1 /* $FreeBSD: releng/11.0/sys/fs/msdosfs/msdosfs_vnops.c 298806 2016-04-29 20:51:24Z pfg $ */
2 /* $NetBSD: msdosfs_vnops.c,v 1.68 1998/02/10 14:10:04 mrg Exp $ */
3
4 /*-
5 * Copyright (C) 1994, 1995, 1997 Wolfgang Solfrank.
6 * Copyright (C) 1994, 1995, 1997 TooLs GmbH.
7 * All rights reserved.
8 * Original code by Paul Popelka (paulp@uts.amdahl.com) (see below).
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by TooLs GmbH.
21 * 4. The name of TooLs GmbH may not be used to endorse or promote products
22 * derived from this software without specific prior written permission.
23 *
24 * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
25 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
26 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
27 * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
28 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
29 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
30 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
31 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
32 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
33 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34 */
35 /*-
36 * Written by Paul Popelka (paulp@uts.amdahl.com)
37 *
38 * You can do anything you want with this software, just don't say you wrote
39 * it, and don't remove this notice.
40 *
41 * This software is provided "as is".
42 *
43 * The author supplies this software to be publicly redistributed on the
44 * understanding that the author is not responsible for the correct
45 * functioning of this software in any circumstances and is not liable for
46 * any damages caused by this software.
47 *
48 * October 1992
49 */
50
51 #include <sys/param.h>
52 #include <sys/systm.h>
53 #include <sys/bio.h>
54 #include <sys/buf.h>
55 #include <sys/clock.h>
56 #include <sys/dirent.h>
57 #include <sys/lock.h>
58 #include <sys/lockf.h>
59 #include <sys/malloc.h>
60 #include <sys/mount.h>
61 #include <sys/mutex.h>
62 #include <sys/namei.h>
63 #include <sys/priv.h>
64 #include <sys/stat.h>
65 #include <sys/unistd.h>
66 #include <sys/vnode.h>
67
68 #include <vm/vm.h>
69 #include <vm/vm_extern.h>
70
71 #include <fs/msdosfs/bpb.h>
72 #include <fs/msdosfs/direntry.h>
73 #include <fs/msdosfs/denode.h>
74 #include <fs/msdosfs/fat.h>
75 #include <fs/msdosfs/msdosfsmount.h>
76
77 #define DOS_FILESIZE_MAX 0xffffffff
78
79 /*
80 * Prototypes for MSDOSFS vnode operations
81 */
82 static vop_create_t msdosfs_create;
83 static vop_mknod_t msdosfs_mknod;
84 static vop_open_t msdosfs_open;
85 static vop_close_t msdosfs_close;
86 static vop_access_t msdosfs_access;
87 static vop_getattr_t msdosfs_getattr;
88 static vop_setattr_t msdosfs_setattr;
89 static vop_read_t msdosfs_read;
90 static vop_write_t msdosfs_write;
91 static vop_fsync_t msdosfs_fsync;
92 static vop_remove_t msdosfs_remove;
93 static vop_link_t msdosfs_link;
94 static vop_rename_t msdosfs_rename;
95 static vop_mkdir_t msdosfs_mkdir;
96 static vop_rmdir_t msdosfs_rmdir;
97 static vop_symlink_t msdosfs_symlink;
98 static vop_readdir_t msdosfs_readdir;
99 static vop_bmap_t msdosfs_bmap;
100 static vop_strategy_t msdosfs_strategy;
101 static vop_print_t msdosfs_print;
102 static vop_pathconf_t msdosfs_pathconf;
103 static vop_vptofh_t msdosfs_vptofh;
104
105 /*
106 * Some general notes:
107 *
108 * In the ufs filesystem the inodes, superblocks, and indirect blocks are
109 * read/written using the vnode for the filesystem. Blocks that represent
110 * the contents of a file are read/written using the vnode for the file
111 * (including directories when they are read/written as files). This
112 * presents problems for the dos filesystem because data that should be in
113 * an inode (if dos had them) resides in the directory itself. Since we
114 * must update directory entries without the benefit of having the vnode
115 * for the directory we must use the vnode for the filesystem. This means
116 * that when a directory is actually read/written (via read, write, or
117 * readdir, or seek) we must use the vnode for the filesystem instead of
118 * the vnode for the directory as would happen in ufs. This is to insure we
119 * retrieve the correct block from the buffer cache since the hash value is
120 * based upon the vnode address and the desired block number.
121 */
122
123 /*
124 * Create a regular file. On entry the directory to contain the file being
125 * created is locked. We must release before we return. We must also free
126 * the pathname buffer pointed at by cnp->cn_pnbuf, always on error, or
127 * only if the SAVESTART bit in cn_flags is clear on success.
128 */
129 static int
130 msdosfs_create(struct vop_create_args *ap)
131 {
132 struct componentname *cnp = ap->a_cnp;
133 struct denode ndirent;
134 struct denode *dep;
135 struct denode *pdep = VTODE(ap->a_dvp);
136 struct timespec ts;
137 int error;
138
139 #ifdef MSDOSFS_DEBUG
140 printf("msdosfs_create(cnp %p, vap %p\n", cnp, ap->a_vap);
141 #endif
142
143 /*
144 * If this is the root directory and there is no space left we
145 * can't do anything. This is because the root directory can not
146 * change size.
147 */
148 if (pdep->de_StartCluster == MSDOSFSROOT
149 && pdep->de_fndoffset >= pdep->de_FileSize) {
150 error = ENOSPC;
151 goto bad;
152 }
153
154 /*
155 * Create a directory entry for the file, then call createde() to
156 * have it installed. NOTE: DOS files are always executable. We
157 * use the absence of the owner write bit to make the file
158 * readonly.
159 */
160 #ifdef DIAGNOSTIC
161 if ((cnp->cn_flags & HASBUF) == 0)
162 panic("msdosfs_create: no name");
163 #endif
164 bzero(&ndirent, sizeof(ndirent));
165 error = uniqdosname(pdep, cnp, ndirent.de_Name);
166 if (error)
167 goto bad;
168
169 ndirent.de_Attributes = ATTR_ARCHIVE;
170 ndirent.de_LowerCase = 0;
171 ndirent.de_StartCluster = 0;
172 ndirent.de_FileSize = 0;
173 ndirent.de_pmp = pdep->de_pmp;
174 ndirent.de_flag = DE_ACCESS | DE_CREATE | DE_UPDATE;
175 getnanotime(&ts);
176 DETIMES(&ndirent, &ts, &ts, &ts);
177 error = createde(&ndirent, pdep, &dep, cnp);
178 if (error)
179 goto bad;
180 *ap->a_vpp = DETOV(dep);
181 if ((cnp->cn_flags & MAKEENTRY) != 0)
182 cache_enter(ap->a_dvp, *ap->a_vpp, cnp);
183 return (0);
184
185 bad:
186 return (error);
187 }
188
189 static int
190 msdosfs_mknod(struct vop_mknod_args *ap)
191 {
192
193 return (EINVAL);
194 }
195
196 static int
197 msdosfs_open(struct vop_open_args *ap)
198 {
199 struct denode *dep = VTODE(ap->a_vp);
200 vnode_create_vobject(ap->a_vp, dep->de_FileSize, ap->a_td);
201 return 0;
202 }
203
204 static int
205 msdosfs_close(struct vop_close_args *ap)
206 {
207 struct vnode *vp = ap->a_vp;
208 struct denode *dep = VTODE(vp);
209 struct timespec ts;
210
211 VI_LOCK(vp);
212 if (vp->v_usecount > 1) {
213 getnanotime(&ts);
214 DETIMES(dep, &ts, &ts, &ts);
215 }
216 VI_UNLOCK(vp);
217 return 0;
218 }
219
220 static int
221 msdosfs_access(struct vop_access_args *ap)
222 {
223 struct vnode *vp = ap->a_vp;
224 struct denode *dep = VTODE(ap->a_vp);
225 struct msdosfsmount *pmp = dep->de_pmp;
226 mode_t file_mode;
227 accmode_t accmode = ap->a_accmode;
228
229 file_mode = S_IRWXU|S_IRWXG|S_IRWXO;
230 file_mode &= (vp->v_type == VDIR ? pmp->pm_dirmask : pmp->pm_mask);
231
232 /*
233 * Disallow writing to directories and regular files if the
234 * filesystem is read-only.
235 */
236 if (accmode & VWRITE) {
237 switch (vp->v_type) {
238 case VREG:
239 case VDIR:
240 if (vp->v_mount->mnt_flag & MNT_RDONLY)
241 return (EROFS);
242 break;
243 default:
244 break;
245 }
246 }
247
248 return (vaccess(vp->v_type, file_mode, pmp->pm_uid, pmp->pm_gid,
249 ap->a_accmode, ap->a_cred, NULL));
250 }
251
252 static int
253 msdosfs_getattr(struct vop_getattr_args *ap)
254 {
255 struct denode *dep = VTODE(ap->a_vp);
256 struct msdosfsmount *pmp = dep->de_pmp;
257 struct vattr *vap = ap->a_vap;
258 mode_t mode;
259 struct timespec ts;
260 u_long dirsperblk = pmp->pm_BytesPerSec / sizeof(struct direntry);
261 uint64_t fileid;
262
263 getnanotime(&ts);
264 DETIMES(dep, &ts, &ts, &ts);
265 vap->va_fsid = dev2udev(pmp->pm_dev);
266 /*
267 * The following computation of the fileid must be the same as that
268 * used in msdosfs_readdir() to compute d_fileno. If not, pwd
269 * doesn't work.
