2 /* $NetBSD: msdosfs_vnops.c,v 1.68 1998/02/10 14:10:04 mrg Exp $ */
5 * SPDX-License-Identifier: BSD-4-Clause
7 * Copyright (C) 1994, 1995, 1997 Wolfgang Solfrank.
8 * Copyright (C) 1994, 1995, 1997 TooLs GmbH.
10 * Original code by Paul Popelka (paulp@uts.amdahl.com) (see below).
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. All advertising materials mentioning features or use of this software
21 * must display the following acknowledgement:
22 * This product includes software developed by TooLs GmbH.
23 * 4. The name of TooLs GmbH may not be used to endorse or promote products
24 * derived from this software without specific prior written permission.
26 * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
27 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
28 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
29 * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
30 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
31 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
32 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
33 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
34 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
35 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
38 * Written by Paul Popelka (paulp@uts.amdahl.com)
40 * You can do anything you want with this software, just don't say you wrote
41 * it, and don't remove this notice.
43 * This software is provided "as is".
45 * The author supplies this software to be publicly redistributed on the
46 * understanding that the author is not responsible for the correct
47 * functioning of this software in any circumstances and is not liable for
48 * any damages caused by this software.
53 #include <sys/param.h>
54 #include <sys/systm.h>
57 #include <sys/clock.h>
58 #include <sys/dirent.h>
60 #include <sys/lockf.h>
61 #include <sys/malloc.h>
62 #include <sys/mount.h>
63 #include <sys/mutex.h>
64 #include <sys/namei.h>
67 #include <sys/sysctl.h>
68 #include <sys/unistd.h>
69 #include <sys/vmmeter.h>
70 #include <sys/vnode.h>
73 #include <vm/vm_extern.h>
74 #include <vm/vnode_pager.h>
76 #include <fs/msdosfs/bpb.h>
77 #include <fs/msdosfs/direntry.h>
78 #include <fs/msdosfs/denode.h>
79 #include <fs/msdosfs/fat.h>
80 #include <fs/msdosfs/msdosfsmount.h>
83 * Prototypes for MSDOSFS vnode operations
85 static vop_create_t msdosfs_create;
86 static vop_mknod_t msdosfs_mknod;
87 static vop_open_t msdosfs_open;
88 static vop_close_t msdosfs_close;
89 static vop_access_t msdosfs_access;
90 static vop_getattr_t msdosfs_getattr;
91 static vop_setattr_t msdosfs_setattr;
92 static vop_read_t msdosfs_read;
93 static vop_write_t msdosfs_write;
94 static vop_fsync_t msdosfs_fsync;
95 static vop_remove_t msdosfs_remove;
96 static vop_link_t msdosfs_link;
97 static vop_rename_t msdosfs_rename;
98 static vop_mkdir_t msdosfs_mkdir;
99 static vop_rmdir_t msdosfs_rmdir;
100 static vop_symlink_t msdosfs_symlink;
101 static vop_readdir_t msdosfs_readdir;
102 static vop_bmap_t msdosfs_bmap;
103 static vop_getpages_t msdosfs_getpages;
104 static vop_strategy_t msdosfs_strategy;
105 static vop_print_t msdosfs_print;
106 static vop_pathconf_t msdosfs_pathconf;
107 static vop_vptofh_t msdosfs_vptofh;
110 * Some general notes:
112 * In the ufs filesystem the inodes, superblocks, and indirect blocks are
113 * read/written using the vnode for the filesystem. Blocks that represent
114 * the contents of a file are read/written using the vnode for the file
115 * (including directories when they are read/written as files). This
116 * presents problems for the dos filesystem because data that should be in
117 * an inode (if dos had them) resides in the directory itself. Since we
118 * must update directory entries without the benefit of having the vnode
119 * for the directory we must use the vnode for the filesystem. This means
120 * that when a directory is actually read/written (via read, write, or
121 * readdir, or seek) we must use the vnode for the filesystem instead of
122 * the vnode for the directory as would happen in ufs. This is to insure we
123 * retrieve the correct block from the buffer cache since the hash value is
124 * based upon the vnode address and the desired block number.
128 * Create a regular file. On entry the directory to contain the file being
129 * created is locked. We must release before we return. We must also free
130 * the pathname buffer pointed at by cnp->cn_pnbuf, always on error, or
131 * only if the SAVESTART bit in cn_flags is clear on success.
134 msdosfs_create(struct vop_create_args *ap)
136 struct componentname *cnp = ap->a_cnp;
137 struct denode ndirent;
139 struct denode *pdep = VTODE(ap->a_dvp);
144 printf("msdosfs_create(cnp %p, vap %p\n", cnp, ap->a_vap);
148 * If this is the root directory and there is no space left we
149 * can't do anything. This is because the root directory can not
152 if (pdep->de_StartCluster == MSDOSFSROOT
153 && pdep->de_fndoffset >= pdep->de_FileSize) {
159 * Create a directory entry for the file, then call createde() to
160 * have it installed. NOTE: DOS files are always executable. We
161 * use the absence of the owner write bit to make the file
165 if ((cnp->cn_flags & HASBUF) == 0)
166 panic("msdosfs_create: no name");
168 memset(&ndirent, 0, sizeof(ndirent));
169 error = uniqdosname(pdep, cnp, ndirent.de_Name);
173 ndirent.de_Attributes = ATTR_ARCHIVE;
174 ndirent.de_LowerCase = 0;
175 ndirent.de_StartCluster = 0;
176 ndirent.de_FileSize = 0;
177 ndirent.de_pmp = pdep->de_pmp;
178 ndirent.de_flag = DE_ACCESS | DE_CREATE | DE_UPDATE;
180 DETIMES(&ndirent, &ts, &ts, &ts);
181 error = createde(&ndirent, pdep, &dep, cnp);
184 *ap->a_vpp = DETOV(dep);
185 if ((cnp->cn_flags & MAKEENTRY) != 0)
186 cache_enter(ap->a_dvp, *ap->a_vpp, cnp);
194 msdosfs_mknod(struct vop_mknod_args *ap)
201 msdosfs_open(struct vop_open_args *ap)
203 struct denode *dep = VTODE(ap->a_vp);
204 vnode_create_vobject(ap->a_vp, dep->de_FileSize, ap->a_td);
209 msdosfs_close(struct vop_close_args *ap)
211 struct vnode *vp = ap->a_vp;
212 struct denode *dep = VTODE(vp);
216 if (vp->v_usecount > 1) {
218 DETIMES(dep, &ts, &ts, &ts);
225 msdosfs_access(struct vop_access_args *ap)
227 struct vnode *vp = ap->a_vp;
228 struct denode *dep = VTODE(ap->a_vp);
229 struct msdosfsmount *pmp = dep->de_pmp;
231 accmode_t accmode = ap->a_accmode;
233 file_mode = S_IRWXU|S_IRWXG|S_IRWXO;
234 file_mode &= (vp->v_type == VDIR ? pmp->pm_dirmask : pmp->pm_mask);
237 * Disallow writing to directories and regular files if the
238 * filesystem is read-only.
