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>
82 #define DOS_FILESIZE_MAX 0xffffffff
85 * Prototypes for MSDOSFS vnode operations
87 static vop_create_t msdosfs_create;
88 static vop_mknod_t msdosfs_mknod;
89 static vop_open_t msdosfs_open;
90 static vop_close_t msdosfs_close;
91 static vop_access_t msdosfs_access;
92 static vop_getattr_t msdosfs_getattr;
93 static vop_setattr_t msdosfs_setattr;
94 static vop_read_t msdosfs_read;
95 static vop_write_t msdosfs_write;
96 static vop_fsync_t msdosfs_fsync;
97 static vop_remove_t msdosfs_remove;
98 static vop_link_t msdosfs_link;
99 static vop_rename_t msdosfs_rename;
100 static vop_mkdir_t msdosfs_mkdir;
101 static vop_rmdir_t msdosfs_rmdir;
102 static vop_symlink_t msdosfs_symlink;
103 static vop_readdir_t msdosfs_readdir;
104 static vop_bmap_t msdosfs_bmap;
105 static vop_getpages_t msdosfs_getpages;
106 static vop_strategy_t msdosfs_strategy;
107 static vop_print_t msdosfs_print;
108 static vop_pathconf_t msdosfs_pathconf;
109 static vop_vptofh_t msdosfs_vptofh;
112 * Some general notes:
114 * In the ufs filesystem the inodes, superblocks, and indirect blocks are
115 * read/written using the vnode for the filesystem. Blocks that represent
116 * the contents of a file are read/written using the vnode for the file
117 * (including directories when they are read/written as files). This
118 * presents problems for the dos filesystem because data that should be in
119 * an inode (if dos had them) resides in the directory itself. Since we
120 * must update directory entries without the benefit of having the vnode
121 * for the directory we must use the vnode for the filesystem. This means
122 * that when a directory is actually read/written (via read, write, or
123 * readdir, or seek) we must use the vnode for the filesystem instead of
124 * the vnode for the directory as would happen in ufs. This is to insure we
125 * retrieve the correct block from the buffer cache since the hash value is
126 * based upon the vnode address and the desired block number.
130 * Create a regular file. On entry the directory to contain the file being
131 * created is locked. We must release before we return. We must also free
132 * the pathname buffer pointed at by cnp->cn_pnbuf, always on error, or
133 * only if the SAVESTART bit in cn_flags is clear on success.
136 msdosfs_create(struct vop_create_args *ap)
138 struct componentname *cnp = ap->a_cnp;
139 struct denode ndirent;
141 struct denode *pdep = VTODE(ap->a_dvp);
146 printf("msdosfs_create(cnp %p, vap %p\n", cnp, ap->a_vap);
150 * If this is the root directory and there is no space left we
151 * can't do anything. This is because the root directory can not
154 if (pdep->de_StartCluster == MSDOSFSROOT
155 && pdep->de_fndoffset >= pdep->de_FileSize) {
161 * Create a directory entry for the file, then call createde() to
162 * have it installed. NOTE: DOS files are always executable. We
163 * use the absence of the owner write bit to make the file
167 if ((cnp->cn_flags & HASBUF) == 0)
168 panic("msdosfs_create: no name");
170 memset(&ndirent, 0, sizeof(ndirent));
171 error = uniqdosname(pdep, cnp, ndirent.de_Name);
175 ndirent.de_Attributes = ATTR_ARCHIVE;
176 ndirent.de_LowerCase = 0;
177 ndirent.de_StartCluster = 0;
178 ndirent.de_FileSize = 0;
179 ndirent.de_pmp = pdep->de_pmp;
180 ndirent.de_flag = DE_ACCESS | DE_CREATE | DE_UPDATE;
182 DETIMES(&ndirent, &ts, &ts, &ts);
183 error = createde(&ndirent, pdep, &dep, cnp);
186 *ap->a_vpp = DETOV(dep);
187 if ((cnp->cn_flags & MAKEENTRY) != 0)
188 cache_enter(ap->a_dvp, *ap->a_vpp, cnp);
196 msdosfs_mknod(struct vop_mknod_args *ap)
203 msdosfs_open(struct vop_open_args *ap)
205 struct denode *dep = VTODE(ap->a_vp);
206 vnode_create_vobject(ap->a_vp, dep->de_FileSize, ap->a_td);
211 msdosfs_close(struct vop_close_args *ap)
213 struct vnode *vp = ap->a_vp;
214 struct denode *dep = VTODE(vp);
218 if (vp->v_usecount > 1) {
220 DETIMES(dep, &ts, &ts, &ts);
227 msdosfs_access(struct vop_access_args *ap)
229 struct vnode *vp = ap->a_vp;
230 struct denode *dep = VTODE(ap->a_vp);
231 struct msdosfsmount *pmp = dep->de_pmp;
233 accmode_t accmode = ap->a_accmode;
235 file_mode = S_IRWXU|S_IRWXG|S_IRWXO;
236 file_mode &= (vp->v_type == VDIR ? pmp->pm_dirmask : pmp->pm_mask);
239 * Disallow writing to directories and regular files if the
240 * filesystem is read-only.
