912 /*
913 * Be sure we are not renaming ".", "..", or an alias of ".". This
914 * leads to a crippled directory tree. It's pretty tough to do a
915 * "ls" or "pwd" with the "." directory entry missing, and "cd .."
916 * doesn't work if the ".." entry is missing.
917 */
918 if (ip->de_Attributes & ATTR_DIRECTORY) {
919 /*
920 * Avoid ".", "..", and aliases of "." for obvious reasons.
921 */
922 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
923 dp == ip ||
924 (fcnp->cn_flags & ISDOTDOT) ||
925 (tcnp->cn_flags & ISDOTDOT) ||
926 (ip->de_flag & DE_RENAME)) {
927 VOP_UNLOCK(fvp, 0, p);
928 error = EINVAL;
929 goto abortit;
930 }
931 ip->de_flag |= DE_RENAME;
932 doingdirectory++;
933 }
934
935 /*
936 * When the target exists, both the directory
937 * and target vnodes are returned locked.
938 */
939 dp = VTODE(tdvp);
940 xp = tvp ? VTODE(tvp) : NULL;
941 /*
942 * Remember direntry place to use for destination
943 */
944 to_diroffset = dp->de_fndoffset;
945 to_count = dp->de_fndcnt;
946
947 /*
948 * If ".." must be changed (ie the directory gets a new
949 * parent) then the source directory must not be in the
950 * directory hierarchy above the target, as this would
951 * orphan everything below the source directory. Also
952 * the user must have write permission in the source so
953 * as to be able to change "..". We must repeat the call
954 * to namei, as the parent directory is unlocked by the
955 * call to doscheckpath().
956 */
957 error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_proc);
958 VOP_UNLOCK(fvp, 0, p);
959 if (VTODE(fdvp)->de_StartCluster != VTODE(tdvp)->de_StartCluster)
960 newparent = 1;
961 vrele(fdvp);
962 if (doingdirectory && newparent) {
963 if (error) /* write access check above */
964 goto bad1;
965 if (xp != NULL)
966 vput(tvp);
967 /*
968 * doscheckpath() vput()'s dp,
969 * so we have to do a relookup afterwards
970 */
971 if ((error = doscheckpath(ip, dp)) != 0)
972 goto out;
973 if ((tcnp->cn_flags & SAVESTART) == 0)
974 panic("msdosfs_rename: lost to startdir");
975 if ((error = relookup(tdvp, &tvp, tcnp)) != 0)
976 goto out;
977 dp = VTODE(tdvp);
978 xp = tvp ? VTODE(tvp) : NULL;
979 }
980
981 if (xp != NULL) {
982 /*
983 * Target must be empty if a directory and have no links
984 * to it. Also, ensure source and target are compatible
985 * (both directories, or both not directories).
986 */
987 if (xp->de_Attributes & ATTR_DIRECTORY) {
988 if (!dosdirempty(xp)) {
989 error = ENOTEMPTY;
990 goto bad1;
991 }
992 if (!doingdirectory) {
993 error = ENOTDIR;
994 goto bad1;
995 }
996 cache_purge(tdvp);
997 } else if (doingdirectory) {
998 error = EISDIR;
999 goto bad1;
1000 }
1001 if ((error = removede(dp, xp)) != 0)
1002 goto bad1;
1003 vput(tvp);
1004 xp = NULL;
1005 }
1006
1007 /*
1008 * Convert the filename in tcnp into a dos filename. We copy this
1009 * into the denode and directory entry for the destination
1010 * file/directory.
1011 */
1012 if ((error = uniqdosname(VTODE(tdvp), tcnp, toname)) != 0)
1013 goto bad1;
1014
1015 /*
1016 * Since from wasn't locked at various places above,
1017 * have to do a relookup here.
1018 */
1019 fcnp->cn_flags &= ~MODMASK;
1020 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
1021 if ((fcnp->cn_flags & SAVESTART) == 0)
1022 panic("msdosfs_rename: lost from startdir");
1023 if (!newparent)
1024 VOP_UNLOCK(tdvp, 0, p);
1025 (void) relookup(fdvp, &fvp, fcnp);
1026 if (fvp == NULL) {
1027 /*
1028 * From name has disappeared.
1029 */
1030 if (doingdirectory)
1031 panic("rename: lost dir entry");
1032 vrele(ap->a_fvp);
1033 if (newparent)
1034 VOP_UNLOCK(tdvp, 0, p);
1035 vrele(tdvp);
1036 return 0;
1037 }
1038 xp = VTODE(fvp);
1039 zp = VTODE(fdvp);
1040 from_diroffset = zp->de_fndoffset;
1041
1042 /*
1043 * Ensure that the directory entry still exists and has not
1044 * changed till now. If the source is a file the entry may
1045 * have been unlinked or renamed. In either case there is
1046 * no further work to be done. If the source is a directory
1047 * then it cannot have been rmdir'ed or renamed; this is
1048 * prohibited by the DE_RENAME flag.