240 if (accmode & VWRITE) {
241 switch (vp->v_type) {
244 if (vp->v_mount->mnt_flag & MNT_RDONLY)
252 return (vaccess(vp->v_type, file_mode, pmp->pm_uid, pmp->pm_gid,
253 ap->a_accmode, ap->a_cred));
257 msdosfs_getattr(struct vop_getattr_args *ap)
259 struct denode *dep = VTODE(ap->a_vp);
260 struct msdosfsmount *pmp = dep->de_pmp;
261 struct vattr *vap = ap->a_vap;
264 u_long dirsperblk = pmp->pm_BytesPerSec / sizeof(struct direntry);
268 DETIMES(dep, &ts, &ts, &ts);
269 vap->va_fsid = dev2udev(pmp->pm_dev);
271 * The following computation of the fileid must be the same as that
272 * used in msdosfs_readdir() to compute d_fileno. If not, pwd
275 if (dep->de_Attributes & ATTR_DIRECTORY) {
276 fileid = (uint64_t)cntobn(pmp, dep->de_StartCluster) *
278 if (dep->de_StartCluster == MSDOSFSROOT)
281 fileid = (uint64_t)cntobn(pmp, dep->de_dirclust) *
283 if (dep->de_dirclust == MSDOSFSROOT)
284 fileid = (uint64_t)roottobn(pmp, 0) * dirsperblk;
285 fileid += (uoff_t)dep->de_diroffset / sizeof(struct direntry);
287 vap->va_fileid = fileid;
289 mode = S_IRWXU|S_IRWXG|S_IRWXO;
290 if (dep->de_Attributes & ATTR_READONLY)
291 mode &= ~(S_IWUSR|S_IWGRP|S_IWOTH);
292 vap->va_mode = mode &
293 (ap->a_vp->v_type == VDIR ? pmp->pm_dirmask : pmp->pm_mask);
294 vap->va_uid = pmp->pm_uid;
295 vap->va_gid = pmp->pm_gid;
297 vap->va_rdev = NODEV;
298 vap->va_size = dep->de_FileSize;
299 fattime2timespec(dep->de_MDate, dep->de_MTime, 0, 0, &vap->va_mtime);
300 vap->va_ctime = vap->va_mtime;
301 if (pmp->pm_flags & MSDOSFSMNT_LONGNAME) {
302 fattime2timespec(dep->de_ADate, 0, 0, 0, &vap->va_atime);
303 fattime2timespec(dep->de_CDate, dep->de_CTime, dep->de_CHun,
304 0, &vap->va_birthtime);
306 vap->va_atime = vap->va_mtime;
307 vap->va_birthtime.tv_sec = -1;
308 vap->va_birthtime.tv_nsec = 0;
311 if (dep->de_Attributes & ATTR_ARCHIVE)
312 vap->va_flags |= UF_ARCHIVE;
313 if (dep->de_Attributes & ATTR_HIDDEN)
314 vap->va_flags |= UF_HIDDEN;
315 if (dep->de_Attributes & ATTR_READONLY)
316 vap->va_flags |= UF_READONLY;
317 if (dep->de_Attributes & ATTR_SYSTEM)
318 vap->va_flags |= UF_SYSTEM;
320 vap->va_blocksize = pmp->pm_bpcluster;
322 (dep->de_FileSize + pmp->pm_crbomask) & ~pmp->pm_crbomask;
323 vap->va_type = ap->a_vp->v_type;
324 vap->va_filerev = dep->de_modrev;
329 msdosfs_setattr(struct vop_setattr_args *ap)
331 struct vnode *vp = ap->a_vp;
332 struct denode *dep = VTODE(ap->a_vp);
333 struct msdosfsmount *pmp = dep->de_pmp;
334 struct vattr *vap = ap->a_vap;
335 struct ucred *cred = ap->a_cred;
336 struct thread *td = curthread;
340 printf("msdosfs_setattr(): vp %p, vap %p, cred %p\n",
341 ap->a_vp, vap, cred);
345 * Check for unsettable attributes.
347 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
348 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
349 (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
350 (vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
352 printf("msdosfs_setattr(): returning EINVAL\n");
353 printf(" va_type %d, va_nlink %llx, va_fsid %llx, va_fileid %llx\n",
354 vap->va_type, (unsigned long long)vap->va_nlink,
355 (unsigned long long)vap->va_fsid,
356 (unsigned long long)vap->va_fileid);
357 printf(" va_blocksize %lx, va_rdev %llx, va_bytes %llx, va_gen %lx\n",
358 vap->va_blocksize, (unsigned long long)vap->va_rdev,
359 (unsigned long long)vap->va_bytes, vap->va_gen);
360 printf(" va_uid %x, va_gid %x\n",
361 vap->va_uid, vap->va_gid);
367 * We don't allow setting attributes on the root directory.
368 * The special case for the root directory is because before
369 * FAT32, the root directory didn't have an entry for itself
370 * (and was otherwise special). With FAT32, the root
371 * directory is not so special, but still doesn't have an
374 if (vp->v_vflag & VV_ROOT)
377 if (vap->va_flags != VNOVAL) {
378 if (vp->v_mount->mnt_flag & MNT_RDONLY)
380 if (cred->cr_uid != pmp->pm_uid) {
381 error = priv_check_cred(cred, PRIV_VFS_ADMIN);
386 * We are very inconsistent about handling unsupported
387 * attributes. We ignored the access time and the
388 * read and execute bits. We were strict for the other
391 if (vap->va_flags & ~(UF_ARCHIVE | UF_HIDDEN | UF_READONLY |
394 if (vap->va_flags & UF_ARCHIVE)
395 dep->de_Attributes |= ATTR_ARCHIVE;
397 dep->de_Attributes &= ~ATTR_ARCHIVE;
398 if (vap->va_flags & UF_HIDDEN)
399 dep->de_Attributes |= ATTR_HIDDEN;
401 dep->de_Attributes &= ~ATTR_HIDDEN;
402 /* We don't allow changing the readonly bit on directories. */
403 if (vp->v_type != VDIR) {
404 if (vap->va_flags & UF_READONLY)
405 dep->de_Attributes |= ATTR_READONLY;
407 dep->de_Attributes &= ~ATTR_READONLY;
409 if (vap->va_flags & UF_SYSTEM)
410 dep->de_Attributes |= ATTR_SYSTEM;
412 dep->de_Attributes &= ~ATTR_SYSTEM;
413 dep->de_flag |= DE_MODIFIED;
416 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
420 if (vp->v_mount->mnt_flag & MNT_RDONLY)
423 if (uid == (uid_t)VNOVAL)
426 if (gid == (gid_t)VNOVAL)
428 if (cred->cr_uid != pmp->pm_uid || uid != pmp->pm_uid ||
429 (gid != pmp->pm_gid && !groupmember(gid, cred))) {
430 error = priv_check_cred(cred, PRIV_VFS_CHOWN);
434 if (uid != pmp->pm_uid || gid != pmp->pm_gid)
438 if (vap->va_size != VNOVAL) {
439 switch (vp->v_type) {
444 * Truncation is only supported for regular files,
445 * Disallow it if the filesystem is read-only.