270 */
271 if (dep->de_Attributes & ATTR_DIRECTORY) {
272 fileid = (uint64_t)cntobn(pmp, dep->de_StartCluster) *
273 dirsperblk;
274 if (dep->de_StartCluster == MSDOSFSROOT)
275 fileid = 1;
276 } else {
277 fileid = (uint64_t)cntobn(pmp, dep->de_dirclust) *
278 dirsperblk;
279 if (dep->de_dirclust == MSDOSFSROOT)
280 fileid = (uint64_t)roottobn(pmp, 0) * dirsperblk;
281 fileid += (uoff_t)dep->de_diroffset / sizeof(struct direntry);
282 }
283
284 if (pmp->pm_flags & MSDOSFS_LARGEFS)
285 vap->va_fileid = msdosfs_fileno_map(pmp->pm_mountp, fileid);
286 else
287 vap->va_fileid = (long)fileid;
288
289 mode = S_IRWXU|S_IRWXG|S_IRWXO;
290 vap->va_mode = mode &
291 (ap->a_vp->v_type == VDIR ? pmp->pm_dirmask : pmp->pm_mask);
292 vap->va_uid = pmp->pm_uid;
293 vap->va_gid = pmp->pm_gid;
294 vap->va_nlink = 1;
295 vap->va_rdev = NODEV;
296 vap->va_size = dep->de_FileSize;
297 fattime2timespec(dep->de_MDate, dep->de_MTime, 0, 0, &vap->va_mtime);
298 vap->va_ctime = vap->va_mtime;
299 if (pmp->pm_flags & MSDOSFSMNT_LONGNAME) {
300 fattime2timespec(dep->de_ADate, 0, 0, 0, &vap->va_atime);
301 fattime2timespec(dep->de_CDate, dep->de_CTime, dep->de_CHun,
302 0, &vap->va_birthtime);
303 } else {
304 vap->va_atime = vap->va_mtime;
305 vap->va_birthtime.tv_sec = -1;
306 vap->va_birthtime.tv_nsec = 0;
307 }
308 vap->va_flags = 0;
309 if (dep->de_Attributes & ATTR_ARCHIVE)
310 vap->va_flags |= UF_ARCHIVE;
311 if (dep->de_Attributes & ATTR_HIDDEN)
312 vap->va_flags |= UF_HIDDEN;
313 if (dep->de_Attributes & ATTR_READONLY)
314 vap->va_flags |= UF_READONLY;
315 if (dep->de_Attributes & ATTR_SYSTEM)
316 vap->va_flags |= UF_SYSTEM;
317 vap->va_gen = 0;
318 vap->va_blocksize = pmp->pm_bpcluster;
319 vap->va_bytes =
320 (dep->de_FileSize + pmp->pm_crbomask) & ~pmp->pm_crbomask;
321 vap->va_type = ap->a_vp->v_type;
322 vap->va_filerev = dep->de_modrev;
323 return (0);
324 }
325
326 static int
327 msdosfs_setattr(struct vop_setattr_args *ap)
328 {
329 struct vnode *vp = ap->a_vp;
330 struct denode *dep = VTODE(ap->a_vp);
331 struct msdosfsmount *pmp = dep->de_pmp;
332 struct vattr *vap = ap->a_vap;
333 struct ucred *cred = ap->a_cred;
334 struct thread *td = curthread;
335 int error = 0;
336
337 #ifdef MSDOSFS_DEBUG
338 printf("msdosfs_setattr(): vp %p, vap %p, cred %p\n",
339 ap->a_vp, vap, cred);
340 #endif
341
342 /*
343 * Check for unsettable attributes.
344 */
345 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
346 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
347 (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
348 (vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
349 #ifdef MSDOSFS_DEBUG
350 printf("msdosfs_setattr(): returning EINVAL\n");
351 printf(" va_type %d, va_nlink %x, va_fsid %lx, va_fileid %lx\n",
352 vap->va_type, vap->va_nlink, vap->va_fsid, vap->va_fileid);
353 printf(" va_blocksize %lx, va_rdev %x, va_bytes %qx, va_gen %lx\n",
354 vap->va_blocksize, vap->va_rdev, vap->va_bytes, vap->va_gen);
355 printf(" va_uid %x, va_gid %x\n",
356 vap->va_uid, vap->va_gid);
357 #endif
358 return (EINVAL);
359 }
360
361 /*
362 * We don't allow setting attributes on the root directory.
363 * The special case for the root directory is because before
364 * FAT32, the root directory didn't have an entry for itself
365 * (and was otherwise special). With FAT32, the root
366 * directory is not so special, but still doesn't have an
367 * entry for itself.
368 */
369 if (vp->v_vflag & VV_ROOT)
370 return (EINVAL);
371
372 if (vap->va_flags != VNOVAL) {
373 if (vp->v_mount->mnt_flag & MNT_RDONLY)
374 return (EROFS);
375 if (cred->cr_uid != pmp->pm_uid) {
376 error = priv_check_cred(cred, PRIV_VFS_ADMIN, 0);
377 if (error)
378 return (error);
379 }
380 /*
381 * We are very inconsistent about handling unsupported
382 * attributes. We ignored the access time and the
383 * read and execute bits. We were strict for the other
384 * attributes.
385 */
386 if (vap->va_flags & ~(UF_ARCHIVE | UF_HIDDEN | UF_READONLY |
387 UF_SYSTEM))
388 return EOPNOTSUPP;
389 if (vap->va_flags & UF_ARCHIVE)
390 dep->de_Attributes |= ATTR_ARCHIVE;
391 else
392 dep->de_Attributes &= ~ATTR_ARCHIVE;
393 if (vap->va_flags & UF_HIDDEN)
394 dep->de_Attributes |= ATTR_HIDDEN;
395 else
396 dep->de_Attributes &= ~ATTR_HIDDEN;
397 /* We don't allow changing the readonly bit on directories. */
398 if (vp->v_type != VDIR) {
399 if (vap->va_flags & UF_READONLY)
400 dep->de_Attributes |= ATTR_READONLY;
401 else
402 dep->de_Attributes &= ~ATTR_READONLY;
403 }
404 if (vap->va_flags & UF_SYSTEM)
405 dep->de_Attributes |= ATTR_SYSTEM;
406 else
407 dep->de_Attributes &= ~ATTR_SYSTEM;
408 dep->de_flag |= DE_MODIFIED;
409 }
410
411 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
412 uid_t uid;
413 gid_t gid;
414
415 if (vp->v_mount->mnt_flag & MNT_RDONLY)
416 return (EROFS);
417 uid = vap->va_uid;
418 if (uid == (uid_t)VNOVAL)
419 uid = pmp->pm_uid;
420 gid = vap->va_gid;
421 if (gid == (gid_t)VNOVAL)
422 gid = pmp->pm_gid;
423 if (cred->cr_uid != pmp->pm_uid || uid != pmp->pm_uid ||
424 (gid != pmp->pm_gid && !groupmember(gid, cred))) {
425 error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0);
426 if (error)
427 return (error);
428 }
429 if (uid != pmp->pm_uid || gid != pmp->pm_gid)
430 return EINVAL;
431 }
432
433 if (vap->va_size != VNOVAL) {
434 switch (vp->v_type) {
435 case VDIR:
436 return (EISDIR);
437 case VREG:
438 /*
439 * Truncation is only supported for regular files,
440 * Disallow it if the filesystem is read-only.
441 */
442 if (vp->v_mount->mnt_flag & MNT_RDONLY)
443 return (EROFS);
444 break;
445 default:
446 /*
447 * According to POSIX, the result is unspecified
448 * for file types other than regular files,
449 * directories and shared memory objects. We
450 * don't support any file types except regular
451 * files and directories in this file system, so
452 * this (default) case is unreachable and can do
453 * anything. Keep falling through to detrunc()
454 * for now.
455 */
456 break;
457 }
458 error = detrunc(dep, vap->va_size, 0, cred);
459 if (error)
460 return error;
461 }
462 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
463 if (vp->v_mount->mnt_flag & MNT_RDONLY)
464 return (EROFS);
465 error = vn_utimes_perm(vp, vap, cred, td);
466 if (error != 0)
467 return (error);
468 if ((pmp->pm_flags & MSDOSFSMNT_NOWIN95) == 0 &&
469 vap->va_atime.tv_sec != VNOVAL) {
470 dep->de_flag &= ~DE_ACCESS;
471 timespec2fattime(&vap->va_atime, 0,
472 &dep->de_ADate, NULL, NULL);
473 }
474 if (vap->va_mtime.tv_sec != VNOVAL) {
475 dep->de_flag &= ~DE_UPDATE;
476 timespec2fattime(&vap->va_mtime, 0,
477 &dep->de_MDate, &dep->de_MTime, NULL);
478 }
479 /*
480 * We don't set the archive bit when modifying the time of
481 * a directory to emulate the Windows/DOS behavior.
