2 /* $NetBSD: msdosfs_vnops.c,v 1.68 1998/02/10 14:10:04 mrg Exp $ */
5 * Copyright (C) 1994, 1995, 1997 Wolfgang Solfrank.
6 * Copyright (C) 1994, 1995, 1997 TooLs GmbH.
8 * Original code by Paul Popelka (paulp@uts.amdahl.com) (see below).
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by TooLs GmbH.
21 * 4. The name of TooLs GmbH may not be used to endorse or promote products
22 * derived from this software without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
25 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
26 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
27 * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
28 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
29 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
30 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
31 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
32 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
33 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
36 * Written by Paul Popelka (paulp@uts.amdahl.com)
38 * You can do anything you want with this software, just don't say you wrote
39 * it, and don't remove this notice.
41 * This software is provided "as is".
43 * The author supplies this software to be publicly redistributed on the
44 * understanding that the author is not responsible for the correct
45 * functioning of this software in any circumstances and is not liable for
46 * any damages caused by this software.
51 #include <sys/param.h>
52 #include <sys/systm.h>
55 #include <sys/clock.h>
56 #include <sys/dirent.h>
58 #include <sys/lockf.h>
59 #include <sys/malloc.h>
60 #include <sys/mount.h>
61 #include <sys/mutex.h>
62 #include <sys/namei.h>
65 #include <sys/sysctl.h>
66 #include <sys/unistd.h>
67 #include <sys/vnode.h>
70 #include <vm/vm_extern.h>
71 #include <vm/vnode_pager.h>
73 #include <fs/msdosfs/bpb.h>
74 #include <fs/msdosfs/direntry.h>
75 #include <fs/msdosfs/denode.h>
76 #include <fs/msdosfs/fat.h>
77 #include <fs/msdosfs/msdosfsmount.h>
79 #define DOS_FILESIZE_MAX 0xffffffff
82 * Prototypes for MSDOSFS vnode operations
84 static vop_create_t msdosfs_create;
85 static vop_mknod_t msdosfs_mknod;
86 static vop_open_t msdosfs_open;
87 static vop_close_t msdosfs_close;
88 static vop_access_t msdosfs_access;
89 static vop_getattr_t msdosfs_getattr;
90 static vop_setattr_t msdosfs_setattr;
91 static vop_read_t msdosfs_read;
92 static vop_write_t msdosfs_write;
93 static vop_fsync_t msdosfs_fsync;
94 static vop_remove_t msdosfs_remove;
95 static vop_link_t msdosfs_link;
96 static vop_rename_t msdosfs_rename;
97 static vop_mkdir_t msdosfs_mkdir;
98 static vop_rmdir_t msdosfs_rmdir;
99 static vop_symlink_t msdosfs_symlink;
100 static vop_readdir_t msdosfs_readdir;
101 static vop_bmap_t msdosfs_bmap;
102 static vop_getpages_t msdosfs_getpages;
103 static vop_strategy_t msdosfs_strategy;
104 static vop_print_t msdosfs_print;
105 static vop_pathconf_t msdosfs_pathconf;
106 static vop_vptofh_t msdosfs_vptofh;
109 * Some general notes:
111 * In the ufs filesystem the inodes, superblocks, and indirect blocks are
112 * read/written using the vnode for the filesystem. Blocks that represent
113 * the contents of a file are read/written using the vnode for the file
114 * (including directories when they are read/written as files). This
115 * presents problems for the dos filesystem because data that should be in
116 * an inode (if dos had them) resides in the directory itself. Since we
117 * must update directory entries without the benefit of having the vnode
118 * for the directory we must use the vnode for the filesystem. This means
119 * that when a directory is actually read/written (via read, write, or
120 * readdir, or seek) we must use the vnode for the filesystem instead of
121 * the vnode for the directory as would happen in ufs. This is to insure we
122 * retrieve the correct block from the buffer cache since the hash value is
123 * based upon the vnode address and the desired block number.
127 * Create a regular file. On entry the directory to contain the file being
128 * created is locked. We must release before we return. We must also free
129 * the pathname buffer pointed at by cnp->cn_pnbuf, always on error, or
130 * only if the SAVESTART bit in cn_flags is clear on success.
133 msdosfs_create(struct vop_create_args *ap)
135 struct componentname *cnp = ap->a_cnp;
136 struct denode ndirent;
138 struct denode *pdep = VTODE(ap->a_dvp);
143 printf("msdosfs_create(cnp %p, vap %p\n", cnp, ap->a_vap);
147 * If this is the root directory and there is no space left we
148 * can't do anything. This is because the root directory can not
151 if (pdep->de_StartCluster == MSDOSFSROOT
152 && pdep->de_fndoffset >= pdep->de_FileSize) {
158 * Create a directory entry for the file, then call createde() to
159 * have it installed. NOTE: DOS files are always executable. We
160 * use the absence of the owner write bit to make the file
164 if ((cnp->cn_flags & HASBUF) == 0)
165 panic("msdosfs_create: no name");
167 bzero(&ndirent, sizeof(ndirent));
168 error = uniqdosname(pdep, cnp, ndirent.de_Name);
172 ndirent.de_Attributes = ATTR_ARCHIVE;
173 ndirent.de_LowerCase = 0;
174 ndirent.de_StartCluster = 0;
175 ndirent.de_FileSize = 0;
176 ndirent.de_pmp = pdep->de_pmp;
177 ndirent.de_flag = DE_ACCESS | DE_CREATE | DE_UPDATE;
179 DETIMES(&ndirent, &ts, &ts, &ts);
180 error = createde(&ndirent, pdep, &dep, cnp);
183 *ap->a_vpp = DETOV(dep);
184 if ((cnp->cn_flags & MAKEENTRY) != 0)
185 cache_enter(ap->a_dvp, *ap->a_vpp, cnp);
193 msdosfs_mknod(struct vop_mknod_args *ap)
200 msdosfs_open(struct vop_open_args *ap)
202 struct denode *dep = VTODE(ap->a_vp);
203 vnode_create_vobject(ap->a_vp, dep->de_FileSize, ap->a_td);
208 msdosfs_close(struct vop_close_args *ap)
210 struct vnode *vp = ap->a_vp;
211 struct denode *dep = VTODE(vp);
215 if (vp->v_usecount > 1) {
217 DETIMES(dep, &ts, &ts, &ts);
224 msdosfs_access(struct vop_access_args *ap)
226 struct vnode *vp = ap->a_vp;
227 struct denode *dep = VTODE(ap->a_vp);
228 struct msdosfsmount *pmp = dep->de_pmp;
230 accmode_t accmode = ap->a_accmode;
232 file_mode = S_IRWXU|S_IRWXG|S_IRWXO;
233 file_mode &= (vp->v_type == VDIR ? pmp->pm_dirmask : pmp->pm_mask);
236 * Disallow writing to directories and regular files if the
237 * filesystem is read-only.