447 if (vp->v_mount->mnt_flag & MNT_RDONLY)
452 * According to POSIX, the result is unspecified
453 * for file types other than regular files,
454 * directories and shared memory objects. We
455 * don't support any file types except regular
456 * files and directories in this file system, so
457 * this (default) case is unreachable and can do
458 * anything. Keep falling through to detrunc()
463 error = detrunc(dep, vap->va_size, 0, cred);
467 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
468 if (vp->v_mount->mnt_flag & MNT_RDONLY)
470 error = vn_utimes_perm(vp, vap, cred, td);
473 if ((pmp->pm_flags & MSDOSFSMNT_NOWIN95) == 0 &&
474 vap->va_atime.tv_sec != VNOVAL) {
475 dep->de_flag &= ~DE_ACCESS;
476 timespec2fattime(&vap->va_atime, 0,
477 &dep->de_ADate, NULL, NULL);
479 if (vap->va_mtime.tv_sec != VNOVAL) {
480 dep->de_flag &= ~DE_UPDATE;
481 timespec2fattime(&vap->va_mtime, 0,
482 &dep->de_MDate, &dep->de_MTime, NULL);
485 * We don't set the archive bit when modifying the time of
486 * a directory to emulate the Windows/DOS behavior.
488 if (vp->v_type != VDIR)
489 dep->de_Attributes |= ATTR_ARCHIVE;
490 dep->de_flag |= DE_MODIFIED;
493 * DOS files only have the ability to have their writability
494 * attribute set, so we use the owner write bit to set the readonly
497 if (vap->va_mode != (mode_t)VNOVAL) {
498 if (vp->v_mount->mnt_flag & MNT_RDONLY)
500 if (cred->cr_uid != pmp->pm_uid) {
501 error = priv_check_cred(cred, PRIV_VFS_ADMIN);
505 if (vp->v_type != VDIR) {
506 /* We ignore the read and execute bits. */
507 if (vap->va_mode & S_IWUSR)
508 dep->de_Attributes &= ~ATTR_READONLY;
510 dep->de_Attributes |= ATTR_READONLY;
511 dep->de_Attributes |= ATTR_ARCHIVE;
512 dep->de_flag |= DE_MODIFIED;
515 return (deupdat(dep, 0));
519 msdosfs_read(struct vop_read_args *ap)
533 struct vnode *vp = ap->a_vp;
534 struct denode *dep = VTODE(vp);
535 struct msdosfsmount *pmp = dep->de_pmp;
536 struct uio *uio = ap->a_uio;
539 * If they didn't ask for any data, then we are done.
541 orig_resid = uio->uio_resid;
546 * The caller is supposed to ensure that
547 * uio->uio_offset >= 0 and uio->uio_resid >= 0.
548 * We don't need to check for large offsets as in ffs because
549 * dep->de_FileSize <= MSDOSFS_FILESIZE_MAX < OFF_MAX, so large
550 * offsets cannot cause overflow even in theory.
553 seqcount = ap->a_ioflag >> IO_SEQSHIFT;
555 isadir = dep->de_Attributes & ATTR_DIRECTORY;
557 if (uio->uio_offset >= dep->de_FileSize)
559 lbn = de_cluster(pmp, uio->uio_offset);
561 blsize = pmp->pm_bpcluster;
562 on = uio->uio_offset & pmp->pm_crbomask;
564 * If we are operating on a directory file then be sure to
565 * do i/o with the vnode for the filesystem instead of the
566 * vnode for the directory.
569 /* convert cluster # to block # */
570 error = pcbmap(dep, lbn, &lbn, 0, &blsize);
571 if (error == E2BIG) {
576 error = bread(pmp->pm_devvp, lbn, blsize, NOCRED, &bp);
577 } else if (de_cn2off(pmp, rablock) >= dep->de_FileSize) {
578 error = bread(vp, lbn, blsize, NOCRED, &bp);
579 } else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
580 error = cluster_read(vp, dep->de_FileSize, lbn, blsize,
581 NOCRED, on + uio->uio_resid, seqcount, 0, &bp);
582 } else if (seqcount > 1) {
584 error = breadn(vp, lbn,
585 blsize, &rablock, &rasize, 1, NOCRED, &bp);
587 error = bread(vp, lbn, blsize, NOCRED, &bp);
593 diff = pmp->pm_bpcluster - on;
594 n = diff > uio->uio_resid ? uio->uio_resid : diff;
595 diff = dep->de_FileSize - uio->uio_offset;
598 diff = blsize - bp->b_resid;
601 error = vn_io_fault_uiomove(bp->b_data + on, (int) n, uio);
603 } while (error == 0 && uio->uio_resid > 0 && n != 0);
604 if (!isadir && (error == 0 || uio->uio_resid != orig_resid) &&
605 (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
606 dep->de_flag |= DE_ACCESS;
611 * Write data to a file or directory.
614 msdosfs_write(struct vop_write_args *ap)
625 int ioflag = ap->a_ioflag;
626 struct uio *uio = ap->a_uio;
627 struct vnode *vp = ap->a_vp;
628 struct vnode *thisvp;
629 struct denode *dep = VTODE(vp);
630 struct msdosfsmount *pmp = dep->de_pmp;
631 struct ucred *cred = ap->a_cred;
634 printf("msdosfs_write(vp %p, uio %p, ioflag %x, cred %p\n",
635 vp, uio, ioflag, cred);
636 printf("msdosfs_write(): diroff %lu, dirclust %lu, startcluster %lu\n",
637 dep->de_diroffset, dep->de_dirclust, dep->de_StartCluster);
640 switch (vp->v_type) {
642 if (ioflag & IO_APPEND)
643 uio->uio_offset = dep->de_FileSize;
649 panic("msdosfs_write(): bad file type");
653 * This is needed (unlike in ffs_write()) because we extend the
654 * file outside of the loop but we don't want to extend the file
655 * for writes of 0 bytes.
657 if (uio->uio_resid == 0)
661 * The caller is supposed to ensure that
662 * uio->uio_offset >= 0 and uio->uio_resid >= 0.
664 if ((uoff_t)uio->uio_offset + uio->uio_resid > MSDOSFS_FILESIZE_MAX)
668 * If they've exceeded their filesize limit, tell them about it.
670 if (vn_rlimit_fsize(vp, uio, uio->uio_td))
674 * If the offset we are starting the write at is beyond the end of
675 * the file, then they've done a seek. Unix filesystems allow
676 * files with holes in them, DOS doesn't so we must fill the hole
677 * with zeroed blocks.
679 if (uio->uio_offset > dep->de_FileSize) {
680 error = deextend(dep, uio->uio_offset, cred);
686 * Remember some values in case the write fails.
688 resid = uio->uio_resid;
689 osize = dep->de_FileSize;
692 * If we write beyond the end of the file, extend it to its ultimate
693 * size ahead of the time to hopefully get a contiguous area.