482 */
483 if (vp->v_type != VDIR)
484 dep->de_Attributes |= ATTR_ARCHIVE;
485 dep->de_flag |= DE_MODIFIED;
486 }
487 /*
488 * DOS files only have the ability to have their writability
489 * attribute set, so we use the owner write bit to set the readonly
490 * attribute.
491 */
492 if (vap->va_mode != (mode_t)VNOVAL) {
493 if (vp->v_mount->mnt_flag & MNT_RDONLY)
494 return (EROFS);
495 if (cred->cr_uid != pmp->pm_uid) {
496 error = priv_check_cred(cred, PRIV_VFS_ADMIN, 0);
497 if (error)
498 return (error);
499 }
500 if (vp->v_type != VDIR) {
501 /* We ignore the read and execute bits. */
502 if (vap->va_mode & VWRITE)
503 dep->de_Attributes &= ~ATTR_READONLY;
504 else
505 dep->de_Attributes |= ATTR_READONLY;
506 dep->de_Attributes |= ATTR_ARCHIVE;
507 dep->de_flag |= DE_MODIFIED;
508 }
509 }
510 return (deupdat(dep, 0));
511 }
512
513 static int
514 msdosfs_read(struct vop_read_args *ap)
515 {
516 int error = 0;
517 int blsize;
518 int isadir;
519 ssize_t orig_resid;
520 u_int n;
521 u_long diff;
522 u_long on;
523 daddr_t lbn;
524 daddr_t rablock;
525 int rasize;
526 int seqcount;
527 struct buf *bp;
528 struct vnode *vp = ap->a_vp;
529 struct denode *dep = VTODE(vp);
530 struct msdosfsmount *pmp = dep->de_pmp;
531 struct uio *uio = ap->a_uio;
532
533 /*
534 * If they didn't ask for any data, then we are done.
535 */
536 orig_resid = uio->uio_resid;
537 if (orig_resid == 0)
538 return (0);
539
540 /*
541 * The caller is supposed to ensure that
542 * uio->uio_offset >= 0 and uio->uio_resid >= 0.
543 * We don't need to check for large offsets as in ffs because
544 * dep->de_FileSize <= DOS_FILESIZE_MAX < OFF_MAX, so large
545 * offsets cannot cause overflow even in theory.
546 */
547
548 seqcount = ap->a_ioflag >> IO_SEQSHIFT;
549
550 isadir = dep->de_Attributes & ATTR_DIRECTORY;
551 do {
552 if (uio->uio_offset >= dep->de_FileSize)
553 break;
554 lbn = de_cluster(pmp, uio->uio_offset);
555 rablock = lbn + 1;
556 blsize = pmp->pm_bpcluster;
557 on = uio->uio_offset & pmp->pm_crbomask;
558 /*
559 * If we are operating on a directory file then be sure to
560 * do i/o with the vnode for the filesystem instead of the
561 * vnode for the directory.
562 */
563 if (isadir) {
564 /* convert cluster # to block # */
565 error = pcbmap(dep, lbn, &lbn, 0, &blsize);
566 if (error == E2BIG) {
567 error = EINVAL;
568 break;
569 } else if (error)
570 break;
571 error = bread(pmp->pm_devvp, lbn, blsize, NOCRED, &bp);
572 } else if (de_cn2off(pmp, rablock) >= dep->de_FileSize) {
573 error = bread(vp, lbn, blsize, NOCRED, &bp);
574 } else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
575 error = cluster_read(vp, dep->de_FileSize, lbn, blsize,
576 NOCRED, on + uio->uio_resid, seqcount, 0, &bp);
577 } else if (seqcount > 1) {
578 rasize = blsize;
579 error = breadn(vp, lbn,
580 blsize, &rablock, &rasize, 1, NOCRED, &bp);
581 } else {
582 error = bread(vp, lbn, blsize, NOCRED, &bp);
583 }
584 if (error) {
585 brelse(bp);
586 break;
587 }
588 diff = pmp->pm_bpcluster - on;
589 n = diff > uio->uio_resid ? uio->uio_resid : diff;
590 diff = dep->de_FileSize - uio->uio_offset;
591 if (diff < n)
592 n = diff;
593 diff = blsize - bp->b_resid;
594 if (diff < n)
595 n = diff;
596 error = uiomove(bp->b_data + on, (int) n, uio);
597 brelse(bp);
598 } while (error == 0 && uio->uio_resid > 0 && n != 0);
599 if (!isadir && (error == 0 || uio->uio_resid != orig_resid) &&
600 (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
601 dep->de_flag |= DE_ACCESS;
602 return (error);
603 }
604
605 /*
606 * Write data to a file or directory.
607 */
608 static int
609 msdosfs_write(struct vop_write_args *ap)
610 {
611 int n;
612 int croffset;
613 ssize_t resid;
614 u_long osize;
615 int error = 0;
616 u_long count;
617 int seqcount;
618 daddr_t bn, lastcn;
619 struct buf *bp;
620 int ioflag = ap->a_ioflag;
621 struct uio *uio = ap->a_uio;
622 struct vnode *vp = ap->a_vp;
623 struct vnode *thisvp;
624 struct denode *dep = VTODE(vp);
625 struct msdosfsmount *pmp = dep->de_pmp;
626 struct ucred *cred = ap->a_cred;
627
628 #ifdef MSDOSFS_DEBUG
629 printf("msdosfs_write(vp %p, uio %p, ioflag %x, cred %p\n",
630 vp, uio, ioflag, cred);
631 printf("msdosfs_write(): diroff %lu, dirclust %lu, startcluster %lu\n",
632 dep->de_diroffset, dep->de_dirclust, dep->de_StartCluster);
633 #endif
634
635 switch (vp->v_type) {
636 case VREG:
637 if (ioflag & IO_APPEND)
638 uio->uio_offset = dep->de_FileSize;
639 thisvp = vp;
640 break;
641 case VDIR:
642 return EISDIR;
643 default:
644 panic("msdosfs_write(): bad file type");
645 }
646
647 /*
648 * This is needed (unlike in ffs_write()) because we extend the
649 * file outside of the loop but we don't want to extend the file
650 * for writes of 0 bytes.
651 */
652 if (uio->uio_resid == 0)
653 return (0);
654
655 /*
656 * The caller is supposed to ensure that
657 * uio->uio_offset >= 0 and uio->uio_resid >= 0.
658 */
659 if ((uoff_t)uio->uio_offset + uio->uio_resid > DOS_FILESIZE_MAX)
660 return (EFBIG);
661
662 /*
663 * If they've exceeded their filesize limit, tell them about it.
664 */
665 if (vn_rlimit_fsize(vp, uio, uio->uio_td))
666 return (EFBIG);
667
668 /*
669 * If the offset we are starting the write at is beyond the end of
670 * the file, then they've done a seek. Unix filesystems allow
671 * files with holes in them, DOS doesn't so we must fill the hole
672 * with zeroed blocks.
673 */
674 if (uio->uio_offset > dep->de_FileSize) {
675 error = deextend(dep, uio->uio_offset, cred);
676 if (error)
677 return (error);
678 }
679
680 /*
681 * Remember some values in case the write fails.
682 */
683 resid = uio->uio_resid;
684 osize = dep->de_FileSize;
685
686 /*
687 * If we write beyond the end of the file, extend it to its ultimate
688 * size ahead of the time to hopefully get a contiguous area.
689 */
690 if (uio->uio_offset + resid > osize) {
691 count = de_clcount(pmp, uio->uio_offset + resid) -
692 de_clcount(pmp, osize);
693 error = extendfile(dep, count, NULL, NULL, 0);
694 if (error && (error != ENOSPC || (ioflag & IO_UNIT)))
695 goto errexit;
696 lastcn = dep->de_fc[FC_LASTFC].fc_frcn;
697 } else
698 lastcn = de_clcount(pmp, osize) - 1;
699
700 seqcount = ioflag >> IO_SEQSHIFT;
701 do {
702 if (de_cluster(pmp, uio->uio_offset) > lastcn) {
703 error = ENOSPC;
704 break;
705 }
706
707 croffset = uio->uio_offset & pmp->pm_crbomask;
708 n = min(uio->uio_resid, pmp->pm_bpcluster - croffset);
709 if (uio->uio_offset + n > dep->de_FileSize) {
710 dep->de_FileSize = uio->uio_offset + n;
711 /* The object size needs to be set before buffer is allocated */
712 vnode_pager_setsize(vp, dep->de_FileSize);
713 }
714
715 bn = de_cluster(pmp, uio->uio_offset);
716 if ((uio->uio_offset & pmp->pm_crbomask) == 0
717 && (de_cluster(pmp, uio->uio_offset + uio->uio_resid)
718 > de_cluster(pmp, uio->uio_offset)
719 || uio->uio_offset + uio->uio_resid >= dep->de_FileSize)) {
720 /*
721 * If either the whole cluster gets written,
722 * or we write the cluster from its start beyond EOF,
723 * then no need to read data from disk.