1049 */
1050 if (xp != ip) {
1051 if (doingdirectory)
1052 panic("rename: lost dir entry");
1053 vrele(ap->a_fvp);
1054 if (newparent)
1055 VOP_UNLOCK(fdvp, 0, p);
1056 xp = NULL;
1057 } else {
1058 vrele(fvp);
1059 xp = NULL;
1060
1061 /*
1062 * First write a new entry in the destination
1063 * directory and mark the entry in the source directory
1064 * as deleted. Then move the denode to the correct hash
1065 * chain for its new location in the filesystem. And, if
1066 * we moved a directory, then update its .. entry to point
1067 * to the new parent directory.
1068 */
1069 bcopy(ip->de_Name, oldname, 11);
1070 bcopy(toname, ip->de_Name, 11); /* update denode */
1071 dp->de_fndoffset = to_diroffset;
1072 dp->de_fndcnt = to_count;
1073 error = createde(ip, dp, (struct denode **)0, tcnp);
1074 if (error) {
1075 bcopy(oldname, ip->de_Name, 11);
1076 if (newparent)
1077 VOP_UNLOCK(fdvp, 0, p);
1078 goto bad;
1079 }
1080 ip->de_refcnt++;
1081 zp->de_fndoffset = from_diroffset;
1082 if ((error = removede(zp, ip)) != 0) {
1083 /* XXX should really panic here, fs is corrupt */
1084 if (newparent)
1085 VOP_UNLOCK(fdvp, 0, p);
1086 goto bad;
1087 }
1088
1089 cache_purge(fvp);
1090
1091 if (!doingdirectory) {
1092 error = pcbmap(dp, de_cluster(pmp, to_diroffset), 0,
1093 &ip->de_dirclust, 0);
1094 if (error) {
1095 /* XXX should really panic here, fs is corrupt */
1096 if (newparent)
1097 VOP_UNLOCK(fdvp, 0, p);
1098 goto bad;
1099 }
1100 if (ip->de_dirclust != MSDOSFSROOT)
1101 ip->de_diroffset = to_diroffset & pmp->pm_crbomask;
1102 }
1103 reinsert(ip);
1104 if (newparent)
1105 VOP_UNLOCK(fdvp, 0, p);
1106 }
1107
1108 /*
1109 * If we moved a directory to a new parent directory, then we must
1110 * fixup the ".." entry in the moved directory.
1111 */
1112 if (doingdirectory && newparent) {
1113 cn = ip->de_StartCluster;
1114 if (cn == MSDOSFSROOT) {
1115 /* this should never happen */
1116 panic("msdosfs_rename: updating .. in root directory?");
1117 } else
1118 bn = cntobn(pmp, cn);
1119 error = bread(pmp->pm_devvp, bn, pmp->pm_bpcluster,
1120 NOCRED, &bp);
1121 if (error) {
1122 /* XXX should really panic here, fs is corrupt */
1123 brelse(bp);
1124 goto bad;
1125 }
1126 dotdotp = (struct direntry *)bp->b_data;
1127 putushort(dotdotp[0].deStartCluster, cn);
1128 pcl = dp->de_StartCluster;
1129 if (FAT32(pmp) && pcl == pmp->pm_rootdirblk)
1130 pcl = 0;
1131 putushort(dotdotp[1].deStartCluster, pcl);
1132 if (FAT32(pmp)) {
1133 putushort(dotdotp[0].deHighClust, cn >> 16);
1134 putushort(dotdotp[1].deHighClust, pcl >> 16);
1135 }
1136 if ((error = bwrite(bp)) != 0) {
1137 /* XXX should really panic here, fs is corrupt */
1138 goto bad;
1139 }
1140 }
1141
1142 bad:
1143 VOP_UNLOCK(fvp, 0, p);
1144 vrele(fdvp);
1145 bad1:
1146 if (xp)
1147 vput(tvp);
1148 vput(tdvp);
1149 out:
1150 ip->de_flag &= ~DE_RENAME;
1151 vrele(fvp);
1152 return (error);
1153
1154 }
1155
1156 struct {
1157 struct direntry dot;
1158 struct direntry dotdot;
1159 } dosdirtemplate = {
1160 { ". ", " ", /* the . entry */
1161 ATTR_DIRECTORY, /* file attribute */
1162 CASE_LOWER_BASE | CASE_LOWER_EXT, /* lower case */
1163 0, /* create time 100ths */
1164 { 0, 0 }, { 0, 0 }, /* create time & date */
1165 { 0, 0 }, /* access date */
1166 { 0, 0 }, /* high bits of start cluster */
1167 { 210, 4 }, { 210, 4 }, /* modify time & date */