695 if (uio->uio_offset + resid > osize) {
696 count = de_clcount(pmp, uio->uio_offset + resid) -
697 de_clcount(pmp, osize);
698 error = extendfile(dep, count, NULL, NULL, 0);
699 if (error && (error != ENOSPC || (ioflag & IO_UNIT)))
701 lastcn = dep->de_fc[FC_LASTFC].fc_frcn;
703 lastcn = de_clcount(pmp, osize) - 1;
705 seqcount = ioflag >> IO_SEQSHIFT;
707 if (de_cluster(pmp, uio->uio_offset) > lastcn) {
712 croffset = uio->uio_offset & pmp->pm_crbomask;
713 n = min(uio->uio_resid, pmp->pm_bpcluster - croffset);
714 if (uio->uio_offset + n > dep->de_FileSize) {
715 dep->de_FileSize = uio->uio_offset + n;
716 /* The object size needs to be set before buffer is allocated */
717 vnode_pager_setsize(vp, dep->de_FileSize);
720 bn = de_cluster(pmp, uio->uio_offset);
721 if ((uio->uio_offset & pmp->pm_crbomask) == 0
722 && (de_cluster(pmp, uio->uio_offset + uio->uio_resid)
723 > de_cluster(pmp, uio->uio_offset)
724 || uio->uio_offset + uio->uio_resid >= dep->de_FileSize)) {
726 * If either the whole cluster gets written,
727 * or we write the cluster from its start beyond EOF,
728 * then no need to read data from disk.
730 bp = getblk(thisvp, bn, pmp->pm_bpcluster, 0, 0, 0);
732 * This call to vfs_bio_clrbuf() ensures that
733 * even if vn_io_fault_uiomove() below faults,
734 * garbage from the newly instantiated buffer
735 * is not exposed to the userspace via mmap().
739 * Do the bmap now, since pcbmap needs buffers
740 * for the FAT table. (see msdosfs_strategy)
742 if (bp->b_blkno == bp->b_lblkno) {
743 error = pcbmap(dep, bp->b_lblkno, &bn, 0, 0);
749 if (bp->b_blkno == -1) {
752 error = EIO; /* XXX */
757 * The block we need to write into exists, so read it in.
759 error = bread(thisvp, bn, pmp->pm_bpcluster, cred, &bp);
766 * Should these vnode_pager_* functions be done on dir
771 * Copy the data from user space into the buf header.
773 error = vn_io_fault_uiomove(bp->b_data + croffset, n, uio);
779 /* Prepare for clustered writes in some else clauses. */
780 if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0)
781 bp->b_flags |= B_CLUSTEROK;
784 * If IO_SYNC, then each buffer is written synchronously.
785 * Otherwise, if we have a severe page deficiency then
786 * write the buffer asynchronously. Otherwise, if on a
787 * cluster boundary then write the buffer asynchronously,
788 * combining it with contiguous clusters if permitted and
789 * possible, since we don't expect more writes into this
790 * buffer soon. Otherwise, do a delayed write because we
791 * expect more writes into this buffer soon.
793 if (ioflag & IO_SYNC)
795 else if (vm_page_count_severe() || buf_dirty_count_severe())
797 else if (n + croffset == pmp->pm_bpcluster) {
798 if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0)
799 cluster_write(vp, bp, dep->de_FileSize,
805 dep->de_flag |= DE_UPDATE;
806 } while (error == 0 && uio->uio_resid > 0);
809 * If the write failed and they want us to, truncate the file back
810 * to the size it was before the write was attempted.
814 if (ioflag & IO_UNIT) {
815 detrunc(dep, osize, ioflag & IO_SYNC, NOCRED);
816 uio->uio_offset -= resid - uio->uio_resid;
817 uio->uio_resid = resid;
819 detrunc(dep, dep->de_FileSize, ioflag & IO_SYNC, NOCRED);
820 if (uio->uio_resid != resid)
823 } else if (ioflag & IO_SYNC)
824 error = deupdat(dep, 1);
829 * Flush the blocks of a file to disk.
832 msdosfs_fsync(struct vop_fsync_args *ap)
840 * If the syncing request comes from fsync(2), sync the entire
841 * FAT and any other metadata that happens to be on devvp. We
842 * need this mainly for the FAT. We write the FAT sloppily, and
843 * syncing it all now is the best we can easily do to get all
844 * directory entries associated with the file (not just the file)
845 * fully synced. The other metadata includes critical metadata
846 * for all directory entries, but only in the MNT_ASYNC case. We
847 * will soon sync all metadata in the file's directory entry.
848 * Non-critical metadata for associated directory entries only
849 * gets synced accidentally, as in most file systems.
851 if (ap->a_waitfor != MNT_NOWAIT) {
852 devvp = VTODE(ap->a_vp)->de_pmp->pm_devvp;
853 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
854 allerror = VOP_FSYNC(devvp, MNT_WAIT, ap->a_td);
859 error = deupdat(VTODE(ap->a_vp), ap->a_waitfor != MNT_NOWAIT);
866 msdosfs_remove(struct vop_remove_args *ap)
868 struct denode *dep = VTODE(ap->a_vp);
869 struct denode *ddep = VTODE(ap->a_dvp);
872 if (ap->a_vp->v_type == VDIR)
875 error = removede(ddep, dep);
877 printf("msdosfs_remove(), dep %p, v_usecount %d\n", dep, ap->a_vp->v_usecount);
883 * DOS filesystems don't know what links are.
886 msdosfs_link(struct vop_link_args *ap)
892 * Renames on files require moving the denode to a new hash queue since the
893 * denode's location is used to compute which hash queue to put the file
894 * in. Unless it is a rename in place. For example "mv a b".
896 * What follows is the basic algorithm:
899 * if (dest file exists) {
902 * if (dest and src in same directory) {
903 * rewrite name in existing directory slot
905 * write new entry in dest directory
906 * update offset and dirclust in denode
907 * move denode to new hash chain
908 * clear old directory entry
912 * if (dest directory exists) {
913 * if (dest is not empty) {
916 * remove dest directory
918 * if (dest and src in same directory) {
919 * rewrite name in existing entry
921 * be sure dest is not a child of src directory
922 * write entry in dest directory
923 * update "." and ".." in moved directory
924 * clear old directory entry for moved directory
929 * source's parent directory is unlocked
930 * source file or directory is unlocked
931 * destination's parent directory is locked
932 * destination file or directory is locked if it exists
935 * all denodes should be released
938 msdosfs_rename(struct vop_rename_args *ap)
940 struct vnode *tdvp = ap->a_tdvp;
941 struct vnode *fvp = ap->a_fvp;
942 struct vnode *fdvp = ap->a_fdvp;
943 struct vnode *tvp = ap->a_tvp;
944 struct componentname *tcnp = ap->a_tcnp;
945 struct componentname *fcnp = ap->a_fcnp;
946 struct denode *ip, *xp, *dp, *zp;
947 u_char toname[12], oldname[11];
948 u_long from_diroffset, to_diroffset;
950 int doingdirectory = 0, newparent = 0;
954 struct msdosfsmount *pmp;
955 struct direntry *dotdotp;
958 pmp = VFSTOMSDOSFS(fdvp->v_mount);
961 if ((tcnp->cn_flags & HASBUF) == 0 ||
962 (fcnp->cn_flags & HASBUF) == 0)
963 panic("msdosfs_rename: no name");
966 * Check for cross-device rename.