242 if (accmode & VWRITE) {
243 switch (vp->v_type) {
246 if (vp->v_mount->mnt_flag & MNT_RDONLY)
254 return (vaccess(vp->v_type, file_mode, pmp->pm_uid, pmp->pm_gid,
255 ap->a_accmode, ap->a_cred, NULL));
259 msdosfs_getattr(struct vop_getattr_args *ap)
261 struct denode *dep = VTODE(ap->a_vp);
262 struct msdosfsmount *pmp = dep->de_pmp;
263 struct vattr *vap = ap->a_vap;
266 u_long dirsperblk = pmp->pm_BytesPerSec / sizeof(struct direntry);
270 DETIMES(dep, &ts, &ts, &ts);
271 vap->va_fsid = dev2udev(pmp->pm_dev);
273 * The following computation of the fileid must be the same as that
274 * used in msdosfs_readdir() to compute d_fileno. If not, pwd
277 if (dep->de_Attributes & ATTR_DIRECTORY) {
278 fileid = (uint64_t)cntobn(pmp, dep->de_StartCluster) *
280 if (dep->de_StartCluster == MSDOSFSROOT)
283 fileid = (uint64_t)cntobn(pmp, dep->de_dirclust) *
285 if (dep->de_dirclust == MSDOSFSROOT)
286 fileid = (uint64_t)roottobn(pmp, 0) * dirsperblk;
287 fileid += (uoff_t)dep->de_diroffset / sizeof(struct direntry);
289 vap->va_fileid = fileid;
291 mode = S_IRWXU|S_IRWXG|S_IRWXO;
292 if (dep->de_Attributes & ATTR_READONLY)
293 mode &= ~(S_IWUSR|S_IWGRP|S_IWOTH);
294 vap->va_mode = mode &
295 (ap->a_vp->v_type == VDIR ? pmp->pm_dirmask : pmp->pm_mask);
296 vap->va_uid = pmp->pm_uid;
297 vap->va_gid = pmp->pm_gid;
299 vap->va_rdev = NODEV;
300 vap->va_size = dep->de_FileSize;
301 fattime2timespec(dep->de_MDate, dep->de_MTime, 0, 0, &vap->va_mtime);
302 vap->va_ctime = vap->va_mtime;
303 if (pmp->pm_flags & MSDOSFSMNT_LONGNAME) {
304 fattime2timespec(dep->de_ADate, 0, 0, 0, &vap->va_atime);
305 fattime2timespec(dep->de_CDate, dep->de_CTime, dep->de_CHun,
306 0, &vap->va_birthtime);
308 vap->va_atime = vap->va_mtime;
309 vap->va_birthtime.tv_sec = -1;
310 vap->va_birthtime.tv_nsec = 0;
313 if (dep->de_Attributes & ATTR_ARCHIVE)
314 vap->va_flags |= UF_ARCHIVE;
315 if (dep->de_Attributes & ATTR_HIDDEN)
316 vap->va_flags |= UF_HIDDEN;
317 if (dep->de_Attributes & ATTR_READONLY)
318 vap->va_flags |= UF_READONLY;
319 if (dep->de_Attributes & ATTR_SYSTEM)
320 vap->va_flags |= UF_SYSTEM;
322 vap->va_blocksize = pmp->pm_bpcluster;
324 (dep->de_FileSize + pmp->pm_crbomask) & ~pmp->pm_crbomask;
325 vap->va_type = ap->a_vp->v_type;
326 vap->va_filerev = dep->de_modrev;
331 msdosfs_setattr(struct vop_setattr_args *ap)
333 struct vnode *vp = ap->a_vp;
334 struct denode *dep = VTODE(ap->a_vp);
335 struct msdosfsmount *pmp = dep->de_pmp;
336 struct vattr *vap = ap->a_vap;
337 struct ucred *cred = ap->a_cred;
338 struct thread *td = curthread;
342 printf("msdosfs_setattr(): vp %p, vap %p, cred %p\n",
343 ap->a_vp, vap, cred);
347 * Check for unsettable attributes.
349 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
350 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
351 (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
352 (vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
354 printf("msdosfs_setattr(): returning EINVAL\n");
355 printf(" va_type %d, va_nlink %llx, va_fsid %llx, va_fileid %llx\n",
356 vap->va_type, (unsigned long long)vap->va_nlink,
357 (unsigned long long)vap->va_fsid,
358 (unsigned long long)vap->va_fileid);
359 printf(" va_blocksize %lx, va_rdev %llx, va_bytes %llx, va_gen %lx\n",
360 vap->va_blocksize, (unsigned long long)vap->va_rdev,
361 (unsigned long long)vap->va_bytes, vap->va_gen);
362 printf(" va_uid %x, va_gid %x\n",
363 vap->va_uid, vap->va_gid);
369 * We don't allow setting attributes on the root directory.
370 * The special case for the root directory is because before
371 * FAT32, the root directory didn't have an entry for itself
372 * (and was otherwise special). With FAT32, the root
373 * directory is not so special, but still doesn't have an
376 if (vp->v_vflag & VV_ROOT)
379 if (vap->va_flags != VNOVAL) {
380 if (vp->v_mount->mnt_flag & MNT_RDONLY)
382 if (cred->cr_uid != pmp->pm_uid) {
383 error = priv_check_cred(cred, PRIV_VFS_ADMIN, 0);
388 * We are very inconsistent about handling unsupported
389 * attributes. We ignored the access time and the
390 * read and execute bits. We were strict for the other
393 if (vap->va_flags & ~(UF_ARCHIVE | UF_HIDDEN | UF_READONLY |
396 if (vap->va_flags & UF_ARCHIVE)
397 dep->de_Attributes |= ATTR_ARCHIVE;
399 dep->de_Attributes &= ~ATTR_ARCHIVE;
400 if (vap->va_flags & UF_HIDDEN)
401 dep->de_Attributes |= ATTR_HIDDEN;
403 dep->de_Attributes &= ~ATTR_HIDDEN;
404 /* We don't allow changing the readonly bit on directories. */
405 if (vp->v_type != VDIR) {
406 if (vap->va_flags & UF_READONLY)
407 dep->de_Attributes |= ATTR_READONLY;
409 dep->de_Attributes &= ~ATTR_READONLY;
411 if (vap->va_flags & UF_SYSTEM)
412 dep->de_Attributes |= ATTR_SYSTEM;
414 dep->de_Attributes &= ~ATTR_SYSTEM;
415 dep->de_flag |= DE_MODIFIED;
418 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
422 if (vp->v_mount->mnt_flag & MNT_RDONLY)
425 if (uid == (uid_t)VNOVAL)
428 if (gid == (gid_t)VNOVAL)
430 if (cred->cr_uid != pmp->pm_uid || uid != pmp->pm_uid ||
431 (gid != pmp->pm_gid && !groupmember(gid, cred))) {
432 error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0);
436 if (uid != pmp->pm_uid || gid != pmp->pm_gid)
440 if (vap->va_size != VNOVAL) {
441 switch (vp->v_type) {
446 * Truncation is only supported for regular files,
447 * Disallow it if the filesystem is read-only.
449 if (vp->v_mount->mnt_flag & MNT_RDONLY)
454 * According to POSIX, the result is unspecified
455 * for file types other than regular files,
456 * directories and shared memory objects. We
457 * don't support any file types except regular
458 * files and directories in this file system, so
459 * this (default) case is unreachable and can do
460 * anything. Keep falling through to detrunc()
465 error = detrunc(dep, vap->va_size, 0, cred);
469 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
470 if (vp->v_mount->mnt_flag & MNT_RDONLY)
472 error = vn_utimes_perm(vp, vap, cred, td);
475 if ((pmp->pm_flags & MSDOSFSMNT_NOWIN95) == 0 &&
476 vap->va_atime.tv_sec != VNOVAL) {
477 dep->de_flag &= ~DE_ACCESS;
478 timespec2fattime(&vap->va_atime, 0,
479 &dep->de_ADate, NULL, NULL);
481 if (vap->va_mtime.tv_sec != VNOVAL) {
482 dep->de_flag &= ~DE_UPDATE;
483 timespec2fattime(&vap->va_mtime, 0,
484 &dep->de_MDate, &dep->de_MTime, NULL);
487 * We don't set the archive bit when modifying the time of
488 * a directory to emulate the Windows/DOS behavior.