1168 { 0, 0 }, /* startcluster */
1169 { 0, 0, 0, 0 } /* filesize */
1170 },
1171 { ".. ", " ", /* the .. entry */
1172 ATTR_DIRECTORY, /* file attribute */
1173 CASE_LOWER_BASE | CASE_LOWER_EXT, /* lower case */
1174 0, /* create time 100ths */
1175 { 0, 0 }, { 0, 0 }, /* create time & date */
1176 { 0, 0 }, /* access date */
1177 { 0, 0 }, /* high bits of start cluster */
1178 { 210, 4 }, { 210, 4 }, /* modify time & date */
1179 { 0, 0 }, /* startcluster */
1180 { 0, 0, 0, 0 } /* filesize */
1181 }
1182 };
1183
1184 int
1185 msdosfs_mkdir(v)
1186 void *v;
1187 {
1188 struct vop_mkdir_args *ap = v;
1189 struct componentname *cnp = ap->a_cnp;
1190 struct denode ndirent;
1191 struct denode *dep;
1192 struct denode *pdep = VTODE(ap->a_dvp);
1193 int error;
1194 daddr64_t bn;
1195 uint32_t newcluster, pcl;
1196 struct direntry *denp;
1197 struct msdosfsmount *pmp = pdep->de_pmp;
1198 struct buf *bp;
1199 struct timespec ts;
1200
1201 /*
1202 * If this is the root directory and there is no space left we
1203 * can't do anything. This is because the root directory can not
1204 * change size.
1205 */
1206 if (pdep->de_StartCluster == MSDOSFSROOT
1207 && pdep->de_fndoffset >= pdep->de_FileSize) {
1208 error = ENOSPC;
1209 goto bad2;
1210 }
1211
1212 /*
1213 * Allocate a cluster to hold the about to be created directory.
1214 */
1215 error = clusteralloc(pmp, 0, 1, CLUST_EOFE, &newcluster, NULL);
1216 if (error)
1217 goto bad2;
1218
1219 bzero(&ndirent, sizeof(ndirent));
1220 ndirent.de_pmp = pmp;
1221 ndirent.de_flag = DE_ACCESS | DE_CREATE | DE_UPDATE;
1222 getnanotime(&ts);
1223 DETIMES(&ndirent, &ts, &ts, &ts);
1224
1225 /*
1226 * Now fill the cluster with the "." and ".." entries. And write
1227 * the cluster to disk. This way it is there for the parent
1228 * directory to be pointing at if there were a crash.
1229 */
1230 bn = cntobn(pmp, newcluster);
1231 /* always succeeds */
1232 bp = getblk(pmp->pm_devvp, bn, pmp->pm_bpcluster, 0, 0);
1233 bzero(bp->b_data, pmp->pm_bpcluster);
1234 bcopy(&dosdirtemplate, bp->b_data, sizeof dosdirtemplate);
1235 denp = (struct direntry *)bp->b_data;
1236 putushort(denp[0].deStartCluster, newcluster);
1237 putushort(denp[0].deCDate, ndirent.de_CDate);
1238 putushort(denp[0].deCTime, ndirent.de_CTime);
1239 denp[0].deCTimeHundredth = ndirent.de_CTimeHundredth;
1240 putushort(denp[0].deADate, ndirent.de_ADate);
1241 putushort(denp[0].deMDate, ndirent.de_MDate);
1242 putushort(denp[0].deMTime, ndirent.de_MTime);
1243 pcl = pdep->de_StartCluster;
1244 if (FAT32(pmp) && pcl == pmp->pm_rootdirblk)
1245 pcl = 0;
1246 putushort(denp[1].deStartCluster, pcl);
1247 putushort(denp[1].deCDate, ndirent.de_CDate);
1248 putushort(denp[1].deCTime, ndirent.de_CTime);
1249 denp[1].deCTimeHundredth = ndirent.de_CTimeHundredth;
1250 putushort(denp[1].deADate, ndirent.de_ADate);
1251 putushort(denp[1].deMDate, ndirent.de_MDate);
1252 putushort(denp[1].deMTime, ndirent.de_MTime);
1253 if (FAT32(pmp)) {
1254 putushort(denp[0].deHighClust, newcluster >> 16);
1255 putushort(denp[1].deHighClust, pdep->de_StartCluster >> 16);
1256 }
1257
1258 if ((error = bwrite(bp)) != 0)
1259 goto bad;
1260
1261 /*
1262 * Now build up a directory entry pointing to the newly allocated
1263 * cluster. This will be written to an empty slot in the parent
1264 * directory.
