sys/vfs/msdosfs/msdosfs_vnops.c

   1 /* $FreeBSD: src/sys/msdosfs/msdosfs_vnops.c,v 1.95.2.4 2003/06/13 15:05:47 trhodes Exp $ */
   2 /* $DragonFly: src/sys/vfs/msdosfs/msdosfs_vnops.c,v 1.54 2007/11/20 21:03:50 dillon Exp $ */
   3 /*      $NetBSD: msdosfs_vnops.c,v 1.68 1998/02/10 14:10:04 mrg Exp $   */
   4
   5 /*-
   6  * Copyright (C) 1994, 1995, 1997 Wolfgang Solfrank.
   7  * Copyright (C) 1994, 1995, 1997 TooLs GmbH.
   8  * All rights reserved.
   9  * Original code by Paul Popelka (paulp@uts.amdahl.com) (see below).
  10  *
  11  * Redistribution and use in source and binary forms, with or without
  12  * modification, are permitted provided that the following conditions
  13  * are met:
  14  * 1. Redistributions of source code must retain the above copyright
  15  *    notice, this list of conditions and the following disclaimer.
  16  * 2. Redistributions in binary form must reproduce the above copyright
  17  *    notice, this list of conditions and the following disclaimer in the
  18  *    documentation and/or other materials provided with the distribution.
  19  * 3. All advertising materials mentioning features or use of this software
  20  *    must display the following acknowledgement:
  21  *      This product includes software developed by TooLs GmbH.
  22  * 4. The name of TooLs GmbH may not be used to endorse or promote products
  23  *    derived from this software without specific prior written permission.
  24  *
  25  * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
  26  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  27  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  28  * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  29  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
  30  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
  31  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
  32  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
  33  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
  34  * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  35  */
  36 /*
  37  * Written by Paul Popelka (paulp@uts.amdahl.com)
  38  *
  39  * You can do anything you want with this software, just don't say you wrote
  40  * it, and don't remove this notice.
  41  *
  42  * This software is provided "as is".
  43  *
  44  * The author supplies this software to be publicly redistributed on the
  45  * understanding that the author is not responsible for the correct
  46  * functioning of this software in any circumstances and is not liable for
  47  * any damages caused by this software.
  48  *
  49  * October 1992
  50  */
  51
  52 #include <sys/param.h>
  53 #include <sys/systm.h>
  54 #include <sys/resourcevar.h>    /* defines plimit structure in proc struct */
  55 #include <sys/kernel.h>
  56 #include <sys/stat.h>
  57 #include <sys/buf.h>
  58 #include <sys/proc.h>
  59 #include <sys/namei.h>
  60 #include <sys/mount.h>
  61 #include <sys/unistd.h>
  62 #include <sys/vnode.h>
  63 #include <sys/malloc.h>
  64 #include <sys/dirent.h>
  65 #include <sys/signalvar.h>
  66
  67 #include <vm/vm.h>
  68 #include <vm/vm_extern.h>
  69 #include <vm/vm_zone.h>
  70 #include <vm/vnode_pager.h>
  71
  72 #include <sys/buf2.h>
  73
  74 #include "bpb.h"
  75 #include "direntry.h"
  76 #include "denode.h"
  77 #include "msdosfsmount.h"
  78 #include "fat.h"
  79
  80 #define DOS_FILESIZE_MAX        0xffffffff
  81
  82 /*
  83  * Prototypes for MSDOSFS vnode operations
  84  */
  85 static int msdosfs_create (struct vop_old_create_args *);
  86 static int msdosfs_mknod (struct vop_old_mknod_args *);
  87 static int msdosfs_open (struct vop_open_args *);
  88 static int msdosfs_close (struct vop_close_args *);
  89 static int msdosfs_access (struct vop_access_args *);
  90 static int msdosfs_getattr (struct vop_getattr_args *);
  91 static int msdosfs_setattr (struct vop_setattr_args *);
  92 static int msdosfs_read (struct vop_read_args *);
  93 static int msdosfs_write (struct vop_write_args *);
  94 static int msdosfs_fsync (struct vop_fsync_args *);
  95 static int msdosfs_remove (struct vop_old_remove_args *);
  96 static int msdosfs_link (struct vop_old_link_args *);
  97 static int msdosfs_rename (struct vop_old_rename_args *);
  98 static int msdosfs_mkdir (struct vop_old_mkdir_args *);
  99 static int msdosfs_rmdir (struct vop_old_rmdir_args *);
 100 static int msdosfs_symlink (struct vop_old_symlink_args *);
 101 static int msdosfs_readdir (struct vop_readdir_args *);
 102 static int msdosfs_bmap (struct vop_bmap_args *);
 103 static int msdosfs_strategy (struct vop_strategy_args *);
 104 static int msdosfs_print (struct vop_print_args *);
 105 static int msdosfs_pathconf (struct vop_pathconf_args *ap);
 106
 107 /*
 108  * Some general notes:
 109  *
 110  * In the ufs filesystem the inodes, superblocks, and indirect blocks are
 111  * read/written using the vnode for the filesystem. Blocks that represent
 112  * the contents of a file are read/written using the vnode for the file
 113  * (including directories when they are read/written as files). This
 114  * presents problems for the dos filesystem because data that should be in
 115  * an inode (if dos had them) resides in the directory itself.  Since we
 116  * must update directory entries without the benefit of having the vnode
 117  * for the directory we must use the vnode for the filesystem.  This means
 118  * that when a directory is actually read/written (via read, write, or
 119  * readdir, or seek) we must use the vnode for the filesystem instead of
 120  * the vnode for the directory as would happen in ufs. This is to insure we
 121  * retreive the correct block from the buffer cache since the hash value is
 122  * based upon the vnode address and the desired block number.
 123  */
 124
 125 /*
 126  * Create a regular file. On entry the directory to contain the file being
 127  * created is locked.  We must release before we return.
 128  *
 129  * msdosfs_create(struct vnode *a_dvp, struct vnode **a_vpp,
 130  *                struct componentname *a_cnp, struct vattr *a_vap)
 131  */
 132 static int
 133 msdosfs_create(struct vop_old_create_args *ap)
 134 {
 135         struct componentname *cnp = ap->a_cnp;
 136         struct denode ndirent;
 137         struct denode *dep;
 138         struct denode *pdep = VTODE(ap->a_dvp);
 139         struct timespec ts;
 140         int error;
 141
 142 #ifdef MSDOSFS_DEBUG
 143         kprintf("msdosfs_create(cnp %p, vap %p\n", cnp, ap->a_vap);
 144 #endif
 145
 146         /*
 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
 149          * change size.
 150          */
 151         if (pdep->de_StartCluster == MSDOSFSROOT
 152             && pdep->de_fndoffset >= pdep->de_FileSize) {
 153                 error = ENOSPC;
 154                 goto bad;
 155         }
 156
 157         /*
 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
 161          * readonly.
 162          */
 163         bzero(&ndirent, sizeof(ndirent));
 164         error = uniqdosname(pdep, cnp, ndirent.de_Name);
 165         if (error)
 166                 goto bad;
 167
 168         ndirent.de_Attributes = (ap->a_vap->va_mode & VWRITE) ?
 169                                 ATTR_ARCHIVE : ATTR_ARCHIVE | ATTR_READONLY;
 170         ndirent.de_LowerCase = 0;
 171         ndirent.de_StartCluster = 0;
 172         ndirent.de_FileSize = 0;
 173         ndirent.de_dev = pdep->de_dev;
 174         ndirent.de_devvp = pdep->de_devvp;
 175         ndirent.de_pmp = pdep->de_pmp;
 176         ndirent.de_flag = DE_ACCESS | DE_CREATE | DE_UPDATE;
 177         getnanotime(&ts);
 178         DETIMES(&ndirent, &ts, &ts, &ts);
 179         error = createde(&ndirent, pdep, &dep, cnp);
 180         if (error)
 181                 goto bad;
 182         *ap->a_vpp = DETOV(dep);
 183         return (0);
 184
 185 bad:
 186         return (error);
 187 }
 188
 189 /*
 190  * msdosfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp,
 191  *               struct componentname *a_cnp, struct vattr *a_vap)
 192  */
 193 static int
 194 msdosfs_mknod(struct vop_old_mknod_args *ap)
 195 {
 196         switch (ap->a_vap->va_type) {
 197         case VDIR:
 198                 return (msdosfs_mkdir((struct vop_old_mkdir_args *)ap));
 199                 break;
 200
 201         case VREG:
 202                 return (msdosfs_create((struct vop_old_create_args *)ap));
 203                 break;
 204
 205         default:
 206                 return (EINVAL);
 207         }
 208         /* NOTREACHED */
 209 }
 210
 211 /*
 212  * msdosfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
 213  *              struct file *a_fp)
 214  */
 215 static int
 216 msdosfs_open(struct vop_open_args *ap)
 217 {
 218         return(vop_stdopen(ap));
 219 }
 220
 221 /*
 222  * msdosfs_close(struct vnode *a_vp, int a_fflag)
 223  */
 224 static int
 225 msdosfs_close(struct vop_close_args *ap)
 226 {
 227         struct vnode *vp = ap->a_vp;
 228         struct denode *dep = VTODE(vp);
 229         struct timespec ts;
 230
 231         if (vp->v_sysref.refcnt > 1) {
 232                 getnanotime(&ts);
 233                 DETIMES(dep, &ts, &ts, &ts);
 234         }
 235         return (vop_stdclose(ap));
 236 }
 237
 238 /*
 239  * msdosfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
 240  */
 241 static int
 242 msdosfs_access(struct vop_access_args *ap)
 243 {
 244         struct vnode *vp = ap->a_vp;
 245         struct denode *dep = VTODE(ap->a_vp);
 246         struct msdosfsmount *pmp = dep->de_pmp;
 247         struct ucred *cred = ap->a_cred;
 248         mode_t mask, file_mode, mode = ap->a_mode;
 249         gid_t *gp;
 250         int i;
 251
 252         file_mode = (S_IXUSR|S_IXGRP|S_IXOTH) | (S_IRUSR|S_IRGRP|S_IROTH) |
 253             ((dep->de_Attributes & ATTR_READONLY) ? 0 : (S_IWUSR|S_IWGRP|S_IWOTH));
 254         file_mode &= pmp->pm_mask;
 255
 256         /*
 257          * Disallow write attempts on read-only file systems;
 258          * unless the file is a socket, fifo, or a block or
 259          * character device resident on the file system.