239 if (accmode & VWRITE) {
240 switch (vp->v_type) {
243 if (vp->v_mount->mnt_flag & MNT_RDONLY)
251 return (vaccess(vp->v_type, file_mode, pmp->pm_uid, pmp->pm_gid,
252 ap->a_accmode, ap->a_cred, NULL));
256 msdosfs_getattr(struct vop_getattr_args *ap)
258 struct denode *dep = VTODE(ap->a_vp);
259 struct msdosfsmount *pmp = dep->de_pmp;
260 struct vattr *vap = ap->a_vap;
263 u_long dirsperblk = pmp->pm_BytesPerSec / sizeof(struct direntry);
267 DETIMES(dep, &ts, &ts, &ts);
268 vap->va_fsid = dev2udev(pmp->pm_dev);
270 * The following computation of the fileid must be the same as that
271 * used in msdosfs_readdir() to compute d_fileno. If not, pwd
274 if (dep->de_Attributes & ATTR_DIRECTORY) {
275 fileid = (uint64_t)cntobn(pmp, dep->de_StartCluster) *
277 if (dep->de_StartCluster == MSDOSFSROOT)
280 fileid = (uint64_t)cntobn(pmp, dep->de_dirclust) *
282 if (dep->de_dirclust == MSDOSFSROOT)
283 fileid = (uint64_t)roottobn(pmp, 0) * dirsperblk;
284 fileid += (uoff_t)dep->de_diroffset / sizeof(struct direntry);
287 if (pmp->pm_flags & MSDOSFS_LARGEFS)
288 vap->va_fileid = msdosfs_fileno_map(pmp->pm_mountp, fileid);
290 vap->va_fileid = (long)fileid;
292 mode = S_IRWXU|S_IRWXG|S_IRWXO;
293 vap->va_mode = mode &
294 (ap->a_vp->v_type == VDIR ? pmp->pm_dirmask : pmp->pm_mask);
295 vap->va_uid = pmp->pm_uid;
296 vap->va_gid = pmp->pm_gid;
298 vap->va_rdev = NODEV;
299 vap->va_size = dep->de_FileSize;
300 fattime2timespec(dep->de_MDate, dep->de_MTime, 0, 0, &vap->va_mtime);
301 vap->va_ctime = vap->va_mtime;
302 if (pmp->pm_flags & MSDOSFSMNT_LONGNAME) {
303 fattime2timespec(dep->de_ADate, 0, 0, 0, &vap->va_atime);
304 fattime2timespec(dep->de_CDate, dep->de_CTime, dep->de_CHun,
305 0, &vap->va_birthtime);
307 vap->va_atime = vap->va_mtime;
308 vap->va_birthtime.tv_sec = -1;
309 vap->va_birthtime.tv_nsec = 0;
312 if (dep->de_Attributes & ATTR_ARCHIVE)
313 vap->va_flags |= UF_ARCHIVE;
314 if (dep->de_Attributes & ATTR_HIDDEN)
315 vap->va_flags |= UF_HIDDEN;
316 if (dep->de_Attributes & ATTR_READONLY)
317 vap->va_flags |= UF_READONLY;
318 if (dep->de_Attributes & ATTR_SYSTEM)
319 vap->va_flags |= UF_SYSTEM;
321 vap->va_blocksize = pmp->pm_bpcluster;
323 (dep->de_FileSize + pmp->pm_crbomask) & ~pmp->pm_crbomask;
324 vap->va_type = ap->a_vp->v_type;
325 vap->va_filerev = dep->de_modrev;
330 msdosfs_setattr(struct vop_setattr_args *ap)
332 struct vnode *vp = ap->a_vp;
333 struct denode *dep = VTODE(ap->a_vp);
334 struct msdosfsmount *pmp = dep->de_pmp;
335 struct vattr *vap = ap->a_vap;
336 struct ucred *cred = ap->a_cred;
337 struct thread *td = curthread;
341 printf("msdosfs_setattr(): vp %p, vap %p, cred %p\n",
342 ap->a_vp, vap, cred);
346 * Check for unsettable attributes.
348 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
349 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
350 (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
351 (vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
353 printf("msdosfs_setattr(): returning EINVAL\n");
354 printf(" va_type %d, va_nlink %x, va_fsid %lx, va_fileid %lx\n",
355 vap->va_type, vap->va_nlink, vap->va_fsid, vap->va_fileid);
356 printf(" va_blocksize %lx, va_rdev %x, va_bytes %qx, va_gen %lx\n",
357 vap->va_blocksize, vap->va_rdev, vap->va_bytes, vap->va_gen);
358 printf(" va_uid %x, va_gid %x\n",
359 vap->va_uid, vap->va_gid);
365 * We don't allow setting attributes on the root directory.