968 if (fvp->v_mount != tdvp->v_mount ||
969 (tvp && fvp->v_mount != tvp->v_mount)) {
984 * If source and dest are the same, do nothing.
991 error = vn_lock(fvp, LK_EXCLUSIVE);
998 * Be sure we are not renaming ".", "..", or an alias of ".". This
999 * leads to a crippled directory tree. It's pretty tough to do a
1000 * "ls" or "pwd" with the "." directory entry missing, and "cd .."
1001 * doesn't work if the ".." entry is missing.
1003 if (ip->de_Attributes & ATTR_DIRECTORY) {
1005 * Avoid ".", "..", and aliases of "." for obvious reasons.
1007 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
1009 (fcnp->cn_flags & ISDOTDOT) ||
1010 (tcnp->cn_flags & ISDOTDOT) ||
1011 (ip->de_flag & DE_RENAME)) {
1016 ip->de_flag |= DE_RENAME;
1021 * When the target exists, both the directory
1022 * and target vnodes are returned locked.
1025 xp = tvp ? VTODE(tvp) : NULL;
1027 * Remember direntry place to use for destination
1029 to_diroffset = dp->de_fndoffset;
1030 to_count = dp->de_fndcnt;
1033 * If ".." must be changed (ie the directory gets a new
1034 * parent) then the source directory must not be in the
1035 * directory hierarchy above the target, as this would
1036 * orphan everything below the source directory. Also
1037 * the user must have write permission in the source so
1038 * as to be able to change "..". We must repeat the call
1039 * to namei, as the parent directory is unlocked by the
1040 * call to doscheckpath().
1042 error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread);
1044 if (VTODE(fdvp)->de_StartCluster != VTODE(tdvp)->de_StartCluster)
1046 if (doingdirectory && newparent) {
1047 if (error) /* write access check above */
1052 * doscheckpath() vput()'s dp,
1053 * so we have to do a relookup afterwards
1055 error = doscheckpath(ip, dp);
1058 if ((tcnp->cn_flags & SAVESTART) == 0)
1059 panic("msdosfs_rename: lost to startdir");
1060 error = relookup(tdvp, &tvp, tcnp);
1064 xp = tvp ? VTODE(tvp) : NULL;
1069 * Target must be empty if a directory and have no links
1070 * to it. Also, ensure source and target are compatible
1071 * (both directories, or both not directories).
1073 if (xp->de_Attributes & ATTR_DIRECTORY) {
1074 if (!dosdirempty(xp)) {
1078 if (!doingdirectory) {
1083 } else if (doingdirectory) {
1087 error = removede(dp, xp);
1095 * Convert the filename in tcnp into a dos filename. We copy this
1096 * into the denode and directory entry for the destination
1099 error = uniqdosname(VTODE(tdvp), tcnp, toname);
1104 * Since from wasn't locked at various places above,
1105 * have to do a relookup here.
1107 fcnp->cn_flags &= ~MODMASK;
1108 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
1109 if ((fcnp->cn_flags & SAVESTART) == 0)
1110 panic("msdosfs_rename: lost from startdir");
1113 if (relookup(fdvp, &fvp, fcnp) == 0)
1117 * From name has disappeared.
1120 panic("rename: lost dir entry");
1129 from_diroffset = zp->de_fndoffset;
1132 * Ensure that the directory entry still exists and has not
1133 * changed till now. If the source is a file the entry may
1134 * have been unlinked or renamed. In either case there is
1135 * no further work to be done. If the source is a directory
1136 * then it cannot have been rmdir'ed or renamed; this is
1137 * prohibited by the DE_RENAME flag.
1141 panic("rename: lost dir entry");
1152 * First write a new entry in the destination
1153 * directory and mark the entry in the source directory
1154 * as deleted. Then move the denode to the correct hash
1155 * chain for its new location in the filesystem. And, if
1156 * we moved a directory, then update its .. entry to point
1157 * to the new parent directory.
1159 memcpy(oldname, ip->de_Name, 11);
1160 memcpy(ip->de_Name, toname, 11); /* update denode */
1161 dp->de_fndoffset = to_diroffset;
1162 dp->de_fndcnt = to_count;
1163 error = createde(ip, dp, (struct denode **)0, tcnp);
1165 memcpy(ip->de_Name, oldname, 11);
1172 * If ip is for a directory, then its name should always
1173 * be "." since it is for the directory entry in the
1174 * directory itself (msdosfs_lookup() always translates
1175 * to the "." entry so as to get a unique denode, except
1176 * for the root directory there are different
1177 * complications). However, we just corrupted its name
1178 * to pass the correct name to createde(). Undo this.
1180 if ((ip->de_Attributes & ATTR_DIRECTORY) != 0)
1181 memcpy(ip->de_Name, oldname, 11);
1183 zp->de_fndoffset = from_diroffset;
1184 error = removede(zp, ip);
1186 /* XXX should downgrade to ro here, fs is corrupt */
1192 if (!doingdirectory) {
1193 error = pcbmap(dp, de_cluster(pmp, to_diroffset), 0,
1194 &ip->de_dirclust, 0);
1196 /* XXX should downgrade to ro here, fs is corrupt */
1202 if (ip->de_dirclust == MSDOSFSROOT)
1203 ip->de_diroffset = to_diroffset;
1205 ip->de_diroffset = to_diroffset & pmp->pm_crbomask;
1213 * If we moved a directory to a new parent directory, then we must
1214 * fixup the ".." entry in the moved directory.
1216 if (doingdirectory && newparent) {
1217 cn = ip->de_StartCluster;
1218 if (cn == MSDOSFSROOT) {
1219 /* this should never happen */
1220 panic("msdosfs_rename(): updating .. in root directory?");
1222 bn = cntobn(pmp, cn);
1223 error = bread(pmp->pm_devvp, bn, pmp->pm_bpcluster,
1226 /* XXX should downgrade to ro here, fs is corrupt */
1230 dotdotp = (struct direntry *)bp->b_data + 1;
1231 pcl = dp->de_StartCluster;
1232 if (FAT32(pmp) && pcl == pmp->pm_rootdirblk)
1234 putushort(dotdotp->deStartCluster, pcl);
1236 putushort(dotdotp->deHighClust, pcl >> 16);
1237 if (DOINGASYNC(fvp))
1239 else if ((error = bwrite(bp)) != 0) {
1240 /* XXX should downgrade to ro here, fs is corrupt */
1247 * The msdosfs lookup is case insensitive. Several aliases may
1248 * be inserted for a single directory entry. As a consequnce,
1249 * name cache purge done by lookup for fvp when DELETE op for
1250 * namei is specified, might be not enough to expunge all
1251 * namecache entries that were installed for this direntry.