 260          */
 261         if (mode & VWRITE) {
 262                 switch (vp->v_type) {
 263                 case VDIR:
 264                 case VLNK:
 265                 case VREG:
 266                         if (vp->v_mount->mnt_flag & MNT_RDONLY)
 267                                 return (EROFS);
 268                         break;
 269                 default:
 270                         break;
 271                 }
 272         }
 273
 274         /* User id 0 always gets access. */
 275         if (cred->cr_uid == 0)
 276                 return 0;
 277
 278         mask = 0;
 279
 280         /* Otherwise, check the owner. */
 281         if (cred->cr_uid == pmp->pm_uid) {
 282                 if (mode & VEXEC)
 283                         mask |= S_IXUSR;
 284                 if (mode & VREAD)
 285                         mask |= S_IRUSR;
 286                 if (mode & VWRITE)
 287                         mask |= S_IWUSR;
 288                 return (file_mode & mask) == mask ? 0 : EACCES;
 289         }
 290
 291         /* Otherwise, check the groups. */
 292         for (i = 0, gp = cred->cr_groups; i < cred->cr_ngroups; i++, gp++)
 293                 if (pmp->pm_gid == *gp) {
 294                         if (mode & VEXEC)
 295                                 mask |= S_IXGRP;
 296                         if (mode & VREAD)
 297                                 mask |= S_IRGRP;
 298                         if (mode & VWRITE)
 299                                 mask |= S_IWGRP;
 300                         return (file_mode & mask) == mask ? 0 : EACCES;
 301                 }
 302
 303         /* Otherwise, check everyone else. */
 304         if (mode & VEXEC)
 305                 mask |= S_IXOTH;
 306         if (mode & VREAD)
 307                 mask |= S_IROTH;
 308         if (mode & VWRITE)
 309                 mask |= S_IWOTH;
 310         return (file_mode & mask) == mask ? 0 : EACCES;
 311 }
 312
 313 /*
 314  * msdosfs_getattr(struct vnode *a_vp, struct vattr *a_vap)
 315  */
 316 static int
 317 msdosfs_getattr(struct vop_getattr_args *ap)
 318 {
 319         struct denode *dep = VTODE(ap->a_vp);
 320         struct msdosfsmount *pmp = dep->de_pmp;
 321         struct vattr *vap = ap->a_vap;
 322         mode_t mode;
 323         struct timespec ts;
 324         u_long dirsperblk = pmp->pm_BytesPerSec / sizeof(struct direntry);
 325         u_long fileid;
 326
 327         getnanotime(&ts);
 328         DETIMES(dep, &ts, &ts, &ts);
 329         vap->va_fsid = dev2udev(dep->de_dev);
 330         /*
 331          * The following computation of the fileid must be the same as that
 332          * used in msdosfs_readdir() to compute d_fileno. If not, pwd
 333          * doesn't work.
 334          */
 335         if (dep->de_Attributes & ATTR_DIRECTORY) {
 336                 fileid = xcntobn(pmp, dep->de_StartCluster) * dirsperblk;
 337                 if (dep->de_StartCluster == MSDOSFSROOT)
 338                         fileid = 1;
 339         } else {
 340                 fileid = xcntobn(pmp, dep->de_dirclust) * dirsperblk;
 341                 if (dep->de_dirclust == MSDOSFSROOT)
 342                         fileid = roottobn(pmp, 0) * dirsperblk;
 343                 fileid += dep->de_diroffset / sizeof(struct direntry);
 344         }
 345         vap->va_fileid = fileid;
 346         if ((dep->de_Attributes & ATTR_READONLY) == 0)
 347                 mode = S_IRWXU|S_IRWXG|S_IRWXO;
 348         else
 349                 mode = S_IRUSR|S_IXUSR|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH;
 350         vap->va_mode = mode & pmp->pm_mask;
 351         vap->va_uid = pmp->pm_uid;
 352         vap->va_gid = pmp->pm_gid;
 353         vap->va_nlink = 1;
 354         vap->va_rmajor = VNOVAL;
 355         vap->va_rminor = VNOVAL;
 356         vap->va_size = dep->de_FileSize;
 357         dos2unixtime(dep->de_MDate, dep->de_MTime, 0, &vap->va_mtime);
 358         if (pmp->pm_flags & MSDOSFSMNT_LONGNAME) {
 359                 dos2unixtime(dep->de_ADate, 0, 0, &vap->va_atime);
 360                 dos2unixtime(dep->de_CDate, dep->de_CTime, dep->de_CHun, &vap->va_ctime);
 361         } else {
 362                 vap->va_atime = vap->va_mtime;
 363                 vap->va_ctime = vap->va_mtime;
 364         }
 365         vap->va_flags = 0;
 366         if ((dep->de_Attributes & ATTR_ARCHIVE) == 0)
 367                 vap->va_flags |= SF_ARCHIVED;
 368         vap->va_gen = 0;
 369         vap->va_blocksize = pmp->pm_bpcluster;
 370         vap->va_bytes =
 371             (dep->de_FileSize + pmp->pm_crbomask) & ~pmp->pm_crbomask;
 372         vap->va_type = ap->a_vp->v_type;
 373         vap->va_filerev = dep->de_modrev;
 374         return (0);
 375 }
 376
 377 /*
 378  * msdosfs_setattr(struct vnode *a_vp, struct vattr *a_vap,
 379  *                 struct ucred *a_cred)
 380  */
 381 static int
 382 msdosfs_setattr(struct vop_setattr_args *ap)
 383 {
 384         struct vnode *vp = ap->a_vp;
 385         struct denode *dep = VTODE(ap->a_vp);
 386         struct msdosfsmount *pmp = dep->de_pmp;
 387         struct vattr *vap = ap->a_vap;
 388         struct ucred *cred = ap->a_cred;
 389         int error = 0;
 390
 391 #ifdef MSDOSFS_DEBUG
 392         kprintf("msdosfs_setattr(): vp %p, vap %p, cred %p\n",
 393             ap->a_vp, vap, cred);
 394 #endif
 395
 396         /*
 397          * Check for unsettable attributes.
 398          */
 399         if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
 400             (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
 401             (vap->va_blocksize != VNOVAL) || (vap->va_rmajor != VNOVAL) ||
 402             (vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
 403 #ifdef MSDOSFS_DEBUG
 404                 kprintf("msdosfs_setattr(): returning EINVAL\n");
 405                 kprintf("    va_type %d, va_nlink %llx, va_fsid %x, va_fileid %llx\n",
 406                     vap->va_type, vap->va_nlink, vap->va_fsid, vap->va_fileid);
 407                 kprintf("    va_blocksize %lx, va_rmajor %x, va_bytes %qx, va_gen %llx\n",
 408                     vap->va_blocksize, vap->va_rmajor, vap->va_bytes, vap->va_gen);
 409                 kprintf("    va_uid %x, va_gid %x\n",
 410                     vap->va_uid, vap->va_gid);
 411 #endif
 412                 return (EINVAL);
 413         }
 414         if (vap->va_flags != VNOVAL) {
 415                 if (vp->v_mount->mnt_flag & MNT_RDONLY)
 416                         return (EROFS);
 417                 if (cred->cr_uid != pmp->pm_uid &&
 418                     (error = suser_cred(cred, PRISON_ROOT)))
 419                         return (error);
 420                 /*
 421                  * We are very inconsistent about handling unsupported
 422                  * attributes.  We ignored the access time and the
 423                  * read and execute bits.  We were strict for the other
 424                  * attributes.
 425                  *
 426                  * Here we are strict, stricter than ufs in not allowing
 427                  * users to attempt to set SF_SETTABLE bits or anyone to
 428                  * set unsupported bits.  However, we ignore attempts to
 429                  * set ATTR_ARCHIVE for directories `cp -pr' from a more
 430                  * sensible file system attempts it a lot.
 431                  */
 432                 if (cred->cr_uid != 0) {
 433                         if (vap->va_flags & SF_SETTABLE)
 434                                 return EPERM;
 435                 }
 436                 if (vap->va_flags & ~SF_ARCHIVED)
 437                         return EOPNOTSUPP;
 438                 if (vap->va_flags & SF_ARCHIVED)
 439                         dep->de_Attributes &= ~ATTR_ARCHIVE;
 440                 else if (!(dep->de_Attributes & ATTR_DIRECTORY))
 441                         dep->de_Attributes |= ATTR_ARCHIVE;
 442                 dep->de_flag |= DE_MODIFIED;
 443         }
 444
 445         if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
 446                 uid_t uid;
 447                 gid_t gid;
 448
 449                 if (vp->v_mount->mnt_flag & MNT_RDONLY)
 450                         return (EROFS);
 451                 uid = vap->va_uid;
 452                 if (uid == (uid_t)VNOVAL)
 453                         uid = pmp->pm_uid;
 454                 gid = vap->va_gid;
 455                 if (gid == (gid_t)VNOVAL)
 456                         gid = pmp->pm_gid;
 457                 if ((cred->cr_uid != pmp->pm_uid || uid != pmp->pm_uid ||
 458                     (gid != pmp->pm_gid && !groupmember(gid, cred))) &&
 459                     (error = suser_cred(cred, PRISON_ROOT)))
 460                         return error;
 461                 if (uid != pmp->pm_uid || gid != pmp->pm_gid)
 462                         return EINVAL;
 463         }
 464
 465         if (vap->va_size != VNOVAL) {
 466                 /*
 467                  * Disallow write attempts on read-only file systems;
 468                  * unless the file is a socket, fifo, or a block or
 469                  * character device resident on the file system.