1265 */
1266 #ifdef DIAGNOSTIC
1267 if ((cnp->cn_flags & HASBUF) == 0)
1268 panic("msdosfs_mkdir: no name");
1269 #endif
1270 if ((error = uniqdosname(pdep, cnp, ndirent.de_Name)) != 0)
1271 goto bad;
1272
1273 ndirent.de_Attributes = ATTR_DIRECTORY;
1274 ndirent.de_StartCluster = newcluster;
1275 ndirent.de_FileSize = 0;
1276 ndirent.de_dev = pdep->de_dev;
1277 ndirent.de_devvp = pdep->de_devvp;
1278 if ((error = createde(&ndirent, pdep, &dep, cnp)) != 0)
1279 goto bad;
1280 if ((cnp->cn_flags & SAVESTART) == 0)
1281 pool_put(&namei_pool, cnp->cn_pnbuf);
1282 vput(ap->a_dvp);
1283 *ap->a_vpp = DETOV(dep);
1284 return (0);
1285
1286 bad:
1287 clusterfree(pmp, newcluster, NULL);
1288 bad2:
1289 pool_put(&namei_pool, cnp->cn_pnbuf);
1290 vput(ap->a_dvp);
1291 return (error);
1292 }
1293
1294 int
1295 msdosfs_rmdir(v)
1296 void *v;
1297 {
1298 struct vop_rmdir_args *ap = v;
1299 register struct vnode *vp = ap->a_vp;
1300 register struct vnode *dvp = ap->a_dvp;
1301 register struct componentname *cnp = ap->a_cnp;
1302 register struct denode *ip, *dp;
1303 int error;
1304
1305 ip = VTODE(vp);
1306 dp = VTODE(dvp);
1307 /*
1308 * No rmdir "." please.
1309 */
1310 if (dp == ip) {
1311 vrele(dvp);
1312 vput(vp);
1313 return (EINVAL);
1314 }
1315 /*
1316 * Verify the directory is empty (and valid).
1317 * (Rmdir ".." won't be valid since
1318 * ".." will contain a reference to
1319 * the current directory and thus be
1320 * non-empty.)
1321 */
1322 error = 0;
1323 if (!dosdirempty(ip) || ip->de_flag & DE_RENAME) {
1324 error = ENOTEMPTY;
1325 goto out;
1326 }
1327 /*
1328 * Delete the entry from the directory. For dos filesystems this
1329 * gets rid of the directory entry on disk, the in memory copy
1330 * still exists but the de_refcnt is <= 0. This prevents it from
1331 * being found by deget(). When the vput() on dep is done we give
1332 * up access and eventually msdosfs_reclaim() will be called which
1333 * will remove it from the denode cache.
1334 */
1335 if ((error = removede(dp, ip)) != 0)
1336 goto out;
1337 /*
1338 * This is where we decrement the link count in the parent
1339 * directory. Since dos filesystems don't do this we just purge
1340 * the name cache and let go of the parent directory denode.
1341 */
1342 cache_purge(dvp);
1343 vput(dvp);
1344 dvp = NULL;
1345 /*
1346 * Truncate the directory that is being deleted.
1347 */
1348 error = detrunc(ip, (uint32_t)0, IO_SYNC, cnp->cn_cred, cnp->cn_proc);
1349 cache_purge(vp);
1350 out:
1351 if (dvp)
1352 vput(dvp);
1353 vput(vp);
1354 return (error);
1355 }
1356
1357 /*
1358 * DOS filesystems don't know what symlinks are.
1359 */
1360 int
1361 msdosfs_symlink(v)
1362 void *v;
1363 {
1364 struct vop_symlink_args *ap = v;
1365
1366 VOP_ABORTOP(ap->a_dvp, ap->a_cnp);
1367 vput(ap->a_dvp);
1368 return (EOPNOTSUPP);
1369 }
1370
1371 int
1372 msdosfs_readdir(v)
1373 void *v;
1374 {
1375 struct vop_readdir_args *ap = v;
1376 int error = 0;
1377 int diff;
1378 long n;
1379 int blsize;
1380 long on;
1381 long lost;
1382 long count;
1383 uint32_t dirsperblk;
1384 uint32_t cn, lbn;
1385 uint32_t fileno;
1386 long bias = 0;
1387 daddr64_t bn;
1388 struct buf *bp;
1389 struct denode *dep = VTODE(ap->a_vp);
1390 struct msdosfsmount *pmp = dep->de_pmp;
1391 struct direntry *dentp;
1392 struct dirent dirbuf;
1393 struct uio *uio = ap->a_uio;
1394 u_long *cookies = NULL;
1395 int ncookies = 0;
1396 off_t offset, wlast = -1;
1397 int chksum = -1;
1398
1399 #ifdef MSDOSFS_DEBUG
1400 printf("msdosfs_readdir(): vp %08x, uio %08x, cred %08x, eofflagp %08x\n",
1401 ap->a_vp, uio, ap->a_cred, ap->a_eofflag);
1402 #endif
1403
1404 /*
1405 * msdosfs_readdir() won't operate properly on regular files since
1406 * it does i/o only with the filesystem vnode, and hence can
1407 * retrieve the wrong block from the buffer cache for a plain file.