1260 ip->de_flag &= ~DE_RENAME;
1268 struct direntry dot;
1269 struct direntry dotdot;
1270 } dosdirtemplate = {
1271 { ". ", /* the . entry */
1272 ATTR_DIRECTORY, /* file attribute */
1274 0, { 0, 0 }, { 0, 0 }, /* create time & date */
1275 { 0, 0 }, /* access date */
1276 { 0, 0 }, /* high bits of start cluster */
1277 { 210, 4 }, { 210, 4 }, /* modify time & date */
1278 { 0, 0 }, /* startcluster */
1279 { 0, 0, 0, 0 } /* filesize */
1281 { ".. ", /* the .. entry */
1282 ATTR_DIRECTORY, /* file attribute */
1284 0, { 0, 0 }, { 0, 0 }, /* create time & date */
1285 { 0, 0 }, /* access date */
1286 { 0, 0 }, /* high bits of start cluster */
1287 { 210, 4 }, { 210, 4 }, /* modify time & date */
1288 { 0, 0 }, /* startcluster */
1289 { 0, 0, 0, 0 } /* filesize */
1294 msdosfs_mkdir(struct vop_mkdir_args *ap)
1296 struct componentname *cnp = ap->a_cnp;
1298 struct denode *pdep = VTODE(ap->a_dvp);
1299 struct direntry *denp;
1300 struct msdosfsmount *pmp = pdep->de_pmp;
1302 u_long newcluster, pcl;
1305 struct denode ndirent;
1309 * If this is the root directory and there is no space left we
1310 * can't do anything. This is because the root directory can not
1313 if (pdep->de_StartCluster == MSDOSFSROOT
1314 && pdep->de_fndoffset >= pdep->de_FileSize) {
1320 * Allocate a cluster to hold the about to be created directory.
1322 error = clusteralloc(pmp, 0, 1, CLUST_EOFE, &newcluster, NULL);
1326 memset(&ndirent, 0, sizeof(ndirent));
1327 ndirent.de_pmp = pmp;
1328 ndirent.de_flag = DE_ACCESS | DE_CREATE | DE_UPDATE;
1330 DETIMES(&ndirent, &ts, &ts, &ts);
1333 * Now fill the cluster with the "." and ".." entries. And write
1334 * the cluster to disk. This way it is there for the parent
1335 * directory to be pointing at if there were a crash.
1337 bn = cntobn(pmp, newcluster);
1338 /* always succeeds */
1339 bp = getblk(pmp->pm_devvp, bn, pmp->pm_bpcluster, 0, 0, 0);
1340 memset(bp->b_data, 0, pmp->pm_bpcluster);
1341 memcpy(bp->b_data, &dosdirtemplate, sizeof dosdirtemplate);
1342 denp = (struct direntry *)bp->b_data;
1343 putushort(denp[0].deStartCluster, newcluster);
1344 putushort(denp[0].deCDate, ndirent.de_CDate);
1345 putushort(denp[0].deCTime, ndirent.de_CTime);
1346 denp[0].deCHundredth = ndirent.de_CHun;
1347 putushort(denp[0].deADate, ndirent.de_ADate);
1348 putushort(denp[0].deMDate, ndirent.de_MDate);
1349 putushort(denp[0].deMTime, ndirent.de_MTime);
1350 pcl = pdep->de_StartCluster;
1352 * Although the root directory has a non-magic starting cluster
1353 * number for FAT32, chkdsk and fsck_msdosfs still require
1354 * references to it in dotdot entries to be magic.
1356 if (FAT32(pmp) && pcl == pmp->pm_rootdirblk)
1358 putushort(denp[1].deStartCluster, pcl);
1359 putushort(denp[1].deCDate, ndirent.de_CDate);
1360 putushort(denp[1].deCTime, ndirent.de_CTime);
1361 denp[1].deCHundredth = ndirent.de_CHun;
1362 putushort(denp[1].deADate, ndirent.de_ADate);
1363 putushort(denp[1].deMDate, ndirent.de_MDate);
1364 putushort(denp[1].deMTime, ndirent.de_MTime);
1366 putushort(denp[0].deHighClust, newcluster >> 16);
1367 putushort(denp[1].deHighClust, pcl >> 16);
1370 if (DOINGASYNC(ap->a_dvp))
1372 else if ((error = bwrite(bp)) != 0)
1376 * Now build up a directory entry pointing to the newly allocated
1377 * cluster. This will be written to an empty slot in the parent
1381 if ((cnp->cn_flags & HASBUF) == 0)
1382 panic("msdosfs_mkdir: no name");
1384 error = uniqdosname(pdep, cnp, ndirent.de_Name);
1388 ndirent.de_Attributes = ATTR_DIRECTORY;
1389 ndirent.de_LowerCase = 0;
1390 ndirent.de_StartCluster = newcluster;
1391 ndirent.de_FileSize = 0;
1392 error = createde(&ndirent, pdep, &dep, cnp);
1395 *ap->a_vpp = DETOV(dep);
1399 clusterfree(pmp, newcluster, NULL);
1405 msdosfs_rmdir(struct vop_rmdir_args *ap)
1407 struct vnode *vp = ap->a_vp;
1408 struct vnode *dvp = ap->a_dvp;
1409 struct componentname *cnp = ap->a_cnp;
1410 struct denode *ip, *dp;
1417 * Verify the directory is empty (and valid).
1418 * (Rmdir ".." won't be valid since
1419 * ".." will contain a reference to
1420 * the current directory and thus be
1424 if (!dosdirempty(ip) || ip->de_flag & DE_RENAME) {
1429 * Delete the entry from the directory. For dos filesystems this
1430 * gets rid of the directory entry on disk, the in memory copy
1431 * still exists but the de_refcnt is <= 0. This prevents it from
1432 * being found by deget(). When the vput() on dep is done we give
1433 * up access and eventually msdosfs_reclaim() will be called which
1434 * will remove it from the denode cache.
1436 error = removede(dp, ip);
1440 * This is where we decrement the link count in the parent
1441 * directory. Since dos filesystems don't do this we just purge
1446 * Truncate the directory that is being deleted.
1448 error = detrunc(ip, (u_long)0, IO_SYNC, cnp->cn_cred);
1456 * DOS filesystems don't know what symlinks are.
1459 msdosfs_symlink(struct vop_symlink_args *ap)
1461 return (EOPNOTSUPP);
1465 msdosfs_readdir(struct vop_readdir_args *ap)
1478 struct denode *dep = VTODE(ap->a_vp);
1479 struct msdosfsmount *pmp = dep->de_pmp;
1480 struct direntry *dentp;
1481 struct dirent dirbuf;
1482 struct uio *uio = ap->a_uio;
1483 u_long *cookies = NULL;
1488 #ifdef MSDOSFS_DEBUG
1489 printf("msdosfs_readdir(): vp %p, uio %p, cred %p, eofflagp %p\n",
1490 ap->a_vp, uio, ap->a_cred, ap->a_eofflag);
1494 * msdosfs_readdir() won't operate properly on regular files since
1495 * it does i/o only with the filesystem vnode, and hence can
1496 * retrieve the wrong block from the buffer cache for a plain file.
1497 * So, fail attempts to readdir() on a plain file.
1499 if ((dep->de_Attributes & ATTR_DIRECTORY) == 0)
1503 * To be safe, initialize dirbuf
1505 memset(dirbuf.d_name, 0, sizeof(dirbuf.d_name));
1508 * If the user buffer is smaller than the size of one dos directory
1509 * entry or the file offset is not a multiple of the size of a
1510 * directory entry, then we fail the read.