366 * The special case for the root directory is because before
367 * FAT32, the root directory didn't have an entry for itself
368 * (and was otherwise special). With FAT32, the root
369 * directory is not so special, but still doesn't have an
372 if (vp->v_vflag & VV_ROOT)
375 if (vap->va_flags != VNOVAL) {
376 if (vp->v_mount->mnt_flag & MNT_RDONLY)
378 if (cred->cr_uid != pmp->pm_uid) {
379 error = priv_check_cred(cred, PRIV_VFS_ADMIN, 0);
384 * We are very inconsistent about handling unsupported
385 * attributes. We ignored the access time and the
386 * read and execute bits. We were strict for the other
389 if (vap->va_flags & ~(UF_ARCHIVE | UF_HIDDEN | UF_READONLY |
392 if (vap->va_flags & UF_ARCHIVE)
393 dep->de_Attributes |= ATTR_ARCHIVE;
395 dep->de_Attributes &= ~ATTR_ARCHIVE;
396 if (vap->va_flags & UF_HIDDEN)
397 dep->de_Attributes |= ATTR_HIDDEN;
399 dep->de_Attributes &= ~ATTR_HIDDEN;
400 /* We don't allow changing the readonly bit on directories. */
401 if (vp->v_type != VDIR) {
402 if (vap->va_flags & UF_READONLY)
403 dep->de_Attributes |= ATTR_READONLY;
405 dep->de_Attributes &= ~ATTR_READONLY;
407 if (vap->va_flags & UF_SYSTEM)
408 dep->de_Attributes |= ATTR_SYSTEM;
410 dep->de_Attributes &= ~ATTR_SYSTEM;
411 dep->de_flag |= DE_MODIFIED;
414 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
418 if (vp->v_mount->mnt_flag & MNT_RDONLY)
421 if (uid == (uid_t)VNOVAL)
424 if (gid == (gid_t)VNOVAL)
426 if (cred->cr_uid != pmp->pm_uid || uid != pmp->pm_uid ||
427 (gid != pmp->pm_gid && !groupmember(gid, cred))) {
428 error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0);
432 if (uid != pmp->pm_uid || gid != pmp->pm_gid)
436 if (vap->va_size != VNOVAL) {
437 switch (vp->v_type) {
442 * Truncation is only supported for regular files,
443 * Disallow it if the filesystem is read-only.
445 if (vp->v_mount->mnt_flag & MNT_RDONLY)
450 * According to POSIX, the result is unspecified
451 * for file types other than regular files,
452 * directories and shared memory objects. We
453 * don't support any file types except regular
454 * files and directories in this file system, so
455 * this (default) case is unreachable and can do
456 * anything. Keep falling through to detrunc()
461 error = detrunc(dep, vap->va_size, 0, cred);
465 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
466 if (vp->v_mount->mnt_flag & MNT_RDONLY)
468 error = vn_utimes_perm(vp, vap, cred, td);
471 if ((pmp->pm_flags & MSDOSFSMNT_NOWIN95) == 0 &&
472 vap->va_atime.tv_sec != VNOVAL) {
473 dep->de_flag &= ~DE_ACCESS;
474 timespec2fattime(&vap->va_atime, 0,
475 &dep->de_ADate, NULL, NULL);
477 if (vap->va_mtime.tv_sec != VNOVAL) {
478 dep->de_flag &= ~DE_UPDATE;
479 timespec2fattime(&vap->va_mtime, 0,
480 &dep->de_MDate, &dep->de_MTime, NULL);
483 * We don't set the archive bit when modifying the time of
484 * a directory to emulate the Windows/DOS behavior.
486 if (vp->v_type != VDIR)
487 dep->de_Attributes |= ATTR_ARCHIVE;
488 dep->de_flag |= DE_MODIFIED;
491 * DOS files only have the ability to have their writability
492 * attribute set, so we use the owner write bit to set the readonly
495 if (vap->va_mode != (mode_t)VNOVAL) {
496 if (vp->v_mount->mnt_flag & MNT_RDONLY)
498 if (cred->cr_uid != pmp->pm_uid) {
499 error = priv_check_cred(cred, PRIV_VFS_ADMIN, 0);
503 if (vp->v_type != VDIR) {
504 /* We ignore the read and execute bits. */
505 if (vap->va_mode & VWRITE)
506 dep->de_Attributes &= ~ATTR_READONLY;
508 dep->de_Attributes |= ATTR_READONLY;
509 dep->de_Attributes |= ATTR_ARCHIVE;
510 dep->de_flag |= DE_MODIFIED;
513 return (deupdat(dep, 0));
517 msdosfs_read(struct vop_read_args *ap)
531 struct vnode *vp = ap->a_vp;
532 struct denode *dep = VTODE(vp);
533 struct msdosfsmount *pmp = dep->de_pmp;
534 struct uio *uio = ap->a_uio;
537 * If they didn't ask for any data, then we are done.
539 orig_resid = uio->uio_resid;
544 * The caller is supposed to ensure that
545 * uio->uio_offset >= 0 and uio->uio_resid >= 0.
546 * We don't need to check for large offsets as in ffs because
547 * dep->de_FileSize <= DOS_FILESIZE_MAX < OFF_MAX, so large
548 * offsets cannot cause overflow even in theory.
551 seqcount = ap->a_ioflag >> IO_SEQSHIFT;
553 isadir = dep->de_Attributes & ATTR_DIRECTORY;
555 if (uio->uio_offset >= dep->de_FileSize)
557 lbn = de_cluster(pmp, uio->uio_offset);
559 blsize = pmp->pm_bpcluster;
560 on = uio->uio_offset & pmp->pm_crbomask;
562 * If we are operating on a directory file then be sure to
563 * do i/o with the vnode for the filesystem instead of the
564 * vnode for the directory.
567 /* convert cluster # to block # */
568 error = pcbmap(dep, lbn, &lbn, 0, &blsize);
569 if (error == E2BIG) {
574 error = bread(pmp->pm_devvp, lbn, blsize, NOCRED, &bp);
575 } else if (de_cn2off(pmp, rablock) >= dep->de_FileSize) {
576 error = bread(vp, lbn, blsize, NOCRED, &bp);
577 } else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
578 error = cluster_read(vp, dep->de_FileSize, lbn, blsize,
579 NOCRED, on + uio->uio_resid, seqcount, 0, &bp);
580 } else if (seqcount > 1) {
582 error = breadn(vp, lbn,
583 blsize, &rablock, &rasize, 1, NOCRED, &bp);
585 error = bread(vp, lbn, blsize, NOCRED, &bp);
591 diff = pmp->pm_bpcluster - on;
592 n = diff > uio->uio_resid ? uio->uio_resid : diff;
593 diff = dep->de_FileSize - uio->uio_offset;
596 diff = blsize - bp->b_resid;
599 error = vn_io_fault_uiomove(bp->b_data + on, (int) n, uio);
601 } while (error == 0 && uio->uio_resid > 0 && n != 0);
602 if (!isadir && (error == 0 || uio->uio_resid != orig_resid) &&
603 (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
604 dep->de_flag |= DE_ACCESS;
609 * Write data to a file or directory.