 470                  */
 471                 switch (vp->v_type) {
 472                 case VDIR:
 473                         return (EISDIR);
 474                         /* NOT REACHED */
 475                 case VLNK:
 476                 case VREG:
 477                         if (vp->v_mount->mnt_flag & MNT_RDONLY)
 478                                 return (EROFS);
 479                         break;
 480                 default:
 481                         break;
 482                 }
 483                 error = detrunc(dep, vap->va_size, 0);
 484                 if (error)
 485                         return error;
 486         }
 487         if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
 488                 if (vp->v_mount->mnt_flag & MNT_RDONLY)
 489                         return (EROFS);
 490                 if (cred->cr_uid != pmp->pm_uid &&
 491                     (error = suser_cred(cred, PRISON_ROOT)) &&
 492                     ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
 493                     (error = VOP_ACCESS(ap->a_vp, VWRITE, cred))))
 494                         return (error);
 495                 if (vp->v_type != VDIR) {
 496                         if ((pmp->pm_flags & MSDOSFSMNT_NOWIN95) == 0 &&
 497                             vap->va_atime.tv_sec != VNOVAL) {
 498                                 dep->de_flag &= ~DE_ACCESS;
 499                                 unix2dostime(&vap->va_atime, &dep->de_ADate,
 500                                     NULL, NULL);
 501                         }
 502                         if (vap->va_mtime.tv_sec != VNOVAL) {
 503                                 dep->de_flag &= ~DE_UPDATE;
 504                                 unix2dostime(&vap->va_mtime, &dep->de_MDate,
 505                                     &dep->de_MTime, NULL);
 506                         }
 507                         dep->de_Attributes |= ATTR_ARCHIVE;
 508                         dep->de_flag |= DE_MODIFIED;
 509                 }
 510         }
 511         /*
 512          * DOS files only have the ability to have their writability
 513          * attribute set, so we use the owner write bit to set the readonly
 514          * attribute.
 515          */
 516         if (vap->va_mode != (mode_t)VNOVAL) {
 517                 if (vp->v_mount->mnt_flag & MNT_RDONLY)
 518                         return (EROFS);
 519                 if (cred->cr_uid != pmp->pm_uid &&
 520                     (error = suser_cred(cred, PRISON_ROOT)))
 521                         return (error);
 522                 if (vp->v_type != VDIR) {
 523                         /* We ignore the read and execute bits. */
 524                         if (vap->va_mode & VWRITE)
 525                                 dep->de_Attributes &= ~ATTR_READONLY;
 526                         else
 527                                 dep->de_Attributes |= ATTR_READONLY;
 528                         dep->de_Attributes |= ATTR_ARCHIVE;
 529                         dep->de_flag |= DE_MODIFIED;
 530                 }
 531         }
 532         return (deupdat(dep, 1));
 533 }
 534
 535 /*
 536  * msdosfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
 537  *              struct ucred *a_cred)
 538  */
 539 static int
 540 msdosfs_read(struct vop_read_args *ap)
 541 {
 542         int error = 0;
 543         int blsize;
 544         int isadir;
 545         int orig_resid;
 546         u_int n;
 547         u_long diff;
 548         u_long on;
 549         daddr_t lbn;
 550         daddr_t rablock;
 551         off_t raoffset;
 552         off_t loffset;
 553         int rasize;
 554         int seqcount;
 555         struct buf *bp;
 556         struct vnode *vp = ap->a_vp;
 557         struct denode *dep = VTODE(vp);
 558         struct msdosfsmount *pmp = dep->de_pmp;
 559         struct uio *uio = ap->a_uio;
 560
 561         if (uio->uio_offset < 0)
 562                 return (EINVAL);
 563
 564         if ((uoff_t)uio->uio_offset > DOS_FILESIZE_MAX)
 565                 return (0);
 566         /*
 567          * If they didn't ask for any data, then we are done.
 568          */
 569         orig_resid = uio->uio_resid;
 570         if (orig_resid <= 0)
 571                 return (0);
 572
 573         seqcount = ap->a_ioflag >> IO_SEQSHIFT;
 574
 575         isadir = dep->de_Attributes & ATTR_DIRECTORY;
 576         do {
 577                 if (uio->uio_offset >= dep->de_FileSize)
 578                         break;
 579
 580                 /*
 581                  * note: lbn is a cluster number, not a device block number.
 582                  */
 583                 lbn = de_off2cn(pmp, uio->uio_offset);
 584                 loffset = de_cn2doff(pmp, lbn);
 585
 586                 /*
 587                  * If we are operating on a directory file then be sure to
 588                  * do i/o with the vnode for the filesystem instead of the
 589                  * vnode for the directory.
 590                  */
 591                 if (isadir) {
 592                         /*
 593                          * convert cluster # to block #.  lbn is a
 594                          * device block number after this.
 595                          */
 596                         error = pcbmap(dep, lbn, &lbn, NULL, &blsize);
 597                         loffset = de_bntodoff(pmp, lbn);
 598                         if (error == E2BIG) {
 599                                 error = EINVAL;
 600                                 break;
 601                         } else if (error)
 602                                 break;
 603                         error = bread(pmp->pm_devvp, loffset, blsize, &bp);
 604                 } else {
 605                         blsize = pmp->pm_bpcluster;
 606                         rablock = lbn + 1;
 607                         raoffset = de_cn2doff(pmp, rablock);
 608                         if (seqcount > 1 &&
 609                             raoffset < dep->de_FileSize) {
 610                                 rasize = pmp->pm_bpcluster;
 611                                 error = breadn(vp, loffset, blsize,
 612                                                 &raoffset, &rasize, 1, &bp);
 613                         } else {
 614                                 error = bread(vp, loffset, blsize, &bp);
 615                         }
 616                 }
 617                 if (error) {
 618                         brelse(bp);
 619                         break;
 620                 }
 621                 on = uio->uio_offset & pmp->pm_crbomask;
 622                 diff = pmp->pm_bpcluster - on;
 623                 n = diff > uio->uio_resid ? uio->uio_resid : diff;
 624                 diff = dep->de_FileSize - uio->uio_offset;
 625                 if (diff < n)
 626                         n = diff;
 627                 diff = blsize - bp->b_resid;
 628                 if (diff < n)
 629                         n = diff;
 630                 error = uiomove(bp->b_data + on, (int) n, uio);
 631                 brelse(bp);
 632         } while (error == 0 && uio->uio_resid > 0 && n != 0);
 633         if (!isadir && (error == 0 || uio->uio_resid != orig_resid) &&
 634             (vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
 635                 dep->de_flag |= DE_ACCESS;
 636         return (error);
 637 }
 638
 639 /*
 640  * Write data to a file or directory.
 641  *
 642  * msdosfs_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
 643  *               struct ucred *a_cred)
 644  */
 645 static int
 646 msdosfs_write(struct vop_write_args *ap)
 647 {
 648         int n;
 649         int croffset;
 650         int resid;
 651         u_long osize;
 652         int error = 0;
 653         u_long count;
 654         daddr_t cn, lastcn;
 655         struct buf *bp;
 656         int ioflag = ap->a_ioflag;
 657         struct uio *uio = ap->a_uio;
 658         struct thread *td = uio->uio_td;
 659         struct vnode *vp = ap->a_vp;
 660         struct vnode *thisvp;
 661         struct denode *dep = VTODE(vp);
 662         struct msdosfsmount *pmp = dep->de_pmp;
 663         struct proc *p = (td ? td->td_proc : NULL);
 664         struct lwp *lp = (td ? td->td_lwp : NULL);
 665
 666 #ifdef MSDOSFS_DEBUG
 667         kprintf("msdosfs_write(vp %p, uio %p, ioflag %x, cred %p\n",
 668             vp, uio, ioflag, ap->a_cred);
 669         kprintf("msdosfs_write(): diroff %lu, dirclust %lu, startcluster %lu\n",
 670             dep->de_diroffset, dep->de_dirclust, dep->de_StartCluster);
 671 #endif
 672
 673         switch (vp->v_type) {
 674         case VREG:
 675                 if (ioflag & IO_APPEND)
 676                         uio->uio_offset = dep->de_FileSize;
 677                 thisvp = vp;
 678                 break;
 679         case VDIR:
 680                 return EISDIR;
 681         default:
 682                 panic("msdosfs_write(): bad file type");
 683         }
 684
 685         if (uio->uio_offset < 0)
 686                 return (EFBIG);
 687
 688         if (uio->uio_resid == 0)
 689                 return (0);
 690
 691         /*
 692          * If they've exceeded their filesize limit, tell them about it.
 693          */
 694         if (p &&
 695             ((uoff_t)uio->uio_offset + uio->uio_resid >
 696             p->p_rlimit[RLIMIT_FSIZE].rlim_cur)) {
 697                 lwpsignal(p, lp, SIGXFSZ);
 698                 return (EFBIG);
 699         }
 700
 701         if ((uoff_t)uio->uio_offset + uio->uio_resid > DOS_FILESIZE_MAX)
 702                 return (EFBIG);
 703
 704         /*
 705          * If the offset we are starting the write at is beyond the end of
 706          * the file, then they've done a seek.  Unix filesystems allow
 707          * files with holes in them, DOS doesn't so we must fill the hole
 708          * with zeroed blocks.
 709          */
 710         if (uio->uio_offset > dep->de_FileSize) {
 711                 error = deextend(dep, uio->uio_offset);
 712                 if (error)
 713                         return (error);
 714         }
 715
 716         /*
 717          * Remember some values in case the write fails.
 718          */
 719         resid = uio->uio_resid;
 720         osize = dep->de_FileSize;
 721
 722         /*
 723          * If we write beyond the end of the file, extend it to its ultimate
 724          * size ahead of the time to hopefully get a contiguous area.