1512 off = offset = uio->uio_offset;
1513 if (uio->uio_resid < sizeof(struct direntry) ||
1514 (offset & (sizeof(struct direntry) - 1)))
1517 if (ap->a_ncookies) {
1518 ncookies = uio->uio_resid / 16;
1519 cookies = malloc(ncookies * sizeof(u_long), M_TEMP,
1521 *ap->a_cookies = cookies;
1522 *ap->a_ncookies = ncookies;
1525 dirsperblk = pmp->pm_BytesPerSec / sizeof(struct direntry);
1528 * If they are reading from the root directory then, we simulate
1529 * the . and .. entries since these don't exist in the root
1530 * directory. We also set the offset bias to make up for having to
1531 * simulate these entries. By this I mean that at file offset 64 we
1532 * read the first entry in the root directory that lives on disk.
1534 if (dep->de_StartCluster == MSDOSFSROOT
1535 || (FAT32(pmp) && dep->de_StartCluster == pmp->pm_rootdirblk)) {
1537 printf("msdosfs_readdir(): going after . or .. in root dir, offset %d\n",
1540 bias = 2 * sizeof(struct direntry);
1541 if (offset < bias) {
1542 for (n = (int)offset / sizeof(struct direntry);
1544 dirbuf.d_fileno = FAT32(pmp) ?
1545 (uint64_t)cntobn(pmp, pmp->pm_rootdirblk) *
1547 dirbuf.d_type = DT_DIR;
1550 dirbuf.d_namlen = 1;
1551 dirbuf.d_name[0] = '.';
1554 dirbuf.d_namlen = 2;
1555 dirbuf.d_name[0] = '.';
1556 dirbuf.d_name[1] = '.';
1559 dirbuf.d_reclen = GENERIC_DIRSIZ(&dirbuf);
1560 /* NOTE: d_off is the offset of the *next* entry. */
1561 dirbuf.d_off = offset + sizeof(struct direntry);
1562 dirent_terminate(&dirbuf);
1563 if (uio->uio_resid < dirbuf.d_reclen)
1565 error = uiomove(&dirbuf, dirbuf.d_reclen, uio);
1568 offset += sizeof(struct direntry);
1571 *cookies++ = offset;
1572 if (--ncookies <= 0)
1581 while (uio->uio_resid > 0) {
1582 lbn = de_cluster(pmp, offset - bias);
1583 on = (offset - bias) & pmp->pm_crbomask;
1584 n = min(pmp->pm_bpcluster - on, uio->uio_resid);
1585 diff = dep->de_FileSize - (offset - bias);
1589 error = pcbmap(dep, lbn, &bn, &cn, &blsize);
1592 error = bread(pmp->pm_devvp, bn, blsize, NOCRED, &bp);
1596 n = min(n, blsize - bp->b_resid);
1603 * Convert from dos directory entries to fs-independent
1604 * directory entries.
1606 for (dentp = (struct direntry *)(bp->b_data + on);
1607 (char *)dentp < bp->b_data + on + n;
1608 dentp++, offset += sizeof(struct direntry)) {
1610 printf("rd: dentp %08x prev %08x crnt %08x deName %02x attr %02x\n",
1611 dentp, prev, crnt, dentp->deName[0], dentp->deAttributes);
1614 * If this is an unused entry, we can stop.
1616 if (dentp->deName[0] == SLOT_EMPTY) {
1621 * Skip deleted entries.
1623 if (dentp->deName[0] == SLOT_DELETED) {
1630 * Handle Win95 long directory entries
1632 if (dentp->deAttributes == ATTR_WIN95) {
1633 if (pmp->pm_flags & MSDOSFSMNT_SHORTNAME)
1635 chksum = win2unixfn(&nb,
1636 (struct winentry *)dentp, chksum, pmp);
1641 * Skip volume labels
1643 if (dentp->deAttributes & ATTR_VOLUME) {
1649 * This computation of d_fileno must match
1650 * the computation of va_fileid in
1653 if (dentp->deAttributes & ATTR_DIRECTORY) {
1654 cn = getushort(dentp->deStartCluster);
1656 cn |= getushort(dentp->deHighClust) <<
1658 if (cn == MSDOSFSROOT)
1659 cn = pmp->pm_rootdirblk;
1661 if (cn == MSDOSFSROOT && !FAT32(pmp))
1662 dirbuf.d_fileno = 1;
1664 dirbuf.d_fileno = cntobn(pmp, cn) *
1666 dirbuf.d_type = DT_DIR;
1668 dirbuf.d_fileno = (uoff_t)offset /
1669 sizeof(struct direntry);
1670 dirbuf.d_type = DT_REG;
1673 if (chksum != winChksum(dentp->deName)) {
1674 dirbuf.d_namlen = dos2unixfn(dentp->deName,
1675 (u_char *)dirbuf.d_name,
1676 dentp->deLowerCase |
1677 ((pmp->pm_flags & MSDOSFSMNT_SHORTNAME) ?
1678 (LCASE_BASE | LCASE_EXT) : 0),
1682 mbnambuf_flush(&nb, &dirbuf);
1684 dirbuf.d_reclen = GENERIC_DIRSIZ(&dirbuf);
1685 /* NOTE: d_off is the offset of the *next* entry. */
1686 dirbuf.d_off = offset + sizeof(struct direntry);
1687 if (uio->uio_resid < dirbuf.d_reclen) {
1691 error = uiomove(&dirbuf, dirbuf.d_reclen, uio);
1697 *cookies++ = offset + sizeof(struct direntry);
1698 if (--ncookies <= 0) {
1703 off = offset + sizeof(struct direntry);
1708 /* Subtract unused cookies */
1710 *ap->a_ncookies -= ncookies;
1712 uio->uio_offset = off;
1715 * Set the eofflag (NFS uses it)
1717 if (ap->a_eofflag) {
1718 if (dep->de_FileSize - (offset - bias) <= 0)
1727 * a_vp - pointer to the file's vnode
1728 * a_bn - logical block number within the file (cluster number for us)
1729 * a_bop - where to return the bufobj of the special file containing the fs
1730 * a_bnp - where to return the "physical" block number corresponding to a_bn
1731 * (relative to the special file; units are blocks of size DEV_BSIZE)
1732 * a_runp - where to return the "run past" a_bn. This is the count of logical
1733 * blocks whose physical blocks (together with a_bn's physical block)
1735 * a_runb - where to return the "run before" a_bn.