612 msdosfs_write(struct vop_write_args *ap)
623 int ioflag = ap->a_ioflag;
624 struct uio *uio = ap->a_uio;
625 struct vnode *vp = ap->a_vp;
626 struct vnode *thisvp;
627 struct denode *dep = VTODE(vp);
628 struct msdosfsmount *pmp = dep->de_pmp;
629 struct ucred *cred = ap->a_cred;
632 printf("msdosfs_write(vp %p, uio %p, ioflag %x, cred %p\n",
633 vp, uio, ioflag, cred);
634 printf("msdosfs_write(): diroff %lu, dirclust %lu, startcluster %lu\n",
635 dep->de_diroffset, dep->de_dirclust, dep->de_StartCluster);
638 switch (vp->v_type) {
640 if (ioflag & IO_APPEND)
641 uio->uio_offset = dep->de_FileSize;
647 panic("msdosfs_write(): bad file type");
651 * This is needed (unlike in ffs_write()) because we extend the
652 * file outside of the loop but we don't want to extend the file
653 * for writes of 0 bytes.
655 if (uio->uio_resid == 0)
659 * The caller is supposed to ensure that
660 * uio->uio_offset >= 0 and uio->uio_resid >= 0.
662 if ((uoff_t)uio->uio_offset + uio->uio_resid > DOS_FILESIZE_MAX)
666 * If they've exceeded their filesize limit, tell them about it.
668 if (vn_rlimit_fsize(vp, uio, uio->uio_td))
672 * If the offset we are starting the write at is beyond the end of
673 * the file, then they've done a seek. Unix filesystems allow
674 * files with holes in them, DOS doesn't so we must fill the hole
675 * with zeroed blocks.
677 if (uio->uio_offset > dep->de_FileSize) {
678 error = deextend(dep, uio->uio_offset, cred);
684 * Remember some values in case the write fails.
686 resid = uio->uio_resid;
687 osize = dep->de_FileSize;
690 * If we write beyond the end of the file, extend it to its ultimate
691 * size ahead of the time to hopefully get a contiguous area.
693 if (uio->uio_offset + resid > osize) {
694 count = de_clcount(pmp, uio->uio_offset + resid) -
695 de_clcount(pmp, osize);
696 error = extendfile(dep, count, NULL, NULL, 0);
697 if (error && (error != ENOSPC || (ioflag & IO_UNIT)))
699 lastcn = dep->de_fc[FC_LASTFC].fc_frcn;
701 lastcn = de_clcount(pmp, osize) - 1;
703 seqcount = ioflag >> IO_SEQSHIFT;
705 if (de_cluster(pmp, uio->uio_offset) > lastcn) {
710 croffset = uio->uio_offset & pmp->pm_crbomask;
711 n = min(uio->uio_resid, pmp->pm_bpcluster - croffset);
712 if (uio->uio_offset + n > dep->de_FileSize) {
713 dep->de_FileSize = uio->uio_offset + n;
714 /* The object size needs to be set before buffer is allocated */
715 vnode_pager_setsize(vp, dep->de_FileSize);
718 bn = de_cluster(pmp, uio->uio_offset);
719 if ((uio->uio_offset & pmp->pm_crbomask) == 0
720 && (de_cluster(pmp, uio->uio_offset + uio->uio_resid)
721 > de_cluster(pmp, uio->uio_offset)
722 || uio->uio_offset + uio->uio_resid >= dep->de_FileSize)) {
724 * If either the whole cluster gets written,
725 * or we write the cluster from its start beyond EOF,
726 * then no need to read data from disk.
728 bp = getblk(thisvp, bn, pmp->pm_bpcluster, 0, 0, 0);
730 * This call to vfs_bio_clrbuf() ensures that
731 * even if vn_io_fault_uiomove() below faults,
732 * garbage from the newly instantiated buffer
733 * is not exposed to the userspace via mmap().
737 * Do the bmap now, since pcbmap needs buffers
738 * for the fat table. (see msdosfs_strategy)
740 if (bp->b_blkno == bp->b_lblkno) {
741 error = pcbmap(dep, bp->b_lblkno, &bn, 0, 0);
747 if (bp->b_blkno == -1) {
750 error = EIO; /* XXX */
755 * The block we need to write into exists, so read it in.
757 error = bread(thisvp, bn, pmp->pm_bpcluster, cred, &bp);
765 * Should these vnode_pager_* functions be done on dir
770 * Copy the data from user space into the buf header.
772 error = vn_io_fault_uiomove(bp->b_data + croffset, n, uio);
778 /* Prepare for clustered writes in some else clauses. */
779 if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0)
780 bp->b_flags |= B_CLUSTEROK;
783 * If IO_SYNC, then each buffer is written synchronously.
784 * Otherwise, if we have a severe page deficiency then
785 * write the buffer asynchronously. Otherwise, if on a
786 * cluster boundary then write the buffer asynchronously,
787 * combining it with contiguous clusters if permitted and
788 * possible, since we don't expect more writes into this
789 * buffer soon. Otherwise, do a delayed write because we
790 * expect more writes into this buffer soon.
792 if (ioflag & IO_SYNC)
794 else if (vm_page_count_severe() || buf_dirty_count_severe())
796 else if (n + croffset == pmp->pm_bpcluster) {
797 if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0)
798 cluster_write(vp, bp, dep->de_FileSize,
804 dep->de_flag |= DE_UPDATE;
805 } while (error == 0 && uio->uio_resid > 0);
808 * If the write failed and they want us to, truncate the file back
809 * to the size it was before the write was attempted.