724 */
725 bp = getblk(thisvp, bn, pmp->pm_bpcluster, 0, 0, 0);
726 vfs_bio_clrbuf(bp);
727 /*
728 * Do the bmap now, since pcbmap needs buffers
729 * for the fat table. (see msdosfs_strategy)
730 */
731 if (bp->b_blkno == bp->b_lblkno) {
732 error = pcbmap(dep, bp->b_lblkno, &bn, 0, 0);
733 if (error)
734 bp->b_blkno = -1;
735 else
736 bp->b_blkno = bn;
737 }
738 if (bp->b_blkno == -1) {
739 brelse(bp);
740 if (!error)
741 error = EIO; /* XXX */
742 break;
743 }
744 } else {
745 /*
746 * The block we need to write into exists, so read it in.
747 */
748 error = bread(thisvp, bn, pmp->pm_bpcluster, cred, &bp);
749 if (error) {
750 brelse(bp);
751 break;
752 }
753 }
754
755 /*
756 * Should these vnode_pager_* functions be done on dir
757 * files?
758 */
759
760 /*
761 * Copy the data from user space into the buf header.
762 */
763 error = uiomove(bp->b_data + croffset, n, uio);
764 if (error) {
765 brelse(bp);
766 break;
767 }
768
769 /* Prepare for clustered writes in some else clauses. */
770 if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0)
771 bp->b_flags |= B_CLUSTEROK;
772
773 /*
774 * If IO_SYNC, then each buffer is written synchronously.
775 * Otherwise, if we have a severe page deficiency then
776 * write the buffer asynchronously. Otherwise, if on a
777 * cluster boundary then write the buffer asynchronously,
778 * combining it with contiguous clusters if permitted and
779 * possible, since we don't expect more writes into this
780 * buffer soon. Otherwise, do a delayed write because we
781 * expect more writes into this buffer soon.
782 */
783 if (ioflag & IO_SYNC)
784 (void)bwrite(bp);
785 else if (vm_page_count_severe() || buf_dirty_count_severe())
786 bawrite(bp);
787 else if (n + croffset == pmp->pm_bpcluster) {
788 if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0)
789 cluster_write(vp, bp, dep->de_FileSize,
790 seqcount, 0);
791 else
792 bawrite(bp);
793 } else
794 bdwrite(bp);
795 dep->de_flag |= DE_UPDATE;
796 } while (error == 0 && uio->uio_resid > 0);
797
798 /*
799 * If the write failed and they want us to, truncate the file back
800 * to the size it was before the write was attempted.
801 */
802 errexit:
803 if (error) {
804 if (ioflag & IO_UNIT) {
805 detrunc(dep, osize, ioflag & IO_SYNC, NOCRED);
806 uio->uio_offset -= resid - uio->uio_resid;
807 uio->uio_resid = resid;
808 } else {
809 detrunc(dep, dep->de_FileSize, ioflag & IO_SYNC, NOCRED);
810 if (uio->uio_resid != resid)
811 error = 0;
812 }
813 } else if (ioflag & IO_SYNC)
814 error = deupdat(dep, 1);
815 return (error);
816 }
817
818 /*
819 * Flush the blocks of a file to disk.
820 */
821 static int
822 msdosfs_fsync(struct vop_fsync_args *ap)
823 {
824 struct vnode *devvp;
825 int allerror, error;
826
827 vop_stdfsync(ap);
828
829 /*
830 * If the syncing request comes from fsync(2), sync the entire
831 * FAT and any other metadata that happens to be on devvp. We
832 * need this mainly for the FAT. We write the FAT sloppily, and
833 * syncing it all now is the best we can easily do to get all
834 * directory entries associated with the file (not just the file)
835 * fully synced. The other metadata includes critical metadata
836 * for all directory entries, but only in the MNT_ASYNC case. We
837 * will soon sync all metadata in the file's directory entry.
838 * Non-critical metadata for associated directory entries only
839 * gets synced accidentally, as in most file systems.
840 */
841 if (ap->a_waitfor == MNT_WAIT) {
842 devvp = VTODE(ap->a_vp)->de_pmp->pm_devvp;
843 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
844 allerror = VOP_FSYNC(devvp, MNT_WAIT, ap->a_td);
845 VOP_UNLOCK(devvp, 0);
846 } else
847 allerror = 0;
848
849 error = deupdat(VTODE(ap->a_vp), ap->a_waitfor == MNT_WAIT);
850 if (allerror == 0)
851 allerror = error;
852 return (allerror);
853 }
854
855 static int
856 msdosfs_remove(struct vop_remove_args *ap)
857 {
858 struct denode *dep = VTODE(ap->a_vp);
859 struct denode *ddep = VTODE(ap->a_dvp);
860 int error;
861
862 if (ap->a_vp->v_type == VDIR)
863 error = EPERM;
864 else
865 error = removede(ddep, dep);
866 #ifdef MSDOSFS_DEBUG
867 printf("msdosfs_remove(), dep %p, v_usecount %d\n", dep, ap->a_vp->v_usecount);
868 #endif
869 return (error);
870 }
871
872 /*
873 * DOS filesystems don't know what links are.
874 */
875 static int
876 msdosfs_link(struct vop_link_args *ap)
877 {
878 return (EOPNOTSUPP);
879 }
880
881 /*
882 * Renames on files require moving the denode to a new hash queue since the
883 * denode's location is used to compute which hash queue to put the file
884 * in. Unless it is a rename in place. For example "mv a b".
885 *
886 * What follows is the basic algorithm:
887 *
888 * if (file move) {
889 * if (dest file exists) {
890 * remove dest file
891 * }
892 * if (dest and src in same directory) {
893 * rewrite name in existing directory slot
894 * } else {
895 * write new entry in dest directory
896 * update offset and dirclust in denode
897 * move denode to new hash chain
898 * clear old directory entry
899 * }
900 * } else {
901 * directory move
902 * if (dest directory exists) {
903 * if (dest is not empty) {
904 * return ENOTEMPTY
905 * }
906 * remove dest directory
907 * }
908 * if (dest and src in same directory) {
909 * rewrite name in existing entry
910 * } else {
911 * be sure dest is not a child of src directory
912 * write entry in dest directory
913 * update "." and ".." in moved directory
914 * clear old directory entry for moved directory
915 * }
916 * }
917 *
918 * On entry:
919 * source's parent directory is unlocked
920 * source file or directory is unlocked
921 * destination's parent directory is locked
922 * destination file or directory is locked if it exists
923 *
924 * On exit:
925 * all denodes should be released
926 */
927 static int
928 msdosfs_rename(struct vop_rename_args *ap)
929 {
930 struct vnode *tdvp = ap->a_tdvp;
931 struct vnode *fvp = ap->a_fvp;
932 struct vnode *fdvp = ap->a_fdvp;
933 struct vnode *tvp = ap->a_tvp;
934 struct componentname *tcnp = ap->a_tcnp;
935 struct componentname *fcnp = ap->a_fcnp;
936 struct denode *ip, *xp, *dp, *zp;
937 u_char toname[12], oldname[11];
938 u_long from_diroffset, to_diroffset;
939 u_char to_count;
940 int doingdirectory = 0, newparent = 0;
941 int error;
942 u_long cn, pcl;
943 daddr_t bn;
944 struct msdosfsmount *pmp;
945 struct direntry *dotdotp;
946 struct buf *bp;
947
948 pmp = VFSTOMSDOSFS(fdvp->v_mount);
949
950 #ifdef DIAGNOSTIC
951 if ((tcnp->cn_flags & HASBUF) == 0 ||
952 (fcnp->cn_flags & HASBUF) == 0)
953 panic("msdosfs_rename: no name");
954 #endif
955 /*
956 * Check for cross-device rename.
957 */
958 if (fvp->v_mount != tdvp->v_mount ||
959 (tvp && fvp->v_mount != tvp->v_mount)) {
960 error = EXDEV;
961 abortit:
962 if (tdvp == tvp)
963 vrele(tdvp);
964 else
965 vput(tdvp);
966 if (tvp)
967 vput(tvp);
968 vrele(fdvp);
969 vrele(fvp);
970 return (error);
971 }
972
973 /*
974 * If source and dest are the same, do nothing.
975 */
976 if (tvp == fvp) {
977 error = 0;
978 goto abortit;
979 }
980
981 error = vn_lock(fvp, LK_EXCLUSIVE);
982 if (error)
983 goto abortit;
984 dp = VTODE(fdvp);
985 ip = VTODE(fvp);
986
987 /*
988 * Be sure we are not renaming ".", "..", or an alias of ".". This
989 * leads to a crippled directory tree. It's pretty tough to do a
990 * "ls" or "pwd" with the "." directory entry missing, and "cd .."