1408 * So, fail attempts to readdir() on a plain file.
1409 */
1410 if ((dep->de_Attributes & ATTR_DIRECTORY) == 0)
1411 return (ENOTDIR);
1412
1413 /*
1414 * To be safe, initialize dirbuf
1415 */
1416 bzero(dirbuf.d_name, sizeof(dirbuf.d_name));
1417
1418 /*
1419 * If the user buffer is smaller than the size of one dos directory
1420 * entry or the file offset is not a multiple of the size of a
1421 * directory entry, then we fail the read.
1422 */
1423 count = uio->uio_resid & ~(sizeof(struct direntry) - 1);
1424 offset = uio->uio_offset;
1425 if (count < sizeof(struct direntry) ||
1426 (offset & (sizeof(struct direntry) - 1)))
1427 return (EINVAL);
1428 lost = uio->uio_resid - count;
1429 uio->uio_resid = count;
1430
1431 if (ap->a_ncookies) {
1432 ncookies = uio->uio_resid / sizeof(struct direntry) + 3;
1433 MALLOC(cookies, u_long *, ncookies * sizeof(u_long), M_TEMP,
1434 M_WAITOK);
1435 *ap->a_cookies = cookies;
1436 *ap->a_ncookies = ncookies;
1437 }
1438
1439 dirsperblk = pmp->pm_BytesPerSec / sizeof(struct direntry);
1440
1441 /*
1442 * If they are reading from the root directory then, we simulate
1443 * the . and .. entries since these don't exist in the root
1444 * directory. We also set the offset bias to make up for having to
1445 * simulate these entries. By this I mean that at file offset 64 we
1446 * read the first entry in the root directory that lives on disk.
1447 */
1448 if (dep->de_StartCluster == MSDOSFSROOT
1449 || (FAT32(pmp) && dep->de_StartCluster == pmp->pm_rootdirblk)) {
1450 #if 0
1451 printf("msdosfs_readdir(): going after . or .. in root dir, offset %d\n",
1452 offset);
1453 #endif
1454 bias = 2 * sizeof(struct direntry);
1455 if (offset < bias) {
1456 for (n = (int)offset / sizeof(struct direntry);
1457 n < 2; n++) {
1458 if (FAT32(pmp))
1459 dirbuf.d_fileno = pmp->pm_rootdirblk;
1460 else
1461 dirbuf.d_fileno = 1;
1462 dirbuf.d_type = DT_DIR;
1463 switch (n) {
1464 case 0:
1465 dirbuf.d_namlen = 1;
1466 strlcpy(dirbuf.d_name, ".",
1467 sizeof dirbuf.d_name);
1468 break;
1469 case 1:
1470 dirbuf.d_namlen = 2;
1471 strlcpy(dirbuf.d_name, "..",
1472 sizeof dirbuf.d_name);
1473 break;
1474 }
1475 dirbuf.d_reclen = DIRENT_SIZE(&dirbuf);
1476 if (uio->uio_resid < dirbuf.d_reclen)
1477 goto out;
1478 error = uiomove((caddr_t) &dirbuf,
1479 dirbuf.d_reclen, uio);
1480 if (error)
1481 goto out;
1482 offset += sizeof(struct direntry);
1483 if (cookies) {
1484 *cookies++ = offset;
1485 if (--ncookies <= 0)
1486 goto out;
1487 }
1488 }
1489 }
1490 }
1491
1492 while (uio->uio_resid > 0) {
1493 lbn = de_cluster(pmp, offset - bias);
1494 on = (offset - bias) & pmp->pm_crbomask;
1495 n = min(pmp->pm_bpcluster - on, uio->uio_resid);
1496 diff = dep->de_FileSize - (offset - bias);
1497 if (diff <= 0)
1498 break;
1499 n = min(n, diff);
1500 if ((error = pcbmap(dep, lbn, &bn, &cn, &blsize)) != 0)
1501 break;
1502 error = bread(pmp->pm_devvp, bn, blsize, NOCRED, &bp);
1503 if (error) {
1504 brelse(bp);
1505 return (error);
1506 }
1507 n = min(n, blsize - bp->b_resid);
1508
1509 /*
1510 * Convert from dos directory entries to fs-independent
1511 * directory entries.
1512 */
1513 for (dentp = (struct direntry *)(bp->b_data + on);
1514 (char *)dentp < bp->b_data + on + n;
1515 dentp++, offset += sizeof(struct direntry)) {
1516 #if 0
1517 printf("rd: dentp %08x prev %08x crnt %08x deName %02x attr %02x\n",
1518 dentp, prev, crnt, dentp->deName[0], dentp->deAttributes);
1519 #endif
1520 /*
1521 * If this is an unused entry, we can stop.