490 if (vp->v_type != VDIR)
491 dep->de_Attributes |= ATTR_ARCHIVE;
492 dep->de_flag |= DE_MODIFIED;
495 * DOS files only have the ability to have their writability
496 * attribute set, so we use the owner write bit to set the readonly
499 if (vap->va_mode != (mode_t)VNOVAL) {
500 if (vp->v_mount->mnt_flag & MNT_RDONLY)
502 if (cred->cr_uid != pmp->pm_uid) {
503 error = priv_check_cred(cred, PRIV_VFS_ADMIN, 0);
507 if (vp->v_type != VDIR) {
508 /* We ignore the read and execute bits. */
509 if (vap->va_mode & S_IWUSR)
510 dep->de_Attributes &= ~ATTR_READONLY;
512 dep->de_Attributes |= ATTR_READONLY;
513 dep->de_Attributes |= ATTR_ARCHIVE;
514 dep->de_flag |= DE_MODIFIED;
517 return (deupdat(dep, 0));
521 msdosfs_read(struct vop_read_args *ap)
535 struct vnode *vp = ap->a_vp;
536 struct denode *dep = VTODE(vp);
537 struct msdosfsmount *pmp = dep->de_pmp;
538 struct uio *uio = ap->a_uio;
541 * If they didn't ask for any data, then we are done.
543 orig_resid = uio->uio_resid;
548 * The caller is supposed to ensure that
549 * uio->uio_offset >= 0 and uio->uio_resid >= 0.
550 * We don't need to check for large offsets as in ffs because
551 * dep->de_FileSize <= DOS_FILESIZE_MAX < OFF_MAX, so large
552 * offsets cannot cause overflow even in theory.
555 seqcount = ap->a_ioflag >> IO_SEQSHIFT;
557 isadir = dep->de_Attributes & ATTR_DIRECTORY;
559 if (uio->uio_offset >= dep->de_FileSize)
561 lbn = de_cluster(pmp, uio->uio_offset);
563 blsize = pmp->pm_bpcluster;
564 on = uio->uio_offset & pmp->pm_crbomask;
566 * If we are operating on a directory file then be sure to
567 * do i/o with the vnode for the filesystem instead of the
568 * vnode for the directory.
571 /* convert cluster # to block # */
572 error = pcbmap(dep, lbn, &lbn, 0, &blsize);
573 if (error == E2BIG) {
578 error = bread(pmp->pm_devvp, lbn, blsize, NOCRED, &bp);
579 } else if (de_cn2off(pmp, rablock) >= dep->de_FileSize) {
580 error = bread(vp, lbn, blsize, NOCRED, &bp);
581 } else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
582 error = cluster_read(vp, dep->de_FileSize, lbn, blsize,
583 NOCRED, on + uio->uio_resid, seqcount, 0, &bp);
584 } else if (seqcount > 1) {
586 error = breadn(vp, lbn,
587 blsize, &rablock, &rasize, 1, NOCRED, &bp);
589 error = bread(vp, lbn, blsize, NOCRED, &bp);
595 diff = pmp->pm_bpcluster - on;
596 n = diff > uio->uio_resid ? uio->uio_resid : diff;
597 diff = dep->de_FileSize - uio->uio_offset;
600 diff = blsize - bp->b_resid;
603 error = vn_io_fault_uiomove(bp->b_data + on, (int) n, uio);
605 } while (error == 0 && uio->uio_resid > 0 && n != 0);
606 if (!isadir && (error == 0 || uio->uio_resid != orig_resid) &&
607 (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
608 dep->de_flag |= DE_ACCESS;
613 * Write data to a file or directory.
616 msdosfs_write(struct vop_write_args *ap)
627 int ioflag = ap->a_ioflag;
628 struct uio *uio = ap->a_uio;
629 struct vnode *vp = ap->a_vp;
630 struct vnode *thisvp;
631 struct denode *dep = VTODE(vp);
632 struct msdosfsmount *pmp = dep->de_pmp;
633 struct ucred *cred = ap->a_cred;
636 printf("msdosfs_write(vp %p, uio %p, ioflag %x, cred %p\n",
637 vp, uio, ioflag, cred);
638 printf("msdosfs_write(): diroff %lu, dirclust %lu, startcluster %lu\n",
639 dep->de_diroffset, dep->de_dirclust, dep->de_StartCluster);
642 switch (vp->v_type) {
644 if (ioflag & IO_APPEND)
645 uio->uio_offset = dep->de_FileSize;
651 panic("msdosfs_write(): bad file type");
655 * This is needed (unlike in ffs_write()) because we extend the
656 * file outside of the loop but we don't want to extend the file
657 * for writes of 0 bytes.
659 if (uio->uio_resid == 0)
663 * The caller is supposed to ensure that
664 * uio->uio_offset >= 0 and uio->uio_resid >= 0.
666 if ((uoff_t)uio->uio_offset + uio->uio_resid > DOS_FILESIZE_MAX)
670 * If they've exceeded their filesize limit, tell them about it.
672 if (vn_rlimit_fsize(vp, uio, uio->uio_td))
676 * If the offset we are starting the write at is beyond the end of
677 * the file, then they've done a seek. Unix filesystems allow
678 * files with holes in them, DOS doesn't so we must fill the hole
679 * with zeroed blocks.
681 if (uio->uio_offset > dep->de_FileSize) {
682 error = deextend(dep, uio->uio_offset, cred);
688 * Remember some values in case the write fails.
690 resid = uio->uio_resid;
691 osize = dep->de_FileSize;
694 * If we write beyond the end of the file, extend it to its ultimate
695 * size ahead of the time to hopefully get a contiguous area.
697 if (uio->uio_offset + resid > osize) {
698 count = de_clcount(pmp, uio->uio_offset + resid) -
699 de_clcount(pmp, osize);
700 error = extendfile(dep, count, NULL, NULL, 0);
701 if (error && (error != ENOSPC || (ioflag & IO_UNIT)))
703 lastcn = dep->de_fc[FC_LASTFC].fc_frcn;
705 lastcn = de_clcount(pmp, osize) - 1;
707 seqcount = ioflag >> IO_SEQSHIFT;
709 if (de_cluster(pmp, uio->uio_offset) > lastcn) {
714 croffset = uio->uio_offset & pmp->pm_crbomask;
715 n = min(uio->uio_resid, pmp->pm_bpcluster - croffset);
716 if (uio->uio_offset + n > dep->de_FileSize) {
717 dep->de_FileSize = uio->uio_offset + n;
718 /* The object size needs to be set before buffer is allocated */
719 vnode_pager_setsize(vp, dep->de_FileSize);
722 bn = de_cluster(pmp, uio->uio_offset);
723 if ((uio->uio_offset & pmp->pm_crbomask) == 0
724 && (de_cluster(pmp, uio->uio_offset + uio->uio_resid)
725 > de_cluster(pmp, uio->uio_offset)
726 || uio->uio_offset + uio->uio_resid >= dep->de_FileSize)) {
728 * If either the whole cluster gets written,
729 * or we write the cluster from its start beyond EOF,
730 * then no need to read data from disk.