991 * doesn't work if the ".." entry is missing.
992 */
993 if (ip->de_Attributes & ATTR_DIRECTORY) {
994 /*
995 * Avoid ".", "..", and aliases of "." for obvious reasons.
996 */
997 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
998 dp == ip ||
999 (fcnp->cn_flags & ISDOTDOT) ||
1000 (tcnp->cn_flags & ISDOTDOT) ||
1001 (ip->de_flag & DE_RENAME)) {
1002 VOP_UNLOCK(fvp, 0);
1003 error = EINVAL;
1004 goto abortit;
1005 }
1006 ip->de_flag |= DE_RENAME;
1007 doingdirectory++;
1008 }
1009
1010 /*
1011 * When the target exists, both the directory
1012 * and target vnodes are returned locked.
1013 */
1014 dp = VTODE(tdvp);
1015 xp = tvp ? VTODE(tvp) : NULL;
1016 /*
1017 * Remember direntry place to use for destination
1018 */
1019 to_diroffset = dp->de_fndoffset;
1020 to_count = dp->de_fndcnt;
1021
1022 /*
1023 * If ".." must be changed (ie the directory gets a new
1024 * parent) then the source directory must not be in the
1025 * directory hierarchy above the target, as this would
1026 * orphan everything below the source directory. Also
1027 * the user must have write permission in the source so
1028 * as to be able to change "..". We must repeat the call
1029 * to namei, as the parent directory is unlocked by the
1030 * call to doscheckpath().
1031 */
1032 error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread);
1033 VOP_UNLOCK(fvp, 0);
1034 if (VTODE(fdvp)->de_StartCluster != VTODE(tdvp)->de_StartCluster)
1035 newparent = 1;
1036 if (doingdirectory && newparent) {
1037 if (error) /* write access check above */
1038 goto bad;
1039 if (xp != NULL)
1040 vput(tvp);
1041 /*
1042 * doscheckpath() vput()'s dp,
1043 * so we have to do a relookup afterwards
1044 */
1045 error = doscheckpath(ip, dp);
1046 if (error)
1047 goto out;
1048 if ((tcnp->cn_flags & SAVESTART) == 0)
1049 panic("msdosfs_rename: lost to startdir");
1050 error = relookup(tdvp, &tvp, tcnp);
1051 if (error)
1052 goto out;
1053 dp = VTODE(tdvp);
1054 xp = tvp ? VTODE(tvp) : NULL;
1055 }
1056
1057 if (xp != NULL) {
1058 /*
1059 * Target must be empty if a directory and have no links
1060 * to it. Also, ensure source and target are compatible
1061 * (both directories, or both not directories).
1062 */
1063 if (xp->de_Attributes & ATTR_DIRECTORY) {
1064 if (!dosdirempty(xp)) {
1065 error = ENOTEMPTY;
1066 goto bad;
1067 }
1068 if (!doingdirectory) {
1069 error = ENOTDIR;
1070 goto bad;
1071 }
1072 cache_purge(tdvp);
1073 } else if (doingdirectory) {
1074 error = EISDIR;
1075 goto bad;
1076 }
1077 error = removede(dp, xp);
1078 if (error)
1079 goto bad;
1080 vput(tvp);
1081 xp = NULL;
1082 }
1083
1084 /*
1085 * Convert the filename in tcnp into a dos filename. We copy this
1086 * into the denode and directory entry for the destination
1087 * file/directory.
1088 */
1089 error = uniqdosname(VTODE(tdvp), tcnp, toname);
1090 if (error)
1091 goto abortit;
1092
1093 /*
1094 * Since from wasn't locked at various places above,
1095 * have to do a relookup here.
1096 */
1097 fcnp->cn_flags &= ~MODMASK;
1098 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
1099 if ((fcnp->cn_flags & SAVESTART) == 0)
1100 panic("msdosfs_rename: lost from startdir");
1101 if (!newparent)
1102 VOP_UNLOCK(tdvp, 0);
1103 if (relookup(fdvp, &fvp, fcnp) == 0)
1104 vrele(fdvp);
1105 if (fvp == NULL) {
1106 /*
1107 * From name has disappeared.
1108 */
1109 if (doingdirectory)
1110 panic("rename: lost dir entry");
1111 if (newparent)
1112 VOP_UNLOCK(tdvp, 0);
1113 vrele(tdvp);
1114 vrele(ap->a_fvp);
1115 return 0;
1116 }
1117 xp = VTODE(fvp);
1118 zp = VTODE(fdvp);
1119 from_diroffset = zp->de_fndoffset;
1120
1121 /*
1122 * Ensure that the directory entry still exists and has not
1123 * changed till now. If the source is a file the entry may
1124 * have been unlinked or renamed. In either case there is
1125 * no further work to be done. If the source is a directory
1126 * then it cannot have been rmdir'ed or renamed; this is
1127 * prohibited by the DE_RENAME flag.
1128 */
1129 if (xp != ip) {
1130 if (doingdirectory)
1131 panic("rename: lost dir entry");
1132 VOP_UNLOCK(fvp, 0);
1133 if (newparent)
1134 VOP_UNLOCK(fdvp, 0);
1135 vrele(ap->a_fvp);
1136 xp = NULL;
1137 } else {
1138 vrele(fvp);
1139 xp = NULL;
1140
1141 /*
1142 * First write a new entry in the destination
1143 * directory and mark the entry in the source directory
1144 * as deleted. Then move the denode to the correct hash
1145 * chain for its new location in the filesystem. And, if
1146 * we moved a directory, then update its .. entry to point
1147 * to the new parent directory.
1148 */
1149 bcopy(ip->de_Name, oldname, 11);
1150 bcopy(toname, ip->de_Name, 11); /* update denode */
1151 dp->de_fndoffset = to_diroffset;
1152 dp->de_fndcnt = to_count;
1153 error = createde(ip, dp, (struct denode **)0, tcnp);
1154 if (error) {
1155 bcopy(oldname, ip->de_Name, 11);
1156 if (newparent)
1157 VOP_UNLOCK(fdvp, 0);
1158 VOP_UNLOCK(fvp, 0);
1159 goto bad;
1160 }
1161 /*
1162 * If ip is for a directory, then its name should always
1163 * be "." since it is for the directory entry in the
1164 * directory itself (msdosfs_lookup() always translates
1165 * to the "." entry so as to get a unique denode, except
1166 * for the root directory there are different
1167 * complications). However, we just corrupted its name
1168 * to pass the correct name to createde(). Undo this.
1169 */
1170 if ((ip->de_Attributes & ATTR_DIRECTORY) != 0)
1171 bcopy(oldname, ip->de_Name, 11);
1172 ip->de_refcnt++;
1173 zp->de_fndoffset = from_diroffset;
1174 error = removede(zp, ip);
1175 if (error) {
1176 /* XXX should downgrade to ro here, fs is corrupt */
1177 if (newparent)
1178 VOP_UNLOCK(fdvp, 0);
1179 VOP_UNLOCK(fvp, 0);
1180 goto bad;
1181 }
1182 if (!doingdirectory) {
1183 error = pcbmap(dp, de_cluster(pmp, to_diroffset), 0,
1184 &ip->de_dirclust, 0);
1185 if (error) {
1186 /* XXX should downgrade to ro here, fs is corrupt */
1187 if (newparent)
1188 VOP_UNLOCK(fdvp, 0);
1189 VOP_UNLOCK(fvp, 0);
1190 goto bad;
1191 }
1192 if (ip->de_dirclust == MSDOSFSROOT)
1193 ip->de_diroffset = to_diroffset;
1194 else
1195 ip->de_diroffset = to_diroffset & pmp->pm_crbomask;
1196 }
1197 reinsert(ip);
1198 if (newparent)
1199 VOP_UNLOCK(fdvp, 0);
1200 }
1201
1202 /*
1203 * If we moved a directory to a new parent directory, then we must
1204 * fixup the ".." entry in the moved directory.
1205 */
1206 if (doingdirectory && newparent) {
1207 cn = ip->de_StartCluster;
1208 if (cn == MSDOSFSROOT) {
1209 /* this should never happen */
1210 panic("msdosfs_rename(): updating .. in root directory?");
1211 } else
1212 bn = cntobn(pmp, cn);
1213 error = bread(pmp->pm_devvp, bn, pmp->pm_bpcluster,
1214 NOCRED, &bp);
1215 if (error) {
1216 /* XXX should downgrade to ro here, fs is corrupt */
1217 brelse(bp);
1218 VOP_UNLOCK(fvp, 0);
1219 goto bad;
1220 }
1221 dotdotp = (struct direntry *)bp->b_data + 1;
1222 pcl = dp->de_StartCluster;
1223 if (FAT32(pmp) && pcl == pmp->pm_rootdirblk)
1224 pcl = MSDOSFSROOT;
1225 putushort(dotdotp->deStartCluster, pcl);
1226 if (FAT32(pmp))
1227 putushort(dotdotp->deHighClust, pcl >> 16);
1228 if (DOINGASYNC(fvp))
1229 bdwrite(bp);
1230 else if ((error = bwrite(bp)) != 0) {
1231 /* XXX should downgrade to ro here, fs is corrupt */
1232 VOP_UNLOCK(fvp, 0);
1233 goto bad;
1234 }
1235 }
1236
1237 /*
1238 * The msdosfs lookup is case insensitive. Several aliases may
1239 * be inserted for a single directory entry. As a consequnce,
1240 * name cache purge done by lookup for fvp when DELETE op for
1241 * namei is specified, might be not enough to expunge all
1242 * namecache entries that were installed for this direntry.