1522 */
1523 if (dentp->deName[0] == SLOT_EMPTY) {
1524 brelse(bp);
1525 goto out;
1526 }
1527 /*
1528 * Skip deleted entries.
1529 */
1530 if (dentp->deName[0] == SLOT_DELETED) {
1531 chksum = -1;
1532 wlast = -1;
1533 continue;
1534 }
1535
1536 /*
1537 * Handle Win95 long directory entries
1538 */
1539 if (dentp->deAttributes == ATTR_WIN95) {
1540 struct winentry *wep;
1541 if (pmp->pm_flags & MSDOSFSMNT_SHORTNAME)
1542 continue;
1543 wep = (struct winentry *)dentp;
1544 chksum = win2unixfn(wep, &dirbuf, chksum);
1545 if (wep->weCnt & WIN_LAST)
1546 wlast = offset;
1547 continue;
1548 }
1549
1550 /*
1551 * Skip volume labels
1552 */
1553 if (dentp->deAttributes & ATTR_VOLUME) {
1554 chksum = -1;
1555 wlast = -1;
1556 continue;
1557 }
1558
1559 /*
1560 * This computation of d_fileno must match
1561 * the computation of va_fileid in
1562 * msdosfs_getattr.
1563 */
1564 fileno = getushort(dentp->deStartCluster);
1565 if (FAT32(pmp))
1566 fileno |= getushort(dentp->deHighClust) << 16;
1567
1568 if (dentp->deAttributes & ATTR_DIRECTORY) {
1569 /* Special-case root */
1570 if (fileno == MSDOSFSROOT) {
1571 fileno = FAT32(pmp) ?
1572 pmp->pm_rootdirblk : 1;
1573 }
1574
1575 dirbuf.d_fileno = fileno;
1576 dirbuf.d_type = DT_DIR;
1577 } else {
1578 if (getulong(dentp->deFileSize) == 0) {
1579 uint64_t fileno64;
1580
1581 fileno64 = (cn == MSDOSFSROOT) ?
1582 roottobn(pmp, 0) : cntobn(pmp, cn);
1583
1584 fileno64 *= dirsperblk;
1585 fileno64 += dentp -
1586 (struct direntry *)bp->b_data;
1587
1588 fileno = fileidhash(fileno64);
1589 }
1590
1591 dirbuf.d_fileno = fileno;
1592 dirbuf.d_type = DT_REG;
1593 }
1594
1595 if (chksum != winChksum(dentp->deName))
1596 dirbuf.d_namlen = dos2unixfn(dentp->deName,
1597 (u_char *)dirbuf.d_name,
1598 pmp->pm_flags & MSDOSFSMNT_SHORTNAME);
1599 else
1600 dirbuf.d_name[dirbuf.d_namlen] = 0;
1601 chksum = -1;
1602 dirbuf.d_reclen = DIRENT_SIZE(&dirbuf);
1603 if (uio->uio_resid < dirbuf.d_reclen) {
1604 brelse(bp);
1605 /* Remember long-name offset. */
1606 if (wlast != -1)
1607 offset = wlast;
1608 goto out;
1609 }
1610 wlast = -1;
1611 error = uiomove((caddr_t) &dirbuf,
1612 dirbuf.d_reclen, uio);
1613 if (error) {
1614 brelse(bp);
1615 goto out;
1616 }
1617 if (cookies) {
1618 *cookies++ = offset + sizeof(struct direntry);
1619 if (--ncookies <= 0) {
1620 brelse(bp);
1621 goto out;
1622 }
1623 }
1624 }
1625 brelse(bp);
1626 }
1627
1628 out:
1629 /* Subtract unused cookies */
1630 if (ap->a_ncookies)
1631 *ap->a_ncookies -= ncookies;
1632
1633 uio->uio_offset = offset;
1634 uio->uio_resid += lost;
1635 if (dep->de_FileSize - (offset - bias) <= 0)
1636 *ap->a_eofflag = 1;
1637 else
1638 *ap->a_eofflag = 0;
1639 return (error);
1640 }
1641
1642 /*
1643 * DOS filesystems don't know what symlinks are.