 725          */
 726         if (uio->uio_offset + resid > osize) {
 727                 count = de_clcount(pmp, uio->uio_offset + resid) -
 728                         de_clcount(pmp, osize);
 729                 error = extendfile(dep, count, NULL, NULL, 0);
 730                 if (error &&  (error != ENOSPC || (ioflag & IO_UNIT)))
 731                         goto errexit;
 732                 lastcn = dep->de_fc[FC_LASTFC].fc_frcn;
 733         } else
 734                 lastcn = de_clcount(pmp, osize) - 1;
 735
 736         do {
 737                 if (de_off2cn(pmp, uio->uio_offset) > lastcn) {
 738                         error = ENOSPC;
 739                         break;
 740                 }
 741
 742                 croffset = uio->uio_offset & pmp->pm_crbomask;
 743                 n = min(uio->uio_resid, pmp->pm_bpcluster - croffset);
 744                 if (uio->uio_offset + n > dep->de_FileSize) {
 745                         dep->de_FileSize = uio->uio_offset + n;
 746                         /* The object size needs to be set before buffer is allocated */
 747                         vnode_pager_setsize(vp, dep->de_FileSize);
 748                 }
 749
 750                 /*
 751                  * If either the whole cluster gets written, or we write
 752                  * the cluster from its start beyond EOF, then no need to
 753                  * read data from disk.
 754                  *
 755                  * If UIO_NOCOPY is set we have to do a read-before-write
 756                  * to fill in any missing pieces of the buffer since no
 757                  * actual overwrite will occur.
 758                  */
 759                 cn = de_off2cn(pmp, uio->uio_offset);
 760                 if ((uio->uio_offset & pmp->pm_crbomask) == 0
 761                     && uio->uio_segflg != UIO_NOCOPY
 762                     && (de_off2cn(pmp, uio->uio_offset + uio->uio_resid)
 763                         > de_off2cn(pmp, uio->uio_offset)
 764                         || uio->uio_offset + uio->uio_resid >= dep->de_FileSize)) {
 765                         bp = getblk(thisvp, de_cn2doff(pmp, cn),
 766                                     pmp->pm_bpcluster, 0, 0);
 767                         clrbuf(bp);
 768                         /*
 769                          * Do the bmap now, since pcbmap needs buffers
 770                          * for the fat table. (see msdosfs_strategy)
 771                          */
 772                         if (bp->b_bio2.bio_offset == NOOFFSET) {
 773                                 daddr_t lblkno = de_off2cn(pmp, bp->b_loffset);
 774                                 daddr_t dblkno;
 775
 776                                 error = pcbmap(dep, lblkno,
 777                                                &dblkno, NULL, NULL);
 778                                 if (error || dblkno == (daddr_t)-1) {
 779                                         bp->b_bio2.bio_offset = NOOFFSET;
 780                                 } else {
 781                                         bp->b_bio2.bio_offset = de_bntodoff(pmp, dblkno);
 782                                 }
 783                         }
 784                         if (bp->b_bio2.bio_offset == NOOFFSET) {
 785                                 brelse(bp);
 786                                 if (!error)
 787                                         error = EIO;            /* XXX */
 788                                 break;
 789                         }
 790                 } else {
 791                         /*
 792                          * The block we need to write into exists, so read
 793                          * it in.
 794                          */
 795                         error = bread(thisvp, de_cn2doff(pmp, cn),
 796                                       pmp->pm_bpcluster, &bp);
 797                         if (error) {
 798                                 brelse(bp);
 799                                 break;
 800                         }
 801                 }
 802
 803                 /*
 804                  * Should these vnode_pager_* functions be done on dir
 805                  * files?
 806                  */
 807
 808                 /*
 809                  * Copy the data from user space into the buf header.
 810                  */
 811                 error = uiomove(bp->b_data + croffset, n, uio);
 812                 if (error) {
 813                         brelse(bp);
 814                         break;
 815                 }
 816
 817                 /*
 818                  * If they want this synchronous then write it and wait for
 819                  * it.  Otherwise, if on a cluster boundary write it
 820                  * asynchronously so we can move on to the next block
 821                  * without delay.  Otherwise do a delayed write because we
 822                  * may want to write somemore into the block later.
 823                  */
 824                 if (ioflag & IO_SYNC)
 825                         bwrite(bp);
 826                 else if (n + croffset == pmp->pm_bpcluster)
 827                         bawrite(bp);
 828                 else
 829                         bdwrite(bp);
 830                 dep->de_flag |= DE_UPDATE;
 831         } while (error == 0 && uio->uio_resid > 0);
 832
 833         /*
 834          * If the write failed and they want us to, truncate the file back
 835          * to the size it was before the write was attempted.
 836          */
 837 errexit:
 838         if (error) {
 839                 if (ioflag & IO_UNIT) {
 840                         detrunc(dep, osize, ioflag & IO_SYNC);
 841                         uio->uio_offset -= resid - uio->uio_resid;
 842                         uio->uio_resid = resid;
 843                 } else {
 844                         detrunc(dep, dep->de_FileSize, ioflag & IO_SYNC);
 845                         if (uio->uio_resid != resid)
 846                                 error = 0;
 847                 }
 848         } else if (ioflag & IO_SYNC)
 849                 error = deupdat(dep, 1);
 850         return (error);
 851 }
 852
 853 /*
 854  * Flush the blocks of a file to disk.
 855  *
 856  * This function is worthless for vnodes that represent directories. Maybe we
 857  * could just do a sync if they try an fsync on a directory file.
 858  *
 859  * msdosfs_fsync(struct vnode *a_vp, int a_waitfor)
 860  */
 861 static int
 862 msdosfs_fsync(struct vop_fsync_args *ap)
 863 {
 864         struct vnode *vp = ap->a_vp;
 865
 866         /*
 867          * Flush all dirty buffers associated with a vnode.
 868          */
 869 #ifdef DIAGNOSTIC
 870 loop:
 871 #endif
 872         vfsync(vp, ap->a_waitfor, 0, NULL, NULL);
 873 #ifdef DIAGNOSTIC
 874         if (ap->a_waitfor == MNT_WAIT && !RB_EMPTY(&vp->v_rbdirty_tree)) {
 875                 vprint("msdosfs_fsync: dirty", vp);
 876                 goto loop;
 877         }
 878 #endif
 879         return (deupdat(VTODE(vp), ap->a_waitfor == MNT_WAIT));
 880 }
 881
 882 /*
 883  * msdosfs_remove(struct vnode *a_dvp, struct vnode *a_vp,
 884  *                struct componentname *a_cnp)
 885  */
 886 static int
 887 msdosfs_remove(struct vop_old_remove_args *ap)
 888 {
 889         struct denode *dep = VTODE(ap->a_vp);
 890         struct denode *ddep = VTODE(ap->a_dvp);
 891         int error;
 892
 893         if (ap->a_vp->v_type == VDIR)
 894                 error = EPERM;
 895         else
 896                 error = removede(ddep, dep);
 897 #ifdef MSDOSFS_DEBUG
 898         kprintf("msdosfs_remove(), dep %p, v_sysrefs %d\n",
 899                 dep, ap->a_vp->v_sysref.refcnt);
 900 #endif
 901         return (error);
 902 }
 903
 904 /*
 905  * DOS filesystems don't know what links are. But since we already called
 906  * msdosfs_lookup() with create and lockparent, the parent is locked so we
 907  * have to free it before we return the error.
 908  *
 909  * msdosfs_link(struct vnode *a_tdvp, struct vnode *a_vp,
 910  *              struct componentname *a_cnp)
 911  */
 912 static int
 913 msdosfs_link(struct vop_old_link_args *ap)
 914 {
 915         return (EOPNOTSUPP);
 916 }
 917
 918 /*
 919  * Renames on files require moving the denode to a new hash queue since the
 920  * denode's location is used to compute which hash queue to put the file
 921  * in. Unless it is a rename in place.  For example "mv a b".
 922  *
 923  * What follows is the basic algorithm:
 924  *
 925  * if (file move) {
 926  *      if (dest file exists) {
 927  *              remove dest file
 928  *      }
 929  *      if (dest and src in same directory) {
 930  *              rewrite name in existing directory slot
 931  *      } else {
 932  *              write new entry in dest directory
 933  *              update offset and dirclust in denode
 934  *              move denode to new hash chain
 935  *              clear old directory entry
 936  *      }
 937  * } else {
 938  *      directory move
 939  *      if (dest directory exists) {
 940  *              if (dest is not empty) {
 941  *                      return ENOTEMPTY
 942  *              }
 943  *              remove dest directory
 944  *      }
 945  *      if (dest and src in same directory) {
 946  *              rewrite name in existing entry
 947  *      } else {
 948  *              be sure dest is not a child of src directory
 949  *              write entry in dest directory
 950  *              update "." and ".." in moved directory
 951  *              clear old directory entry for moved directory
 952  *      }
 953  * }
 954  *
 955  * On entry:
 956  *      source's parent directory is unlocked
 957  *      source file or directory is unlocked
 958  *      destination's parent directory is locked
 959  *      destination file or directory is locked if it exists
 960  *
 961  * On exit:
 962  *      all denodes should be released
 963  *
 964  * Notes:
 965  * I'm not sure how the memory containing the pathnames pointed at by the
 966  * componentname structures is freed, there may be some memory bleeding
 967  * for each rename done.