1738 msdosfs_bmap(struct vop_bmap_args *ap)
1740 struct fatcache savefc;
1743 struct msdosfsmount *pmp;
1747 int bnpercn, error, maxio, maxrun, run;
1752 if (ap->a_bop != NULL)
1753 *ap->a_bop = &pmp->pm_devvp->v_bufobj;
1754 if (ap->a_bnp == NULL)
1756 if (ap->a_runp != NULL)
1758 if (ap->a_runb != NULL)
1763 error = pcbmap(dep, cn, ap->a_bnp, NULL, NULL);
1764 if (error != 0 || (ap->a_runp == NULL && ap->a_runb == NULL))
1768 * Prepare to back out updates of the fatchain cache after the one
1769 * for the first block done by pcbmap() above. Without the backout,
1770 * then whenever the caller doesn't do i/o to all of the blocks that
1771 * we find, the single useful cache entry would be too far in advance
1772 * of the actual i/o to work for the next sequential i/o. Then the
1773 * FAT would be searched from the beginning. With the backout, the
1774 * FAT is searched starting at most a few blocks early. This wastes
1775 * much less time. Time is also wasted finding more blocks than the
1776 * caller will do i/o to. This is necessary because the runlength
1777 * parameters are output-only.
1779 savefc = dep->de_fc[FC_LASTMAP];
1782 maxio = mp->mnt_iosize_max / mp->mnt_stat.f_iosize;
1783 bnpercn = de_cn2bn(pmp, 1);
1784 if (ap->a_runp != NULL) {
1785 maxrun = ulmin(maxio - 1, pmp->pm_maxcluster - cn);
1786 for (run = 1; run <= maxrun; run++) {
1787 if (pcbmap(dep, cn + run, &runbn, NULL, NULL) != 0 ||
1788 runbn != *ap->a_bnp + run * bnpercn)
1791 *ap->a_runp = run - 1;
1793 if (ap->a_runb != NULL) {
1794 maxrun = ulmin(maxio - 1, cn);
1795 for (run = 1; run < maxrun; run++) {
1796 if (pcbmap(dep, cn - run, &runbn, NULL, NULL) != 0 ||
1797 runbn != *ap->a_bnp - run * bnpercn)
1800 *ap->a_runb = run - 1;
1802 dep->de_fc[FC_LASTMAP] = savefc;
1806 SYSCTL_NODE(_vfs, OID_AUTO, msdosfs, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
1807 "msdos filesystem");
1808 static int use_buf_pager = 1;
1809 SYSCTL_INT(_vfs_msdosfs, OID_AUTO, use_buf_pager, CTLFLAG_RWTUN,
1811 "Use buffer pager instead of bmap");
1814 msdosfs_gbp_getblkno(struct vnode *vp, vm_ooffset_t off)
1817 return (de_cluster(VTODE(vp)->de_pmp, off));
1821 msdosfs_gbp_getblksz(struct vnode *vp, daddr_t lbn)
1824 return (VTODE(vp)->de_pmp->pm_bpcluster);
1828 msdosfs_getpages(struct vop_getpages_args *ap)
1832 return (vfs_bio_getpages(ap->a_vp, ap->a_m, ap->a_count,
1833 ap->a_rbehind, ap->a_rahead, msdosfs_gbp_getblkno,
1834 msdosfs_gbp_getblksz));
1835 return (vnode_pager_generic_getpages(ap->a_vp, ap->a_m, ap->a_count,
1836 ap->a_rbehind, ap->a_rahead, NULL, NULL));
1840 msdosfs_strategy(struct vop_strategy_args *ap)
1842 struct buf *bp = ap->a_bp;
1843 struct denode *dep = VTODE(ap->a_vp);
1849 * If we don't already know the filesystem relative block number
1850 * then get it using pcbmap(). If pcbmap() returns the block
1851 * number as -1 then we've got a hole in the file. DOS filesystems
1852 * don't allow files with holes, so we shouldn't ever see this.
1854 if (bp->b_blkno == bp->b_lblkno) {
1855 error = pcbmap(dep, bp->b_lblkno, &blkno, 0, 0);
1856 bp->b_blkno = blkno;
1858 bp->b_error = error;
1859 bp->b_ioflags |= BIO_ERROR;
1863 if ((long)bp->b_blkno == -1)
1866 if (bp->b_blkno == -1) {
1871 * Read/write the block from/to the disk that contains the desired
1874 bp->b_iooffset = dbtob(bp->b_blkno);
1875 bo = dep->de_pmp->pm_bo;
1876 BO_STRATEGY(bo, bp);
1881 msdosfs_print(struct vop_print_args *ap)
1883 struct denode *dep = VTODE(ap->a_vp);
1885 printf("\tstartcluster %lu, dircluster %lu, diroffset %lu, ",
1886 dep->de_StartCluster, dep->de_dirclust, dep->de_diroffset);
1887 printf("on dev %s\n", devtoname(dep->de_pmp->pm_dev));
1892 msdosfs_pathconf(struct vop_pathconf_args *ap)
1894 struct msdosfsmount *pmp = VTODE(ap->a_vp)->de_pmp;
1896 switch (ap->a_name) {
1897 case _PC_FILESIZEBITS:
1904 *ap->a_retval = pmp->pm_flags & MSDOSFSMNT_LONGNAME ? WIN_MAXLEN : 12;
1906 case _PC_CHOWN_RESTRICTED:
1913 return (vop_stdpathconf(ap));
1919 msdosfs_vptofh(struct vop_vptofh_args *ap)
1922 struct defid *defhp;
1924 dep = VTODE(ap->a_vp);
1925 defhp = (struct defid *)ap->a_fhp;
1926 defhp->defid_len = sizeof(struct defid);
1927 defhp->defid_dirclust = dep->de_dirclust;
1928 defhp->defid_dirofs = dep->de_diroffset;
1929 /* defhp->defid_gen = dep->de_gen; */
1933 /* Global vfs data structures for msdosfs */
1934 struct vop_vector msdosfs_vnodeops = {
1935 .vop_default = &default_vnodeops,
1937 .vop_access = msdosfs_access,
1938 .vop_bmap = msdosfs_bmap,
1939 .vop_getpages = msdosfs_getpages,
1940 .vop_cachedlookup = msdosfs_lookup,
1941 .vop_open = msdosfs_open,
1942 .vop_close = msdosfs_close,
1943 .vop_create = msdosfs_create,
1944 .vop_fsync = msdosfs_fsync,
1945 .vop_fdatasync = vop_stdfdatasync_buf,
1946 .vop_getattr = msdosfs_getattr,
1947 .vop_inactive = msdosfs_inactive,
1948 .vop_link = msdosfs_link,
1949 .vop_lookup = vfs_cache_lookup,
1950 .vop_mkdir = msdosfs_mkdir,
1951 .vop_mknod = msdosfs_mknod,
1952 .vop_pathconf = msdosfs_pathconf,
1953 .vop_print = msdosfs_print,
1954 .vop_read = msdosfs_read,
1955 .vop_readdir = msdosfs_readdir,
1956 .vop_reclaim = msdosfs_reclaim,
1957 .vop_remove = msdosfs_remove,
1958 .vop_rename = msdosfs_rename,
1959 .vop_rmdir = msdosfs_rmdir,
1960 .vop_setattr = msdosfs_setattr,
1961 .vop_strategy = msdosfs_strategy,
1962 .vop_symlink = msdosfs_symlink,
1963 .vop_write = msdosfs_write,
1964 .vop_vptofh = msdosfs_vptofh,
1966 VFS_VOP_VECTOR_REGISTER(msdosfs_vnodeops);