732 bp = getblk(thisvp, bn, pmp->pm_bpcluster, 0, 0, 0);
734 * This call to vfs_bio_clrbuf() ensures that
735 * even if vn_io_fault_uiomove() below faults,
736 * garbage from the newly instantiated buffer
737 * is not exposed to the userspace via mmap().
741 * Do the bmap now, since pcbmap needs buffers
742 * for the FAT table. (see msdosfs_strategy)
744 if (bp->b_blkno == bp->b_lblkno) {
745 error = pcbmap(dep, bp->b_lblkno, &bn, 0, 0);
751 if (bp->b_blkno == -1) {
754 error = EIO; /* XXX */
759 * The block we need to write into exists, so read it in.
761 error = bread(thisvp, bn, pmp->pm_bpcluster, cred, &bp);
769 * Should these vnode_pager_* functions be done on dir
774 * Copy the data from user space into the buf header.
776 error = vn_io_fault_uiomove(bp->b_data + croffset, n, uio);
782 /* Prepare for clustered writes in some else clauses. */
783 if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0)
784 bp->b_flags |= B_CLUSTEROK;
787 * If IO_SYNC, then each buffer is written synchronously.
788 * Otherwise, if we have a severe page deficiency then
789 * write the buffer asynchronously. Otherwise, if on a
790 * cluster boundary then write the buffer asynchronously,
791 * combining it with contiguous clusters if permitted and
792 * possible, since we don't expect more writes into this
793 * buffer soon. Otherwise, do a delayed write because we
794 * expect more writes into this buffer soon.
796 if (ioflag & IO_SYNC)
798 else if (vm_page_count_severe() || buf_dirty_count_severe())
800 else if (n + croffset == pmp->pm_bpcluster) {
801 if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0)
802 cluster_write(vp, bp, dep->de_FileSize,
808 dep->de_flag |= DE_UPDATE;
809 } while (error == 0 && uio->uio_resid > 0);
812 * If the write failed and they want us to, truncate the file back
813 * to the size it was before the write was attempted.
817 if (ioflag & IO_UNIT) {
818 detrunc(dep, osize, ioflag & IO_SYNC, NOCRED);
819 uio->uio_offset -= resid - uio->uio_resid;
820 uio->uio_resid = resid;
822 detrunc(dep, dep->de_FileSize, ioflag & IO_SYNC, NOCRED);
823 if (uio->uio_resid != resid)
826 } else if (ioflag & IO_SYNC)
827 error = deupdat(dep, 1);
832 * Flush the blocks of a file to disk.
835 msdosfs_fsync(struct vop_fsync_args *ap)
843 * If the syncing request comes from fsync(2), sync the entire
844 * FAT and any other metadata that happens to be on devvp. We
845 * need this mainly for the FAT. We write the FAT sloppily, and
846 * syncing it all now is the best we can easily do to get all
847 * directory entries associated with the file (not just the file)
848 * fully synced. The other metadata includes critical metadata
849 * for all directory entries, but only in the MNT_ASYNC case. We
850 * will soon sync all metadata in the file's directory entry.
851 * Non-critical metadata for associated directory entries only
852 * gets synced accidentally, as in most file systems.
854 if (ap->a_waitfor == MNT_WAIT) {
855 devvp = VTODE(ap->a_vp)->de_pmp->pm_devvp;
856 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
857 allerror = VOP_FSYNC(devvp, MNT_WAIT, ap->a_td);
858 VOP_UNLOCK(devvp, 0);
862 error = deupdat(VTODE(ap->a_vp), ap->a_waitfor == MNT_WAIT);
869 msdosfs_remove(struct vop_remove_args *ap)
871 struct denode *dep = VTODE(ap->a_vp);
872 struct denode *ddep = VTODE(ap->a_dvp);
875 if (ap->a_vp->v_type == VDIR)
878 error = removede(ddep, dep);
880 printf("msdosfs_remove(), dep %p, v_usecount %d\n", dep, ap->a_vp->v_usecount);
886 * DOS filesystems don't know what links are.
889 msdosfs_link(struct vop_link_args *ap)
895 * Renames on files require moving the denode to a new hash queue since the
896 * denode's location is used to compute which hash queue to put the file
897 * in. Unless it is a rename in place. For example "mv a b".
899 * What follows is the basic algorithm:
902 * if (dest file exists) {
905 * if (dest and src in same directory) {
906 * rewrite name in existing directory slot
908 * write new entry in dest directory
909 * update offset and dirclust in denode
910 * move denode to new hash chain
911 * clear old directory entry
915 * if (dest directory exists) {
916 * if (dest is not empty) {
919 * remove dest directory
921 * if (dest and src in same directory) {
922 * rewrite name in existing entry
924 * be sure dest is not a child of src directory
925 * write entry in dest directory
926 * update "." and ".." in moved directory
927 * clear old directory entry for moved directory
932 * source's parent directory is unlocked
933 * source file or directory is unlocked
934 * destination's parent directory is locked
935 * destination file or directory is locked if it exists
938 * all denodes should be released
941 msdosfs_rename(struct vop_rename_args *ap)
943 struct vnode *tdvp = ap->a_tdvp;
944 struct vnode *fvp = ap->a_fvp;
945 struct vnode *fdvp = ap->a_fdvp;
946 struct vnode *tvp = ap->a_tvp;
947 struct componentname *tcnp = ap->a_tcnp;
948 struct componentname *fcnp = ap->a_fcnp;
949 struct denode *ip, *xp, *dp, *zp;
950 u_char toname[12], oldname[11];
951 u_long from_diroffset, to_diroffset;
953 int doingdirectory = 0, newparent = 0;
957 struct msdosfsmount *pmp;
958 struct direntry *dotdotp;
961 pmp = VFSTOMSDOSFS(fdvp->v_mount);
964 if ((tcnp->cn_flags & HASBUF) == 0 ||
965 (fcnp->cn_flags & HASBUF) == 0)
966 panic("msdosfs_rename: no name");
969 * Check for cross-device rename.
971 if (fvp->v_mount != tdvp->v_mount ||
972 (tvp && fvp->v_mount != tvp->v_mount)) {
987 * If source and dest are the same, do nothing.