 968  *
 969  * msdosfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp,
 970  *                struct componentname *a_fcnp, struct vnode *a_tdvp,
 971  *                struct vnode *a_tvp, struct componentname *a_tcnp)
 972  */
 973 static int
 974 msdosfs_rename(struct vop_old_rename_args *ap)
 975 {
 976         struct vnode *tdvp = ap->a_tdvp;
 977         struct vnode *fvp = ap->a_fvp;
 978         struct vnode *fdvp = ap->a_fdvp;
 979         struct vnode *tvp = ap->a_tvp;
 980         struct componentname *tcnp = ap->a_tcnp;
 981         struct componentname *fcnp = ap->a_fcnp;
 982         struct denode *ip, *xp, *dp, *zp;
 983         u_char toname[11], oldname[11];
 984         u_long from_diroffset, to_diroffset;
 985         u_char to_count;
 986         int doingdirectory = 0, newparent = 0;
 987         int error;
 988         u_long cn;
 989         daddr_t bn;
 990         struct denode *fddep;   /* from file's parent directory  */
 991         struct msdosfsmount *pmp;
 992         struct direntry *dotdotp;
 993         struct buf *bp;
 994
 995         fddep = VTODE(ap->a_fdvp);
 996         pmp = fddep->de_pmp;
 997
 998         pmp = VFSTOMSDOSFS(fdvp->v_mount);
 999
1000         /*
1001          * Check for cross-device rename.
1002          */
1003         if ((fvp->v_mount != tdvp->v_mount) ||
1004             (tvp && (fvp->v_mount != tvp->v_mount))) {
1005                 error = EXDEV;
1006 abortit:
1007                 if (tdvp == tvp)
1008                         vrele(tdvp);
1009                 else
1010                         vput(tdvp);
1011                 if (tvp)
1012                         vput(tvp);
1013                 vrele(fdvp);
1014                 vrele(fvp);
1015                 return (error);
1016         }
1017
1018         /*
1019          * If source and dest are the same, do nothing.
1020          */
1021         if (tvp == fvp) {
1022                 error = 0;
1023                 goto abortit;
1024         }
1025
1026         /*
1027          * fvp, fdvp are unlocked, tvp, tdvp are locked.  Lock fvp and note
1028          * that we have to unlock it to use the abortit target.
1029          */
1030         error = vn_lock(fvp, LK_EXCLUSIVE);
1031         if (error)
1032                 goto abortit;
1033         dp = VTODE(fdvp);
1034         ip = VTODE(fvp);
1035
1036         /*
1037          * Be sure we are not renaming ".", "..", or an alias of ".". This
1038          * leads to a crippled directory tree.  It's pretty tough to do a
1039          * "ls" or "pwd" with the "." directory entry missing, and "cd .."
1040          * doesn't work if the ".." entry is missing.
1041          */
1042         if (ip->de_Attributes & ATTR_DIRECTORY) {
1043                 /*
1044                  * Avoid ".", "..", and aliases of "." for obvious reasons.
1045                  */
1046                 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
1047                     dp == ip ||
1048                     (fcnp->cn_flags & CNP_ISDOTDOT) ||
1049                     (tcnp->cn_flags & CNP_ISDOTDOT) ||
1050                     (ip->de_flag & DE_RENAME)) {
1051                         vn_unlock(fvp);
1052                         error = EINVAL;
1053                         goto abortit;
1054                 }
1055                 ip->de_flag |= DE_RENAME;
1056                 doingdirectory++;
1057         }
1058
1059         /*
1060          * fvp locked, fdvp unlocked, tvp, tdvp locked.  DE_RENAME only
1061          * set if doingdirectory.  We will get fvp unlocked in fairly
1062          * short order.  dp and xp must be setup and fvp must be unlocked
1063          * for the out and bad targets to work properly.
1064          */
1065         dp = VTODE(tdvp);
1066         xp = tvp ? VTODE(tvp) : NULL;
1067
1068         /*
1069          * Remember direntry place to use for destination
1070          */
1071         to_diroffset = dp->de_fndoffset;
1072         to_count = dp->de_fndcnt;
1073
1074         /*
1075          * If ".." must be changed (ie the directory gets a new
1076          * parent) then the source directory must not be in the
1077          * directory heirarchy above the target, as this would
1078          * orphan everything below the source directory. Also
1079          * the user must have write permission in the source so
1080          * as to be able to change "..". We must repeat the call
1081          * to namei, as the parent directory is unlocked by the
1082          * call to doscheckpath().
1083          */
1084         error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred);
1085         vn_unlock(fvp);
1086         if (VTODE(fdvp)->de_StartCluster != VTODE(tdvp)->de_StartCluster)
1087                 newparent = 1;
1088
1089         /*
1090          * ok. fvp, fdvp unlocked, tvp, tdvp locked. tvp may be NULL.
1091          * DE_RENAME only set if doingdirectory.
1092          */
1093         if (doingdirectory && newparent) {
1094                 if (error)      /* write access check above */
1095                         goto bad;
1096                 if (xp != NULL) {
1097                         vput(tvp);
1098                         xp = NULL;
1099                 }
1100                 /*
1101                  * checkpath vput's tdvp (VTOI(dp)) on return no matter what,
1102                  * get an extra ref so we wind up with just an unlocked, ref'd
1103                  * tdvp.  The 'out' target skips tvp and tdvp cleanups (tdvp
1104                  * isn't locked so we can't use the out target).
1105                  */
1106                 vref(tdvp);
1107                 error = doscheckpath(ip, dp);
1108                 tcnp->cn_flags |= CNP_PDIRUNLOCK;
1109                 if (error) {
1110                         vrele(tdvp);
1111                         goto out;
1112                 }
1113                 /*
1114                  * relookup no longer messes with the ref count.  tdvp must
1115                  * be unlocked on entry and on success will be locked on
1116                  * return.
1117                  */
1118                 error = relookup(tdvp, &tvp, tcnp);
1119                 if (error) {
1120                         if (tcnp->cn_flags & CNP_PDIRUNLOCK)
1121                                 vrele(tdvp);
1122                         else
1123                                 vput(tdvp);
1124                         goto out;
1125                 }
1126
1127                 /*
1128                  * tvp and tdvp are now locked again.
1129                  */
1130                 dp = VTODE(tdvp);
1131                 xp = tvp ? VTODE(tvp) : NULL;
1132         }
1133
1134         /*
1135          * tvp and tdvp are now locked again, the 'bad' target can be used
1136          * to clean them up again.  Delete an existant target and clean
1137          * up tvp.  Set xp to NULL to indicate that tvp has been cleaned up.
1138          */
1139         if (xp != NULL) {
1140                 /*
1141                  * Target must be empty if a directory and have no links
1142                  * to it. Also, ensure source and target are compatible
1143                  * (both directories, or both not directories).
1144                  */
1145                 if (xp->de_Attributes & ATTR_DIRECTORY) {
1146                         if (!dosdirempty(xp)) {
1147                                 error = ENOTEMPTY;
1148                                 goto bad;
1149                         }
1150                         if (!doingdirectory) {
1151                                 error = ENOTDIR;
1152                                 goto bad;
1153                         }
1154                 } else if (doingdirectory) {
1155                         error = EISDIR;
1156                         goto bad;
1157                 }
1158                 error = removede(dp, xp);
1159                 if (error)
1160                         goto bad;
1161                 vput(tvp);
1162                 xp = NULL;
1163                 tvp = NULL;
1164         }
1165
1166         /*
1167          * Convert the filename in tcnp into a dos filename. We copy this
1168          * into the denode and directory entry for the destination
1169          * file/directory.
1170          */
1171         error = uniqdosname(VTODE(tdvp), tcnp, toname);
1172         if (error)
1173                 goto bad;
1174
1175         /*
1176          * Since from wasn't locked at various places above, we have to do
1177          * a relookup here.  If the target and source are the same directory
1178          * we have to unlock the target directory in order to safely relookup
1179          * the source, because relookup expects its directory to be unlocked.
1180          *
1181          * Note that ap->a_fvp is still valid and ref'd.  Any cleanup must
1182          * now take that into account.
1183          *
1184          * The tdvp locking issues make this a real mess.
1185          */
1186         fcnp->cn_flags &= ~CNP_MODMASK;
1187         fcnp->cn_flags |= CNP_LOCKPARENT;
1188         if (newparent == 0)
1189                 vn_unlock(tdvp);
1190         error = relookup(fdvp, &fvp, fcnp);
1191         if (error || fvp == NULL) {
1192                 /*
1193                  * From name has disappeared.  Note: fdvp might == tdvp.
1194                  *
1195                  * DE_RENAME is only set if doingdirectory.
1196                  */
1197                 if (doingdirectory)
1198                         panic("rename: lost dir entry");
1199                 if (fcnp->cn_flags & CNP_PDIRUNLOCK)
1200                         vrele(fdvp);
1201                 else
1202                         vput(fdvp);
1203                 if (newparent == 0)
1204                         vrele(tdvp);
1205                 else
1206                         vput(tdvp);
1207                 vrele(ap->a_fvp);
1208                 return(0);
1209         }
1210
1211         /*
1212          * No error occured.  tdvp, fdvp and fvp are all locked.  If
1213          * newparent was 0 be aware that fdvp == tdvp.  tvp has been cleaned
1214          * up.  ap->a_fvp is still refd.
1215          */
1216         xp = VTODE(fvp);
1217         zp = VTODE(fdvp);
1218         from_diroffset = zp->de_fndoffset;
1219
1220         /*
1221          * Ensure that the directory entry still exists and has not
1222          * changed till now. If the source is a file the entry may
1223          * have been unlinked or renamed. In either case there is
1224          * no further work to be done. If the source is a directory
1225          * then it cannot have been rmdir'ed or renamed; this is
1226          * prohibited by the DE_RENAME flag.
1227          *
1228          * DE_RENAME is only set if doingdirectory.
1229          */
1230         if (xp != ip) {
1231                 if (doingdirectory)
1232                         panic("rename: lost dir entry");
1233                 goto done;
1234         } else {
1235                 u_long new_dirclust;
1236                 u_long new_diroffset;
1237
1238                 /*
1239                  * First write a new entry in the destination
1240                  * directory and mark the entry in the source directory
1241                  * as deleted.  Then move the denode to the correct hash
1242                  * chain for its new location in the filesystem.  And, if
1243                  * we moved a directory, then update its .. entry to point
1244                  * to the new parent directory.