1243 */
1244 cache_purge(fvp);
1245 VOP_UNLOCK(fvp, 0);
1246 bad:
1247 if (xp)
1248 vput(tvp);
1249 vput(tdvp);
1250 out:
1251 ip->de_flag &= ~DE_RENAME;
1252 vrele(fdvp);
1253 vrele(fvp);
1254 return (error);
1255
1256 }
1257
1258 static struct {
1259 struct direntry dot;
1260 struct direntry dotdot;
1261 } dosdirtemplate = {
1262 { ". ", /* the . entry */
1263 ATTR_DIRECTORY, /* file attribute */
1264 0, /* reserved */
1265 0, { 0, 0 }, { 0, 0 }, /* create time & date */
1266 { 0, 0 }, /* access date */
1267 { 0, 0 }, /* high bits of start cluster */
1268 { 210, 4 }, { 210, 4 }, /* modify time & date */
1269 { 0, 0 }, /* startcluster */
1270 { 0, 0, 0, 0 } /* filesize */
1271 },
1272 { ".. ", /* the .. entry */
1273 ATTR_DIRECTORY, /* file attribute */
1274 0, /* reserved */
1275 0, { 0, 0 }, { 0, 0 }, /* create time & date */
1276 { 0, 0 }, /* access date */
1277 { 0, 0 }, /* high bits of start cluster */
1278 { 210, 4 }, { 210, 4 }, /* modify time & date */
1279 { 0, 0 }, /* startcluster */
1280 { 0, 0, 0, 0 } /* filesize */
1281 }
1282 };
1283
1284 static int
1285 msdosfs_mkdir(struct vop_mkdir_args *ap)
1286 {
1287 struct componentname *cnp = ap->a_cnp;
1288 struct denode *dep;
1289 struct denode *pdep = VTODE(ap->a_dvp);
1290 struct direntry *denp;
1291 struct msdosfsmount *pmp = pdep->de_pmp;
1292 struct buf *bp;
1293 u_long newcluster, pcl;
1294 int bn;
1295 int error;
1296 struct denode ndirent;
1297 struct timespec ts;
1298
1299 /*
1300 * If this is the root directory and there is no space left we
1301 * can't do anything. This is because the root directory can not
1302 * change size.
1303 */
1304 if (pdep->de_StartCluster == MSDOSFSROOT
1305 && pdep->de_fndoffset >= pdep->de_FileSize) {
1306 error = ENOSPC;
1307 goto bad2;
1308 }
1309
1310 /*
1311 * Allocate a cluster to hold the about to be created directory.
1312 */
1313 error = clusteralloc(pmp, 0, 1, CLUST_EOFE, &newcluster, NULL);
1314 if (error)
1315 goto bad2;
1316
1317 bzero(&ndirent, sizeof(ndirent));
1318 ndirent.de_pmp = pmp;
1319 ndirent.de_flag = DE_ACCESS | DE_CREATE | DE_UPDATE;
1320 getnanotime(&ts);
1321 DETIMES(&ndirent, &ts, &ts, &ts);
1322
1323 /*
1324 * Now fill the cluster with the "." and ".." entries. And write
1325 * the cluster to disk. This way it is there for the parent
1326 * directory to be pointing at if there were a crash.
1327 */
1328 bn = cntobn(pmp, newcluster);
1329 /* always succeeds */
1330 bp = getblk(pmp->pm_devvp, bn, pmp->pm_bpcluster, 0, 0, 0);
1331 bzero(bp->b_data, pmp->pm_bpcluster);
1332 bcopy(&dosdirtemplate, bp->b_data, sizeof dosdirtemplate);
1333 denp = (struct direntry *)bp->b_data;
1334 putushort(denp[0].deStartCluster, newcluster);
1335 putushort(denp[0].deCDate, ndirent.de_CDate);
1336 putushort(denp[0].deCTime, ndirent.de_CTime);
1337 denp[0].deCHundredth = ndirent.de_CHun;
1338 putushort(denp[0].deADate, ndirent.de_ADate);
1339 putushort(denp[0].deMDate, ndirent.de_MDate);
1340 putushort(denp[0].deMTime, ndirent.de_MTime);
1341 pcl = pdep->de_StartCluster;
1342 /*
1343 * Although the root directory has a non-magic starting cluster
1344 * number for FAT32, chkdsk and fsck_msdosfs still require
1345 * references to it in dotdot entries to be magic.
1346 */
1347 if (FAT32(pmp) && pcl == pmp->pm_rootdirblk)
1348 pcl = MSDOSFSROOT;
1349 putushort(denp[1].deStartCluster, pcl);
1350 putushort(denp[1].deCDate, ndirent.de_CDate);
1351 putushort(denp[1].deCTime, ndirent.de_CTime);
1352 denp[1].deCHundredth = ndirent.de_CHun;
1353 putushort(denp[1].deADate, ndirent.de_ADate);
1354 putushort(denp[1].deMDate, ndirent.de_MDate);
1355 putushort(denp[1].deMTime, ndirent.de_MTime);
1356 if (FAT32(pmp)) {
1357 putushort(denp[0].deHighClust, newcluster >> 16);
1358 putushort(denp[1].deHighClust, pcl >> 16);
1359 }
1360
1361 if (DOINGASYNC(ap->a_dvp))
1362 bdwrite(bp);
1363 else if ((error = bwrite(bp)) != 0)
1364 goto bad;
1365
1366 /*
1367 * Now build up a directory entry pointing to the newly allocated
1368 * cluster. This will be written to an empty slot in the parent
1369 * directory.
1370 */
1371 #ifdef DIAGNOSTIC
1372 if ((cnp->cn_flags & HASBUF) == 0)
1373 panic("msdosfs_mkdir: no name");
1374 #endif
1375 error = uniqdosname(pdep, cnp, ndirent.de_Name);
1376 if (error)
1377 goto bad;
1378
1379 ndirent.de_Attributes = ATTR_DIRECTORY;
1380 ndirent.de_LowerCase = 0;
1381 ndirent.de_StartCluster = newcluster;
1382 ndirent.de_FileSize = 0;
1383 error = createde(&ndirent, pdep, &dep, cnp);
1384 if (error)
1385 goto bad;
1386 *ap->a_vpp = DETOV(dep);
1387 return (0);
1388
1389 bad:
1390 clusterfree(pmp, newcluster, NULL);
1391 bad2:
1392 return (error);
1393 }
1394
1395 static int
1396 msdosfs_rmdir(struct vop_rmdir_args *ap)
1397 {
1398 struct vnode *vp = ap->a_vp;
1399 struct vnode *dvp = ap->a_dvp;
1400 struct componentname *cnp = ap->a_cnp;
1401 struct denode *ip, *dp;
1402 int error;
1403
1404 ip = VTODE(vp);
1405 dp = VTODE(dvp);
1406
1407 /*
1408 * Verify the directory is empty (and valid).
1409 * (Rmdir ".." won't be valid since
1410 * ".." will contain a reference to
1411 * the current directory and thus be
1412 * non-empty.)
1413 */
1414 error = 0;
1415 if (!dosdirempty(ip) || ip->de_flag & DE_RENAME) {
1416 error = ENOTEMPTY;
1417 goto out;
1418 }
1419 /*
1420 * Delete the entry from the directory. For dos filesystems this
1421 * gets rid of the directory entry on disk, the in memory copy
1422 * still exists but the de_refcnt is <= 0. This prevents it from
1423 * being found by deget(). When the vput() on dep is done we give
1424 * up access and eventually msdosfs_reclaim() will be called which
1425 * will remove it from the denode cache.
1426 */
1427 error = removede(dp, ip);
1428 if (error)
1429 goto out;
1430 /*
1431 * This is where we decrement the link count in the parent
1432 * directory. Since dos filesystems don't do this we just purge
1433 * the name cache.
1434 */
1435 cache_purge(dvp);
1436 /*
1437 * Truncate the directory that is being deleted.
1438 */
1439 error = detrunc(ip, (u_long)0, IO_SYNC, cnp->cn_cred);
1440 cache_purge(vp);
1441
1442 out:
1443 return (error);
1444 }
1445
1446 /*
1447 * DOS filesystems don't know what symlinks are.