994 error = vn_lock(fvp, LK_EXCLUSIVE);
1001 * Be sure we are not renaming ".", "..", or an alias of ".". This
1002 * leads to a crippled directory tree. It's pretty tough to do a
1003 * "ls" or "pwd" with the "." directory entry missing, and "cd .."
1004 * doesn't work if the ".." entry is missing.
1006 if (ip->de_Attributes & ATTR_DIRECTORY) {
1008 * Avoid ".", "..", and aliases of "." for obvious reasons.
1010 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
1012 (fcnp->cn_flags & ISDOTDOT) ||
1013 (tcnp->cn_flags & ISDOTDOT) ||
1014 (ip->de_flag & DE_RENAME)) {
1019 ip->de_flag |= DE_RENAME;
1024 * When the target exists, both the directory
1025 * and target vnodes are returned locked.
1028 xp = tvp ? VTODE(tvp) : NULL;
1030 * Remember direntry place to use for destination
1032 to_diroffset = dp->de_fndoffset;
1033 to_count = dp->de_fndcnt;
1036 * If ".." must be changed (ie the directory gets a new
1037 * parent) then the source directory must not be in the
1038 * directory hierarchy above the target, as this would
1039 * orphan everything below the source directory. Also
1040 * the user must have write permission in the source so
1041 * as to be able to change "..". We must repeat the call
1042 * to namei, as the parent directory is unlocked by the
1043 * call to doscheckpath().
1045 error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread);
1047 if (VTODE(fdvp)->de_StartCluster != VTODE(tdvp)->de_StartCluster)
1049 if (doingdirectory && newparent) {
1050 if (error) /* write access check above */
1055 * doscheckpath() vput()'s dp,
1056 * so we have to do a relookup afterwards
1058 error = doscheckpath(ip, dp);
1061 if ((tcnp->cn_flags & SAVESTART) == 0)
1062 panic("msdosfs_rename: lost to startdir");
1063 error = relookup(tdvp, &tvp, tcnp);
1067 xp = tvp ? VTODE(tvp) : NULL;
1072 * Target must be empty if a directory and have no links
1073 * to it. Also, ensure source and target are compatible
1074 * (both directories, or both not directories).
1076 if (xp->de_Attributes & ATTR_DIRECTORY) {
1077 if (!dosdirempty(xp)) {
1081 if (!doingdirectory) {
1086 } else if (doingdirectory) {
1090 error = removede(dp, xp);
1098 * Convert the filename in tcnp into a dos filename. We copy this
1099 * into the denode and directory entry for the destination
1102 error = uniqdosname(VTODE(tdvp), tcnp, toname);
1107 * Since from wasn't locked at various places above,
1108 * have to do a relookup here.
1110 fcnp->cn_flags &= ~MODMASK;
1111 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
1112 if ((fcnp->cn_flags & SAVESTART) == 0)
1113 panic("msdosfs_rename: lost from startdir");
1115 VOP_UNLOCK(tdvp, 0);
1116 if (relookup(fdvp, &fvp, fcnp) == 0)
1120 * From name has disappeared.
1123 panic("rename: lost dir entry");
1125 VOP_UNLOCK(tdvp, 0);
1132 from_diroffset = zp->de_fndoffset;
1135 * Ensure that the directory entry still exists and has not
1136 * changed till now. If the source is a file the entry may
1137 * have been unlinked or renamed. In either case there is
1138 * no further work to be done. If the source is a directory
1139 * then it cannot have been rmdir'ed or renamed; this is
1140 * prohibited by the DE_RENAME flag.
1144 panic("rename: lost dir entry");
1147 VOP_UNLOCK(fdvp, 0);
1155 * First write a new entry in the destination
1156 * directory and mark the entry in the source directory
1157 * as deleted. Then move the denode to the correct hash
1158 * chain for its new location in the filesystem. And, if
1159 * we moved a directory, then update its .. entry to point
1160 * to the new parent directory.
1162 memcpy(oldname, ip->de_Name, 11);
1163 memcpy(ip->de_Name, toname, 11); /* update denode */
1164 dp->de_fndoffset = to_diroffset;
1165 dp->de_fndcnt = to_count;
1166 error = createde(ip, dp, (struct denode **)0, tcnp);
1168 memcpy(ip->de_Name, oldname, 11);
1170 VOP_UNLOCK(fdvp, 0);
1175 * If ip is for a directory, then its name should always
1176 * be "." since it is for the directory entry in the
1177 * directory itself (msdosfs_lookup() always translates
1178 * to the "." entry so as to get a unique denode, except
1179 * for the root directory there are different
1180 * complications). However, we just corrupted its name
1181 * to pass the correct name to createde(). Undo this.
1183 if ((ip->de_Attributes & ATTR_DIRECTORY) != 0)
1184 memcpy(ip->de_Name, oldname, 11);
1186 zp->de_fndoffset = from_diroffset;
1187 error = removede(zp, ip);
1189 /* XXX should downgrade to ro here, fs is corrupt */
1191 VOP_UNLOCK(fdvp, 0);
1195 if (!doingdirectory) {
1196 error = pcbmap(dp, de_cluster(pmp, to_diroffset), 0,
1197 &ip->de_dirclust, 0);
1199 /* XXX should downgrade to ro here, fs is corrupt */
1201 VOP_UNLOCK(fdvp, 0);
1205 if (ip->de_dirclust == MSDOSFSROOT)
1206 ip->de_diroffset = to_diroffset;
1208 ip->de_diroffset = to_diroffset & pmp->pm_crbomask;
1212 VOP_UNLOCK(fdvp, 0);
1216 * If we moved a directory to a new parent directory, then we must
1217 * fixup the ".." entry in the moved directory.
1219 if (doingdirectory && newparent) {
1220 cn = ip->de_StartCluster;
1221 if (cn == MSDOSFSROOT) {
1222 /* this should never happen */
1223 panic("msdosfs_rename(): updating .. in root directory?");
1225 bn = cntobn(pmp, cn);
1226 error = bread(pmp->pm_devvp, bn, pmp->pm_bpcluster,
1229 /* XXX should downgrade to ro here, fs is corrupt */
1234 dotdotp = (struct direntry *)bp->b_data + 1;
1235 pcl = dp->de_StartCluster;
1236 if (FAT32(pmp) && pcl == pmp->pm_rootdirblk)
1238 putushort(dotdotp->deStartCluster, pcl);
1240 putushort(dotdotp->deHighClust, pcl >> 16);
1241 if (DOINGASYNC(fvp))
1243 else if ((error = bwrite(bp)) != 0) {
1244 /* XXX should downgrade to ro here, fs is corrupt */
1251 * The msdosfs lookup is case insensitive. Several aliases may
1252 * be inserted for a single directory entry. As a consequnce,
1253 * name cache purge done by lookup for fvp when DELETE op for
1254 * namei is specified, might be not enough to expunge all
1255 * namecache entries that were installed for this direntry.