1644 */
1645 int
1646 msdosfs_readlink(v)
1647 void *v;
1648 {
1649 #if 0
1650 struct vop_readlink_args /* {
1651 struct vnode *a_vp;
1652 struct uio *a_uio;
1653 struct ucred *a_cred;
1654 } */ *ap;
1655 #endif
1656
1657 return (EINVAL);
1658 }
1659
1660 int
1661 msdosfs_lock(v)
1662 void *v;
1663 {
1664 struct vop_lock_args *ap = v;
1665 struct vnode *vp = ap->a_vp;
1666
1667 return (lockmgr(&VTODE(vp)->de_lock, ap->a_flags, NULL));
1668 }
1669
1670 int
1671 msdosfs_unlock(v)
1672 void *v;
1673 {
1674 struct vop_unlock_args *ap = v;
1675 struct vnode *vp = ap->a_vp;
1676
1677 return (lockmgr(&VTODE(vp)->de_lock, ap->a_flags | LK_RELEASE, NULL));
1678 }
1679
1680 int
1681 msdosfs_islocked(v)
1682 void *v;
1683 {
1684 struct vop_islocked_args *ap = v;
1685
1686 return (lockstatus(&VTODE(ap->a_vp)->de_lock));
1687 }
1688
1689 /*
1690 * vp - address of vnode file the file
1691 * bn - which cluster we are interested in mapping to a filesystem block number.
1692 * vpp - returns the vnode for the block special file holding the filesystem
1693 * containing the file of interest
1694 * bnp - address of where to return the filesystem relative block number
1695 */
1696 int
1697 msdosfs_bmap(v)
1698 void *v;
1699 {
1700 struct vop_bmap_args *ap = v;
1701 struct denode *dep = VTODE(ap->a_vp);
1702 struct msdosfsmount *pmp = dep->de_pmp;
1703
1704 if (ap->a_vpp != NULL)
1705 *ap->a_vpp = dep->de_devvp;
1706 if (ap->a_bnp == NULL)
1707 return (0);
1708 if (ap->a_runp) {
1709 /*
1710 * Sequential clusters should be counted here.
1711 */
1712 *ap->a_runp = 0;
1713 }
1714 return (pcbmap(dep, de_bn2cn(pmp, ap->a_bn), ap->a_bnp, 0, 0));
1715 }
1716
1717 int
1718 msdosfs_strategy(v)
1719 void *v;
1720 {
1721 struct vop_strategy_args *ap = v;
1722 struct buf *bp = ap->a_bp;
1723 struct denode *dep = VTODE(bp->b_vp);
1724 struct vnode *vp;
1725 int error = 0;
1726 int s;
1727
1728 if (bp->b_vp->v_type == VBLK || bp->b_vp->v_type == VCHR)
1729 panic("msdosfs_strategy: spec");
1730 /*
1731 * If we don't already know the filesystem relative block number
1732 * then get it using pcbmap(). If pcbmap() returns the block
1733 * number as -1 then we've got a hole in the file. DOS filesystems
1734 * don't allow files with holes, so we shouldn't ever see this.
1735 */
1736 if (bp->b_blkno == bp->b_lblkno) {
1737 error = pcbmap(dep, de_bn2cn(dep->de_pmp, bp->b_lblkno),
1738 &bp->b_blkno, 0, 0);
1739 if (error)
1740 bp->b_blkno = -1;
1741 if (bp->b_blkno == -1)
1742 clrbuf(bp);
1743 }
1744 if (bp->b_blkno == -1) {
1745 s = splbio();
1746 biodone(bp);
1747 splx(s);
1748 return (error);
1749 }
1750
1751 /*
1752 * Read/write the block from/to the disk that contains the desired
1753 * file block.
1754 */
1755
1756 vp = dep->de_devvp;
1757 bp->b_dev = vp->v_rdev;
1758 VOCALL(vp->v_op, VOFFSET(vop_strategy), ap);
1759 return (0);
1760 }
1761
1762 int
1763 msdosfs_print(v)
1764 void *v;
1765 {
1766 struct vop_print_args *ap = v;
1767 struct denode *dep = VTODE(ap->a_vp);
1768
1769 printf(
1770 "tag VT_MSDOSFS, startcluster %ld, dircluster %ld, diroffset %ld ",
1771 dep->de_StartCluster, dep->de_dirclust, dep->de_diroffset);
1772 printf(" dev %d, %d, %s\n",
1773 major(dep->de_dev), minor(dep->de_dev),
1774 VOP_ISLOCKED(ap->a_vp) ? "(LOCKED)" : "");
1775 #ifdef DIAGNOSTIC
1776 lockmgr_printinfo(&dep->de_lock);
1777 #endif
1778
1779 return (0);
1780 }
1781
1782 int
1783 msdosfs_advlock(v)
1784 void *v;
1785 {
1786 struct vop_advlock_args *ap = v;
1787 register struct denode *dep = VTODE(ap->a_vp);
1788
1789 return (lf_advlock(&dep->de_lockf, dep->de_FileSize, ap->a_id, ap->a_op,
1790 ap->a_fl, ap->a_flags));
1791 }
1792
1793 int
1794 msdosfs_pathconf(v)
1795 void *v;
1796 {
1797 struct vop_pathconf_args *ap = v;
1798 struct msdosfsmount *pmp = VTODE(ap->a_vp)->de_pmp;
1799
1800 switch (ap->a_name) {
1801 case _PC_LINK_MAX:
1802 *ap->a_retval = 1;
1803 return (0);
1804 case _PC_NAME_MAX:
1805 *ap->a_retval = pmp->pm_flags & MSDOSFSMNT_LONGNAME ? WIN_MAXLEN : 12;
1806 return (0);
1807 case _PC_PATH_MAX:
1808 *ap->a_retval = PATH_MAX;
1809 return (0);
1810 case _PC_CHOWN_RESTRICTED:
1811 *ap->a_retval = 1;
1812 return (0);
1813 case _PC_NO_TRUNC:
1814 *ap->a_retval = 0;
1815 return (0);
1816 default:
1817 return (EINVAL);
1818 }
1819 /* NOTREACHED */
1820 }
1821
1822 /*
1823 * Thomas Wang's hash function, severely hacked to always set the high
1824 * bit on the number it returns (so no longer a proper hash function).