813 if (ioflag & IO_UNIT) {
814 detrunc(dep, osize, ioflag & IO_SYNC, NOCRED);
815 uio->uio_offset -= resid - uio->uio_resid;
816 uio->uio_resid = resid;
818 detrunc(dep, dep->de_FileSize, ioflag & IO_SYNC, NOCRED);
819 if (uio->uio_resid != resid)
822 } else if (ioflag & IO_SYNC)
823 error = deupdat(dep, 1);
828 * Flush the blocks of a file to disk.
831 msdosfs_fsync(struct vop_fsync_args *ap)
839 * If the syncing request comes from fsync(2), sync the entire
840 * FAT and any other metadata that happens to be on devvp. We
841 * need this mainly for the FAT. We write the FAT sloppily, and
842 * syncing it all now is the best we can easily do to get all
843 * directory entries associated with the file (not just the file)
844 * fully synced. The other metadata includes critical metadata
845 * for all directory entries, but only in the MNT_ASYNC case. We
846 * will soon sync all metadata in the file's directory entry.
847 * Non-critical metadata for associated directory entries only
848 * gets synced accidentally, as in most file systems.
850 if (ap->a_waitfor == MNT_WAIT) {
851 devvp = VTODE(ap->a_vp)->de_pmp->pm_devvp;
852 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
853 allerror = VOP_FSYNC(devvp, MNT_WAIT, ap->a_td);
854 VOP_UNLOCK(devvp, 0);
858 error = deupdat(VTODE(ap->a_vp), ap->a_waitfor == MNT_WAIT);
865 msdosfs_remove(struct vop_remove_args *ap)
867 struct denode *dep = VTODE(ap->a_vp);
868 struct denode *ddep = VTODE(ap->a_dvp);
871 if (ap->a_vp->v_type == VDIR)
874 error = removede(ddep, dep);
876 printf("msdosfs_remove(), dep %p, v_usecount %d\n", dep, ap->a_vp->v_usecount);
882 * DOS filesystems don't know what links are.
885 msdosfs_link(struct vop_link_args *ap)
891 * Renames on files require moving the denode to a new hash queue since the
892 * denode's location is used to compute which hash queue to put the file
893 * in. Unless it is a rename in place. For example "mv a b".
895 * What follows is the basic algorithm:
898 * if (dest file exists) {
901 * if (dest and src in same directory) {
902 * rewrite name in existing directory slot
904 * write new entry in dest directory
905 * update offset and dirclust in denode
906 * move denode to new hash chain
907 * clear old directory entry
911 * if (dest directory exists) {
912 * if (dest is not empty) {
915 * remove dest directory
917 * if (dest and src in same directory) {
918 * rewrite name in existing entry
920 * be sure dest is not a child of src directory
921 * write entry in dest directory
922 * update "." and ".." in moved directory
923 * clear old directory entry for moved directory
928 * source's parent directory is unlocked
929 * source file or directory is unlocked
930 * destination's parent directory is locked
931 * destination file or directory is locked if it exists
934 * all denodes should be released
937 msdosfs_rename(struct vop_rename_args *ap)
939 struct vnode *tdvp = ap->a_tdvp;
940 struct vnode *fvp = ap->a_fvp;
941 struct vnode *fdvp = ap->a_fdvp;
942 struct vnode *tvp = ap->a_tvp;
943 struct componentname *tcnp = ap->a_tcnp;
944 struct componentname *fcnp = ap->a_fcnp;
945 struct denode *ip, *xp, *dp, *zp;
946 u_char toname[12], oldname[11];
947 u_long from_diroffset, to_diroffset;
949 int doingdirectory = 0, newparent = 0;
953 struct msdosfsmount *pmp;
954 struct direntry *dotdotp;
957 pmp = VFSTOMSDOSFS(fdvp->v_mount);
960 if ((tcnp->cn_flags & HASBUF) == 0 ||
961 (fcnp->cn_flags & HASBUF) == 0)
962 panic("msdosfs_rename: no name");
965 * Check for cross-device rename.
967 if (fvp->v_mount != tdvp->v_mount ||
968 (tvp && fvp->v_mount != tvp->v_mount)) {
983 * If source and dest are the same, do nothing.
990 error = vn_lock(fvp, LK_EXCLUSIVE);
997 * Be sure we are not renaming ".", "..", or an alias of ".". This
998 * leads to a crippled directory tree. It's pretty tough to do a
999 * "ls" or "pwd" with the "." directory entry missing, and "cd .."
1000 * doesn't work if the ".." entry is missing.
1002 if (ip->de_Attributes & ATTR_DIRECTORY) {
1004 * Avoid ".", "..", and aliases of "." for obvious reasons.
1006 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
1008 (fcnp->cn_flags & ISDOTDOT) ||
1009 (tcnp->cn_flags & ISDOTDOT) ||
1010 (ip->de_flag & DE_RENAME)) {
1015 ip->de_flag |= DE_RENAME;
1020 * When the target exists, both the directory
1021 * and target vnodes are returned locked.
1024 xp = tvp ? VTODE(tvp) : NULL;
1026 * Remember direntry place to use for destination
1028 to_diroffset = dp->de_fndoffset;
1029 to_count = dp->de_fndcnt;
1032 * If ".." must be changed (ie the directory gets a new
1033 * parent) then the source directory must not be in the
1034 * directory hierarchy above the target, as this would
1035 * orphan everything below the source directory. Also
1036 * the user must have write permission in the source so
1037 * as to be able to change "..". We must repeat the call
1038 * to namei, as the parent directory is unlocked by the
1039 * call to doscheckpath().
1041 error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread);
1043 if (VTODE(fdvp)->de_StartCluster != VTODE(tdvp)->de_StartCluster)
1045 if (doingdirectory && newparent) {
1046 if (error) /* write access check above */
1051 * doscheckpath() vput()'s dp,
1052 * so we have to do a relookup afterwards
1054 error = doscheckpath(ip, dp);
1057 if ((tcnp->cn_flags & SAVESTART) == 0)
1058 panic("msdosfs_rename: lost to startdir");
1059 error = relookup(tdvp, &tvp, tcnp);
1063 xp = tvp ? VTODE(tvp) : NULL;
1068 * Target must be empty if a directory and have no links
1069 * to it. Also, ensure source and target are compatible
1070 * (both directories, or both not directories).