1448 */
1449 static int
1450 msdosfs_symlink(struct vop_symlink_args *ap)
1451 {
1452 return (EOPNOTSUPP);
1453 }
1454
1455 static int
1456 msdosfs_readdir(struct vop_readdir_args *ap)
1457 {
1458 struct mbnambuf nb;
1459 int error = 0;
1460 int diff;
1461 long n;
1462 int blsize;
1463 long on;
1464 u_long cn;
1465 uint64_t fileno;
1466 u_long dirsperblk;
1467 long bias = 0;
1468 daddr_t bn, lbn;
1469 struct buf *bp;
1470 struct denode *dep = VTODE(ap->a_vp);
1471 struct msdosfsmount *pmp = dep->de_pmp;
1472 struct direntry *dentp;
1473 struct dirent dirbuf;
1474 struct uio *uio = ap->a_uio;
1475 u_long *cookies = NULL;
1476 int ncookies = 0;
1477 off_t offset, off;
1478 int chksum = -1;
1479
1480 #ifdef MSDOSFS_DEBUG
1481 printf("msdosfs_readdir(): vp %p, uio %p, cred %p, eofflagp %p\n",
1482 ap->a_vp, uio, ap->a_cred, ap->a_eofflag);
1483 #endif
1484
1485 /*
1486 * msdosfs_readdir() won't operate properly on regular files since
1487 * it does i/o only with the filesystem vnode, and hence can
1488 * retrieve the wrong block from the buffer cache for a plain file.
1489 * So, fail attempts to readdir() on a plain file.
1490 */
1491 if ((dep->de_Attributes & ATTR_DIRECTORY) == 0)
1492 return (ENOTDIR);
1493
1494 /*
1495 * To be safe, initialize dirbuf
1496 */
1497 bzero(dirbuf.d_name, sizeof(dirbuf.d_name));
1498
1499 /*
1500 * If the user buffer is smaller than the size of one dos directory
1501 * entry or the file offset is not a multiple of the size of a
1502 * directory entry, then we fail the read.
1503 */
1504 off = offset = uio->uio_offset;
1505 if (uio->uio_resid < sizeof(struct direntry) ||
1506 (offset & (sizeof(struct direntry) - 1)))
1507 return (EINVAL);
1508
1509 if (ap->a_ncookies) {
1510 ncookies = uio->uio_resid / 16;
1511 cookies = malloc(ncookies * sizeof(u_long), M_TEMP,
1512 M_WAITOK);
1513 *ap->a_cookies = cookies;
1514 *ap->a_ncookies = ncookies;
1515 }
1516
1517 dirsperblk = pmp->pm_BytesPerSec / sizeof(struct direntry);
1518
1519 /*
1520 * If they are reading from the root directory then, we simulate
1521 * the . and .. entries since these don't exist in the root
1522 * directory. We also set the offset bias to make up for having to
1523 * simulate these entries. By this I mean that at file offset 64 we
1524 * read the first entry in the root directory that lives on disk.
1525 */
1526 if (dep->de_StartCluster == MSDOSFSROOT
1527 || (FAT32(pmp) && dep->de_StartCluster == pmp->pm_rootdirblk)) {
1528 #if 0
1529 printf("msdosfs_readdir(): going after . or .. in root dir, offset %d\n",
1530 offset);
1531 #endif
1532 bias = 2 * sizeof(struct direntry);
1533 if (offset < bias) {
1534 for (n = (int)offset / sizeof(struct direntry);
1535 n < 2; n++) {
1536 if (FAT32(pmp))
1537 fileno = (uint64_t)cntobn(pmp,
1538 pmp->pm_rootdirblk)
1539 * dirsperblk;
1540 else
1541 fileno = 1;
1542 if (pmp->pm_flags & MSDOSFS_LARGEFS) {
1543 dirbuf.d_fileno =
1544 msdosfs_fileno_map(pmp->pm_mountp,
1545 fileno);
1546 } else {
1547
1548 dirbuf.d_fileno = (uint32_t)fileno;
1549 }
1550 dirbuf.d_type = DT_DIR;
1551 switch (n) {
1552 case 0:
1553 dirbuf.d_namlen = 1;
1554 strcpy(dirbuf.d_name, ".");
1555 break;
1556 case 1:
1557 dirbuf.d_namlen = 2;
1558 strcpy(dirbuf.d_name, "..");
1559 break;
1560 }
1561 dirbuf.d_reclen = GENERIC_DIRSIZ(&dirbuf);
1562 if (uio->uio_resid < dirbuf.d_reclen)
1563 goto out;
1564 error = uiomove(&dirbuf, dirbuf.d_reclen, uio);
1565 if (error)
1566 goto out;
1567 offset += sizeof(struct direntry);
1568 off = offset;
1569 if (cookies) {
1570 *cookies++ = offset;
1571 if (--ncookies <= 0)
1572 goto out;
1573 }
1574 }
1575 }
1576 }
1577
1578 mbnambuf_init(&nb);
1579 off = offset;
1580 while (uio->uio_resid > 0) {
1581 lbn = de_cluster(pmp, offset - bias);
1582 on = (offset - bias) & pmp->pm_crbomask;
1583 n = min(pmp->pm_bpcluster - on, uio->uio_resid);
1584 diff = dep->de_FileSize - (offset - bias);
1585 if (diff <= 0)
1586 break;
1587 n = min(n, diff);
1588 error = pcbmap(dep, lbn, &bn, &cn, &blsize);
1589 if (error)
1590 break;
1591 error = bread(pmp->pm_devvp, bn, blsize, NOCRED, &bp);
1592 if (error) {
1593 brelse(bp);
1594 return (error);
1595 }
1596 n = min(n, blsize - bp->b_resid);
1597 if (n == 0) {
1598 brelse(bp);
1599 return (EIO);
1600 }
1601
1602 /*
1603 * Convert from dos directory entries to fs-independent
1604 * directory entries.
1605 */
1606 for (dentp = (struct direntry *)(bp->b_data + on);
1607 (char *)dentp < bp->b_data + on + n;
1608 dentp++, offset += sizeof(struct direntry)) {
1609 #if 0
1610 printf("rd: dentp %08x prev %08x crnt %08x deName %02x attr %02x\n",
1611 dentp, prev, crnt, dentp->deName[0], dentp->deAttributes);
1612 #endif
1613 /*
1614 * If this is an unused entry, we can stop.
1615 */
1616 if (dentp->deName[0] == SLOT_EMPTY) {
1617 brelse(bp);
1618 goto out;
1619 }
1620 /*
1621 * Skip deleted entries.
1622 */
1623 if (dentp->deName[0] == SLOT_DELETED) {
1624 chksum = -1;
1625 mbnambuf_init(&nb);
1626 continue;
1627 }
1628
1629 /*
1630 * Handle Win95 long directory entries
1631 */
1632 if (dentp->deAttributes == ATTR_WIN95) {
1633 if (pmp->pm_flags & MSDOSFSMNT_SHORTNAME)
1634 continue;
1635 chksum = win2unixfn(&nb,
1636 (struct winentry *)dentp, chksum, pmp);
1637 continue;
1638 }
1639
1640 /*
1641 * Skip volume labels
1642 */
1643 if (dentp->deAttributes & ATTR_VOLUME) {
1644 chksum = -1;
1645 mbnambuf_init(&nb);
1646 continue;
1647 }
1648 /*
1649 * This computation of d_fileno must match
1650 * the computation of va_fileid in
1651 * msdosfs_getattr.
1652 */
1653 if (dentp->deAttributes & ATTR_DIRECTORY) {
1654 fileno = getushort(dentp->deStartCluster);
1655 if (FAT32(pmp))
1656 fileno |= getushort(dentp->deHighClust) << 16;
1657 /* if this is the root directory */
1658 if (fileno == MSDOSFSROOT)
1659 if (FAT32(pmp))
1660 fileno = (uint64_t)cntobn(pmp,
1661 pmp->pm_rootdirblk)
1662 * dirsperblk;
1663 else
1664 fileno = 1;
1665 else
1666 fileno = (uint64_t)cntobn(pmp, fileno) *
1667 dirsperblk;
1668 dirbuf.d_type = DT_DIR;
1669 } else {
1670 fileno = (uoff_t)offset /
1671 sizeof(struct direntry);
1672 dirbuf.d_type = DT_REG;
1673 }
1674 if (pmp->pm_flags & MSDOSFS_LARGEFS) {
1675 dirbuf.d_fileno =
1676 msdosfs_fileno_map(pmp->pm_mountp, fileno);
1677 } else
1678 dirbuf.d_fileno = (uint32_t)fileno;
1679
1680 if (chksum != winChksum(dentp->deName)) {
1681 dirbuf.d_namlen = dos2unixfn(dentp->deName,
1682 (u_char *)dirbuf.d_name,
1683 dentp->deLowerCase |
1684 ((pmp->pm_flags & MSDOSFSMNT_SHORTNAME) ?