1825 */
1826 static uint32_t
1827 fileidhash(uint64_t fileid)
1828 {
1829 uint64_t c1 = 0x6e5ea73858134343LL;
1830 uint64_t c2 = 0xb34e8f99a2ec9ef5LL;
1831
1832 /*
1833 * We now have the original fileid value, as 64-bit value.
1834 * We need to reduce it to 32-bits, with the top bit set.
1835 */
1836 fileid ^= ((c1 ^ fileid) >> 32);
1837 fileid *= c1;
1838 fileid ^= ((c2 ^ fileid) >> 31);
1839 fileid *= c2;
1840 fileid ^= ((c1 ^ fileid) >> 32);
1841
1842 return (uint32_t)(fileid | 0x80000000);
1843 }
1844
1845 /* Global vfs data structures for msdosfs */
1846 int (**msdosfs_vnodeop_p)(void *);
1847 struct vnodeopv_entry_desc msdosfs_vnodeop_entries[] = {
1848 { &vop_default_desc, vn_default_error },
1849 { &vop_lookup_desc, msdosfs_lookup }, /* lookup */
1850 { &vop_create_desc, msdosfs_create }, /* create */
1851 { &vop_mknod_desc, msdosfs_mknod }, /* mknod */
1852 { &vop_open_desc, msdosfs_open }, /* open */
1853 { &vop_close_desc, msdosfs_close }, /* close */
1854 { &vop_access_desc, msdosfs_access }, /* access */
1855 { &vop_getattr_desc, msdosfs_getattr }, /* getattr */
1856 { &vop_setattr_desc, msdosfs_setattr }, /* setattr */
1857 { &vop_read_desc, msdosfs_read }, /* read */
1858 { &vop_write_desc, msdosfs_write }, /* write */
1859 { &vop_ioctl_desc, msdosfs_ioctl }, /* ioctl */
1860 { &vop_poll_desc, msdosfs_poll }, /* poll */
1861 { &vop_fsync_desc, msdosfs_fsync }, /* fsync */
1862 { &vop_remove_desc, msdosfs_remove }, /* remove */
1863 { &vop_link_desc, msdosfs_link }, /* link */
1864 { &vop_rename_desc, msdosfs_rename }, /* rename */
1865 { &vop_mkdir_desc, msdosfs_mkdir }, /* mkdir */
1866 { &vop_rmdir_desc, msdosfs_rmdir }, /* rmdir */
1867 { &vop_symlink_desc, msdosfs_symlink }, /* symlink */
1868 { &vop_readdir_desc, msdosfs_readdir }, /* readdir */
1869 { &vop_readlink_desc, msdosfs_readlink }, /* readlink */
1870 { &vop_abortop_desc, vop_generic_abortop }, /* abortop */
1871 { &vop_inactive_desc, msdosfs_inactive }, /* inactive */
1872 { &vop_reclaim_desc, msdosfs_reclaim }, /* reclaim */
1873 { &vop_lock_desc, msdosfs_lock }, /* lock */
1874 { &vop_unlock_desc, msdosfs_unlock }, /* unlock */
1875 { &vop_bmap_desc, msdosfs_bmap }, /* bmap */
1876 { &vop_strategy_desc, msdosfs_strategy }, /* strategy */
1877 { &vop_print_desc, msdosfs_print }, /* print */
1878 { &vop_islocked_desc, msdosfs_islocked }, /* islocked */
1879 { &vop_pathconf_desc, msdosfs_pathconf }, /* pathconf */
1880 { &vop_advlock_desc, msdosfs_advlock }, /* advlock */
1881 { &vop_bwrite_desc, vop_generic_bwrite }, /* bwrite */
1882 { (struct vnodeop_desc *)NULL, (int (*)(void *))NULL }
1883 };
1884 struct vnodeopv_desc msdosfs_vnodeop_opv_desc =
1885 { &msdosfs_vnodeop_p, msdosfs_vnodeop_entries };