1072 if (xp->de_Attributes & ATTR_DIRECTORY) {
1073 if (!dosdirempty(xp)) {
1077 if (!doingdirectory) {
1082 } else if (doingdirectory) {
1086 error = removede(dp, xp);
1094 * Convert the filename in tcnp into a dos filename. We copy this
1095 * into the denode and directory entry for the destination
1098 error = uniqdosname(VTODE(tdvp), tcnp, toname);
1103 * Since from wasn't locked at various places above,
1104 * have to do a relookup here.
1106 fcnp->cn_flags &= ~MODMASK;
1107 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
1108 if ((fcnp->cn_flags & SAVESTART) == 0)
1109 panic("msdosfs_rename: lost from startdir");
1111 VOP_UNLOCK(tdvp, 0);
1112 if (relookup(fdvp, &fvp, fcnp) == 0)
1116 * From name has disappeared.
1119 panic("rename: lost dir entry");
1121 VOP_UNLOCK(tdvp, 0);
1128 from_diroffset = zp->de_fndoffset;
1131 * Ensure that the directory entry still exists and has not
1132 * changed till now. If the source is a file the entry may
1133 * have been unlinked or renamed. In either case there is
1134 * no further work to be done. If the source is a directory
1135 * then it cannot have been rmdir'ed or renamed; this is
1136 * prohibited by the DE_RENAME flag.
1140 panic("rename: lost dir entry");
1143 VOP_UNLOCK(fdvp, 0);
1151 * First write a new entry in the destination
1152 * directory and mark the entry in the source directory
1153 * as deleted. Then move the denode to the correct hash
1154 * chain for its new location in the filesystem. And, if
1155 * we moved a directory, then update its .. entry to point
1156 * to the new parent directory.
1158 bcopy(ip->de_Name, oldname, 11);
1159 bcopy(toname, ip->de_Name, 11); /* update denode */
1160 dp->de_fndoffset = to_diroffset;
1161 dp->de_fndcnt = to_count;
1162 error = createde(ip, dp, (struct denode **)0, tcnp);
1164 bcopy(oldname, ip->de_Name, 11);
1166 VOP_UNLOCK(fdvp, 0);
1171 * If ip is for a directory, then its name should always
1172 * be "." since it is for the directory entry in the
1173 * directory itself (msdosfs_lookup() always translates
1174 * to the "." entry so as to get a unique denode, except
1175 * for the root directory there are different
1176 * complications). However, we just corrupted its name
1177 * to pass the correct name to createde(). Undo this.
1179 if ((ip->de_Attributes & ATTR_DIRECTORY) != 0)
1180 bcopy(oldname, ip->de_Name, 11);
1182 zp->de_fndoffset = from_diroffset;
1183 error = removede(zp, ip);
1185 /* XXX should downgrade to ro here, fs is corrupt */
1187 VOP_UNLOCK(fdvp, 0);
1191 if (!doingdirectory) {
1192 error = pcbmap(dp, de_cluster(pmp, to_diroffset), 0,
1193 &ip->de_dirclust, 0);
1195 /* XXX should downgrade to ro here, fs is corrupt */
1197 VOP_UNLOCK(fdvp, 0);
1201 if (ip->de_dirclust == MSDOSFSROOT)
1202 ip->de_diroffset = to_diroffset;
1204 ip->de_diroffset = to_diroffset & pmp->pm_crbomask;
1208 VOP_UNLOCK(fdvp, 0);
1212 * If we moved a directory to a new parent directory, then we must
1213 * fixup the ".." entry in the moved directory.
1215 if (doingdirectory && newparent) {
1216 cn = ip->de_StartCluster;
1217 if (cn == MSDOSFSROOT) {
1218 /* this should never happen */
1219 panic("msdosfs_rename(): updating .. in root directory?");
1221 bn = cntobn(pmp, cn);
1222 error = bread(pmp->pm_devvp, bn, pmp->pm_bpcluster,
1225 /* 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 bzero(&ndirent, 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 bzero(bp->b_data, pmp->pm_bpcluster);
1341 bcopy(&dosdirtemplate, bp->b_data, 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)
1479 struct denode *dep = VTODE(ap->a_vp);
1480 struct msdosfsmount *pmp = dep->de_pmp;
1481 struct direntry *dentp;
1482 struct dirent dirbuf;
1483 struct uio *uio = ap->a_uio;
1484 u_long *cookies = NULL;
1489 #ifdef MSDOSFS_DEBUG
1490 printf("msdosfs_readdir(): vp %p, uio %p, cred %p, eofflagp %p\n",
1491 ap->a_vp, uio, ap->a_cred, ap->a_eofflag);
1495 * msdosfs_readdir() won't operate properly on regular files since
1496 * it does i/o only with the filesystem vnode, and hence can
1497 * retrieve the wrong block from the buffer cache for a plain file.
1498 * So, fail attempts to readdir() on a plain file.
1500 if ((dep->de_Attributes & ATTR_DIRECTORY) == 0)
1504 * To be safe, initialize dirbuf
1506 bzero(dirbuf.d_name, sizeof(dirbuf.d_name));
1509 * If the user buffer is smaller than the size of one dos directory
1510 * entry or the file offset is not a multiple of the size of a
1511 * directory entry, then we fail the read.
1513 off = offset = uio->uio_offset;
1514 if (uio->uio_resid < sizeof(struct direntry) ||
1515 (offset & (sizeof(struct direntry) - 1)))
1518 if (ap->a_ncookies) {
1519 ncookies = uio->uio_resid / 16;
1520 cookies = malloc(ncookies * sizeof(u_long), M_TEMP,
1522 *ap->a_cookies = cookies;
1523 *ap->a_ncookies = ncookies;
1526 dirsperblk = pmp->pm_BytesPerSec / sizeof(struct direntry);
1529 * If they are reading from the root directory then, we simulate
1530 * the . and .. entries since these don't exist in the root
1531 * directory. We also set the offset bias to make up for having to
1532 * simulate these entries. By this I mean that at file offset 64 we
1533 * read the first entry in the root directory that lives on disk.