1264 ip->de_flag &= ~DE_RENAME;
1272 struct direntry dot;
1273 struct direntry dotdot;
1274 } dosdirtemplate = {
1275 { ". ", /* the . entry */
1276 ATTR_DIRECTORY, /* file attribute */
1278 0, { 0, 0 }, { 0, 0 }, /* create time & date */
1279 { 0, 0 }, /* access date */
1280 { 0, 0 }, /* high bits of start cluster */
1281 { 210, 4 }, { 210, 4 }, /* modify time & date */
1282 { 0, 0 }, /* startcluster */
1283 { 0, 0, 0, 0 } /* filesize */
1285 { ".. ", /* the .. entry */
1286 ATTR_DIRECTORY, /* file attribute */
1288 0, { 0, 0 }, { 0, 0 }, /* create time & date */
1289 { 0, 0 }, /* access date */
1290 { 0, 0 }, /* high bits of start cluster */
1291 { 210, 4 }, { 210, 4 }, /* modify time & date */
1292 { 0, 0 }, /* startcluster */
1293 { 0, 0, 0, 0 } /* filesize */
1298 msdosfs_mkdir(struct vop_mkdir_args *ap)
1300 struct componentname *cnp = ap->a_cnp;
1302 struct denode *pdep = VTODE(ap->a_dvp);
1303 struct direntry *denp;
1304 struct msdosfsmount *pmp = pdep->de_pmp;
1306 u_long newcluster, pcl;
1309 struct denode ndirent;
1313 * If this is the root directory and there is no space left we
1314 * can't do anything. This is because the root directory can not
1317 if (pdep->de_StartCluster == MSDOSFSROOT
1318 && pdep->de_fndoffset >= pdep->de_FileSize) {
1324 * Allocate a cluster to hold the about to be created directory.
1326 error = clusteralloc(pmp, 0, 1, CLUST_EOFE, &newcluster, NULL);
1330 memset(&ndirent, 0, sizeof(ndirent));
1331 ndirent.de_pmp = pmp;
1332 ndirent.de_flag = DE_ACCESS | DE_CREATE | DE_UPDATE;
1334 DETIMES(&ndirent, &ts, &ts, &ts);
1337 * Now fill the cluster with the "." and ".." entries. And write
1338 * the cluster to disk. This way it is there for the parent
1339 * directory to be pointing at if there were a crash.
1341 bn = cntobn(pmp, newcluster);
1342 /* always succeeds */
1343 bp = getblk(pmp->pm_devvp, bn, pmp->pm_bpcluster, 0, 0, 0);
1344 memset(bp->b_data, 0, pmp->pm_bpcluster);
1345 memcpy(bp->b_data, &dosdirtemplate, sizeof dosdirtemplate);
1346 denp = (struct direntry *)bp->b_data;
1347 putushort(denp[0].deStartCluster, newcluster);
1348 putushort(denp[0].deCDate, ndirent.de_CDate);
1349 putushort(denp[0].deCTime, ndirent.de_CTime);
1350 denp[0].deCHundredth = ndirent.de_CHun;
1351 putushort(denp[0].deADate, ndirent.de_ADate);
1352 putushort(denp[0].deMDate, ndirent.de_MDate);
1353 putushort(denp[0].deMTime, ndirent.de_MTime);
1354 pcl = pdep->de_StartCluster;
1356 * Although the root directory has a non-magic starting cluster
1357 * number for FAT32, chkdsk and fsck_msdosfs still require
1358 * references to it in dotdot entries to be magic.
1360 if (FAT32(pmp) && pcl == pmp->pm_rootdirblk)
1362 putushort(denp[1].deStartCluster, pcl);
1363 putushort(denp[1].deCDate, ndirent.de_CDate);
1364 putushort(denp[1].deCTime, ndirent.de_CTime);
1365 denp[1].deCHundredth = ndirent.de_CHun;
1366 putushort(denp[1].deADate, ndirent.de_ADate);
1367 putushort(denp[1].deMDate, ndirent.de_MDate);
1368 putushort(denp[1].deMTime, ndirent.de_MTime);
1370 putushort(denp[0].deHighClust, newcluster >> 16);
1371 putushort(denp[1].deHighClust, pcl >> 16);
1374 if (DOINGASYNC(ap->a_dvp))
1376 else if ((error = bwrite(bp)) != 0)
1380 * Now build up a directory entry pointing to the newly allocated
1381 * cluster. This will be written to an empty slot in the parent
1385 if ((cnp->cn_flags & HASBUF) == 0)
1386 panic("msdosfs_mkdir: no name");
1388 error = uniqdosname(pdep, cnp, ndirent.de_Name);
1392 ndirent.de_Attributes = ATTR_DIRECTORY;
1393 ndirent.de_LowerCase = 0;
1394 ndirent.de_StartCluster = newcluster;
1395 ndirent.de_FileSize = 0;
1396 error = createde(&ndirent, pdep, &dep, cnp);
1399 *ap->a_vpp = DETOV(dep);
1403 clusterfree(pmp, newcluster, NULL);
1409 msdosfs_rmdir(struct vop_rmdir_args *ap)
1411 struct vnode *vp = ap->a_vp;
1412 struct vnode *dvp = ap->a_dvp;
1413 struct componentname *cnp = ap->a_cnp;
1414 struct denode *ip, *dp;
1421 * Verify the directory is empty (and valid).
1422 * (Rmdir ".." won't be valid since
1423 * ".." will contain a reference to
1424 * the current directory and thus be
1428 if (!dosdirempty(ip) || ip->de_flag & DE_RENAME) {
1433 * Delete the entry from the directory. For dos filesystems this
1434 * gets rid of the directory entry on disk, the in memory copy
1435 * still exists but the de_refcnt is <= 0. This prevents it from
1436 * being found by deget(). When the vput() on dep is done we give
1437 * up access and eventually msdosfs_reclaim() will be called which
1438 * will remove it from the denode cache.
1440 error = removede(dp, ip);
1444 * This is where we decrement the link count in the parent
1445 * directory. Since dos filesystems don't do this we just purge
1450 * Truncate the directory that is being deleted.
1452 error = detrunc(ip, (u_long)0, IO_SYNC, cnp->cn_cred);
1460 * DOS filesystems don't know what symlinks are.