1245                  */
1246                 bcopy(ip->de_Name, oldname, 11);
1247                 bcopy(toname, ip->de_Name, 11); /* update denode */
1248                 dp->de_fndoffset = to_diroffset;
1249                 dp->de_fndcnt = to_count;
1250                 error = createde(ip, dp, (struct denode **)0, tcnp);
1251                 if (error) {
1252                         bcopy(oldname, ip->de_Name, 11);
1253                         goto done;
1254                 }
1255                 ip->de_refcnt++;
1256                 zp->de_fndoffset = from_diroffset;
1257                 error = removede(zp, ip);
1258                 if (error) {
1259                         /* XXX should really panic here, fs is corrupt */
1260                         goto done;
1261                 }
1262                 if (!doingdirectory) {
1263                         error = pcbmap(dp, de_cluster(pmp, to_diroffset),
1264                                        NULL, &new_dirclust, NULL);
1265                         if (error) {
1266                                 /* XXX should really panic here, fs is corrupt */
1267                                 goto done;
1268                         }
1269                         if (new_dirclust == MSDOSFSROOT)
1270                                 new_diroffset = to_diroffset;
1271                         else
1272                                 new_diroffset = to_diroffset & pmp->pm_crbomask;
1273                         msdosfs_reinsert(ip, new_dirclust, new_diroffset);
1274                 }
1275         }
1276
1277         /*
1278          * If we moved a directory to a new parent directory, then we must
1279          * fixup the ".." entry in the moved directory.
1280          */
1281         if (doingdirectory && newparent) {
1282                 cn = ip->de_StartCluster;
1283                 if (cn == MSDOSFSROOT) {
1284                         /* this should never happen */
1285                         panic("msdosfs_rename(): updating .. in root directory?");
1286                 } else {
1287                         bn = xcntobn(pmp, cn);
1288                 }
1289                 error = bread(pmp->pm_devvp, de_bntodoff(pmp, bn), pmp->pm_bpcluster, &bp);
1290                 if (error) {
1291                         /* XXX should really panic here, fs is corrupt */
1292                         brelse(bp);
1293                         goto done;
1294                 }
1295                 dotdotp = (struct direntry *)bp->b_data + 1;
1296                 putushort(dotdotp->deStartCluster, dp->de_StartCluster);
1297                 if (FAT32(pmp))
1298                         putushort(dotdotp->deHighClust, dp->de_StartCluster >> 16);
1299                 error = bwrite(bp);
1300                 if (error) {
1301                         /* XXX should really panic here, fs is corrupt */
1302                         goto done;
1303                 }
1304         }
1305
1306         /*
1307          * done case fvp, fdvp, tdvp are locked.  ap->a_fvp is refd
1308          */
1309 done:
1310         if (doingdirectory)
1311                 ip->de_flag &= ~DE_RENAME;      /* XXX fvp not locked */
1312         vput(fvp);
1313         if (newparent)
1314                 vput(fdvp);
1315         else
1316                 vrele(fdvp);
1317         vput(tdvp);
1318         vrele(ap->a_fvp);
1319         return (error);
1320
1321         /*
1322          * 'bad' target: xp governs tvp.  tvp and tdvp arel ocked, fdvp and fvp
1323          * are not locked.  ip points to fvp's inode which may have DE_RENAME
1324          * set.
1325          */
1326 bad:
1327         if (xp)
1328                 vput(tvp);
1329         vput(tdvp);
1330 out:
1331         /*
1332          * 'out' target: tvp and tdvp have already been cleaned up.
1333          */
1334         if (doingdirectory)
1335                 ip->de_flag &= ~DE_RENAME;
1336         vrele(fdvp);
1337         vrele(fvp);
1338         return (error);
1339
1340 }
1341
1342 static struct {
1343         struct direntry dot;
1344         struct direntry dotdot;
1345 } dosdirtemplate = {
1346         {       ".       ", "   ",                      /* the . entry */
1347                 ATTR_DIRECTORY,                         /* file attribute */
1348                 0,                                      /* reserved */
1349                 0, { 0, 0 }, { 0, 0 },                  /* create time & date */
1350                 { 0, 0 },                               /* access date */
1351                 { 0, 0 },                               /* high bits of start cluster */
1352                 { 210, 4 }, { 210, 4 },                 /* modify time & date */
1353                 { 0, 0 },                               /* startcluster */
1354                 { 0, 0, 0, 0 }                          /* filesize */
1355         },
1356         {       "..      ", "   ",                      /* the .. entry */
1357                 ATTR_DIRECTORY,                         /* file attribute */
1358                 0,                                      /* reserved */
1359                 0, { 0, 0 }, { 0, 0 },                  /* create time & date */
1360                 { 0, 0 },                               /* access date */
1361                 { 0, 0 },                               /* high bits of start cluster */
1362                 { 210, 4 }, { 210, 4 },                 /* modify time & date */
1363                 { 0, 0 },                               /* startcluster */
1364                 { 0, 0, 0, 0 }                          /* filesize */
1365         }
1366 };
1367
1368 /*
1369  * msdosfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp,
1370  *               struct componentname *a_cnp, struct vattr *a_vap)
1371  */
1372 static int
1373 msdosfs_mkdir(struct vop_old_mkdir_args *ap)
1374 {
1375         struct componentname *cnp = ap->a_cnp;
1376         struct denode *dep;
1377         struct denode *pdep = VTODE(ap->a_dvp);
1378         struct direntry *denp;
1379         struct msdosfsmount *pmp = pdep->de_pmp;
1380         struct buf *bp;
1381         u_long newcluster, pcl;
1382         int bn;
1383         int error;
1384         struct denode ndirent;
1385         struct timespec ts;
1386
1387         /*
1388          * If this is the root directory and there is no space left we
1389          * can't do anything.  This is because the root directory can not
1390          * change size.
1391          */
1392         if (pdep->de_StartCluster == MSDOSFSROOT
1393             && pdep->de_fndoffset >= pdep->de_FileSize) {
1394                 error = ENOSPC;
1395                 goto bad2;
1396         }
1397
1398         /*
1399          * Allocate a cluster to hold the about to be created directory.
1400          */
1401         error = clusteralloc(pmp, 0, 1, CLUST_EOFE, &newcluster, NULL);
1402         if (error)
1403                 goto bad2;
1404
1405         bzero(&ndirent, sizeof(ndirent));
1406         ndirent.de_pmp = pmp;
1407         ndirent.de_flag = DE_ACCESS | DE_CREATE | DE_UPDATE;
1408         getnanotime(&ts);
1409         DETIMES(&ndirent, &ts, &ts, &ts);
1410
1411         /*
1412          * Now fill the cluster with the "." and ".." entries. And write
1413          * the cluster to disk.  This way it is there for the parent
1414          * directory to be pointing at if there were a crash.
1415          */
1416         bn = xcntobn(pmp, newcluster);
1417         /* always succeeds */
1418         bp = getblk(pmp->pm_devvp, de_bntodoff(pmp, bn),
1419                     pmp->pm_bpcluster, 0, 0);
1420         bzero(bp->b_data, pmp->pm_bpcluster);
1421         bcopy(&dosdirtemplate, bp->b_data, sizeof dosdirtemplate);
1422         denp = (struct direntry *)bp->b_data;
1423         putushort(denp[0].deStartCluster, newcluster);
1424         putushort(denp[0].deCDate, ndirent.de_CDate);
1425         putushort(denp[0].deCTime, ndirent.de_CTime);
1426         denp[0].deCHundredth = ndirent.de_CHun;
1427         putushort(denp[0].deADate, ndirent.de_ADate);
1428         putushort(denp[0].deMDate, ndirent.de_MDate);
1429         putushort(denp[0].deMTime, ndirent.de_MTime);
1430         pcl = pdep->de_StartCluster;
1431         if (FAT32(pmp) && pcl == pmp->pm_rootdirblk)
1432                 pcl = 0;
1433         putushort(denp[1].deStartCluster, pcl);
1434         putushort(denp[1].deCDate, ndirent.de_CDate);
1435         putushort(denp[1].deCTime, ndirent.de_CTime);
1436         denp[1].deCHundredth = ndirent.de_CHun;
1437         putushort(denp[1].deADate, ndirent.de_ADate);
1438         putushort(denp[1].deMDate, ndirent.de_MDate);
1439         putushort(denp[1].deMTime, ndirent.de_MTime);
1440         if (FAT32(pmp)) {
1441                 putushort(denp[0].deHighClust, newcluster >> 16);
1442                 putushort(denp[1].deHighClust, pdep->de_StartCluster >> 16);
1443         }
1444
1445         error = bwrite(bp);
1446         if (error)
1447                 goto bad;
1448
1449         /*
1450          * Now build up a directory entry pointing to the newly allocated
1451          * cluster.  This will be written to an empty slot in the parent
1452          * directory.
1453          */
1454         error = uniqdosname(pdep, cnp, ndirent.de_Name);
1455         if (error)
1456                 goto bad;
1457
1458         ndirent.de_Attributes = ATTR_DIRECTORY;
1459         ndirent.de_LowerCase = 0;
1460         ndirent.de_StartCluster = newcluster;
1461         ndirent.de_FileSize = 0;
1462         ndirent.de_dev = pdep->de_dev;
1463         ndirent.de_devvp = pdep->de_devvp;
1464         error = createde(&ndirent, pdep, &dep, cnp);
1465         if (error)
1466                 goto bad;
1467         *ap->a_vpp = DETOV(dep);
1468         return (0);
1469
1470 bad:
1471         clusterfree(pmp, newcluster, NULL);
1472 bad2:
1473         return (error);
1474 }
1475
1476 /*
1477  * msdosfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp,
1478  *               struct componentname *a_cnp)
1479  */
1480 static int
1481 msdosfs_rmdir(struct vop_old_rmdir_args *ap)
1482 {
1483         struct vnode *vp = ap->a_vp;
1484         struct vnode *dvp = ap->a_dvp;
1485         struct denode *ip, *dp;
1486         int error;
1487
1488         ip = VTODE(vp);
1489         dp = VTODE(dvp);
1490
1491         /*
1492          * Verify the directory is empty (and valid).