1535 if (dep->de_StartCluster == MSDOSFSROOT
1536 || (FAT32(pmp) && dep->de_StartCluster == pmp->pm_rootdirblk)) {
1538 printf("msdosfs_readdir(): going after . or .. in root dir, offset %d\n",
1541 bias = 2 * sizeof(struct direntry);
1542 if (offset < bias) {
1543 for (n = (int)offset / sizeof(struct direntry);
1546 fileno = (uint64_t)cntobn(pmp,
1551 if (pmp->pm_flags & MSDOSFS_LARGEFS) {
1553 msdosfs_fileno_map(pmp->pm_mountp,
1557 dirbuf.d_fileno = (uint32_t)fileno;
1559 dirbuf.d_type = DT_DIR;
1562 dirbuf.d_namlen = 1;
1563 strcpy(dirbuf.d_name, ".");
1566 dirbuf.d_namlen = 2;
1567 strcpy(dirbuf.d_name, "..");
1570 dirbuf.d_reclen = GENERIC_DIRSIZ(&dirbuf);
1571 if (uio->uio_resid < dirbuf.d_reclen)
1573 error = uiomove(&dirbuf, dirbuf.d_reclen, uio);
1576 offset += sizeof(struct direntry);
1579 *cookies++ = offset;
1580 if (--ncookies <= 0)
1589 while (uio->uio_resid > 0) {
1590 lbn = de_cluster(pmp, offset - bias);
1591 on = (offset - bias) & pmp->pm_crbomask;
1592 n = min(pmp->pm_bpcluster - on, uio->uio_resid);
1593 diff = dep->de_FileSize - (offset - bias);
1597 error = pcbmap(dep, lbn, &bn, &cn, &blsize);
1600 error = bread(pmp->pm_devvp, bn, blsize, NOCRED, &bp);
1605 n = min(n, blsize - bp->b_resid);
1612 * Convert from dos directory entries to fs-independent
1613 * directory entries.
1615 for (dentp = (struct direntry *)(bp->b_data + on);
1616 (char *)dentp < bp->b_data + on + n;
1617 dentp++, offset += sizeof(struct direntry)) {
1619 printf("rd: dentp %08x prev %08x crnt %08x deName %02x attr %02x\n",
1620 dentp, prev, crnt, dentp->deName[0], dentp->deAttributes);
1623 * If this is an unused entry, we can stop.
1625 if (dentp->deName[0] == SLOT_EMPTY) {
1630 * Skip deleted entries.
1632 if (dentp->deName[0] == SLOT_DELETED) {
1639 * Handle Win95 long directory entries
1641 if (dentp->deAttributes == ATTR_WIN95) {
1642 if (pmp->pm_flags & MSDOSFSMNT_SHORTNAME)
1644 chksum = win2unixfn(&nb,
1645 (struct winentry *)dentp, chksum, pmp);
1650 * Skip volume labels
1652 if (dentp->deAttributes & ATTR_VOLUME) {
1658 * This computation of d_fileno must match
1659 * the computation of va_fileid in
1662 if (dentp->deAttributes & ATTR_DIRECTORY) {
1663 fileno = getushort(dentp->deStartCluster);
1665 fileno |= getushort(dentp->deHighClust) << 16;
1666 /* if this is the root directory */
1667 if (fileno == MSDOSFSROOT)
1669 fileno = (uint64_t)cntobn(pmp,
1675 fileno = (uint64_t)cntobn(pmp, fileno) *
1677 dirbuf.d_type = DT_DIR;
1679 fileno = (uoff_t)offset /
1680 sizeof(struct direntry);
1681 dirbuf.d_type = DT_REG;
1683 if (pmp->pm_flags & MSDOSFS_LARGEFS) {
1685 msdosfs_fileno_map(pmp->pm_mountp, fileno);
1687 dirbuf.d_fileno = (uint32_t)fileno;
1689 if (chksum != winChksum(dentp->deName)) {
1690 dirbuf.d_namlen = dos2unixfn(dentp->deName,
1691 (u_char *)dirbuf.d_name,
1692 dentp->deLowerCase |
1693 ((pmp->pm_flags & MSDOSFSMNT_SHORTNAME) ?
1694 (LCASE_BASE | LCASE_EXT) : 0),
1698 mbnambuf_flush(&nb, &dirbuf);
1700 dirbuf.d_reclen = GENERIC_DIRSIZ(&dirbuf);
1701 if (uio->uio_resid < dirbuf.d_reclen) {
1705 error = uiomove(&dirbuf, dirbuf.d_reclen, uio);
1711 *cookies++ = offset + sizeof(struct direntry);
1712 if (--ncookies <= 0) {
1717 off = offset + sizeof(struct direntry);
1722 /* Subtract unused cookies */
1724 *ap->a_ncookies -= ncookies;
1726 uio->uio_offset = off;
1729 * Set the eofflag (NFS uses it)
1731 if (ap->a_eofflag) {
1732 if (dep->de_FileSize - (offset - bias) <= 0)
1741 * a_vp - pointer to the file's vnode
1742 * a_bn - logical block number within the file (cluster number for us)
1743 * a_bop - where to return the bufobj of the special file containing the fs
1744 * a_bnp - where to return the "physical" block number corresponding to a_bn
1745 * (relative to the special file; units are blocks of size DEV_BSIZE)
1746 * a_runp - where to return the "run past" a_bn. This is the count of logical
1747 * blocks whose physical blocks (together with a_bn's physical block)
1749 * a_runb - where to return the "run before" a_bn.