1685 (LCASE_BASE | LCASE_EXT) : 0),
1686 pmp);
1687 mbnambuf_init(&nb);
1688 } else
1689 mbnambuf_flush(&nb, &dirbuf);
1690 chksum = -1;
1691 dirbuf.d_reclen = GENERIC_DIRSIZ(&dirbuf);
1692 if (uio->uio_resid < dirbuf.d_reclen) {
1693 brelse(bp);
1694 goto out;
1695 }
1696 error = uiomove(&dirbuf, dirbuf.d_reclen, uio);
1697 if (error) {
1698 brelse(bp);
1699 goto out;
1700 }
1701 if (cookies) {
1702 *cookies++ = offset + sizeof(struct direntry);
1703 if (--ncookies <= 0) {
1704 brelse(bp);
1705 goto out;
1706 }
1707 }
1708 off = offset + sizeof(struct direntry);
1709 }
1710 brelse(bp);
1711 }
1712 out:
1713 /* Subtract unused cookies */
1714 if (ap->a_ncookies)
1715 *ap->a_ncookies -= ncookies;
1716
1717 uio->uio_offset = off;
1718
1719 /*
1720 * Set the eofflag (NFS uses it)
1721 */
1722 if (ap->a_eofflag) {
1723 if (dep->de_FileSize - (offset - bias) <= 0)
1724 *ap->a_eofflag = 1;
1725 else
1726 *ap->a_eofflag = 0;
1727 }
1728 return (error);
1729 }
1730
1731 /*-
1732 * a_vp - pointer to the file's vnode
1733 * a_bn - logical block number within the file (cluster number for us)
1734 * a_bop - where to return the bufobj of the special file containing the fs
1735 * a_bnp - where to return the "physical" block number corresponding to a_bn
1736 * (relative to the special file; units are blocks of size DEV_BSIZE)
1737 * a_runp - where to return the "run past" a_bn. This is the count of logical
1738 * blocks whose physical blocks (together with a_bn's physical block)
1739 * are contiguous.
1740 * a_runb - where to return the "run before" a_bn.
1741 */
1742 static int
1743 msdosfs_bmap(struct vop_bmap_args *ap)
1744 {
1745 struct denode *dep;
1746 struct mount *mp;
1747 struct msdosfsmount *pmp;
1748 struct vnode *vp;
1749 daddr_t runbn;
1750 u_long cn;
1751 int bnpercn, error, maxio, maxrun, run;
1752
1753 vp = ap->a_vp;
1754 dep = VTODE(vp);
1755 pmp = dep->de_pmp;
1756 if (ap->a_bop != NULL)
1757 *ap->a_bop = &pmp->pm_devvp->v_bufobj;
1758 if (ap->a_bnp == NULL)
1759 return (0);
1760 if (ap->a_runp != NULL)
1761 *ap->a_runp = 0;
1762 if (ap->a_runb != NULL)
1763 *ap->a_runb = 0;
1764 cn = ap->a_bn;
1765 if (cn != ap->a_bn)
1766 return (EFBIG);
1767 error = pcbmap(dep, cn, ap->a_bnp, NULL, NULL);
1768 if (error != 0 || (ap->a_runp == NULL && ap->a_runb == NULL))
1769 return (error);
1770
1771 mp = vp->v_mount;
1772 maxio = mp->mnt_iosize_max / mp->mnt_stat.f_iosize;
1773 bnpercn = de_cn2bn(pmp, 1);
1774 if (ap->a_runp != NULL) {
1775 maxrun = ulmin(maxio - 1, pmp->pm_maxcluster - cn);
1776 for (run = 1; run <= maxrun; run++) {
1777 if (pcbmap(dep, cn + run, &runbn, NULL, NULL) != 0 ||
1778 runbn != *ap->a_bnp + run * bnpercn)
1779 break;
1780 }
1781 *ap->a_runp = run - 1;
1782 }
1783 if (ap->a_runb != NULL) {
1784 maxrun = ulmin(maxio - 1, cn);
1785 for (run = 1; run < maxrun; run++) {
1786 if (pcbmap(dep, cn - run, &runbn, NULL, NULL) != 0 ||
1787 runbn != *ap->a_bnp - run * bnpercn)
1788 break;
1789 }
1790 *ap->a_runb = run - 1;
1791 }
1792 return (0);
1793 }
1794
1795 static int
1796 msdosfs_strategy(struct vop_strategy_args *ap)
1797 {
1798 struct buf *bp = ap->a_bp;
1799 struct denode *dep = VTODE(ap->a_vp);
1800 struct bufobj *bo;
1801 int error = 0;
1802 daddr_t blkno;
1803
1804 /*
1805 * If we don't already know the filesystem relative block number
1806 * then get it using pcbmap(). If pcbmap() returns the block
1807 * number as -1 then we've got a hole in the file. DOS filesystems
1808 * don't allow files with holes, so we shouldn't ever see this.
1809 */
1810 if (bp->b_blkno == bp->b_lblkno) {
1811 error = pcbmap(dep, bp->b_lblkno, &blkno, 0, 0);
1812 bp->b_blkno = blkno;
1813 if (error) {
1814 bp->b_error = error;
1815 bp->b_ioflags |= BIO_ERROR;
1816 bufdone(bp);
1817 return (0);
1818 }
1819 if ((long)bp->b_blkno == -1)
1820 vfs_bio_clrbuf(bp);
1821 }
1822 if (bp->b_blkno == -1) {
1823 bufdone(bp);
1824 return (0);
1825 }
1826 /*
1827 * Read/write the block from/to the disk that contains the desired
1828 * file block.
1829 */
1830 bp->b_iooffset = dbtob(bp->b_blkno);
1831 bo = dep->de_pmp->pm_bo;
1832 BO_STRATEGY(bo, bp);
1833 return (0);
1834 }
1835
1836 static int
1837 msdosfs_print(struct vop_print_args *ap)
1838 {
1839 struct denode *dep = VTODE(ap->a_vp);
1840
1841 printf("\tstartcluster %lu, dircluster %lu, diroffset %lu, ",
1842 dep->de_StartCluster, dep->de_dirclust, dep->de_diroffset);
1843 printf("on dev %s\n", devtoname(dep->de_pmp->pm_dev));
1844 return (0);
1845 }
1846
1847 static int
1848 msdosfs_pathconf(struct vop_pathconf_args *ap)
1849 {
1850 struct msdosfsmount *pmp = VTODE(ap->a_vp)->de_pmp;
1851
1852 switch (ap->a_name) {
1853 case _PC_LINK_MAX:
1854 *ap->a_retval = 1;
1855 return (0);
1856 case _PC_NAME_MAX:
1857 *ap->a_retval = pmp->pm_flags & MSDOSFSMNT_LONGNAME ? WIN_MAXLEN : 12;
1858 return (0);
1859 case _PC_PATH_MAX:
1860 *ap->a_retval = PATH_MAX;
1861 return (0);
1862 case _PC_CHOWN_RESTRICTED:
1863 *ap->a_retval = 1;
1864 return (0);
1865 case _PC_NO_TRUNC:
1866 *ap->a_retval = 0;
1867 return (0);
1868 default:
1869 return (EINVAL);
1870 }
1871 /* NOTREACHED */
1872 }
1873
1874 static int
1875 msdosfs_vptofh(struct vop_vptofh_args *ap)
1876 {
1877 struct denode *dep;
1878 struct defid *defhp;
1879
1880 dep = VTODE(ap->a_vp);
1881 defhp = (struct defid *)ap->a_fhp;
1882 defhp->defid_len = sizeof(struct defid);
1883 defhp->defid_dirclust = dep->de_dirclust;
1884 defhp->defid_dirofs = dep->de_diroffset;
1885 /* defhp->defid_gen = dep->de_gen; */
1886 return (0);
1887 }
1888
1889 /* Global vfs data structures for msdosfs */
1890 struct vop_vector msdosfs_vnodeops = {
1891 .vop_default = &default_vnodeops,
1892
1893 .vop_access = msdosfs_access,
1894 .vop_bmap = msdosfs_bmap,
1895 .vop_cachedlookup = msdosfs_lookup,
1896 .vop_open = msdosfs_open,
1897 .vop_close = msdosfs_close,
1898 .vop_create = msdosfs_create,
1899 .vop_fsync = msdosfs_fsync,
1900 .vop_getattr = msdosfs_getattr,
1901 .vop_inactive = msdosfs_inactive,
1902 .vop_link = msdosfs_link,
1903 .vop_lookup = vfs_cache_lookup,
1904 .vop_mkdir = msdosfs_mkdir,
1905 .vop_mknod = msdosfs_mknod,
1906 .vop_pathconf = msdosfs_pathconf,
1907 .vop_print = msdosfs_print,
1908 .vop_read = msdosfs_read,
1909 .vop_readdir = msdosfs_readdir,
1910 .vop_reclaim = msdosfs_reclaim,
1911 .vop_remove = msdosfs_remove,
1912 .vop_rename = msdosfs_rename,
1913 .vop_rmdir = msdosfs_rmdir,
1914 .vop_setattr = msdosfs_setattr,
1915 .vop_strategy = msdosfs_strategy,
1916 .vop_symlink = msdosfs_symlink,
1917 .vop_write = msdosfs_write,
1918 .vop_vptofh = msdosfs_vptofh,
1919 };
Cache object: 19643b73a5cf21fb2429ec5a6cf4462f
|