1493          * (Rmdir ".." won't be valid since
1494          *  ".." will contain a reference to
1495          *  the current directory and thus be
1496          *  non-empty.)
1497          */
1498         error = 0;
1499         if (!dosdirempty(ip) || ip->de_flag & DE_RENAME) {
1500                 error = ENOTEMPTY;
1501                 goto out;
1502         }
1503         /*
1504          * Delete the entry from the directory.  For dos filesystems this
1505          * gets rid of the directory entry on disk, the in memory copy
1506          * still exists but the de_refcnt is <= 0.  This prevents it from
1507          * being found by deget().  When the vput() on dep is done we give
1508          * up access and eventually msdosfs_reclaim() will be called which
1509          * will remove it from the denode cache.
1510          */
1511         error = removede(dp, ip);
1512         if (error)
1513                 goto out;
1514         /*
1515          * This is where we decrement the link count in the parent
1516          * directory.  Since dos filesystems don't do this we just purge
1517          * the name cache.
1518          */
1519         vn_unlock(dvp);
1520         /*
1521          * Truncate the directory that is being deleted.
1522          */
1523         error = detrunc(ip, (u_long)0, IO_SYNC);
1524
1525         vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1526 out:
1527         return (error);
1528 }
1529
1530 /*
1531  * DOS filesystems don't know what symlinks are.
1532  *
1533  * msdosfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp,
1534  *                 struct componentname *a_cnp, struct vattr *a_vap,
1535  *                 char *a_target)
1536  */
1537 static int
1538 msdosfs_symlink(struct vop_old_symlink_args *ap)
1539 {
1540         return (EOPNOTSUPP);
1541 }
1542
1543 /*
1544  * msdosfs_readdir(struct vnode *a_vp, struct uio *a_uio,
1545  *                 struct ucred *a_cred, int *a_eofflag, int *a_ncookies,
1546  *                 off_t **a_cookies)
1547  */
1548 static int
1549 msdosfs_readdir(struct vop_readdir_args *ap)
1550 {
1551         int error = 0;
1552         int diff;
1553         long n;
1554         int blsize;
1555         long on;
1556         u_long cn;
1557         u_long dirsperblk;
1558         long bias = 0;
1559         daddr_t bn, lbn;
1560         struct buf *bp;
1561         struct denode *dep;
1562         struct msdosfsmount *pmp;
1563         struct direntry *dentp;
1564         struct uio *uio = ap->a_uio;
1565         off_t *cookies = NULL;
1566         int ncookies = 0;
1567         off_t offset, off;
1568         int chksum = -1;
1569         ino_t d_ino;
1570         uint16_t d_namlen;
1571         uint8_t d_type;
1572         char *d_name_storage = NULL;
1573         char *d_name;
1574
1575         if ((error = vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
1576                 return (error);
1577
1578         dep = VTODE(ap->a_vp);
1579         pmp = dep->de_pmp;
1580
1581 #ifdef MSDOSFS_DEBUG
1582         kprintf("msdosfs_readdir(): vp %p, uio %p, cred %p, eofflagp %p\n",
1583             ap->a_vp, uio, ap->a_cred, ap->a_eofflag);
1584 #endif
1585
1586         /*
1587          * msdosfs_readdir() won't operate properly on regular files since
1588          * it does i/o only with the the filesystem vnode, and hence can
1589          * retrieve the wrong block from the buffer cache for a plain file.
1590          * So, fail attempts to readdir() on a plain file.
1591          */
1592         if ((dep->de_Attributes & ATTR_DIRECTORY) == 0) {
1593                 error = ENOTDIR;
1594                 goto done;
1595         }
1596
1597         /*
1598          * If the user buffer is smaller than the size of one dos directory
1599          * entry or the file offset is not a multiple of the size of a
1600          * directory entry, then we fail the read.
1601          */
1602         off = offset = uio->uio_offset;
1603         if (uio->uio_resid < sizeof(struct direntry) ||
1604             (offset & (sizeof(struct direntry) - 1))) {
1605                 error = EINVAL;
1606                 goto done;
1607         }
1608
1609         if (ap->a_ncookies) {
1610                 ncookies = uio->uio_resid / 16 + 1;
1611                 if (ncookies > 1024)
1612                         ncookies = 1024;
1613                 MALLOC(cookies, off_t *, ncookies * sizeof(off_t), M_TEMP,
1614                        M_WAITOK);
1615                 *ap->a_cookies = cookies;
1616                 *ap->a_ncookies = ncookies;
1617         }
1618
1619         dirsperblk = pmp->pm_BytesPerSec / sizeof(struct direntry);
1620
1621         /*
1622          * If they are reading from the root directory then, we simulate
1623          * the . and .. entries since these don't exist in the root
1624          * directory.  We also set the offset bias to make up for having to
1625          * simulate these entries. By this I mean that at file offset 64 we
1626          * read the first entry in the root directory that lives on disk.
1627          */
1628         if (dep->de_StartCluster == MSDOSFSROOT
1629             || (FAT32(pmp) && dep->de_StartCluster == pmp->pm_rootdirblk)) {
1630 #if 0
1631                 kprintf("msdosfs_readdir(): going after . or .. in root dir, offset %d\n",
1632                     offset);
1633 #endif
1634                 bias = 2 * sizeof(struct direntry);
1635                 if (offset < bias) {
1636                         for (n = (int)offset / sizeof(struct direntry); n < 2;
1637                              n++) {
1638                                 if (FAT32(pmp))
1639                                         d_ino = xcntobn(pmp, pmp->pm_rootdirblk)
1640                                             * dirsperblk;
1641                                 else
1642                                         d_ino = 1;
1643                                 d_type = DT_DIR;
1644                                 if (n == 0) {
1645                                         d_namlen = 1;
1646                                         d_name = ".";
1647                                 } else /* if (n == 1) */{
1648                                         d_namlen = 2;
1649                                         d_name = "..";
1650                                 }
1651                                 if (vop_write_dirent(&error, uio, d_ino, d_type,
1652                                     d_namlen, d_name))
1653                                         goto out;
1654                                 if (error)
1655                                         goto out;
1656                                 offset += sizeof(struct direntry);
1657                                 off = offset;
1658                                 if (cookies) {
1659                                         *cookies++ = offset;
1660                                         if (--ncookies <= 0)
1661                                                 goto out;
1662                                 }
1663                         }
1664                 }
1665         }
1666
1667         d_name_storage = kmalloc(WIN_MAXLEN, M_TEMP, M_WAITOK);
1668         off = offset;
1669
1670         while (uio->uio_resid > 0) {
1671                 lbn = de_off2cn(pmp, offset - bias);
1672                 on = (offset - bias) & pmp->pm_crbomask;
1673                 n = min(pmp->pm_bpcluster - on, uio->uio_resid);
1674                 diff = dep->de_FileSize - (offset - bias);
1675                 if (diff <= 0)
1676                         break;
1677                 n = min(n, diff);
1678                 error = pcbmap(dep, lbn, &bn, &cn, &blsize);
1679                 if (error)
1680                         break;
1681                 error = bread(pmp->pm_devvp, de_bntodoff(pmp, bn), blsize, &bp);
1682                 if (error) {
1683                         brelse(bp);
1684                         kfree(d_name_storage, M_TEMP);
1685                         goto done;
1686                 }
1687                 n = min(n, blsize - bp->b_resid);
1688
1689                 /*
1690                  * Convert from dos directory entries to fs-independent
1691                  * directory entries.
1692                  */
1693                 for (dentp = (struct direntry *)(bp->b_data + on);
1694                      (char *)dentp < bp->b_data + on + n;
1695                      dentp++, offset += sizeof(struct direntry)) {
1696 #if 0
1697                         kprintf("rd: dentp %08x prev %08x crnt %08x deName %02x attr %02x\n",
1698                             dentp, prev, crnt, dentp->deName[0], dentp->deAttributes);
1699 #endif
1700                         d_name = d_name_storage;
1701                         /*
1702                          * If this is an unused entry, we can stop.
1703                          */
1704                         if (dentp->deName[0] == SLOT_EMPTY) {
1705                                 brelse(bp);
1706                                 goto out;
1707                         }
1708                         /*
1709                          * Skip deleted entries.
1710                          */
1711                         if (dentp->deName[0] == SLOT_DELETED) {
1712                                 chksum = -1;
1713                                 continue;
1714                         }
1715
1716                         /*
1717                          * Handle Win95 long directory entries
1718                          */
1719                         if (dentp->deAttributes == ATTR_WIN95) {
1720                                 if (pmp->pm_flags & MSDOSFSMNT_SHORTNAME)
1721                                         continue;
1722                                 chksum = win2unixfn((struct winentry *)dentp,
1723                                         d_name, &d_namlen, chksum,
1724                                         pmp->pm_flags & MSDOSFSMNT_U2WTABLE,
1725                                         pmp->pm_u2w);
1726                                 continue;
1727                         }
1728
1729                         /*
1730                          * Skip volume labels
1731                          */
1732                         if (dentp->deAttributes & ATTR_VOLUME) {
1733                                 chksum = -1;
1734                                 continue;
1735                         }
1736                         /*
1737                          * This computation of d_ino must match
1738                          * the computation of va_fileid in
1739                          * msdosfs_getattr.