1463 msdosfs_symlink(struct vop_symlink_args *ap)
1465 return (EOPNOTSUPP);
1469 msdosfs_readdir(struct vop_readdir_args *ap)
1482 struct denode *dep = VTODE(ap->a_vp);
1483 struct msdosfsmount *pmp = dep->de_pmp;
1484 struct direntry *dentp;
1485 struct dirent dirbuf;
1486 struct uio *uio = ap->a_uio;
1487 u_long *cookies = NULL;
1492 #ifdef MSDOSFS_DEBUG
1493 printf("msdosfs_readdir(): vp %p, uio %p, cred %p, eofflagp %p\n",
1494 ap->a_vp, uio, ap->a_cred, ap->a_eofflag);
1498 * msdosfs_readdir() won't operate properly on regular files since
1499 * it does i/o only with the filesystem vnode, and hence can
1500 * retrieve the wrong block from the buffer cache for a plain file.
1501 * So, fail attempts to readdir() on a plain file.
1503 if ((dep->de_Attributes & ATTR_DIRECTORY) == 0)
1507 * To be safe, initialize dirbuf
1509 memset(dirbuf.d_name, 0, sizeof(dirbuf.d_name));
1512 * If the user buffer is smaller than the size of one dos directory
1513 * entry or the file offset is not a multiple of the size of a
1514 * directory entry, then we fail the read.
1516 off = offset = uio->uio_offset;
1517 if (uio->uio_resid < sizeof(struct direntry) ||
1518 (offset & (sizeof(struct direntry) - 1)))
1521 if (ap->a_ncookies) {
1522 ncookies = uio->uio_resid / 16;
1523 cookies = malloc(ncookies * sizeof(u_long), M_TEMP,
1525 *ap->a_cookies = cookies;
1526 *ap->a_ncookies = ncookies;
1529 dirsperblk = pmp->pm_BytesPerSec / sizeof(struct direntry);
1532 * If they are reading from the root directory then, we simulate
1533 * the . and .. entries since these don't exist in the root
1534 * directory. We also set the offset bias to make up for having to
1535 * simulate these entries. By this I mean that at file offset 64 we
1536 * read the first entry in the root directory that lives on disk.
1538 if (dep->de_StartCluster == MSDOSFSROOT
1539 || (FAT32(pmp) && dep->de_StartCluster == pmp->pm_rootdirblk)) {
1541 printf("msdosfs_readdir(): going after . or .. in root dir, offset %d\n",
1544 bias = 2 * sizeof(struct direntry);
1545 if (offset < bias) {
1546 for (n = (int)offset / sizeof(struct direntry);
1548 dirbuf.d_fileno = FAT32(pmp) ?
1549 (uint64_t)cntobn(pmp, pmp->pm_rootdirblk) *
1551 dirbuf.d_type = DT_DIR;
1554 dirbuf.d_namlen = 1;
1555 strcpy(dirbuf.d_name, ".");
1558 dirbuf.d_namlen = 2;
1559 strcpy(dirbuf.d_name, "..");
1562 dirbuf.d_reclen = GENERIC_DIRSIZ(&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);
1597 n = min(n, blsize - bp->b_resid);
1604 * Convert from dos directory entries to fs-independent
1605 * directory entries.
1607 for (dentp = (struct direntry *)(bp->b_data + on);
1608 (char *)dentp < bp->b_data + on + n;
1609 dentp++, offset += sizeof(struct direntry)) {
1611 printf("rd: dentp %08x prev %08x crnt %08x deName %02x attr %02x\n",
1612 dentp, prev, crnt, dentp->deName[0], dentp->deAttributes);
1615 * If this is an unused entry, we can stop.
1617 if (dentp->deName[0] == SLOT_EMPTY) {
1622 * Skip deleted entries.
1624 if (dentp->deName[0] == SLOT_DELETED) {
1631 * Handle Win95 long directory entries
1633 if (dentp->deAttributes == ATTR_WIN95) {
1634 if (pmp->pm_flags & MSDOSFSMNT_SHORTNAME)
1636 chksum = win2unixfn(&nb,
1637 (struct winentry *)dentp, chksum, pmp);
1642 * Skip volume labels
1644 if (dentp->deAttributes & ATTR_VOLUME) {
1650 * This computation of d_fileno must match
1651 * the computation of va_fileid in
1654 if (dentp->deAttributes & ATTR_DIRECTORY) {
1655 cn = getushort(dentp->deStartCluster);
1657 cn |= getushort(dentp->deHighClust) <<
1659 if (cn == MSDOSFSROOT)
1660 cn = pmp->pm_rootdirblk;
1662 if (cn == MSDOSFSROOT && !FAT32(pmp))
1663 dirbuf.d_fileno = 1;
1665 dirbuf.d_fileno = cntobn(pmp, cn) *
1667 dirbuf.d_type = DT_DIR;
1669 dirbuf.d_fileno = (uoff_t)offset /
1670 sizeof(struct direntry);
1671 dirbuf.d_type = DT_REG;
1674 if (chksum != winChksum(dentp->deName)) {
1675 dirbuf.d_namlen = dos2unixfn(dentp->deName,
1676 (u_char *)dirbuf.d_name,
1677 dentp->deLowerCase |
1678 ((pmp->pm_flags & MSDOSFSMNT_SHORTNAME) ?
1679 (LCASE_BASE | LCASE_EXT) : 0),
1683 mbnambuf_flush(&nb, &dirbuf);
1685 dirbuf.d_reclen = GENERIC_DIRSIZ(&dirbuf);
1686 if (uio->uio_resid < dirbuf.d_reclen) {
1690 error = uiomove(&dirbuf, dirbuf.d_reclen, uio);
1696 *cookies++ = offset + sizeof(struct direntry);
1697 if (--ncookies <= 0) {
1702 off = offset + sizeof(struct direntry);
1707 /* Subtract unused cookies */
1709 *ap->a_ncookies -= ncookies;
1711 uio->uio_offset = off;
1714 * Set the eofflag (NFS uses it)
1716 if (ap->a_eofflag) {
1717 if (dep->de_FileSize - (offset - bias) <= 0)
1726 * a_vp - pointer to the file's vnode
1727 * a_bn - logical block number within the file (cluster number for us)
1728 * a_bop - where to return the bufobj of the special file containing the fs
1729 * a_bnp - where to return the "physical" block number corresponding to a_bn
1730 * (relative to the special file; units are blocks of size DEV_BSIZE)
1731 * a_runp - where to return the "run past" a_bn. This is the count of logical
1732 * blocks whose physical blocks (together with a_bn's physical block)
1734 * a_runb - where to return the "run before" a_bn.