1752 msdosfs_bmap(struct vop_bmap_args *ap)
1756 struct msdosfsmount *pmp;
1760 int bnpercn, error, maxio, maxrun, run;
1765 if (ap->a_bop != NULL)
1766 *ap->a_bop = &pmp->pm_devvp->v_bufobj;
1767 if (ap->a_bnp == NULL)
1769 if (ap->a_runp != NULL)
1771 if (ap->a_runb != NULL)
1776 error = pcbmap(dep, cn, ap->a_bnp, NULL, NULL);
1777 if (error != 0 || (ap->a_runp == NULL && ap->a_runb == NULL))
1781 maxio = mp->mnt_iosize_max / mp->mnt_stat.f_iosize;
1782 bnpercn = de_cn2bn(pmp, 1);
1783 if (ap->a_runp != NULL) {
1784 maxrun = ulmin(maxio - 1, pmp->pm_maxcluster - cn);
1785 for (run = 1; run <= maxrun; run++) {
1786 if (pcbmap(dep, cn + run, &runbn, NULL, NULL) != 0 ||
1787 runbn != *ap->a_bnp + run * bnpercn)
1790 *ap->a_runp = run - 1;
1792 if (ap->a_runb != NULL) {
1793 maxrun = ulmin(maxio - 1, cn);
1794 for (run = 1; run < maxrun; run++) {
1795 if (pcbmap(dep, cn - run, &runbn, NULL, NULL) != 0 ||
1796 runbn != *ap->a_bnp - run * bnpercn)
1799 *ap->a_runb = run - 1;
1804 SYSCTL_NODE(_vfs, OID_AUTO, msdosfs, CTLFLAG_RW, 0, "msdos filesystem");
1805 static int use_buf_pager = 0;
1806 SYSCTL_INT(_vfs_msdosfs, OID_AUTO, use_buf_pager, CTLFLAG_RWTUN,
1808 "Use buffer pager instead of bmap");
1811 msdosfs_gbp_getblkno(struct vnode *vp, vm_ooffset_t off)
1814 return (de_cluster(VTODE(vp)->de_pmp, off));
1818 msdosfs_gbp_getblksz(struct vnode *vp, daddr_t lbn)
1821 return (VTODE(vp)->de_pmp->pm_bpcluster);
1825 msdosfs_getpages(struct vop_getpages_args *ap)
1829 return (vfs_bio_getpages(ap->a_vp, ap->a_m, ap->a_count,
1830 ap->a_rbehind, ap->a_rahead, msdosfs_gbp_getblkno,
1831 msdosfs_gbp_getblksz));
1832 return (vnode_pager_generic_getpages(ap->a_vp, ap->a_m, ap->a_count,
1833 ap->a_rbehind, ap->a_rahead, NULL, NULL));
1837 msdosfs_strategy(struct vop_strategy_args *ap)
1839 struct buf *bp = ap->a_bp;
1840 struct denode *dep = VTODE(ap->a_vp);
1846 * If we don't already know the filesystem relative block number
1847 * then get it using pcbmap(). If pcbmap() returns the block
1848 * number as -1 then we've got a hole in the file. DOS filesystems
1849 * don't allow files with holes, so we shouldn't ever see this.
1851 if (bp->b_blkno == bp->b_lblkno) {
1852 error = pcbmap(dep, bp->b_lblkno, &blkno, 0, 0);
1853 bp->b_blkno = blkno;
1855 bp->b_error = error;
1856 bp->b_ioflags |= BIO_ERROR;
1860 if ((long)bp->b_blkno == -1)
1863 if (bp->b_blkno == -1) {
1868 * Read/write the block from/to the disk that contains the desired
1871 bp->b_iooffset = dbtob(bp->b_blkno);
1872 bo = dep->de_pmp->pm_bo;
1873 BO_STRATEGY(bo, bp);
1878 msdosfs_print(struct vop_print_args *ap)
1880 struct denode *dep = VTODE(ap->a_vp);
1882 printf("\tstartcluster %lu, dircluster %lu, diroffset %lu, ",
1883 dep->de_StartCluster, dep->de_dirclust, dep->de_diroffset);
1884 printf("on dev %s\n", devtoname(dep->de_pmp->pm_dev));
1889 msdosfs_pathconf(struct vop_pathconf_args *ap)
1891 struct msdosfsmount *pmp = VTODE(ap->a_vp)->de_pmp;
1893 switch (ap->a_name) {
1898 *ap->a_retval = pmp->pm_flags & MSDOSFSMNT_LONGNAME ? WIN_MAXLEN : 12;
1901 *ap->a_retval = PATH_MAX;
1903 case _PC_CHOWN_RESTRICTED:
1916 msdosfs_vptofh(struct vop_vptofh_args *ap)
1919 struct defid *defhp;
1921 dep = VTODE(ap->a_vp);
1922 defhp = (struct defid *)ap->a_fhp;
1923 defhp->defid_len = sizeof(struct defid);
1924 defhp->defid_dirclust = dep->de_dirclust;
1925 defhp->defid_dirofs = dep->de_diroffset;
1926 /* defhp->defid_gen = dep->de_gen; */
1930 /* Global vfs data structures for msdosfs */
1931 struct vop_vector msdosfs_vnodeops = {
1932 .vop_default = &default_vnodeops,
1934 .vop_access = msdosfs_access,
1935 .vop_bmap = msdosfs_bmap,
1936 .vop_getpages = msdosfs_getpages,
1937 .vop_cachedlookup = msdosfs_lookup,
1938 .vop_open = msdosfs_open,
1939 .vop_close = msdosfs_close,
1940 .vop_create = msdosfs_create,
1941 .vop_fsync = msdosfs_fsync,
1942 .vop_fdatasync = vop_stdfdatasync_buf,
1943 .vop_getattr = msdosfs_getattr,
1944 .vop_inactive = msdosfs_inactive,
1945 .vop_link = msdosfs_link,
1946 .vop_lookup = vfs_cache_lookup,
1947 .vop_mkdir = msdosfs_mkdir,
1948 .vop_mknod = msdosfs_mknod,
1949 .vop_pathconf = msdosfs_pathconf,
1950 .vop_print = msdosfs_print,
1951 .vop_read = msdosfs_read,
1952 .vop_readdir = msdosfs_readdir,
1953 .vop_reclaim = msdosfs_reclaim,
1954 .vop_remove = msdosfs_remove,
1955 .vop_rename = msdosfs_rename,
1956 .vop_rmdir = msdosfs_rmdir,
1957 .vop_setattr = msdosfs_setattr,
1958 .vop_strategy = msdosfs_strategy,
1959 .vop_symlink = msdosfs_symlink,
1960 .vop_write = msdosfs_write,
1961 .vop_vptofh = msdosfs_vptofh,