1740                          */
1741                         if (dentp->deAttributes & ATTR_DIRECTORY) {
1742                                 d_ino = getushort(dentp->deStartCluster);
1743                                 if (FAT32(pmp))
1744                                         d_ino |= getushort(dentp->deHighClust) << 16;
1745                                 /* if this is the root directory */
1746                                 if (d_ino != MSDOSFSROOT)
1747                                         d_ino = xcntobn(pmp, d_ino) * dirsperblk;
1748                                 else if (FAT32(pmp))
1749                                         d_ino = xcntobn(pmp, pmp->pm_rootdirblk)
1750                                             * dirsperblk;
1751                                 else
1752                                         d_ino = 1;
1753                                 d_type = DT_DIR;
1754                         } else {
1755                                 d_ino = offset / sizeof(struct direntry);
1756                                 d_type = DT_REG;
1757                         }
1758                         if (chksum != winChksum(dentp->deName)) {
1759                                 d_namlen = dos2unixfn(dentp->deName,
1760                                     (u_char *)d_name,
1761                                     dentp->deLowerCase |
1762                                         ((pmp->pm_flags & MSDOSFSMNT_SHORTNAME) ?
1763                                         (LCASE_BASE | LCASE_EXT) : 0),
1764                                     pmp->pm_flags & MSDOSFSMNT_U2WTABLE,
1765                                     pmp->pm_d2u,
1766                                     pmp->pm_flags & MSDOSFSMNT_ULTABLE,
1767                                     pmp->pm_ul);
1768                         }
1769                         chksum = -1;
1770                         if (vop_write_dirent(&error, uio, d_ino, d_type,
1771                             d_namlen, d_name)) {
1772                                 brelse(bp);
1773                                 goto out;
1774                         }
1775                         if (error) {
1776                                 brelse(bp);
1777                                 goto out;
1778                         }
1779
1780                         if (cookies) {
1781                                 *cookies++ = offset + sizeof(struct direntry);
1782                                 if (--ncookies <= 0) {
1783                                         brelse(bp);
1784                                         goto out;
1785                                 }
1786                         }
1787                         off = offset + sizeof(struct direntry);
1788                 }
1789                 brelse(bp);
1790         }
1791 out:
1792         if (d_name_storage != NULL)
1793                 kfree(d_name_storage, M_TEMP);
1794
1795         /* Subtract unused cookies */
1796         if (ap->a_ncookies)
1797                 *ap->a_ncookies -= ncookies;
1798
1799         uio->uio_offset = off;
1800
1801         /*
1802          * Set the eofflag (NFS uses it)
1803          */
1804         if (ap->a_eofflag) {
1805                 if (dep->de_FileSize - (offset - bias) <= 0)
1806                         *ap->a_eofflag = 1;
1807                 else
1808                         *ap->a_eofflag = 0;
1809         }
1810 done:
1811         vn_unlock(ap->a_vp);
1812         return (error);
1813 }
1814
1815 /*
1816  * vp  - address of vnode file the file
1817  * bn  - which cluster we are interested in mapping to a filesystem block number.
1818  * vpp - returns the vnode for the block special file holding the filesystem
1819  *       containing the file of interest
1820  * bnp - address of where to return the filesystem relative block number
1821  *
1822  * msdosfs_bmap(struct vnode *a_vp, off_t a_loffset,
1823  *              off_t *a_doffsetp, int *a_runp, int *a_runb)
1824  */
1825 static int
1826 msdosfs_bmap(struct vop_bmap_args *ap)
1827 {
1828         struct denode *dep = VTODE(ap->a_vp);
1829         struct msdosfsmount *pmp = dep->de_pmp;
1830         daddr_t lbn;
1831         daddr_t dbn;
1832         int error;
1833
1834         if (ap->a_doffsetp == NULL)
1835                 return (0);
1836         if (ap->a_runp) {
1837                 /*
1838                  * Sequential clusters should be counted here.
1839                  */
1840                 *ap->a_runp = 0;
1841         }
1842         if (ap->a_runb) {
1843                 *ap->a_runb = 0;
1844         }
1845         KKASSERT(((int)ap->a_loffset & ((1 << pmp->pm_cnshift) - 1)) == 0);
1846         lbn = de_off2cn(pmp, ap->a_loffset);
1847         error = pcbmap(dep, lbn, &dbn, NULL, NULL);
1848         if (error || dbn == (daddr_t)-1) {
1849                 *ap->a_doffsetp = NOOFFSET;
1850         } else {
1851                 *ap->a_doffsetp = de_bntodoff(pmp, dbn);
1852         }
1853         return (error);
1854 }
1855
1856 /*
1857  * msdosfs_strategy(struct vnode *a_vp, struct bio *a_bio)
1858  */
1859 static int
1860 msdosfs_strategy(struct vop_strategy_args *ap)
1861 {
1862         struct bio *bio = ap->a_bio;
1863         struct bio *nbio;
1864         struct buf *bp = bio->bio_buf;
1865         struct vnode *vp = ap->a_vp;
1866         struct denode *dep = VTODE(vp);
1867         struct msdosfsmount *pmp = dep->de_pmp;
1868         int error = 0;
1869         daddr_t dblkno;
1870
1871         if (vp->v_type == VBLK || vp->v_type == VCHR)
1872                 panic("msdosfs_strategy: spec");
1873         /*
1874          * If we don't already know the filesystem relative block number
1875          * then get it using pcbmap().  If pcbmap() returns the block
1876          * number as -1 then we've got a hole in the file.  DOS filesystems
1877          * don't allow files with holes, so we shouldn't ever see this.
1878          */
1879         nbio = push_bio(bio);
1880         if (nbio->bio_offset == NOOFFSET) {
1881                 error = pcbmap(dep, de_off2cn(pmp, bio->bio_offset),
1882                                &dblkno, NULL, NULL);
1883                 if (error) {
1884                         bp->b_error = error;
1885                         bp->b_flags |= B_ERROR;
1886                         /* I/O was never started on nbio, must biodone(bio) */
1887                         biodone(bio);
1888                         return (error);
1889                 }
1890                 if (dblkno == (daddr_t)-1) {
1891                         nbio->bio_offset = NOOFFSET;
1892                         vfs_bio_clrbuf(bp);
1893                 } else {
1894                         nbio->bio_offset = de_bntodoff(pmp, dblkno);
1895                 }
1896         }
1897         if (nbio->bio_offset == NOOFFSET) {
1898                 /* I/O was never started on nbio, must biodone(bio) */
1899                 biodone(bio);
1900                 return (0);
1901         }
1902         /*
1903          * Read/write the block from/to the disk that contains the desired
1904          * file block.
1905          */
1906         vn_strategy(dep->de_devvp, nbio);
1907         return (0);
1908 }
1909
1910 /*
1911  * msdosfs_print(struct vnode *vp)
1912  */
1913 static int
1914 msdosfs_print(struct vop_print_args *ap)
1915 {
1916         struct denode *dep = VTODE(ap->a_vp);
1917
1918         kprintf(
1919             "tag VT_MSDOSFS, startcluster %lu, dircluster %lu, diroffset %lu ",
1920                dep->de_StartCluster, dep->de_dirclust, dep->de_diroffset);
1921         kprintf(" dev %d, %d", major(dep->de_dev), minor(dep->de_dev));
1922         lockmgr_printinfo(&ap->a_vp->v_lock);
1923         kprintf("\n");
1924         return (0);
1925 }
1926
1927 /*
1928  * msdosfs_pathconf(struct vnode *a_vp, int a_name, int *a_retval)
1929  */
1930 static int
1931 msdosfs_pathconf(struct vop_pathconf_args *ap)
1932 {
1933         struct msdosfsmount *pmp = VTODE(ap->a_vp)->de_pmp;
1934
1935         switch (ap->a_name) {
1936         case _PC_LINK_MAX:
1937                 *ap->a_retval = 1;
1938                 return (0);
1939         case _PC_NAME_MAX:
1940                 *ap->a_retval = pmp->pm_flags & MSDOSFSMNT_LONGNAME ? WIN_MAXLEN : 12;
1941                 return (0);
1942         case _PC_PATH_MAX:
1943                 *ap->a_retval = PATH_MAX;
1944                 return (0);
1945         case _PC_CHOWN_RESTRICTED:
1946                 *ap->a_retval = 1;
1947                 return (0);
1948         case _PC_NO_TRUNC:
1949                 *ap->a_retval = 0;
1950                 return (0);
1951         default:
1952                 return (EINVAL);
1953         }
1954         /* NOTREACHED */
1955 }
1956
1957 /* Global vfs data structures for msdosfs */
1958 struct vop_ops msdosfs_vnode_vops = {
1959         .vop_default =          vop_defaultop,
1960         .vop_access =           msdosfs_access,
1961         .vop_bmap =             msdosfs_bmap,
1962         .vop_old_lookup =       msdosfs_lookup,
1963         .vop_open =             msdosfs_open,
1964         .vop_close =            msdosfs_close,
1965         .vop_old_create =       msdosfs_create,
1966         .vop_fsync =            msdosfs_fsync,
1967         .vop_getattr =          msdosfs_getattr,
1968         .vop_inactive =         msdosfs_inactive,
1969         .vop_old_link =         msdosfs_link,
1970         .vop_old_mkdir =        msdosfs_mkdir,
1971         .vop_old_mknod =        msdosfs_mknod,
1972         .vop_pathconf =         msdosfs_pathconf,
1973         .vop_print =            msdosfs_print,
1974         .vop_read =             msdosfs_read,
1975         .vop_readdir =          msdosfs_readdir,
1976         .vop_reclaim =          msdosfs_reclaim,
1977         .vop_old_remove =       msdosfs_remove,
1978         .vop_old_rename =       msdosfs_rename,
1979         .vop_old_rmdir =        msdosfs_rmdir,
1980         .vop_setattr =          msdosfs_setattr,
1981         .vop_strategy =         msdosfs_strategy,
1982         .vop_old_symlink =      msdosfs_symlink,
1983         .vop_write =            msdosfs_write,
1984         .vop_getpages =         vop_stdgetpages,
1985         .vop_putpages =         vop_stdputpages
1986 };