1737 msdosfs_bmap(struct vop_bmap_args *ap)
1741 struct msdosfsmount *pmp;
1745 int bnpercn, error, maxio, maxrun, run;
1750 if (ap->a_bop != NULL)
1751 *ap->a_bop = &pmp->pm_devvp->v_bufobj;
1752 if (ap->a_bnp == NULL)
1754 if (ap->a_runp != NULL)
1756 if (ap->a_runb != NULL)
1761 error = pcbmap(dep, cn, ap->a_bnp, NULL, NULL);
1762 if (error != 0 || (ap->a_runp == NULL && ap->a_runb == NULL))
1766 maxio = mp->mnt_iosize_max / mp->mnt_stat.f_iosize;
1767 bnpercn = de_cn2bn(pmp, 1);
1768 if (ap->a_runp != NULL) {
1769 maxrun = ulmin(maxio - 1, pmp->pm_maxcluster - cn);
1770 for (run = 1; run <= maxrun; run++) {
1771 if (pcbmap(dep, cn + run, &runbn, NULL, NULL) != 0 ||
1772 runbn != *ap->a_bnp + run * bnpercn)
1775 *ap->a_runp = run - 1;
1777 if (ap->a_runb != NULL) {
1778 maxrun = ulmin(maxio - 1, cn);
1779 for (run = 1; run < maxrun; run++) {
1780 if (pcbmap(dep, cn - run, &runbn, NULL, NULL) != 0 ||
1781 runbn != *ap->a_bnp - run * bnpercn)
1784 *ap->a_runb = run - 1;
1789 SYSCTL_NODE(_vfs, OID_AUTO, msdosfs, CTLFLAG_RW, 0, "msdos filesystem");
1790 static int use_buf_pager = 1;
1791 SYSCTL_INT(_vfs_msdosfs, OID_AUTO, use_buf_pager, CTLFLAG_RWTUN,
1793 "Use buffer pager instead of bmap");
1796 msdosfs_gbp_getblkno(struct vnode *vp, vm_ooffset_t off)
1799 return (de_cluster(VTODE(vp)->de_pmp, off));
1803 msdosfs_gbp_getblksz(struct vnode *vp, daddr_t lbn)
1806 return (VTODE(vp)->de_pmp->pm_bpcluster);
1810 msdosfs_getpages(struct vop_getpages_args *ap)
1814 return (vfs_bio_getpages(ap->a_vp, ap->a_m, ap->a_count,
1815 ap->a_rbehind, ap->a_rahead, msdosfs_gbp_getblkno,
1816 msdosfs_gbp_getblksz));
1817 return (vnode_pager_generic_getpages(ap->a_vp, ap->a_m, ap->a_count,
1818 ap->a_rbehind, ap->a_rahead, NULL, NULL));
1822 msdosfs_strategy(struct vop_strategy_args *ap)
1824 struct buf *bp = ap->a_bp;
1825 struct denode *dep = VTODE(ap->a_vp);
1831 * If we don't already know the filesystem relative block number
1832 * then get it using pcbmap(). If pcbmap() returns the block
1833 * number as -1 then we've got a hole in the file. DOS filesystems
1834 * don't allow files with holes, so we shouldn't ever see this.
1836 if (bp->b_blkno == bp->b_lblkno) {
1837 error = pcbmap(dep, bp->b_lblkno, &blkno, 0, 0);
1838 bp->b_blkno = blkno;
1840 bp->b_error = error;
1841 bp->b_ioflags |= BIO_ERROR;
1845 if ((long)bp->b_blkno == -1)
1848 if (bp->b_blkno == -1) {
1853 * Read/write the block from/to the disk that contains the desired
1856 bp->b_iooffset = dbtob(bp->b_blkno);
1857 bo = dep->de_pmp->pm_bo;
1858 BO_STRATEGY(bo, bp);
1863 msdosfs_print(struct vop_print_args *ap)
1865 struct denode *dep = VTODE(ap->a_vp);
1867 printf("\tstartcluster %lu, dircluster %lu, diroffset %lu, ",
1868 dep->de_StartCluster, dep->de_dirclust, dep->de_diroffset);
1869 printf("on dev %s\n", devtoname(dep->de_pmp->pm_dev));
1874 msdosfs_pathconf(struct vop_pathconf_args *ap)
1876 struct msdosfsmount *pmp = VTODE(ap->a_vp)->de_pmp;
1878 switch (ap->a_name) {
1879 case _PC_FILESIZEBITS:
1886 *ap->a_retval = pmp->pm_flags & MSDOSFSMNT_LONGNAME ? WIN_MAXLEN : 12;
1888 case _PC_CHOWN_RESTRICTED:
1895 return (vop_stdpathconf(ap));
1901 msdosfs_vptofh(struct vop_vptofh_args *ap)
1904 struct defid *defhp;
1906 dep = VTODE(ap->a_vp);
1907 defhp = (struct defid *)ap->a_fhp;
1908 defhp->defid_len = sizeof(struct defid);
1909 defhp->defid_dirclust = dep->de_dirclust;
1910 defhp->defid_dirofs = dep->de_diroffset;
1911 /* defhp->defid_gen = dep->de_gen; */
1915 /* Global vfs data structures for msdosfs */
1916 struct vop_vector msdosfs_vnodeops = {
1917 .vop_default = &default_vnodeops,
1919 .vop_access = msdosfs_access,
1920 .vop_bmap = msdosfs_bmap,
1921 .vop_getpages = msdosfs_getpages,
1922 .vop_cachedlookup = msdosfs_lookup,
1923 .vop_open = msdosfs_open,
1924 .vop_close = msdosfs_close,
1925 .vop_create = msdosfs_create,
1926 .vop_fsync = msdosfs_fsync,
1927 .vop_fdatasync = vop_stdfdatasync_buf,
1928 .vop_getattr = msdosfs_getattr,
1929 .vop_inactive = msdosfs_inactive,
1930 .vop_link = msdosfs_link,
1931 .vop_lookup = vfs_cache_lookup,
1932 .vop_mkdir = msdosfs_mkdir,
1933 .vop_mknod = msdosfs_mknod,
1934 .vop_pathconf = msdosfs_pathconf,
1935 .vop_print = msdosfs_print,
1936 .vop_read = msdosfs_read,
1937 .vop_readdir = msdosfs_readdir,
1938 .vop_reclaim = msdosfs_reclaim,
1939 .vop_remove = msdosfs_remove,
1940 .vop_rename = msdosfs_rename,
1941 .vop_rmdir = msdosfs_rmdir,
1942 .vop_setattr = msdosfs_setattr,
1943 .vop_strategy = msdosfs_strategy,
1944 .vop_symlink = msdosfs_symlink,
1945 .vop_write = msdosfs_write,
1946 .vop_vptofh = msdosfs_vptofh,